PRECISION FORESTRY
PROCEEDINGS OF
THE FIRST INTERNATIONAL
PRECISION FORESTRY COOPERATIVE SYMPOSIUM
UNIVERSITY OF WASHINGTON COLLEGE OF FOREST RESOURCES
UNIVERSITY OF WASHINGTON COLLEGE OF ENGINEERING
USDA FOREST SERVICE
SEATTLE, WASHINGTON
JUNE 17-20, 2001
PRECISION FORESTRY
PROCEEDINGS OF
THE FIRST INTERNATIONAL
PRECISION FORESTRY COOPERATIVE SYMPOSIUM
UNIVERSITY OF WASHINGTON COLLEGE OF FOREST RESOURCES
UNIVERSITY OF WASHINGTON COLLEGE OF ENGINEERING
USDA FOREST SERVICE
SEATTLE, WASHINGTON
JUNE 17-20, 2001
Printed in the United States of America
All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical,
including photocopy, recording, or any information storage or retrieval system, without permission in writing from the publisher, the College
of Forest Resources.
Institute of Forest Resources
College of Forest Resources
Box 352100
University of Washington
Seattle, WA 98195-2100
(206) 543-2757
Fax: (206) 685-3091
http://www.cfr.washington.edu/Pubs/publist.htm
Proceedings of the First International Precision Forestry Cooperative Symposium, sponsored by the University of Washington College of
Forest Resources, the Precision Forestry Cooperative, the University of Washington College of Engineering, the USDA Forest Service,
Pacific Northwest Research Station, Resource Management and Productivity Program, Portland, Oregon, USDA Forest Service, Research
and Development, Vegetation Management and Protection Research, Washington, DC.
Additional copies of this book maybe purchased from the University of Washington Institute of Forest Resources, Box 352100, Seattle,
Washington 98195-2100.
For addition information on the Precision Forestry Cooperative please visit http://www.precisionforestry.org
Cover photo courtesy of Luke Rogers, University of Washington, College of Forest Resources
TABLE OF CONTENTS
ACKNOWLEDGMENTS
PREFACE
v
vii
KEYNOTE SPEAKERS
Bruce Bare
Francis Pierce
Peter Farnum
Gero Becker
1
1
3
7
PLENARY SESSION A: REMOTE SENSING OF FOREST LAND & VEGETATION
1
Use of Automated Individual Tree Crown Recognition and Measurement Algorithms in
Forest Inventories
Hans-Erik Anderson, Stephen E. Reutebuch, and Gerard F. Schreuder
11
2
An Inventory of Juniper Through the Automated Interpretation of Aerial Digital Imagery
23
Ward W. Carson, Abdullah E. Akay, and Dale Weyermann
3 Spatial Variability Within Managed Forest Stands
Johan Stendahl
35
4 Individual Tree Crown Image Analysis - A Step Toward Precision Forestry
Francois Gougeon, and Donald G. Leckie
43
5 Taking GIS/Remote Sensing into the Field
Rick R. Kerns, Thomas E. Burk, and Marvin E. Bauer
51
6 RTI - Real Time Inventory: A New Approach to an Old Problem
Mark Milligan
61
7 Use of Airborn LASER System for Forest Inventory in Siberia
Igor Danilin, Evgeny Medvedev, and T. Sweda
67
PLENARY SESSION B: SENSING, MEASURING & TAGGING TREES
Diameter Sensing using Radio Frequency Identification for Precision Forestry Applications
Denise Wilson, Sean Hoyt, and Doug St. John
9 Cooperative Use of Advanced Scanning Technology for Low-Volume Wood Processors
Luis G. Occeña, Timothy J. Rayner, Daniel L. Schmoldt, and A. Lynn Abbott
77
10 Applications of an Automated Stem Measurer for Precision Forestry
Neil Clark
93
11 Using Ultrasound to Detect Defects in Trees: Current Knowledge and Future Needs
Theodor D. Leininger, Daniel L. Schmoldt, and Frank H. Tainter
99
8
12 An Automated Log Grading System Based on Computed Tomography
Tim Rayner, Walter Garms, and Elan Scheinman
83
109
13 Evaluating the Performance of GPS Surveying Under Different Forest Conditions in Japan
Seca Gandaseca, Tetsuhiko Yoshimura, and Hisashi Hasegawa
14 Optimization of Forest Road Layout Using a High Resolution Digital Terrain Model
Generated from LIDAR Data
Elizabeth Dodson Coulter, Woodam Chung, Abdullah Akay, and John Sessions
119
125
15 Tightly Coupled Inertial/GPS System for Precision Forestry Surveys Under Canopy: Test Results
Joel Gillet, Bruno M. Scherzinger, and Erik Lithopoulos
131
16 RealTime Harvester: The Future of Logging
Brian H. Holley
139
17 Vehicle Management System for Forest Environmental Conservation
Kazuhiro Aruga, Juang Rata Matangaran, Koji Nakamura, Rin Sakurai,
Masahiro Iwaoka, Toshio Nitami, Hideo Sakai, and Hiroshi Kobayashi
143
18 Using a Laser Range Finder to Assist Harvest Planning
Michael Wing and Loren Kellogg
147
19 Using GPS to Evaluate Productivity and Performance of Forest Machine Systems
Steven E. Taylor, Timothy P. McDonald, Matthew W. Veal, and Ton E. Grift
151
PLENARY SESSION D: DECISION SUPPORT SYSTEMS
20 Global Demands Drive Advances in Data Management and Hierarchical Decision
Support Systems in Northern British Columbia
John Nelson and Dave Harrison
157
21 Precision Log Making for Plantation Operations
Kevin Boston
165
22 A Look to Future Precision Tree Length Stem Analysis and Processing
Phil Araman
171
23 Data Requirements for Precision Visualization of Forest Operations
Robert McGaughey
173
24 Information Needs for Increasing Log Transportation Efficiency
Timothy P. McDonald, Steven E. Taylor, Robert B. Rummer, and Jorge Valenzuela
181
25 Hierarchical Planning: Pathway to the Future?
John Sessions and Pete Bettinger
185
POSTER ABSTRACTS
26 Use of Airborn LASER System for Forest Inventory in Siberia
Igor Danilin, Evgeny Medvedev, and T. Sweda
191
26 Systems Engineering for Biomass Feedstock
Shahab Sokhansanj
191
27 Strategic Research for Inventorying, Monitoring and Modeling our Managed and Natural Ecosystems
Jack Sjostrom, Claire Rutledge, Paul Gessler, Peter Gorsevski, Amy Pocewicz ,
Ashley Covert, Erdene Saikhan, and Jeff Cronce
192
27 Forest Density Classification in Logged-Over Forest Using Airborne Radar Data
Safiah Yusmah M. Y.
193
PRECISION FORESTRY CONTRIBUTORS AND ATTENDEES
194
FINAL AGENDA
200
ACKNOWLEDGMENTS
Many individuals and organizations contributed to the success of this symposium and the collection of papers in this volume.
The conference was sponsored by the University of Washington College of Forest Resources, the Precision Forestry Cooperative, the University of Washington College of Engineering, the USDA Forest Service, Pacific Northwest Research Station, Resource Management and Productivity Program, Portland, Oregon, USDA Forest Service, Research and Development, Vegetation
Management and Protection Research, Washington, DC.
The financial support provided by the USDA Forest Service, Pacific Northwest Research Station, Resource Management and
Productivity Program, Portland, Oregon, USDA Forest Service, Research and Development, Vegetation Management and Protection Research, Washington, DC is gratefully acknowledged.
The program was planned by a committee consisting of:
Chair
Professor David Briggs, College of Forest Resources, University of Washington
Scientific Sub-Committee:
Professor James Fridley, University of Washington
Professor Gerard Schreuder, University of Washington
Professor Bruce Lippke, University of Washington
Steve Reutebuch, USDA Forest Service, PNW Research Station
This committee developed the program and recruited authors for the topics presented. The lead authors, in turn, worked with
coauthors and consulted with others to make this a truly international effort. The time and effort of all these contributors resulted
in excellent presentations and posters. Papers were reviewed before acceptance for publication, and the input of the many
reviewers is much appreciated. The session moderators Mike Renslow, Vice President, Spencer B. Gross, Inc., Portland, OR,
Jamie Barbour, Team Leader, Ecologically Sustainable Production of Forest Resources Team, USDA Forest Service, PNW
Research Station, Portland, OR, Bob Rummer, Project Leader, Forest Operations Research, USDA Forest Service, Southern
Research Station, Auburn, AL, and Bruce Bare, Rachel B. Woods Professor of Forest Management, and Acting Dean, College
of Forest Resources, University of Washington provided program linkage and kept the conference on schedule.
We wish to thank the Washington Department of Natural Resources and the USDA Forest Service, PNW Research Station for
making the Capitol State Forest field trip possible.
Capitol State Forest Field Trip Committee:
Tom Poch, Washington Department of Natural Resources
Steve Reutebuch, USDA Forest Service, PNW Research Station
David Marshall, USDA Forest Service, PNW Research Station
Capitol State Forest Field Tour Speakers:
Tom Poch, Washington Department of Natural Resources
Robert Curtis, USDA Forest Service, PNW Research Station
Steve Reutebuch, USDA Forest Service, PNW Research Station
David Marshall, USDA Forest Service, PNW Research Station
Dean DeBell, USDA Forest Service, PNW Research Station
Hans-Erik Andersen, Graduate Student, University of Washington
Bob McGaughey, USDA Forest Service, PNW Research Station
Gordon Bradley, Professor, University of Washington
v
Flo Damian, Graduate Student, University of Washington
Finn Krogstgaad, Graduate Student, University of Washington
Leslie Chandler-Brodie, USDA Forest Service, PNW Research Station
We would also like to thank the Precision Forestry Board of Directors for their support in the early planning of the symposium.
Chair, Rex McCullough, Weyerhaeuser Company
Wade Boyd, Longview Fibre
Craig D. Campbell, Boise Cascade Corporation
David Crooker, Plum Creek Timber
Suzanne Flagor, Seattle Public Utilities Watersheds Division
Sherry Fox, Washington Forest Protection Association
John Gorman, Simpson Investment Company
Peter Heide, Washington Forest Protection Association
Edwin Lewis, Bureau of Indian Affairs Yakima Agency
John Mankowski, Washington State Department of Fish and Wildlife
John Olsen, Potlach Corporation
Malcom Parks, University of Washington
Charley Peterson, USDA Forest Service
Mike Renslow, Spencer Gross, Inc.
Bryce Stokes, USDA Forest Service
David Warren, The Pacific Forest Trust
Laurie Wayburn, The Pacific Forest Trust
Maurice Wiliamson, Forestry Consultant
Kelly O’Brian, Chrissy Scannell, Kelly Duffield of the University of Washington College of Forest Resources Continuing
Education and Doug St. John, Executive Director, Precision Forestry Cooperative were responsible for conference arrangements
and management. Their efforts and those of the registration workers, projectionists, and other volunteers were critical for the
smooth operation of the conference and are greatly appreciated.
Production of this volume was coordinated through the Institute of Forest Resources; a special thanks to John Haukaas and
Doug St. John for their assistance in editing. Megan O’Shea handled final editing and publication preparation. The capable
efforts of this skilled group are gratefully acknowledged.
The proceedings are reproductions of papers submitted to the organizing committee after peer review. We wish to acknowledge the efforts of all the scientists involved in the anonymous peer reviews of these proceedings papers. No attempt has been
made to verify results. Specific questions regarding papers should be directed to the authors.
David Briggs, Chair
First International Precision Forestry Symposium
vi
PREFACE
The Precision Forestry Cooperative was founded as part of the Washington State Advanced Technology Initiative (ATI) funded
by the Legislature in 1999. The University of Washington’s College of Forest Resources, in collaboration with the College of
Engineering, created the Precision Forestry Cooperative to conduct pioneering research in forest production, management, and
manufacturing at a new scale of resolution and accuracy with the goal of producing economic and environmental benefits.
The Precision Forestry Cooperative is a partnership with the University and private landowners, harvesters, manufacturers,
public agencies, and the public. The goal is to develop tools and processes that increase the precision of forest data collected and
analyzed. This will allow for more exact planning, implementation, and review of management activities.
An important part of the Precision Forestry Cooperative’s mission is to share information and stimulate dialogue among
scientists, engineers, managers, and decision makers involved in the field. The First International Precision Forestry symposium
was part of this directive. Joining the Precision Forestry Cooperative, as sponsors were the University of Washington College of
Engineering and the USDA Forest Service.
The Symposium was organized along four themes:
·
·
·
·
Remote Sensing of Forest Land and Vegetation
Sensing, Measuring and Tagging Trees
Machinery, Monitoring and Road Layout
Decision Support Systems
vii
Opening Remarks and Welcome to the First International
Precision Forestry Symposium
B. BRUCE BARE, ACTING DEAN
The University of Washington College of Forest Resources is pleased to host this path-breaking event. It will help focus
attention on innovative technologies and approaches to guide future industries in Washington State. The Precision Forestry
Cooperative, and the sponsoring of this event, was born from the Washington State Advanced Technology Initiative (ATI). The
State Legislature funded six ATIs during the 1999/2001 biennium to form a partnership between the Legislature, private industry,
and the research universities of the State. Each ATI “cluster” is expected to generate new industries, or transform existing
industries, and create a bridge between research, education, and new economic activity. A nationally-recognized researcher lead
each cluster. Faculty and staff have been chosen for their demonstrated ability to collaborate with private industry and others.
Precision Forestry deploys high-resolution data to support site-specific tactical and operational decision-making. This allows for
highly repeatable measurements, actions, and processes to grow and harvest trees, as well as to protect and enhance riparian
zones, wildlife habitat, esthetics, and other environmental resources. Precision Forestry provides valuable information linkages
between resource managers, the environmental community, and processors. It links the practice of sustainable forestry with
conversion facilities and markets to produce the best economic and environmental returns. This Precision Forestry Symposium
brings together scientists, managers, and developers for the first time. It will provide insights into the current “state of the art”,
and provides a springboard for new ideas and innovations. We hope you enjoy the symposium, the campus, and the city during
your stay with us.
Thank you.
For complete copy of B. Bruce Bare’s Welcome Remarks, please visit the First International Precision Forestry Symposium’s
web page: http://www.cfr.washington.edu/Outreach/Postprefor/Presentations.htm.
Developments and Benefits in Precision Agriculture
FRANCIS J. PIERCE
For complete copy of Keynote speaker, Francis J. Pierce’s remarks, please visit the First International Precision Forestry
Symposium’s web page: http://www.cfr.washington.edu/Outreach/Postprefor/Presentations.htm.
1
Precision Forestry-Finding the Context
PETER FARNUM
INTRODUCTION
In 1960, the production of the 17 most important food,
feed, and fiber crops—virtually all of the important crops
grown in the U.S. at that time and still grown today—
was 252 million tons. By 1990, it had more than doubled,
to 596 million tons, and was produced on 25 million
fewer acres than were cultivated in 1960.
Goodness knows I have tried. However, it is impossible
for me as a trained statistician to ignore the obvious play on
words between “precision” as used in the name of this Symposium and “accuracy” as it is used in my profession.
The theme of this talk is that it doesn’t matter how precise technology is if it is not accurate in the sense of addressing the social and scientific contexts of forestry in the world
today.
Precision refers to the clustering of measurements together,
to repeatability. Accuracy refers to “being on target” – hitting the bulls eye nearly every time.
The question I want you to ask yourselves after this talk
is this: “Is Precision Forestry also accurate forestry?” You
will be able to answer yes if your work addresses the scientific and social contexts that I will discuss.
I will use quotes from three Nobel Prize winners hoping
you may be more impacted by their words than mine.
If we had tried to produce the harvest of 1990 with the
technology of 1960, we would have had to increase the
cultivated area by another 177 million hectares, about
460 million more acres of land of the same quality—
which we didn’t have, and so it would have been much
more.
We would have moved into marginal grazing areas and
plowed up things that wouldn’t be productive in the long
run. We would have had to move into rolling mountainous country and chop down our forests.
It is because we use farmland so effectively now that
President Clinton was recently able to set aside another
50 or 60 million acres of land as wilderness areas. That
would not have been possible had it not been for the
efficiency of modern agriculture.
Interview with Reason Magazine, April 2000
The Social Context
The obvious social context is disappearing forests with
increasing pressure on the remaining forests. Each year nine
million hectares of forests are lost to the world, an area approximately the same size as Western Washington. In the
State of Washington 40 thousand acres are converted each
year to other land uses. To put that in context that is more
land than Weyerhaeuser harvests and replants each year in
this state.
The pressures on the remaining forests come from well
known sources: the need to protect endangered species, the
need for clean water, and the increasing demand for commodity products.
If we just think of the shrinking forest base and the increasing demands the context for this symposium is pretty
bleak and discouraging.
However, I want to offer a more hopeful view of the context we face today. I believe that a solution to these pressures already exists and in fact is being implemented as we
speak.
We know the immediate results of this solution – President Bush proposed new rules, but a court in Idaho issued
an injunction. This is a healthy debate, but it would not be
happening if those forests were not there, and they would
not be there without increases in productivity in agriculture
and forestry. As you know some of the agricultural productivity increases came from technologies that we call “precision agriculture.”
Note that I am not saying we only should be interested in
increased production. We need to be interested in clean water,
endangered species, and other ecological values. We need to
make sure our intensive production is practiced on lands where
it is truly sustainable. What I am saying is increased productivity of commercial forestland is part of the solution to reconcile the apparent conflict between disappearing forests and increasing demands.
Now you can say you do not like this solution, but while we
debate it, it is going forth and in fact functioning quite well as
attested to by Dr. Borlaug’s insight. People are being fed
Norman Borlaug, Nobel Peace Prize winner in 1970 describes this solution.
3
of what the result is going to be, he is in some doubt.
We have found it of paramount importance that in order
to progress we must recognize the ignorance and leave
room for the doubt. Scientific knowledge is a body of
statements of varying degrees of certainty - some most
unsure, some nearly sure, [but] none absolutely certain.
The Pleasure of Finding Things Out.
and housed, and enough forests are being protected so society can debate their use.
Dr. Borlaug gave the national statistics. If we consider
the amount of forestland that has been protected in this state
in the last decade we can see this solution seems to be working locally too.
The Social Challenge
What Feynman is saying here is that if we have not repeatedly subjected our hypotheses to critical tests that have
sufficient power to reject them, and then we cannot claim
the degree of knowledge that science can grant us. In other
words, we cannot claim to be science-based.
But you may say, it is not possible to do this in my field;
it is not possible to subject my ideas to this kind of repeated
and rigorous analysis. Perhaps you think I am hopelessly
conservative when it comes to the scientific method, that I
should recognize that things have changed and that today,
science is different.
But I claim that saying “science is different today” is simply wishful thinking. Just because we find it difficult to
apply the scientific method to problems in forest ecology
does not mean we can change the scientific method and still
claim its authority. And, if we cannot say we are “sciencebased” then all we can say is that we are offering “expert
opinion” based on our scientific experience.
Well is this not just as good? To that I say - clearly no. I
offer as support a quote from my last Nobel Prize winner,
Charles H. Townes, who won the Nobel Prize in Physics in
1964.
Scientific expert opinion comes from our best ideas and
from hypotheses that are just beginning to run the gauntlet
of the scientific method. Our best ideas are likely to be good
ideas, but we must remember what Dr. Townes has said about
good ideas:
So the first challenge is this: How does the technology
that you will be discussing during this symposium fit into
this context? Will it help us increase productivity and protect other ecological values?
If the answer to these questions is not obviously yes to
you, or more importantly, if it is not obviously yes to the
people who manage our forests, then I would certainly challenge you to put your technology in this context quickly.
The Scientific Context
Today it is a cliché to say we want solutions that are science-based. Yet to all of us who work in forests this presents quite a challenge. The scientific method was developed in physics, astronomy, and agricultural research. These
are fields with well-developed theories, where it is practical
to employ rigorous experimental designs, and where exogenous variables can be controlled experimentally or through
replication and randomization. These are fields where others repeat experiments by one scientist before they are accepted.
Unfortunately, often we do not apply this rigor in our research on forest management and ecology. But if we do not
apply this rigor then we cannot claim that our solutions are
science-based. That is unfortunate. Yet what is even more
unfortunate is that many forest scientists have not accepted
the implications of not applying this rigor.
I should note that the challenge here for Precision Forestry is especially acute. Consider a model of slope stability.
New technology allows us to run it using data from 30cm
digital terrain models (DEMs) rather than 10m DEMs. Certainly the results are more precise, but they are in no way any
more science-based. They will not be science-based unless
they are subject to the kind of testing I have described.
Some would say that the rigor of testing that is used in
physics is not possible or relevant to forestry, in particular
that many ecological models are not testable in this way.
To address that claim let me offer a quote from Richard
Feynman, winner of the Nobel Prize in Physics in 1965 for
his work in quantum thermodynamics:
There are thousands of good ideas in science, the only
problem is – most of them are wrong.
Paraphrased from a speech at the Seattle Center
The Scientific Challenge
So to summarize the challenge of the scientific context:
Do you know when your work is really “science-based” and
when it is really “scientific expert opinion”?
The distinction can be described in a fairly simple way.
If we have done one experiment in one watershed and if we
have found a significant relationship in our data then we still
do not know the answer to the problem and we are in the
category that Feynman would call “ignorant.” Townes would
say that most ideas tested this far are likely to be wrong.
Certainly we are still in the zone of expert opinion no matter
how precise our answer.
If we have developed this finding into a hypothesis that
we have tested in five different watersheds over four different weather years and if we still have not rejected it, then I
The scientist has a lot of experience with ignorance and
doubt and uncertainty, and this experience is of very
great importance I think.
When a scientist doesn’t know the answer to a problem,
he is ignorant. When he has a hunch as to what the result is, he is uncertain. And when he is pretty darn sure
4
think we can say we are in Feynman’s category of having a
hunch about the answer. But we still are “uncertain.” We
are on our way to being science-based, but we are not there
yet.
But if we continue this testing and it is repeated by others and still not rejected, and if our hypothesis is based on
theory that correctly predicts unexpected results in watersheds far removed from our sites, then I would guess that
Feynman would say we are “pretty darn sure” but still in
some doubt. Only if our work falls into this last category
can we claim the mantel of being science-based and fully
distinguish ourselves from those giving expert scientific
opinion.
I would challenge each of you to take your technologies
to this level.
CONCLUSION
The technologies that will be discussed at this symposium are critical to the future management of our forests. It
would be a shame to develop them and then not have them
used to their full potential.
I think that potential can be realized only if, as you develop your technologies, you think about the social and scientific contexts in which they will be used, and only if you
meet the challenges inherent in those contexts.
Thank-you
5
Precision Forestry in Central Europe – New Perspectives for a
Classical Management Concept
GERO BECKER
INTRODUCTION
other hand leads to the fact, that this traditional small scale
management concept has to bee improved by introducing
modern Information Technology.
Whereas precision agriculture (or precision farming) as
a management principle and the related tools have been successfully introduced already since some years ago, the concept of precision forestry is still new and a wide range of
different “old” and “new” topics can be included, ranging
from areal photography and remote sensing to GIS, GPS
and logistic chains. It is typical for the European precision
forestry approach that it integrates different technical tools
into a holistic system, using digitized information as a link
and guiding principle. The following example of the use of
precision forestry in Central Europe is focused on this concept of integrated information chains to support management decisions.
Background
Precision Forestry aims at a) defining information relevant
for forest management and forest products precisely and b)
linking this information to locations (coordinates) using advanced methods of information technology. The more diverse
and structured a given forest situation and a forest product
is, the greater will be the advantage which can be achieved
by the application of Precision Forestry. Central European
and especially German forestry shows a great variety and
diversity in many aspects: Climate, site quality and growth
conditions vary in a small- scale pattern. Land use distribution (Forest land, farm land and urban areas as well as protective zones) is equally changing within small distances.
The forest ownership varies too, resulting in a mosaic of all
kinds of private forest ownership from great estates to small
farm holdings, and includes various forms of public and community owned forests.
The forest themselves have a great variety and diversity
of stands and tree species, ranging from uniform and even
aged spruce or beech forests to uneven and highly structured
mixed stands. This results in very different wood quality features, which meet the diverse demands of an equally highly
diversified wood industry.
The existing forest management practices reflect this diversity. Intensive planning and smallscale operations are typical. The management and control units (forest districts) are
relatively small ranging from between some hundreds and
some thousands of hectares only. A lot of information about
the existing forests is available, dating back for decades or
even hundred of years, mostly in the form of written management plans and reports, tables and conventional (analogue)
maps. Combined with a long on-the-spot experience of the
forest management personnel this traditional information
background guarantees a small-scale “precision” forestry
approach using traditional ways and means.
Economic pressure, increasing needs to include conservation aspects into forest management practices as well as a
rising public interest in forest and conservation issues on the
one hand and a growing demand from the wood consuming
industries for a consistent supply chain management on the
Integrated Wood Supply Chains: An example
for Precision Forestry in Central Europe
In Central Europe wood processing industry (saw mills,
pulp and paper) do not have access to own forest resources.
They have to buy every single cubic meter of round wood
they use from a diverse forest ownership on the market. Most
of the wood they purchase at road side, which means that
harvesting and extraction is organized by the forest owner
himself with own capacities or subcontractors, whereas the
transport from the lower landing to the mill is organized by
the buyer (e.g. the wood industry) and it is typically carried
out by small and medium transport companies. Because by
and large over 50% of the total annual cut in Central Europe
comes typically from thinnings or selective cuttings (and
not from clear-cuts) the situation is even more complex and
complicated. This extreme diversity is a great challenge for
an effective organization of the wood supply chain: quantity, dimension and quality as well as the time of delivery of
every purchased unit has to be planned and carried out in a
complex process which typically results in high procurement
costs. Both forest owners and wood industry have therefore
a vital interest to bring these costs down using precision
forestry with its whole set of tools. In this framework a high
added value is therefore a must for both wood industry and
7
forest owners. Consequently the core product in Central
Europe is valuable round wood suitable for saw-milling or
even for veneer whereas wood for pulp and paper or for
particleboard production is only a by-product. Against this
background a typical wood supply chain would consist of
the following elements:
and internal defects automatically and sort the timber according to these results quicker and more precisely.
•
Forwarding, storage, transport
Forwarders as the next part in the delivery chain are also
equipped with GPS, which allows to indicate the exact volume and the location of the delivered timber. There is wireless GMS-communication with the trucks and also with the
central unit (procurement office). The transport truck finds
its way supported by GPS and standard routing systems on
the public road network and furthermore on the digitized
vector map of the individual forest owner. The harvester also
transmits daily (or in future even permanently) harvesting
protocols to the saw mill procurement manager, who can
fine-tune the saw mill operation according to the wood which
will be delivered within the next hours or days.
•
Selecting the most suitable stand
If the objective is to get a maximum of added value, selecting the right stand for the right customer comes first and
has a high priority. Because stands are different in terms of
dimensions, log volumes, taper and wood quality, this information has to be measured or assessed for every stand to be
harvested as precisely as possible, using site adapted yield
tables, stem curves and models for quality prediction. This
information goes into a GIS-supported stand database and
can be matched with specific orders (“demand tables”) from
one or more customers (typically saw mills). The timber to
be harvested will be sorted virtually into the different diameter / length combinations and qualities the customers ask
for and the outcome can be compared and optimized taking
into account volume and value. In a more advanced stage of
operation, with this method stands can even be offered to
interested customers in Internet auctions.
•
Mill
At mill side the timber is typically re-measured by length
and diameter very precisely on automatic measurement stations. The respective volume is basis for the payment of the
forest owner. The measurement results are furthermore used
to optimize the downstream conversion process into boards
or cants, which is based on the exact tree measurement at
the entrance of the mill. In the future advanced chip marking systems will carry automatically all relevant information about a lot of trees or even about a single tree throughout the conversion process. Our institute is partner in an
integrated
EU-sponsored research project named LINESET (Linking Raw Material Characteristics with Industrial Needs for
Environmentally Sustainable and Efficient Transformation
Processes). In this project all technical and organizational
problems which are related to a continuous marking and
tracking concept for the timber are investigated and practical solutions are developed. A poster of this LINESET-project
(Author: Johannes Ressmann) is presented during this conference.
•
Harvesting operation
If the decision “which customer gets which stand” is
taken, the phase of harvesting can be started. Fully mechanized harvester / forwarder systems are more and more used
in terrain conditions up to 30% inclination. The stand database contains coordinates, tree characteristics and other useful information to prepare and carry out the harvesting including the exact location of environmental restrictions
(swamps, waterflows, buffer zones, etc.). The operator sees
all this information on his display in the harvester cabin. If
there are more small private owners whose lots are harvested
in one operation, even the border line of their individual
property is also part of the digital map, so while cutting the
trees, the harvester computer registers automatically which
trees belong to which owner.
During harvesting, every single tree is automatically measured by length and diameter at every ten centimeter creating an individual stem profile. These stem profiles are permanently matched in the harvester computer with the
customer’s demand tables for this respective order so that
the crosscutting of the stem is executed according to the
individual customer’s needs. The demand tables can be
changed on- line according to the actual needs of the mill by
wireless communication. Up to four assortments can be
marked with different colors, so that different customers can
get different pieces of one single stem. In the future this
marking could be done with barcodes or even microchips
instead of colors, which allows to store and carry on more
information. The quality of the logs is today still assessed
by the operator by visual inspection and entered manually
into the system. More advanced measuring systems are under development which will detect knots or other external
Advantages and Challenges
The advantage of these integrated wood delivery chains
is obvious: dimension – and in future even quality – can be
exactly predetermined for a given saw mill product. Quick
changes of the production program can be executed by direct communication between mill and harvester. This allows to minimize stock keeping and gives the chance to have
a maximum conversion factor, which is of advantage to the
forest owner as well as to the saw mill. A short time span
between harvesting, transport and conversion is another big
advantage which does not only save money directly through
reduced interest costs, but also contributes to a higher value
because blue stain and other negative quality features are
minimized. The system allows to harvest and market even
the lots of very small individual landowners without loosing too much of the economies of scale.
In our institute we initiated a pilot project where we try
8
introducing all the precision forestry tools mentioned above,
the economic advantage (reduction of costs and added value)
will exceed 15-20% compared to a conventional operation,
which is standard today.
But the pilot project also shows some difficulties: digital
maps with all relevant terrain and stand data are not yet standard in all forest enterprises. To create these maps is costly
and time consuming. Not all GPS systems do work properly
under a dense canopy and in different (mountainous) terrain
conditions. The existing mechanical measurement systems
of the harvesters are another weak point. Daily calibration
and a well-trained operator is absolutely necessary to produce reliable measurement results. Furthermore not all saw
mills are willing (or able) to define precisely their demand
tables because they were never used to do this. Small logging and transport companies may have difficulties to equip
their machines with costly GPS and computer systems. And
last but not least in many cases the participants along the
delivery chain are reluctant to give their partners open access to “their” relevant production data and information
because they fear they could be “pulled over the table” when
they disclose their “secrets”. It needs a lot of good arguments and time to replace this skeptic attitude by a more
open spirit of cooperation, which in the end serves all participants.
the downstream conversion process, but should be used in a
very extensive manner. This leads to the necessity of marking of a group of logs (if the material is homogeneous) or
even of single logs. We work together with Swedish and
Finnish experts in the joint EU-project (LINESET) already
mentioned above which aims at developing at testing practical ways and means to identify wood along the production
/ conversion chain. The vision is that as a first step of the
harvesting operation or even earlier (during silvicultural
management) every tree will be marked with a chip which
contains all relevant information about the tree: its origin of
growth, silvicultural treatment its exterior and interior qualities, dimensions, etc. The chip would go with the tree from
the forest to the mill and further into the conversion process. The information stored in it would be read and new
information could be added whenever needed. Before the
first cut at the saw mill the data could be stored and support
the further production process (sorting, drying, planing). This
ma rking concept would also allow to trace back every single
log to its stand of origin which is a necessary recondition to
establish a reliable chain of custody as part of a certification
scheme.
Another field of future progress is the interaction between
machine and soil. Harvesters and forwarders as well as trucks
do have GPS already today in many cases. Sensors are available which measure precisely the inclination of the body of
the moving machines, which can be transformed into a micro-profile of the terrain with a high resolution, which is
recorded online while driving. Sensors in the tires allow to
measure actual pressure, which allow to add information
about the baring capacity of the soil. This can help to prevent excessive soil compaction in sensitive areas. These
micro-terrain and soil information would not only allow to
monitor permanently the ongoing operation and its impact
on environment, but also could be added as an additional
layer of information to the existing forest map, so that for
future harvesting operations a more detailed information
would be provided. The vision is that with very management activity, more and more precise information can be
collected automatically and stored for future purposes. This
will not only make future operations more economical but
will also satisfy increasing public demands for a sensitive
treatment of water, soil, landscape and environment as a
sound basis for future sustainable management of our forests as a renewable resource.
Future Developments
In the future the challenge will be to give more precise
information not only about the quantitative, but also the
qualitative aspects of the forest resource. While it is relatively easy to measure dimensions, volume and weight, it is
a technical challenge to scan or measure relevant quality
parameters from outside or even inside the log. X-ray techniques, computertomographs, optical and accustical nondestructive measurement systems are tested, and it seems
likely that not one single tool but a combination of several
approaches will deliver the best results. To install these devices stationary at mill side under controllable conditions
will be relatively easy, but to make these future quality assessment systems mobile as part of heavy equipment (harvester, forwarder, trucks) or even as hand-hold instruments
is more challenging and will lead to substantial higher costs.
Consequently the costly information obtained about the logs
with these advanced mobile devices should not get lost along
9
Chapter 1
Automated Individual Tree Measurement Through
Morphological Analysis of a LIDAR-Based
Canopy Surface Model
HANS-ERIK ANDERSEN
STEPHEN E. REUTEBUCH
GERARD F. SCHREUDER
Abstract—An algorithm for automated individual tree measurement was developed that is driven by a morphological analysis
of a high-resolution LIDAR-based canopy surface model. Binary and grayscale mathematical morphology were used to relate
structure within a three-dimensional forest canopy model to the location of individual tree crown apexes. This information was
used to extract LIDAR measurements of individual tree position and height. Algorithm measurements were compared to photogrammetric measurements from large (1:3000) scale aerial photography. Given a range of “optimal” input parameters, the algorithm was successful in locating and measuring individual tree crown heights. The algorithm identified individual tree crown
apexes in a mature forest with closed canopy within 2 meters of photogrammetrically-measured crown apexes with a User’s
accuracy of 89% and a Producer’s accuracy of 83%. The difference between algorithm and photogrammetric tree crown apex
height measurements was approximately 1 meter in both study areas.
INTRODUCTION
crown image models (Pollock, 1998; Larsen, 1998). Researchers in Scandinavia have attempted to model the relationship between the spatial distribution of individual trees
and the position of spectral maxima in a digital image (Dralle
and Rudemo, 1997; Lund and Rudemo, 2000). Another study
has utilized two-dimensional mathematical morphology to
analyze the spatial structure of individual trees composing
the canopy in color aerial photography (Zheng et al., 1995).
The use of airborne laser scanning for the acquisition of
forest measurement data has also been an active area of research. Research efforts investigating the use of small footprint (< 1 m) LIDAR for forest measurement have primarily
concentrated on estimating forest stand-level parameters
(Naesset, 1997; Nelson, 1988). A study conducted in Oregon demonstrated the use of LIDAR for predicting forest
stand characteristics using plot-level LIDAR heights and
canopy cover percentiles, and found very strong relationships between LIDAR-derived measurements and stand parameters (Means et al., 2000). Researchers in Canada have
used a probabilistic model-based approach to estimate stand
height from LIDAR data (Magnussen et al., 1999). Another
study related the distribution of LIDAR canopy height measurements to the vertical distribution of foliage area
(Magnussen and Boudewyn, 1998). Nelson found that the
shape of the individual tree crowns composing the forest
canopy can have an effect on the LIDAR-based prediction
of stand-level parameters (biomass, basal area, volume), as
LIDAR-based forest height estimates over canopies com-
Forest inventory programs require detailed information
on individual tree characteristics, including height, diameter,
species, volume and condition. In particular, measurement
of predominant tree height is a critical variable in determining stand volume as well as site characteristics (Schreuder et
al., 1993). In addition, information relating to the location
and dimensions of individual trees can support distance-dependent forest modeling and site-specific forest engineering
design. While national and local inventories often utilize remotely-sensed data for stratified sampling and classification
of general forest type, most of these programs remain heavily
reliant upon expensive field data for individual tree-level
information (Czaplewski, 1999). The emergence of commercially available high-resolution active remote sensing technologies, such as airborne laser (LIDAR) scanning, can potentially allow for accurate, precise, and automatic identification and measurement of individual trees composing the
canopy surface.
There has been increasing interest in recent years in the
development of algorithms for identification and measurement of individual trees using high-resolution digital imagery. Probably the most well known algorithm for individual
tree recognition using digital imagery is a valley-following
algorithm developed by the Canadian Forest Service
(Gougeon, 1998). Numerous other studies have used a modelbased approach to locate individual trees using synthetic tree
11
measurements of both forest canopy and ground surface. A
LIDAR sensor system essentially works upon the principle
of measuring the time interval between the emission and
reception of laser pulses, and range measurement is performed by multiplying this time interval by the speed of light,
a known constant (approx. 30 cm/ns). The orientation and
position of the sensor at the time each laser pulse is emitted
is known through the use of an integrated inertial navigation system (INS) and a differential global positioning system (DGPS). The LIDAR data used in this study were acquired in Spring 1999 with a Saab TopEye scanning system
operating from a helicopter platform (see Table 1).
The positional accuracy of LIDAR measurements is approximately 1 m for horizontal positions and 10 cm for vertical positions. LIDAR measurements were provided in
ASCII text format with each data record consisting of a pulse
number, latitude, longitude, elevation (meters, NAVD88),
scan angle, and intensity.
posed of elliptical-shaped crowns will be higher than height
estimates over conical crowns (Nelson, 1997).
If the structural variation within a detailed LIDAR-derived canopy surface model can be related to the positions
and dimensions of individual trees, laser height measurements can be acquired for dominant and codominant trees
composing the canopy surface. The objective of this paper
is to present an approach to automated forest measurement
that utilizes a morphological analysis of the LIDAR-derived
canopy surface model to recognize three-dimensional structural features associated with individual tree crowns, and to
utilize this information in extracting LIDAR measurements
of individual trees.
MATERIAL AND METHODS
Study Site
The data used for this study were acquired over a mature
Douglas-fir (pseudotsuga mensiezii) forest stand in Capitol
State Forest, WA. This site is hilly with elevations varying
from 500 - 1300 feet with ground slopes from 0 - 45 degrees. The study area was 0.4 ha in size and was located in
the control unit for an experimental silvicultural study (Figure 1). This stand exhibits a relatively closed canopy structure, with a dominant height of approximately 48 meters
and a stand density of 280 trees per hectare.
Generation of the Forest Canopy Surface
Model
A digital model of the forest canopy was generated from
raw LIDAR data by extracting probable canopy-level laser
returns from the data set through a filtering operation (Figures 2 & 3). This operation consisted of extracting the maximum laser return within a cell, or window, of a certain size
over the entire area of interest. A canopy surface model was
generated by interpolating an elevation value at each pixel
of the surface model based upon the Delauney triangulation
of these filtered “canopy” returns (Figure 4).
It should be noted that the pixel size of the canopy model
can be different from the size of the filtering cell size. The
pixel size for the canopy model should be small enough to
retain local structural details of the canopy surface. A canopy
surface model generated in this manner is highly sensitive
to the size of the window used in filtering the LIDAR data.
The use of a small filtering cell (less than 1 m) will minimize the loss of information relating to micro-level canopy
features (i.e. small tree crowns) but will increase the probability of extracting laser returns that are not on the canopy
surface. On the other hand, the use of a larger filtering cell
(1 - 2 m) will increase the likelihood that the maximum return is in fact a measurement of the true canopy surface, but
Aerial Photography
Aerial photography at several different scales was acquired over the study area. Normal color photography at
scales of 1:12000 and 1:7000 was acquired from the Washington Department of Transportation (WADOT) in June
1999. Large-scale normal color photography at a scale of
1:3000 was provided by the WADOT in June 2000. This
photography was acquired with a Zeiss LMK aerial camera
with a 305-cm focal length lens. Both paper prints and transparencies were acquired at all scales.
LIDAR Data
LIDAR (LIght Detection And Ranging) is a mature remote sensing technology that can provide highly accurate
Table 1. Flight parameters and LIDAR scanning system settings.
Flying height
Flying speed
Scanning swath width (with max scan angle ± 17 degrees)
Forward tilt
Laser pulse density
Laser pulse rate
Maximum echoes per pulse
12
650 ft
25 m/sec
70 m
8 degrees
3.5 pulses/m2
7,000 pulses/sec
4
Figure 1. Orthophotograph with study area delineated, CapitolForest, WA.
Figure 2. Raw LIDAR data (0.4 ha study area).
13
Figure 3. Filtered canopy returns (0.4 ha study area)
Figure 4. Canopy surface model (0.4 ha study area)
14
will entail the loss of information relating to the morphology of smaller structural components of the canopy surface.
Therefore, canopy models in this study were generated with
a fixed pixel size (0.30 m), but over a range of filtering cell
sizes (0.91 m, 1.22 m, and 1.52 m) in order to investigate
the influence of the canopy surface generation procedure on
the accuracy of individual tree measurement.
Morphological Analysis of the Forest Canopy
Surface Model
Mathematical morphology (or simply morphology) provides a quantitative approach to the analysis of geometric
structure within the canopy surface model. In particular, a
specific sequence of binary and grayscale morphological
image transformations can be used to isolate individual trees
composing the canopy surface, which in turn can drive an
individual tree measurement algorithm.
Although originally developed for the analysis of twodimensional binary images, mathematical morphological
theory has since been extended to three-dimensional
grayscale images, where the grayscale values represent intensity or another pixel attribute, such as elevation of a surface. The operations of mathematical morphology are defined in set theoretic terms. In the morphology context, sets
represent the shapes that collectively make up a binary or
grayscale image. Sets in two dimensions describe the foreground of the image; in three dimensions they can describe
variation within a surface. The goal of any morphological
operation is to gain information relating to the geometric
structure of an image by probing the image with another set,
of specified size and shape, known as a structuring element.
The size and shape of the structuring element is chosen according to the type of shape information to be extracted from
the image. In formal terms, a morphological operation is an
image transformation with the structuring element serving
as the parameter for the transformation. The result of a single
transformation (or morphological operation), carried out with
a given structuring element, conveys information relating to
the shape content of the original image. Varying the size of
the structuring element can result in different image transformations and can therefore provide even more information about image content.
The basic morphological operations are dilation and erosion. If an image is represented as a set A and a structuring
element as another (smaller) set B, the result of the dilation
of image A by structuring element B can be thought of as
showing those areas where the structuring element B hits
the set A (Soille, 1999). In formal, set theoretic terms, if A
and B are subsets of d-dimensional space, the dilation of a
set A by B is defined as:
A€ B \c ‰ E | c a b for some a ‰ A and b ‰ B ^
d
In image processing the dilation operation is often termed
“fill,” “expand,” or “grow”. The dual operation to dilation
is erosion. Using the above notation, the erosion of a set A
by structuring element set B will show those areas where
the structuring element fits the set A. In formal terms, the
erosion of a set A by a structuring element B is defined as:
A $ B = { x ∈ E d | x + b ∈ A for every b ∈ B}
In image processing the erosion operator is often termed
“shrink” or “reduce”. In practice, the dilation and erosion
operations are used together; for example, an erosion followed by a dilation makes up another morphological operation termed an opening. The practical effect of morphological openings is to remove details in the image that are smaller
than the structuring element without distorting the geometric structure of unsuppressed features. Openings therefore
tend to break narrow isthmuses and remove small islands
within a binary image (Haralick et al., 1987).
Grayscale morphology involves extending these notions
from sets in two dimensions to functions in three dimensions. It requires defining top surface of a set and the umbra
of a function. For a set A in three-dimensional space, where
we consider the first two coordinates (x,y) as constituting
the spatial domain and the z coordinate indicating the surface, the top surface T[A] of a set is the highest value z such
that (x,y) Î A. The umbra of a function f, denoted as U[f],
is a set made up of the surface f and everything below the
surface. For a given function (grayscale image) f and threedimensional structuring element k, the grayscale dilation of
f by k is defined as the surface of the dilation of their umbras:
f € k = T U [ f ] € U [ k ]
The grayscale erosion of a function f by a structuring
element k is defined as the surface of the erosion of their
umbras:
f $ k = T U [ f ] $ U [ k ]
We can therefore define a grayscale opening of a function f by structuring element k as
f ok = ( f $k)€k
The grayscale opening operation can be interpreted geometrically as pressing the structuring element up against the
surface and sliding it underneath the entire surface. The opening of the surface by the structuring element is the highest
point reached by any part of the structuring element as it
slides underneath the surface (Haralick et al., 1987).
The Individual Tree Measurement Algorithm
If a flat disk with a specified radius is used as the structuring element in a grayscale morphological opening transformation of the canopy surface model, those areas of the
canopy surface model in which the disk structuring element
does not fit when pressed underneath the surface, such as
the tops of conical or ellipsoidal individual tree crowns, will
15
Figure 5. Morphological opening operation
1. Canopy surface model
2. Morphological opening operation
3. Morphological opening
Figure 6. Morphological opening of canopy surface model with disk of radius 1.2 meters (0.4 ha study area).
smaller crowns within the image. This sequence of grayscale
and binary morphological operations isolates the tops of tree
crowns with distinct conical or elliptical structure that compose the three-dimensional canopy surface model (see Figure
11). With high density LIDAR data (more than 1 return/m2),
the laser measurement with the highest elevation within each
of the areas isolated with this morphological algorithm should
provide an estimate of the location and elevation of the top of
each individual tree crown composing the canopy surface.
The difference between this LIDAR estimate of the location
of an individual tree crown apex and a base elevation interpolated from a LIDAR-derived digital terrain model will allow
for estimation of heights for the individual trees composing
the canopy surface (see Figure 12).
be removed through the opening operation (Figures 5 & 6).
The subtraction of this opened surface from the original
surface, termed the morphological top-hat transformation,
will therefore isolate those areas of the canopy surface that
were removed through the opening, i.e. the apexes of individual tree crowns (Meyer, 1979) (Figures 7 & 8). A
thresholding operation is used to convert this top-hat transform into a binary image (Figure 9). A binary morphological opening transformation with a disk of slightly smaller
size than that used in the previous grayscale opening is carried out to remove noise from this binary top-hat transform
image.
This sequence of morphological operations can be carried out with a range of disk sizes to extract location of trees
with varying crown widths and shapes. If these filtered binary top-hat images are added together the resulting image
will contain information relating to morphological content
at a variety of scales (see Figure 10). This procedure tends
to aggregate areas associated with single large tree crowns,
while still retaining smaller isolated features associated with
Photogrammetric Individual Tree
Measurements
Accuracy assessment was carried out through comparison
to photogrammetric measurements carried out on a Carto In16
Figure 7. Top-hat transformation.
Top-hat transformation
Canopy surface model
Figure 8. Top-hat transformation of canopy surface model (0.4 ha study area).
struments AP190 analytical stereoplotter. This approach
allows direct comparison of the performance of the automated tree measurement algorithm to the results expected
from conventional methods based upon aerial photograph
interpretation. Large-scale (1:3000) normal color photographs were used to maximize the accuracy of the photogrammetric measurements. The tops of all trees visible in
the study area were measured photogrammetrically and
stored as (x,y,z) point coordinates in a file. These photomeasured locations were then compared to the tree crown
measurements generated by the automated algorithm.
2 meters, where the error radius represents the maximum
distance allowed for a “match” between
photogrammetrically-measured trees and algorithm-measured trees. Measures of both omission error (Producer’s
accuracy) and commission error (User’s accuracy) in algorithm tree identification are indicated (Lillesand and Kiefer
1994). Producer’s accuracy was calculated by dividing the
total number of “matched” trees by the total number of trees
measured in the aerial photographs. User’s accuracy is calculated by dividing the total number of “matched” trees by
the total number of trees identified by the algorithm. In addition, the relative accuracy of the algorithm in identifying
trees is shown graphically in Figure 13. In this figure, red
crosses indicate location of trees identified by the algorithm,
while the disks indicates the 2 meter error radius centered
on the photogrammetrically-measured trees.
The elevations of the individual tree crown apexes measured from the photography and generated from the algorithm are compared in Table 2. For all “matched” trees (at
both 1 and 2 meter error radii) the difference in elevation
RESULTS
The results of the accuracy assessment are summarized
in Table 2. This table indicates the relative performance of
the morphology-based tree measurement algorithm as parameters of the canopy surface generation process (filtering
cell size) and morphological analysis (structuring element
radius) are varied. Results are shown for error radii of 1 and
17
Figure 9. Thresholded (binary) image of top-hat transform (0.4
ha study area).
Figure 10. Sum of filtered binary top-hat transforms with disks
of 1.2 m (green) and 1.5 m (white) radii(0.4 ha study area).
Figure 11. Identification of tree crown apexes through morphological operations.
18
Figure 12. Estimation of individual tree locations and heights overlaid on DTM.
Figure 13. Accuracy assessment of algorithm. Red crosses represent algorithm measurements, circles represent 2 meter error radius
surrounding photogrammetrically-measured trees
19
Table 2. Accuracy assessment of individual tree measurement algorithm.
0.9 m filtering cell size
Structuring element
radii (m)
0.6 - 0.9
0.9 - 1.2
1.2 - 1.5
1.5 - 1.8
1.2 m filtering cell size
Structuring element
radii (m)
0.6 - 0.9
0.9 - 1.2
1.2 - 1.5
1.5 - 1.8
1.5 m filtering cell size
Structuring element
radii (m)
0.6 - 0.9
0.9 - 1.2
1.2 - 1.5
1.5 - 1.8
1 meter error radius*
Accuracy*
Ht deviation (m)**
User’s
Producer’s
Mean
St. Dev.
35%
92%
1.11
1.05
60%
80%
1.20
1.02
73%
76%
1.21
1.01
82%
64%
1.29
1.07
2 meter error radius
Accuracy
Ht deviation (m)
User’s
Producer’s
Mean
St. Dev.
42%
95%
0.71
1.53
66%
87%
1.07
1.10
83%
85%
1.01
1.18
91%
72%
1.17
1.14
1 meter error radius
Accuracy
Ht deviation (m)
User’s
Producer’s
Mean
St. Dev.
48%
80%
1.18
1.05
76%
74%
1.24
1.03
83%
78%
1.21
1.03
86%
65%
1.23
1.06
2 meter error radius
Accuracy
Ht deviation (m)
User’s
Producer’s
Mean
St. Dev.
54%
85%
0.94
1.28
83%
81%
1.15
1.08
89%
83%
1.17
1.03
92%
70%
1.19
1.05
1 meter error radius
Accuracy
Ht deviation (m)
User’s
Producer’s
Mean
St. Dev.
53%
77%
1.21
1.01
75%
72%
1.22
1.03
84%
73%
1.20
1.05
88%
58%
1.29
1.05
2 meter error radius
Accuracy
Ht deviation (m)
User’s
Producer’s
Mean
St. Dev.
60%
84%
0.98
1.24
79%
75%
1.14
1.13
89%
78%
1.16
1.04
94%
63%
1.25
1.04
* User’s accuracy = Total # of matches within specified error radius/Total # trees identified by algorithm
Producer’s accuracy = Total # of matches wthin specified error radius /Total # of photo-measured trees
** Ht deviation = Algorithm elevation measurement - photogrammetric elevation measurement (matched trees)
between the algorithm and photogrammetric measurements
were calculated. The mean and standard deviation of the difference are given in the table.
While the results vary considerably across the range of
input parameters, there is evidence that given a range of “optimal” parameters, the algorithm was successful in locating
and measuring individual tree crowns. Given a specified set
of input parameters (1.2 meter bin cell size, structuring element radii of 1.2-1.5 meters), the algorithm identified individual tree crown apexes within 2 meters of
photogrammetrically-measured crown apexes with a User’s
accuracy of 89% and a Producer’s accuracy of 83% in Area
1 (see Table 2).
sional image data (Quackenbush et al. 2000; Stiteler and
Hopkins 2000). The results indicate that the algorithm generally performs much better where tree crowns are larger and
more widely dispersed. It is also apparent that the relationship
between the parameter of the morphological operations – the
size of the structuring element – and the predominant scale of
the tree crowns composing the canopy has a direct influence
on the accuracy of the algorithm. Using a smaller structuring
element results in the extraction of detailed structural features
within the three-dimensional canopy surface. In cases where
most tree crowns making up the canopy are large, a small structuring element is too sensitive to morphological variation and
will tend to extract many features, including multiple features
associated with the same crown. This leads to a high commission error but a low omission error, as it tends to find the
“true” tree crown apex most of the time along with many features not associated with crown apex locations. Similarly, when
the structuring element is large relative to the predominant
size of the tree crowns composing the canopy, the algorithm
extracts large-scale morphological features that can include
DISCUSSION
Individual Tree Identification
The morphology-based tree measurement algorithm
achieves accuracies comparable to other individual tree recognition algorithms that utilize high-resolution two-dimen20
clumps of several tree crowns. This will lead to very low
commission error and a very high omission error, as the algorithm tends to extract too few measurements in areas where
trees are close together. There appears to be an optimal choice
for the morphological parameters where the size of the structuring element corresponds to the dominant scale of the tree
crown structures composing the surface, and where commission error and omission error will be minimized. The
results indicate that the optimal range of structuring element
scale for this mature forest, where tree crowns are relatively
large, is in the range of 1.2 – 1.5 meters.
The results also indicate that the size of the filtering cell
used to generate the canopy surface model does have a subtle
influence on the accuracy of the algorithm. In this mature
forest, with large, widely dispersed tree crowns, the optimal
filtering cell size appears to be 1.2 meters.
measurements to field-based tree measurements. It is expected that future work will integrate remotely-sensed individual tree measurements into sampling designs in order to
optimize forest inventory programs.
LITERATURE CITED
Czaplewski, Raymond L. 1999. Multistage Remote Sensing: Toward an Annual National Inventory. Journal of
Forestry 97(12): 44-48.
Dralle, Kim and Mats Rudemo. 1997. Automatic estimation of individual tree positions from aerial photos. Canadian Journal of Forest Research 27: 1728-1736.
Gougeon, Francois. 1998, Automatic individual tree crown
delineation using a valley-following algorithm and a rulebased system. In Hill and Leckie, eds. 1998. In Proceedings of the International Forum on Automated Interpretation of High Spatial Resolution Digital Imagery for
Forestry. Pacific Forestry Centre, Victoria B.C.
Individual Crown Apex Measurement
The distinct advantage of using actively-sensed high-density LIDAR data over passively-sensed two-dimensional image data for individual tree measurement is the potential for
individual tree stem height measurement. It is expected that
photogrammetric measurement of tree apex locations from
very large scale photography (1:3000) will yield very accurate estimates of the tree stem top (i.e. within a meter). The
results indicate that the tree apex elevation measurements
generated by this algorithm are quite close to the photogrammetric measurements, with a mean difference of approximately +/- 1 meter, with standard deviations of approximately
1-1.5 meters across all parameter ranges. The mean difference between algorithm measurements of tree crown apexes
is approximately + 1 meter (standard deviation of ~ 1 m).
Intuitively, given that the LIDAR was acquired in 1999 and
the photography in 2000, we would expect the LIDAR to
underestimate the elevation of the tree crown apexes by several feet even if all measurements were entirely accurate.
There are a number of possible sources of systematic error
in both LIDAR and photogrammetric elevation measurement.
For example, wind causes movement of the tree top between
photo exposures that can confound the precise parallax measurements that are required for photogrammetric determination of elevation. In addition, bridging ground control from
medium to large scale aerial photography can be a source of
error in photogrammetric height measurement.
Haralick, Robert M., Stanley R. Sternberg and Xinhua
Zhuang. 1987. Image analysis using mathematical morphology. IEEE Transactions on Pattern Analysis and
Machine Intelligence PAMI-9 (4):532-550.
Larsen, Morten. 1998. Finding an optimal match window
for spruce top detection based on an optical tree model.
In Hill and Leckie, eds. 1998. In Proceedings of the International Forum on Automated Interpretation of High
Spatial Resolution Digital Imagery for Forestry. Pacific
Forestry Centre, Victoria B.C.
Lillesand, Thomas, and Ralph Kiefer. 1994. Remote Sensing and Image Interpretation. Wiley, New York.
Lund, J. and M. Rudemo. 2000. Models for point processes
observed with noise. Biometrika 87(2): 235-249.
Magnussen, S. and P. Boudewyn. 1998. Derivations of stand
heights from airborne laser scanner data with canopybased quantile estimators. Canadian Journal of Forest
Research 28: 1016-1031.
CONCLUSIONS
Magnussen, S., P. Eggermont and V.N. LaRiccia. 1999.
Recovering tree heights from airborne laser scanner data.
Forest Science 45(3): 407-422.
This study found that mathematical morphology can be
used to relate the three-dimensional structure within a detailed LIDAR-based forest canopy surface model to the location of tree crowns, and this information can be used to
extract LIDAR measurements of individual tree crown
apexes. The accuracy of this approach is comparable to
methods that use high- resolution image data, and the use of
geometric LIDAR data allows for direct measurement of individual tree heights. Further study is needed to relate these
Means, Joseph E., Steven A. Acker, Brandon J. Fitt, Michael
Renslow, Lisa Emerson and Chad Hendrix. 2000. Predicting forest stand characteristics with airborne scanning lidar. Photogrammetric Engineering and Remote
Sensing 66 (11): 1367-1371.
21
high resolution imagery. In Proceedings of the ASPRS
National Convention, Washington D.C., May 22-26.
Meyer, Fernand. 1979. Iterative image transformations for
an automatic screening of cervical smears. The Journal
of Histochemistry and Cytochemistry 27(1):128-135.
Naesset, Erik. 1997. Determination of mean tree height of
forest stands using airborne laser scanner data. ISPRS
Journal of Photogrammetry and Remote Sensing 52: 4956.
Reutebuch, Stephen E., Kamal Ahmed, Terry Curtis, Dick
Petermann, Michael Wellander, and Michael Froslie.
2000. A test of airborne laser mapping under varying
forest canopy. In Proceedings of the ASPRS National
Convention, Washington D.C., May 22-26.
Nelson, R. 1997. Modeling forest canopy heights: The effects of canopy shape. Remote Sensing of Environment
60: 327-334.
Schreuder, Hans T., Timothy G. Gregoire, and Geoffrey B.
Wood. 1993. Sampling Methods for Multiresource Forest Inventory. Wiley, New York.
Nelson, R. and W. Krabill and J. Tonelli. 1988. Estimating
forest biomass and volume using airborne laser data.
Remote Sensing of the Environment 24: 247-267.
Soille, P.. 1999. Morphological Image Analysis: Principles
and Applications. Springer-Verlag, Berlin
Stiteler, William M. and Paul F. Hopkins. 2000. Using genetic algorithms to select tree crown templates for finding trees in digital imagery. In Proceedings of the ASPRS
National Convention, Washington D.C., May 22-26.
Pollock, Richard. 1998. Individual tree recognition based
upon a synthetic tree crown image model. In Hill and
Leckie, eds. 1998. In Proceedings of the International
Forum on Automated Interpretation of High Spatial Resolution Digital Imagery for Forestry. Pacific Forestry Centre, Victoria B.C.
Zheng, X., P. Gong, and M. Strome. 1995. Characterizing
spatial structure of tree canopy using colour photographs
and mathematical morphology. Canadian Journal of Remote Sensing 21(4): 421-429.
Quackenbush, Lindi, Paul Hopkins, and Gerald Kinn. 2000.
Using template correlation to identify individual trees in
22
Chapter 2
An Inventory of Juniper Through the Automated
Interpretation of Aerial Digital Imagery
WARD W. CARSON
ABDULLAH E. AKAY
DALE WEYERMANN
Abstract—In the summer of 1999, the USDA Forest Service Forest Inventory and Analysis group captured a large number of images
over test plots in eastern Oregon where juniper (juniperus occidentalis) was known to be present. There were color infrared (CIR) images
collected with a Kodak 420 digital camera by the Remote Sensing Applications Center group of the USFS. The juniper trees were generally
apparent on most of these images and estimates of juniper distribution, density, and size could be accomplished manually, however, automatic interpretation was desired to speed and, thereby, develop an inventory technique. This study was designed to examine these images
and attempt to extract juniper density, juniper tree distribution, and juniper crown size information from them automatically through a
combination of image pre-processing and rule-based, crown-delineation techniques. Two patterns—a shadow plot and a crown plot—were
generated as intermediate images through normal image processing techniques. Then, rule-based techniques were developed and applied to
each pattern to specifically delineate the shadows and crowns, and determine associated parameters such the size and centroid of pixel
groups. Finally, the requirement that shadows be associated with the larger junipers was used to select trees and discriminate against purely
terrain-related shadows and smaller trees or vegetation closer to the ground. These techniques and a report of their success is presented.
SUMMARY
shadow/background files and the potential juniper/background files were held to pixel grouping minimums of 50
and 200 pixels respectively. The ground pixel size was nominally 15 cm (approximately 0.5 feet).
The rule-based program designed to delineate objects in
the binary files is very similar to the successful routines first
reported by Gougeon (1995). Gougeon’s primary application was the delineation of crowns in a full canopy (crowns
nearly uniform in size and well-distributed spatially), however, we modified his approach to address the variable sizes
and scattered distribution of juniper expected in eastern Oregon. Again, our program was designed for production—
prepared to treat many images in a batch process with little,
if any human intervention.
Our procedure distinguishes both shadows and potential
junipers—two separate, but spatially related sets of objects
for each original image. These objects are added eventually
into a single image file that then delineated shadow-like features, juniper-like features, areas of shadow-juniper overlap, and the distinctive boundary points and centroid of each
object.
The shadow-juniper image files are the basis for the final
juniper selection from the group of juniper-like objects. The
selection criteria are based upon the expected relationship
between a juniper and its shadow. Briefly, an object is selected as a juniper if it overlaps with a shadow and has a
boundary in the shadow area as well as a portion of its boundary adjacent to the non-shadow, non-juniper background area.
One hundred and twenty-eight color infrared digital images supplied for this study were analyzed while developing
a procedure designed to distinguish the size and frequency
of scattered western juniper (juniperus occidentalis) trees in
areas of eastern Oregon. Junipers were identified through
image pre-processing followed by a series of rule-based,
object delineation and selection techniques. Batch processing routines were implemented, fulfilling a primary goal for
this study, namely, to develop an automatic procedure for
handling large numbers of photos.
The ERDAS image-processing program Imagine was used
to prepare images. Images were processed twice: once to
distinguish shadows and once to distinguish potential juniper. The images were first smoothed to remove the high
frequency of data variations associated with our high spatial-resolutions images. A low-pass convolution, 7x7 kernel
filter applied twice was found to work well. The smoothed
images were then examined to develop specific signatures
for a supervised classification. The Maximum Likelihood
technique was used to place pixels into over 200 parameter
classes. The classified image served as a source for two binary files: a potential juniper on background file and a shadow
on background file.
The final pre-processing step used ERDAS utilities to
clump pixels into groups and to eliminate groups that were
smaller than a minimum size. The binary files for both the
23
As a primary result, a generic binary “juniper object”
image file is produced along with a text file listing parameters that describe each juniper. The image file is passed
back to the ERDAS Imagine program and overlaid upon the
original image to serve as a check upon the selection process. The juniper object parameters are picked up in a spreadsheet to summarize data such as juniper location and crown
size.
The overall accuracy of the detection and size-determination procedure was judged by visual comparisons between
the juniper selected and those missed. The comparison was
made on 39 randomly selected images which gave a true
juniper count average of 27.21 per image; 6.59 (24%) were
missed by our procedure and 0.46 (2%) were miss-identified. The missed percentage is large, however, the figure
should be compared with the results produced by image processing only. For the same (randomly selected) images, the
average number of ‘potential junipers’ (based upon the juniper objects delineated from the binary file as delivered
from the image pre-processing) was 55.13. From this number, our delineation and selection procedure selected 20.16:
74% of the true juniper count. In comparison, the image
processing only approach over counted the true juniper per
image average by 103%.
prehensive inventory of juniper is needed to provide baseline
data.
The Proposal to Inventory Juniper from
Aerial Digital Images
The cost of ground based field inventories is very high:
$900-1000 per plot (an estimate for eastern Oregon;
Weyermann, 1999). Some have speculated that an aerial
sampling method could provide more reliable estimates for
less cost. This study was part of a project designed to investigate the feasibility of aerial sampling based upon the automated processing of digital images.
A literature survey conducted over a period from 1996 to
early 2000 (by senior-author Carson) had revealed specific
studies (Hill and Leckie, 1998) that were successful in the
automatic detection of individual tree crowns in a conifer
forest canopy. The automated “crown-following in high
spatial resolution aerial images” by Francois Gougeon (1995)
was of particular interest. Carson proposed to follow
Gougeon’s approach in an attempt to detect and measure
scattered juniper in a similar manner.
The USDA Forest Service Forest Inventory and Analysis
group had, in the late summer of 1999, captured 128 largescale, color infrared digital images over areas of eastern Oregon. After an extensive examination of these images, we
were encouraged to develop and prove, or disprove the utility of the Gougeon approach while analyzing these images.
INTRODUCTION
The Need for a Juniper Inventory in Eastern
Oregon
METHODS
Reliable, unbiased inventories of natural resources are
critical for formulating public policy. Periodic inventories
have, by Congressional mandate, traditionally been concentrated on forest species of commercial value, and in areas
supporting commercial productivity.
Settlement of the west, suppression of rangeland fires,
the introduction of domesticated grazing stock and non-native forage species have altered much of the sparse, noncommercial forest lands of the western US (Miller, 1995).
These lands are often remote, poorly roaded, and capable of
supporting only low productivity (Garrison et al., 1977).
Changes in public perception of sparsely forested lands have
led to a need for comprehensive inventories of these areas.
Rural economic development issues are causing re-examination of the commercial viability of forest products from
sparsely forested areas, not only for traditional consumptive
uses such as firewood and fence posts, but for spiritual uses
(Miller, 1997), recreation, and to foster development of niche
industries centered around wood working and specialty products.
Considerable interest in western juniper (juniperus
occidentalis) is apparent in eastern Oregon. Commonly held
beliefs are that juniper is dramatically expanding its range
(Miller, 1995), historical stands are increasing in density
(Garrison et al., 1977), and are causing declines in native
understory species (Karl and Leonard, 1995). But estimates
of juniper range and population dynamics are based on fragmentary and often contradictory data (Belsky, 1996). A com-
Our image analysis and juniper selection method proceeds
in four parts: image pre-processing, object delineation, the
creation of shadow-juniper relationships, and the selection
(or rejection) of juniper based upon this relationship. The
image pre-processing relies upon an extensive use of the
ERDAS Imagine program. The object delineation follows
closely the rule-based technique of Gougeon (1995). Shadows and potential juniper are delineated separately and an
array of boundary pixels eventually identifies many such
objects on each original image. These objects are then
brought together to establish shadow-juniper spatial relationships. Our selection rules are based upon parameters
that describe the expected relationship and can be tested for
satisfaction.
Image Pre-processing
The methods used for pre-processing the images are
shown in the ERDAS flow chart that follows. The images
have a high spatial resolution—a square pixel with approximately 5x5 inch (122x122mm) dimension on the ground.
The juniper crown structure, crown density, different branch
structure, shadow effects, and the great variation in background vegetation and soil material produce a high frequency
of data variability in our images (Figure 1). The images
were first smoothed with a low-pass convolution, 7x7 ker24
THE ERDAS FLOW CHART OF IMAGE PROCESSING TECHNIQUE:
Run Batch Mode to
convert input image
from .tif to .img
format
Input
Image
7X7 low
pass
filtering
twice
Run Batch Mode
Run Batch Mode
using final signature
file
Classify
filtered
image
Run the Graphical Model
Shadow=1
Border= 1
Others= 0
Eliminate
clumps
<50 pixels
Shadow=1
Others= 0
Recode
Image for
Shadow
Recode
Image for
Juniper
Clump&
Sieve
Shadow
Clump&
Sieve
Juniper
Final
Recode
of Shadow
Final
Recode
of Juniper
25
Juniper=1
Border= 1
Others= 0
Eliminate
clumps
<200 pixels
Juniper=1
Others= 0
Figure 1. The original image.
Figure 2. Image filtered twice.
Figure 4. The classified image.
nel filter applied twice to reduce the high frequency data
variations (Figure 2).
The ERDAS Model Maker Tool was developed to run
the filtering function. Many of the smoothed images were
then examined with an “Area of Interest” tool to develop
specific signatures for a supervised classification (Figure
3). Once a set of reliable signatures was created, supervised
classification was
Figure 3 Selecting signatures.
performed using the
Maximum Likelihood (statisticallybased classifier)
technique. The classified images (Figure
4) served as a source
for two binary files
from each image: a
shadow on background file
(Figure 5) and a
potential juniper on
background file (Figure 6). Two complex
graphical models
were developed to
generate these intermediate images. The
models allow a quick
run of necessary
functions including
recode, clump, sieve,
and final recode. The
clumps that contain
less than 200 pixels and 50 pixels were filtered out, and
reclassified as a background class in the juniper image and
in the shadow image, respectively.
The Rule-based, Object Delineation Program
As discussed in our introduction, the program that implements one part of the Gougeon crown-delineation method
(Gougeon, 1995) is fundamental in our juniper selection procedure. We depend upon it to distinguish objects (both shadows and potential junipers) and identify the boundary-pixels and internal fill-pixels of each. This section discusses
briefly our implementation of the Gougeon delineation procedure.
Although the ERDAS-based image classification provides
clearly distinguished objects, there is no object-specific list
of boundary-pixels or a count of fill-pixels. The primary
objective of our delineation is to provide such a list for each,
numbered object. This delineation adds clarity to the object
definitions and it often de-couples groups of objects that
may not be distinguished by image processing alone.
The program is built around five rules designed to guide
the creation of a closed polygon beginning with an initial
26
Figure 5. Binary image of shadow.
Figure 6. Binary of potential juniper.
start-pixel and proceeding step by step clockwise around an
object’s perimeter while recording the path as a series of
boundary-pixels. The start-pixel is based upon a center-pixel
located in the interior of the object. Each step moves around
the object in a generally clockwise direction keeping object-values to the right and background-values to the left as
the procedure steps forward from one boundary-pixel to the
next. (Gougeon (1995) includes a detailed discussion of his
rules, and we direct the reader there for such details.)
The rule-based program terminates after all object-value
pixels have been encircled and removed from the binary image that was generated by the ERDAS pre-processing. At
this point, the output image is complete and text files that
hold the boundary-pixel list and other object-defining parameters are available.
Creation of Juniper-Shadow Object Images and Text File:
Since the ERDAS pre-processing delivered initially the
shadow objects and the potential juniper objects from the
same image file, the spatial relationship between delineated
shadows and juniper can be specifically established with
their boundary and fill-pixel coordinates.
Examples of a shadow object and the potentially associated juniper object are shown Figure 7. The shadow object
pixels are labeled with 0 as background, 1 as the interior, 2
as the boundaries, and 3 as the object centroid. The potential juniper object pixels are labeled similarly with 0 as background, 5 as the interior, 6 as the boundaries, and 7 as the
object centroid.
ground pixels, 3 for the shadow object centroids, 4 for the
shadow and juniper objects overlap (note: 4 is assigned to
pixels that had the value 1 or 2 in the original shadow file
and the value of 5 in the potential juniper image file), 5 for
the fill-pixels of potential juniper objects, 6 for those potential juniper objects’ boundary-pixel that interfaces between
the potential juniper and the background pixel, 7 for the
potential juniper centroid, 8 for those potential juniper
boundary-pixels that interface between the potential juniper
object and a shadow object, and 9 for the boundary-pixels
common to both a shadow object and a potential juniper
object.
Juniper Selection—Discrimination Rules:
Our experience has shown that the pre-processing and
rule-based routines distinguish a wide variety of irregular
shadows and potential juniper objects individually. But, for
the actual juniper, one anticipates a combined image similar
to that shown in Figure 8. We call this the classic shadowjuniper combination that should appear around an actual
juniper. If one can expect this, the following discriminates
are apparent:
i.)
there is a shadow area (pixel value 1) separate
from the juniper,
ii.) there is a juniper area (pixel value 5) separate
from the shadow,
iii.) there is a shadow-juniper overlap (pixel values 4)
be tween the two,
iv.) there are background pixels (value 0) just outside
of both the shadow and juniper boundaries in
dividually,
v.) there are shadow boundary-pixels (value 2) just
adjacent to the juniper and/or background areas,
and
vi.) juniper boundary-pixels (value 6) adjacent to the
background, and
vii.) juniper boundary-pixels (value 8) adjacent to the
shadow, and
viii.) there should be at least two common boundaries
The Creation of the Combined-Object
Images:
As will become clear in the next section, our selection of
junipers from the objects delineated as potential junipers
depends upon their relationship to the shadow objects. This
can be quantified as show in Figure 8. Figure 8 is created
by merging the individual image files that defined the objects shown in Figure 7. The ultimate pixel labeling in the
combined image is as shown: 0 for background pixels, 1 for
the fill-pixels of shadow objects, 2 for the shadow objects’
boundary-pixels that interface between shadow and back27
Figure 7. A Shadow object and the associated Juniper Object.
ix.)
Figure 8. Classical Shadow Juniper Relationship.
boundary-pixels with values other than 0 or 1.
Once these sums have been established, an object that fits
the classic pattern of Figure 8 can be identified easily. Specifically, the boundary values must include a significant number of values 6 and 8; the inside_bp counts must include
both 4 and 5 counts; and the outside_bp counts of 0s and 1s
must be significant. If this is true, we have a juniper object.
If any of the counts are zero, we do not have a juniper object. We quantify these considerations into three tests.
(value 9- an overlap of the shadow boundary
(value 2) and juniper (value 6) boundaries), and
there will be a centroid for the shadow (value 3)
and one for the juniper (value 7).
Because the boundary- and fill-pixels have been identified
in the delineation process, these observations can be quantified through pixel counting. We assemble these counts as
follows:
I.) The bp_testing: count the pixel values along the object
boundary. Expected sums:
i.)
kbp6, number of boundary-pixels with value 6,
ii.) kbp8, number of boundary-pixels with value 8,
iii.) kbp9, number of boundary-pixels with value 9,
and
iv.) kbp_other, number of boundary-pixels with val
ues other than 6, 8, or 9.
Primary Discrimination Tests:
Test 1: A juniper object must have non-zero values for kbp6, kbp8,and kbp9.
Test 2: A juniper object must have non-zero values for ki4 and ki5.
Test 3: A juniper object must have non-zero values for ko0 and ko1.
Secondary discrimination rules:
The primary tests provide the minimum requirements
for selection of a juniper. We use also some secondary tests
that trim this juniper list even further:
(Note: These tests are somewhat arbitrary at this point
in time. It was beyond the scope of this study to estab
lish a more well-founded set of tests. How ever, these
rules did seem justified from the experience of our testing to date.)
(Note2: These tests duplicate each other somewhat.
This is done intentionally. We wanted to record inde
pendently the occurrences of failure. The frequency of
secondary failures is recorded in a final spreadsheet
summary. See the RESULTS section.)
II.) The inside_bp testing: count the values of pixels adjacent to the boundary and immediately inside the delineating
polygon. Expected sums:
i.)
ki5, number of immediately-inside-boundary-pix
els with value 5,
ii.) ki4, number of immediately-inside-boundary-pix
els with value 4, and
iii.) ki_other, number of immediately-inside-bound
ary-pixels with values other than 4 or 5.
III.) The outside_bp testing: count the values of pixels adjacent to the boundary and immediately outside the delineating polygon. Expected sums:
i.)
ko0, number of immediately-outside-boundarypixels with value 0,
ii.) ko1, number of immediately-outside-boundarypixels with value 1,
iii.) ko_other, number of immediately-outside-
Secondary Test 1: The pixel-count in a juniper object must
be larger than 200.
This limits the crown size of concern. In the digital
images used here, 200 pixels represent a crown size of
3 square meters: 200 time the area represented by each
pixel on the ground.
28
Table I. The Juniper Selection Statistics (Summaries over all; only five shown).
Image
Name
fia125
fia126
fia127
fia128
fia129
Sums:
Ave:
Juniper Shadow Angle
Size
Size
1215
576
1305
566
1271
673
1426
730
1458
576
153227 166562
1197.09 1301.27
-51
-49
-54
-57
-52
Actual
Juniper
on the
ground
Missed Missed
Not
juniper
Potential Potential Selected Selected
Used
Used
%
Junipers
%
Counted
Juniper
%
Junipers
%
Juniper
s
84
8
10
0
140
167
83
99
76
90
67
15
22
0
157
234
73
109
52
78
1061
27.21
257
6.59
670
17
18
0.46
2167
55.56
14213
364
1004
25.74
5377
138
822
21.08
3409
87
Secondary Test 2: kbp6<15% of total boundary and
ki5<15% of internal pixel total.
This eliminates juniper objects that seem to have a ridiculous amount of associated shadow.
Secondary Test 3: ki4>81% of internal pixel total.
Again, this eliminates the potential juniper that has too
much associated shadow.
Secondary Test 4: kbp6<15% and ki5<15% and fill-pixels
<200.
This test records the fact that both Test 1 and 2 conditions were satisfied.
Secondary Test 5: fill-pixels<200 and ki4>81%.
This rule records the fact that both Test 1 and 3 conditions were satisfied.
Secondary Test 6: kbp6<15% and ki5<15% and fill-pixels<200 and ki4>81%.
This rule records the fact that all three of the Second
ary Tests were satisfied.
RESULTS
The primary results of this study are the original images
overlaid with the delineated boundaries of the selected junipers. These are reported as ERDAS view files (*.vue; e.g.
Figure 9) and these document the success or failure of the
procedure overall. A random set of 39 of these files was
examined to provide statistics. These were assembled in a
spreadsheet (Table I) that summarized statistics over all images as well (although our Table I here does not show all
images; only the last five). There are two other text files
associated with each image. Table II shows an example of
some of the entries in one of these files, namely, p_tot=the
number of boundary-pixels around an object; T_tot=the number of internal (fill) pixels; rc_M=the row and column coordinates of the object’s centroid. Table II shows also a portion of the row,column array of the boundary-pixels that define the object in image space. Our full report (Carson and
Akay, 2001) had a more extensive list of other parameters
designed to distinguish shadow or juniper shape, however,
these have not been used as yet. Table III shows the other
text file and an example of another result: the link between a
potential juniper and its associated shadow. These data ensure that there is a shadow for each juniper, and the geometric angle formed by connecting the centroids of the two objects can ensure that the shadow is cast in the proper direction. Table IV shows another spreadsheet listing other results over all 128 images (although only the last 5 specific
images are shown). It shows in its first column the number
of potential junipers identified on each image by the results
of image pre-processing only. The other columns show the
number of shadows, the number of juniper selected by an
application of our primary rules, and the final number used
after the secondary (Tests 1 through 6) and tertiary (Tests 7,
8, and 9) were applied.
Tertiary Discrimination Rules:
These rules presume that all very large objects are related
with anomalies other than a classic juniper with shadow.
Tertiary Test 1: The number of fill-pixels for a juniper ob
ject must be smaller than 25000 pixels. This would
mean that we expect any object larger than, say, 17
by 7 meters is not a juniper.
Tertiary Test 2: The number of fill-pixels for a shadow
object must be smaller than 50000 pixels. This would
mean that we expect any object larger than, say, 24 by
24 meters is not a shadow caused by a juniper.
So, if the potential juniper object or the related shadow
object fail these tests, the potential juniper is not accepted.
(There is a coded output in the final spreadsheet. Tertiary
Test 1 only 7; Tertiary Test 2 only
8; either of these in
addition to one of the Secondary Tests
9.)
Spreadsheet (Table I) Summaries:
This table shows results for five images, and statistics for
the 39 images chosen randomly to serve as observations on
the success and failure of the selection process. The first
29
Figure 9. The original image with selected junipers delineated (ERDAS *.vue file).
second and third columns show the average size of the number of junipers and shadows (in pixels) on an image. The
product of the juniper size average and the number of juniper Used (second from last column) gives the juniper crown
coverage (again, in total pixels). The fourth column is the
average angle (with respect to the image row, column axes)
between the juniper and shadow centroids. This is a measure of the sun angle and the photo orientation. (Note: We
have not used this value in our selection procedure, but we
could. Please see our recommendations.)
For the 39 test images, the table lists the actual number
of junipers observed in the image. The Missed and Notjuniper Counted were observed as well. There is a column
called the Potential Juniper. This is the number of juniper
objects presented in the binary file that came from the
ERDAS-based, image pre-processing. The next column
shows the percentage of this number based on the actual
number of junipers observed. Following those coulmns are
the number of Selected Juniper (i.e., by the primary tests)
and the number of Used Juniper (after the secondary and
tertiary tests). These numbers are also related as percentages of the actual number of junipers.
The averages of these columns—particularly the Actual
Juniper, Missed, Potential, Selected, and Used—do suggest
the overall success of our selection process. However, the
individual results for any particular image and the range of
success are, perhaps, of most interest.
of the columns hold the same data as displayed in Table I.
In addition, the average size of the related shadows is listed,
and a detailed report of the frequency of rejection by secondary and tertiary tests is listed. (Our Recommendations
suggest a more extensive use of these values.)
Table IV Summary:
Our study was a preliminary investigation of limited
scope. The primary focus was upon image pre-processing
and the implementation of the object delineation and rudimentary selection Tests. In general, our overall findings do
CONCLUSIONS
The accuracy of our procedure for the automatic interpretation of the digital CIR images was established ultimately by comparing observable juniper with the ‘selected
juniper’ objects overlaid on the original image. On the 39
images selected at random (Table I), the average juniper
count per image was 27.21. Of these our analysis missed
an average of 6.59 (24%) and miss-identified 0.46 (2%).
The missed percentage is large, however, the figure should
be compared with the results produced by image processing only. For the same (randomly selected) images, the average number of ‘potential junipers’ (based upon the juniper objects delineated from the binary file as delivered from
the image pre-processing) was 55.13. From this number,
our delineation and selection procedure selected 20.16: 74%
of the true juniper count. In comparison, the image processing results would have over-counted the true count by
103%.
RECOMMENDATIONS
This table was prepared primarily to record the results as
related to the various tests in the selection process. Several
30
The Table II and III shown here are merely extracted from more general files as examples. Every image has output files
of these and other object-related parameters. (See Carson, et.al. 2001).
Table II. The Object Parameter File
Obj#= 5,
p_tot= 55,
Tot_in= 214, rc_M= 26, 293
boundary-pixel (#,row,column)
(1, 19, 290) (2, 18, 291) (3, 18, 292) (4, 17, 293) (5, 17, 294)
(6, 17, 295) (7, 16, 296) (8, 16, 297) (9, 16, 298) …(etc.)
Table III. The Juniper and Related Shadow File
J.Obj# p_tot
5,
11,
29,
47,
78,
55,
110,
161,
37,
37,
T_tot
214,
534,
710,
108,
118,
rc_M
S.Obj#
24, 291
2,
141, 356 22,
321, 78 64,
446, 606 80,
741,1157
p_tot
63,
99,
172,
44,
83,
T_tot
305,
487,
1675,
151,
42,
rc_M
Angle
20, 292,
25, 364,
295, 95,
443, 607,
115, 726,1159,
-14
-26
-33
-18
-7
Table IV. Frequency of Secondary and Tertiary Test usage in the Selections (typical)
Image
Potential
Shadow
Selected
Used
Name
fia125
fia126
fia127
fia128
fia129
Sums:
Ave:
Junipers
150
140
134
157
166
6917
54.04
Polygons
150
124
128
142
146
Junipers
97
83
58
73
66
3134
24.48
Junipers
87
76
49
52
58
2523
19.71
Secondary Elimination Tests
Tertiary
Rule 1 Rule 2 Rule 3 Rule 4 Rule 5 Rule 6 Rule 7 Rule 8 Rule9
4
0
0
1
1
4
0
0
0
3
0
3
0
0
1
0
0
0
5
0
1
0
3
0
0
0
0
12
3
1
1
1
3
0
0
0
5
0
1
0
0
2
0
0
0
224
26
92
24
128
73
9
35
0
1.75
0.20
0.72
0.19
1.00
0.57
0.07
0.27 0.00
Statistical Summary Table over all images
Potential Juniper -Selected Juniper = 54.04 - 24.48 = 29.55
Potential Juniper -Used Juniper = 54.04 - 19.71 = 34.33
Selected Juniper - Used Junipers = 24.48 - 19.71 = 4.77
31
encourage more investigation into this approach to automated
tree recognition. Following are some of the areas of research and development that would appear to be worthwhile.
Several of these are based upon a closer look at the data
already generated; some require further developments.
More statistical investigation: The current statistics
mentioned in the Results section above offer an overview of
success and failure, but they do not suggest reasons. The
study should be extended to correlate the successes and failures.
Extend the study of selection rules: Again, statistics
should be developed to expose the effectiveness of the Primary discrimination rules. For example, there should be
statistics that track the reasons for a “Missed Juniper”. Also,
as noted earlier, some of the test parameters in the Secondary Rules were somewhat arbitrary. These should be examined more closely.
There are opportunities for more detailed discrimination
rules. Proximity rules are, perhaps, the most obvious. For
example, the tree counts per area density of junipers and
their associated shadows are sure to affect the shadow-juniper relationship and, therefore, selections.
Introduction of photo geo-referencing and use of the
shadow angle: The direction of the shadow is an obvious
piece of information that should be used. This would require that the photos be geo-referenced and that the time
and date of exposure be provided. The images of this study
were not accompanied by such data. In the future they should
be. Also, there should be some thought about optimizing
the shadow feature information by planning the date and
time of exposure.
Cross check the final object signatures with the original images: Once the final juniper objects are selected, their
boundary will delineate the pixels on the original image. A
histogram of these pixels could be examined to ensure that
the signature fits the juniper profile. It is likely that such a
study would improve the signatures for pre-processing other
images in the future.
Extending the pre-processing models: The ERDAS
Imagine program offers a Modeler. This was discussed in
our Methods section. We implemented a Modeler for most
of our process, however, we could extend the implementation to the actual Classification. This would improve the
production rate of the procedure.
Introduce stereo-images, DTMs, and the use of stereo-collaboration: There is an opportunity to use geo-referenced stereo-images to provide a second, independent
check on a juniper selection. Stereo-images offer two views
of the same object. Each image could be treated independently as they were here. Then, the independent views could
be combined photogrammetrically to collaborate a selection,
or put it in doubt. (This has been suggested by the seniorauthor for the past several years. The results of this study,
and the introduction of LIDAR and the Direct Geo-Referencing of images, should recommend such a project with a
high priority.)
Investigate the application to images derived by scanning large-format photographs: Developments in recent
years in the traditional Aerial Survey technologies promote
the use of large scale, large-format photographs as a medium for improved sampling techniques. These photos could
be scanned to provide digital images similar to those treated
in this study. The files would be much larger, but the procedures for juniper selection would otherwise be identical to
those used here. Nevertheless, it would be good to study
some cases to ensure that large files did not introduce unforeseen problems.
LITERATURE CITED
Belsky, A.J. 1996. Viewpoint: Western juniper expansion:
Is it a threat to arid northwestern ecosystems? Journal
of Range Management, 49(1), January 1996. pg 53-59.
Carson, W.W., Akay, Abdullah. 2001. Juniper Tree Size and
Frequency Assessment from Large-Scale, Aerial Digital
Images. A report submitted to Dale Weyermann, USDA
Forest Service, PNW Station, Portland, Oregon. June 1,
2001.
ERDAS, Inc. 1997. ERDAS Field Guide, Version 8.3.1,
Atlanta, GA, 655pp.
Garrison, G.A., A.J. Bjustad, D.A. Duncan, M.E. Lewis,
D.R.Smith. 1977. Vegetation and Environmental Features of Forest and range Ecosystems. USDA Forest
Service, Agricultur Handbook No. 475.
Gougeon, Francois. 1995. A Crown-Following Approach to
the Automatic Delineation of Individual Tree Crowns in
High Spatial Resolution Aerial Images. Canadian Journal of Remote Sensing. Vol.21, No.3, August 1995.
Pp.274-284.
Hill,D.A., Leckie,D.G., 1998. Proceedings of the International Forum, Automated Interpretation of High Spatial
Resolution Digital Imagery for Forestry. Natural Resources Canada, Pacific Forestry Centre, Victoria, B.C.,
February 10-12, 1998.
Karl, M.G., S.G. Leonard. 1995. Western Juniper (Juniperus
occidentalis ssp. occidentalis) in the Interior Columbia
Basin: Science Assessment (review draft). Interior Columbia Basin Ecosystem Management Project. Science
Integration Team, Terrestrial Staff, Range Task Group.
Miller, R.F., J.A.Rose. 1995. Historic Expansion of
Juniperus Occidentalis (Western Juniper) in Southeastern Oregon. Great Basin Naturalist 55(1), pp 37-55.
Miller R.F. 1999. Personal communication at planning meeting for 1999 juniper inventory of eastern Oregon. Bend,
Oregon.
32
Miller, R.K. 1997. Cultural Uses of the “Forgotten Forest”.
Journal of Forestry, 1997,95:11. pp 24-28.
Weyermann, D. 1999. An Inventory of Juniper in Eastern
Oregon using Aerial Digital Imagery—A study plan.
USDA Forest Service, PNW Research Station, Portland,
OR. February 16, 1999.
Richards,J.A and Jia, X., 1999: Remote Sensing Digital
Image Analysis; 3rd Edition; Springer-Verlag Berlin
Heidelberg.
33
Chapter 3
Spatial Variability Within Managed Forest Stands
JOHAN STENDAHL
Abstract—In the prospect of precision forestry the need for a more high-resolution description of the forest resource is
evident. A fundamental issue is the within stand variation. How large it is, to what extent it is spatially autocorrelated, and which
ranges of spatial correlation that exists for different properties. This is important both when estimating the effort needed for
resolving the spatial variation by field sampling and when considering the size of the treatment units feasible in a stand-free
forestry planning system.
Among the factors affecting the spatial structure in managed forests are forest operations, such as thinnings, as well as natural
variations in for example soil conditions. This will lead to non-static spatial stand structures, which will be a function of stand
age, forest operations, and site properties. The need for and benefit from a high-resolution spatial description of the forest is
most apparent for the mature forest close to final felling.
Data from three managed stands in different stages of development was used to investigate the within-stand variation. In each
stand, all trees were positioned and measured. The spatial structure was investigated by means of geostatistical methods, which
provide quantitative tools to measure spatial autocorrelation.
In the mature mixed pine-spruce stand, spatial correlation was found at two scales of variation: at 9 m - 38 m, and at 120 m
– 170 m. In the unthinned pine forest the spatial correlation was weak, but ranges at 10 m - 37 m were generally found. In the
unthinned spruce stand, the ranges were rather consistently between 67 m and 83 m. Those properties that had the strongest
spatial correlation were plot mean diameter and tree height, whereas for basal area and stem density it was poor in general. The
spatial correlation in tree diameter for individual trees was weak, which shows the considerable variation at the local level
because of inter-plant competition. Interestingly, this was not the case in the mixed stand when the diameter for pine and spruce
were studied separately.
and Thuresson, 1997; Lu and Eriksson, 2000). The spatial
distribution of a property is the result of many interacting
processes acting at different spatial scales integrated over
time. The result of these complex processes is revealed as
the spatial structure of a property at different scales.
Geostatistics provides the tools to investigate the spatial
structure of properties and estimate the values of properties
at unsampled locations. With the variogram it is possible to
investigate which spatial scale or scales of variation are the
most important in an investigated area, and to what extent
the variation is spatially correlated. Geostatistics has been
applied before on forest data at the tree level by for example
Kuuluvainen et al. (1994 and 1998) and Biondi et al. 1994.
The aim of this study was to investigate the within stand
variation in three managed stands in different stages of development.
INTRODUCTION
If our aim is higher precision in forest operations and better information about the forest resource, then spatial variability in forest properties needs to be considered. The variability in forest properties depends on factors such as stand
age and development, varying site conditions, and how the
forest is being managed. In stand based forestry the major
source of variation is the stand structure. Over time, the landscape will develop into a mosaic of forest stands in different
stages of development. The aim of the stands is to outline
areas with similar conditions, which are assumed to be internally homogenous and spatially stable over time, but this is
not always the case. In response to differences in plant establishment and site conditions, such as soil properties and
physiography, trees will develop at varying rates, which will
lead to the within-stand variation in tree features such as
height and diameter (Ståhl, 1992). Forest operations such as
thinnings will also affect the spatial variation of forest properties, as the aim usually is to reduce the variability in basal
area in the stand. More optimal management of the forest is
now possible with new computer-based models. These allow for continuous spatial variation to be taken into account,
for example a raster description of the forest, which enables
dynamic treatment units (Hof, 1992, 1993, 1996; Holmgren
MATERIAL AND METHODS
The three investigated stands (Figure. 1) are located in
the south of Sweden: one mixed mature stand located at
58°45’N, 16°05’E, one unthinned Scots pine stand (Pinus
sylvestris (L.)) located at 58°42’N, 16°10’E, and one
unthinned Norway Spruce (Picea abies (L.) Karst.) stand
located at 57°08’N, 14°45E. The mixed stand had approxi35
mately 40% pines and 60% spruces. The approximate stand
age for the mixed stand was 95 years, for the pine stand 40
years, and for the spruce stand 31 years. The latter stand had
been thinned sparsely 15 years earlier. The sites were located on Podzol soils (Orthods) formed from tills on granite
bedrock.
An exhaustive forest inventory was carried out whereby
the diameter at 1.3 m height was determined for each tree
larger than 5 cm, and in addition the height was determined
for 25% of the trees in the mixed stand and the pine stand.
The position of each tree was determined in relation to a
reference 25 m square grid. This grid was laid out in each
stand using a levelling instrument, and each node was marked
with a wooden pole. From the nodes the distance and azimuth to each tree was measured and hence the absolute position could be calculated. The location of the reference grid
was determined by DGPS.
Data for individual trees were used for the analyses, and
circular plots were located in the stands (virtually) to investigate local mean values and area-based properties such as
basal area and stem density. In the mixed stand there were
149 plots with radius 7.98 m (200m2), in the pine stand 101
plots with radius 5.64 m (100 m2), and in the spruce stand
163 plots with radius 5.64 m (100 m2). The neighbouring
plots were touching and the plots covered the whole stands.
double spherical:
3


1 h
 3h 1  h  
 3h
γ (h) = c0 + c1 
−    + c2 
− 
2
a
2
a
2
a
 1  
 1
 2 2  a 2
for 0 < h ≤ a1
 3h 1  h 3 
−  
γ (h) = c0 + c1 + c2 
2a 2  a2  
 2

)=
1
2 Μ (h
Μ
(h )
{z (x ) − z (x
)∑
i
i=1
i
+ h
)}2
The mixed mature stand had approximately 450 stems/
ha and the species mixture between pine and spruce was
about 40/60 (Table 1). In general the pines were larger in
diameter than the spruces. The variograms of the diameter
for individual trees had a very large nugget variance (Figure. 2), which indicates a large amount of non-spatially correlated variation, or error component. The range of spatial
variation for diameter in the mixed stand was 65 m. There
was a substantial difference when the variogram was calculated for pine and spruce trees separately (Figure. 3), and
those variograms had a strong spatial correlation, or in other
words better spatial structure. This was due to layering by
tree species, where large pines seemed to coexist with small,
suppressed, spruces and vice versa. This was also seen in
the inverse relation between the mean diameter for pine and
spruce found at plot level (r=-0.82). The large semivariances
observed at small distances were also enhanced by the fact
that the semivariance is based on squared differences. The
range of spatial variation was 16 m for pine diameter, whereas
for spruce there were two ranges of variation at 9 m and 50
m. This nested structure, indicates variability at two distinct
spatial scales. The variogram of the plot mean diameter had
much stronger spatial correlation than the diameter for the
individual trees. The averaging evens out the large differences between spruce and pine trees at small distances and
probably gives rise to the more continuous variability. The
variogram was nested with ranges at about 40 m and 170 m
(Table 2). For tree height the variogram had a fairly small
nugget variance, and thus tree height varied more smoothly
than for example tree diameter. It had two nested structures
with ranges 25 m and 130 m. The spatial correlation for
basal area and stem density was fairly weak and the ranges
were 35 m and 120 m respectively. This is the result of previous thinnings, which reduces the variability in these properties.
,
for 0 < h ≤ a
for h > a
γ (0) = 0
exponential:
γ (h) = c0 + c {1 - exp(-h/r )}
for a1 < h ≤ a2
RESULTS AND DISCUSSION
spherical:
γ (h ) = c0 + c



where γ (h) is the semivariance, c0 is the nugget variance,
i.e. the random error component of the variation, and c is
the sill variance (c1 and c2 for nested models), i.e. the spatially correlated variation. The effective range, a, is the range
of spatial dependence (a1 and a2 for nested models) and corresponds to r·3 for the exponential model.
where γ (h ) is the experimental variogram, γˆ is the
semivariance, M is the number of pairs comparisons, h is
separating distance or lag, and z(xi) and z(xi+h) are the values of a variable at positions x and x+h. From the deciles of
the tree diameter ten binary variables were calculated, and
from these the indicator variograms were calculated as above.
To the experimental variogram, different authorized models
(Webster and Oliver, 2001) were fitted by least square fitting. The following models were used:
 3h 1  h 3 
γ (h) = c0 + c −   
 2a 2  a  
3
γ (0) = 0
For the geostatistical analysis variograms were calculated
using the formula:
γˆ (h




for h > 0
γ (0) = 0
36
Figure 1. Stem map of the three investigated stands. The size of the symbols is proportional to the stem diameter. (MX=Mixed mature
stand, PS=Pine stand, SS=Spruce stand).
37
Table 1. Descriptive statistics of the investigated forest stands.
Variable
Unit
Mixed stand
Pine stand
Spruce stand
Area
Tree data:
Ntree
Pine
Spruce
Diameter
Var.
Diam. pine
Diam. spruce
Height
Var.
Plot data:
Nplot
Mean diam.
Var.
Basal area
Var.
Stem density
Var.
(ha)
5.12
1.92
2.35
(%)
(%)
(mm)
(mm)2
(mm)
(mm)
(cm)
(cm) 2
2390
38
58
264
9031
313
236
225
1564
1639
92
8
185
3004
191
121
157
577
3070
6
91
161
2345
177
162
-
(mm)
(mm)2
(m2/ha)
(m2/ha)2
(stem/ha)
(stem/ha)2
149
281
2060
30
99
453
2.5E4
101
184
656
26
92
913
1.1E5
163
162
413
28
73
1294
1.4E5
Table 2. Variogram models for the investigated properties in the three stands (MX=Mixed mature stand, PS=Pine
stand, SS=Spruce stand). The model parameters are the nugget variance (c ), the sill variance (c for single
0
models, c and c for nested models) and the effective range (a for single models, a and a for nested models),
1
2
1
2
where a=3·r for the exponential model.
Variable
Model type
c0
c (1)
Diameter MX
Diameter pine, MX
Diameter spruce, MX
Diameter PS
Diameter SS
Exponential
Exponential
Double spherical
Double spherical
Spherical
7206
1998
4697
1582
2010
1579
3171
1654
248.9
415.6
Height MX
Height PS
Mean diam. MX
Mean diam. PS
Mean diam. SS
Double spherical
Spherical
Double spherical
Spherical
Spherical
733.7
354.9
440.9
515
202.2
285.5
202.5
1033
208.2
232.7
BA
BA
BA
Spherical
Spherical
58.9
54.7
45.6
26.5
Spherical
Spherical
Spherical
18674
60404
131382
7001
56265
10565
MX
PS
SS
Stem density
Stem density
Stem density
MX
PS
SS
38
c (2)
2249
320.6
460.3
a (1)
65.3
16.2
9.2
9.8
80.0
a (2)
49.7
91.0
94.1
79.3
92.2
95.8
97.8
690.1
25.4
26.1
38.2
118
67.1
170
95.1
75.6
89.8
68.7
97.2
-
35.8
81.7
-
70.7
75.4
120
36.8
82.8
133
r2
62.2
91.3
5.4
Figure 2. Variograms of tree diameter and height for the three stands. The calculations are based on individual trees (MX=Mixed
mature stand, PS=Pine stand, SS=Spruce stand). Tree height was not measured for the spruce thinning.
Figure 3. Variogram for diameter of pine (P), and spruce (S) in the mixed stand. The calculations are based on
individual trees.
39
Figure 4. Variograms of the plot-based forest properties basal area, mean diameter and stem density (MX=Mixed mature stand,
PS=Pine stand, SS=Spruce stand).
Figure 5. Indicator variograms of the deciles of the tree diameter: x =decile 2, Ë% =decile 4, + =decile 6, ¡% =decile 8, “ =decile 10.
(MX=Mixed mature stand, PS=Pine stand, SS=Spruce stand).
40
The unthinned pine stand had 900 stems/ha, which is a
small value for a Swedish thinning. The sparse numbers of
trees also affects the spatial variation within the stand. The
variogram for tree diameter had a very large nugget variance, which was also the case for the mean diameter of the
plots. Both had long ranges of variation at 91 m (diameter
for trees) and 118 m (plot mean diameter). Stronger spatial
correlation was found in the variogram of tree height with
the range 26 m. For basal area no model could be fitted to
the experimental variogram. Although it appeared to be a
range at about 30 m (Figure. 4) the least square fitting was
unsuccessful since the mid-range points were too scattered.
The stem density was the property with the largest amount
of spatial correlation together with tree height. The presence of occasional rock outcrops had caused gaps in the
stand, which explains the spatial correlation structure for
stem density. In general the spatial correlation in the pine
stand was poor, i.e. the stand was rather homogenous.
The unthinned spruce stand had a much larger stem density, 1300 stem/ha, than the pine stand. Similarly to the other
stands, there was a large error component in the variogram
for tree diameter, but when the diameter was averaged over
the plots the spatial correlation became much stronger (Figure. 4). The ranges were similar in the variograms for tree
diameter and mean diameter for the plots (80 and 67 m respectively), which was expected. The stem density was pure
nugget, i.e. lacked any spatial correlation.
The indicator variograms of different diameter classes
(Figure. 5) showed that the larger trees tend to be clustered,
since the variogram for the larger deciles was ascending from
short to longer distances. For the smallest deciles there
seemed to be a negative correlation as the semivariance had
a peak at shorter distances, i.e. small trees tend to be located
close to larger trees rather than in clusters together with other
small trees. For the middle classes (see for example decile 6,
Figure. 5) the variogram was horizontal, i.e. the trees were
evenly spread throughout the stands. This pattern was found
in all three stands although in the thinnings it was not as
apparent as in the mixed mature stand.
To summarize the spatial structure found in the different
stands, the mixed mature stand had ranges of spatial correlation at shorter distances, 9 m - 38 m, and at longer distances, 120 m – 170 m. In the unthinned pine stand, the spatial correlation was weak, but ranges at 10 m - 37 m were
found, except for plot mean diameter at 91 m. In the
unthinned spruce stand, the ranges were rather consistently
between 67 m and 83 m. Those properties that had the strongest spatial correlation were plot mean diameter and tree
height, whereas for basal area and stem density it was poor
in general. The latter properties are affected by thinnings,
which may explain the poor spatial correlation in the mixed
stand. The spatial correlation in tree diameter was weak,
which shows the considerable variation at the local level
because of inter-plant competition. Interestingly, this was
not the case in the mixed stand when the diameter for pine
and spruce were studied separately.
Kuuluvainen et al. (1996) studied 50 m by 50 m plots in
a managed and an unmanaged spruce stand, and calculated
variograms for tree properties. In the managed stand they
found strong spatial correlation for tree diameter and height
with nested variation at about 10 m and 25 m, whereas there
was no spatial correlation in the unmanaged stand. For tree
height this result was similar to the mixed stand (ranges 25
m and 133 m), but for tree diameter of all trees the spatial
correlation was much weaker. The result for spruces and
pines separately was more similar though with strong spatial correlation and similar ranges (16 m for pine and 9 and
50 m for spruce, Table 2).
Knowledge of spatial structures in forests can be of use
in several aspects of forestry. If in the future, forest inventories must provide more precise description of the forest than
today, possibly at the sub-stand level, the spatial correlation
can provide information on how we must sample to resolve
the within stand variation. Spatial estimation by kriging
(Webster and Oliver, 2001) accounts for the spatial correlation, but the knowledge of the variogram is a prerequisite.
Further, the spatial structures may help to decide on the appropriate description units in continuous, raster based, planning systems and estimations of within block variances for
the different units. Spatial structures may also be of importance to understand stand-scale habitat diversity and of aggregation on forest ecosystems (Kuuluvainen et al., 1998).
LITERATURE CITED
Biondi. F., D.E. Myers, and C.C. Aver y. 1994.
Geostatistically modeling stem size and increment in an
old-growth forest. Can J Forest Res 24:1354-1368.
Hof. J., M. Bevers, and J. Pickens. 1996. Chance-constrained
optimization with spatially autocorrelated forest yields.
Forest sci 42(1):118-123.
Hof. J.G., and L.A. Joyce. 1992. Spatial optimization for
wildlife and timber in managed forest ecosystems. Forest sci 38(3):489-508.
Hof. J.G., and L.A. Joyce. 1993. A mixed integer linear programming approach for spatially optimizing wildlife and
timber in managed forest ecosystems. Forest sci
39(4):816-834.
Holmgren. P., and T. Thuresson. 1997. Applying objectively
estimated and spatially continuous forest parameters in
tactical planning to obtain dynamic treatment units. Forest sci 43(3):317-326.
Kuuluvainen. T., E. Järvinen, T.J. Hokkanen, S. Rouvinen,
and K. Heikkinen. 1998. Structural heterogeneity and
spatial autocorrelation in a natural mature Pinus sylvestris
dominated forest. Ecography 21(2):159-174.
41
Stendahl
Ståhl G. 1992. A study on the quality of compartmentwise
forest data acquired by subjective inventory methods.
Dep. Biometry and For. Manage., Swedish Univ. of Agric.
Sci. Rep. 24.
Kuuluvainen. T., A. Penttinen, K. Leinonen, and M. Nygren.
1996. Statistical opportunitiesfor comparing stand structural heterogeneity in managed and primeval forests. An
example fron boreal spruce forest in southern Finland.
Silva Fennica 30:315-328.
Webster, R., and M.A. Oliver. 2001. Geostatistics for Environmental Scientists. J. Wiley & Sons, Chichester.
Lu. F., and L.O. Eriksson. 2000. Formation of harvest units
with genetic algorithms. For Ecol Manage 130:57-67.
ACKNOWLEDGEMENTS
Matérn. B. 1960. Spatial variation. Meddelanden från statens
skogsforskningsinstitut 49(5):1-144.
This work was partly financed by SkogForsk (The Forestry Research Institute of Sweden).
42
Chapter 4
Individual Tree Crown Image Analysis – A Step Towards
Precision Forestry
FRANÇOIS A. GOUGEON
DONALD G. LECKIE
Abstract—Worldwide economy, environmental concerns, and stricter legislation governing forestry practices have put increased demands on forest managers. Riparian zone delineation, helicopter logging, plantation monitoring, selective cuts, just in
time delivery, biodiversity and wildlife management are all various aspects of the same coin. The information requirements
brought on by these activities is staggering. Existing information tools are inadequate and hamper the progress of forest management activities such as precision forestry. The use of high spatial resolution (10-100cm/pixel) remotely sensed images (aerial or
satellite) or scanned aerial photographs, presents possibilities to analyze forested areas on an individual tree crown (ITC) basis.
The Canadian Forest Service is at the forefront of research on individual tree crown based image analysis. We have developed
techniques, methods and processes to separate forested from non-forested areas, delineate individual tree crowns, identify their
species, and if needed, regroup them into forest stands or environmental strata. Eventually, forest managers will forgo static
regroupings in favor of keeping all of the information about the individual tree crowns themselves (e.g., position, crown area,
height, species, and dominance). Regrouping would be done on demand, for each specific application, if done at all. In addition,
the unprecedented level of details afforded by ITC techniques should allow us to extract a variety of additional forest management
information such as: snag locations, forest gap sizes and distribution, highly-valued tree locations, detailed damage and regeneration
assessments. This may also lead to more precise volume and biomass estimates and foster the use of individual tree growth
models.
This paper first presents some of the image analysis concepts, methods and tools behind producing ITC-based forest inventories
and then, reports on some successful applications, limitations, and ongoing research.
Of course, such a transition does not have to be that
drastic. Most of the forestry community now relies on
computers and geographic information systems in their
day-to-day
operations.
Many
traditional
photointerpretation shops have moved to a softcopy environment and are gradually turning into geomatics companies. Transitional states could include: computer assisted interpretation in a softcopy environment or reporting information by forest stand even though the analysis
is done by computer on an individual tree basis. In any
case, interpreters will always remain in the loop, whether
to train a classification process, correct anomalies, or
verify and ascertain the results. In order for automatic
systems to get to the required level of details at the stand
level (e.g., species composition to the nearest 10%), individual tree crowns must be analyzed. Once computers have
enough capacity to store the ITC information for large
areas, foresters will no doubt keep and use the detailed
information, specially knowing that this can lead to application specific on-demand maps and detailed management plans.
The Canadian Forest Service has an ongoing research
project on individual tree crown based image analysis from
INTRODUCTION
A worldwide economy, global and local environmental
concerns and the relatively scarcity of old growth forests
are leading to stricter forestry legislation, certification pressures, and generally, a trend towards more sophisticated
forestry practices. Whether for global, national, provincial,
or regional reporting requirements, local certification or
operational use, there are drastic needs to improve forest
information precision, accuracy, timeliness, completeness,
and cost-effectiveness. Such improvements could come
about gradually via incremental changes to existing methods (e.g., more field visits, more sampling, more detailed
plots), but not without serious cost considerations. Pressing needs and budgetary constraints might force us to opt
for more innovative measures: a significant transition from
mapping relatively homogeneous forest stands and interpreting their content from aerial photographs to using computers to analyze aerial or satellite multispectral images
on an individual tree crown (ITC) basis in order to produce on demand application-specific information at a level
of detail never considered heretofore. In other words, a
drastic shift in paradigm.
43
Individual Tree Crown Delineation
high resolution images (10-100 cm/pixel) from various media (airborne sensors, digitized aerial photos, satellites). We
have developed techniques, methods and processes to separate forested from non-forested areas, delineate individual
tree crowns, identify their species, and if needed, regroup
them into forest strands, either interpreted in the conventional way or automatically generated. From this detailed
information, numerous parameters other than those found
in conventional inventories can be easily extracted (e.g.,
snags locations, forest gaps, health status). This paper describes the techniques and methods that were developed,
their applications to several forestry situations, their current
limitations and the ongoing research directions.
When the illumination image has been selected and the
non-forested areas masked out, the crown delineation is
performed in two main steps. First, from initial local minima
in the illumination image, all possible valleys of shade in
the image are followed pixel by pixel, resulting in a fairly
good, yet often incomplete, separation of tree crowns. Then,
for each block of tree material, a rule-based process tries to
follow the crown boundaries favoring clockwise motions
and aiming for closure. Higher level rules try to detect situations where clusters of trees have been delineated and capitalize on small indentations in the cluster boundary that
may indicate how and where to separate it into individual
tree crowns. Higher level rules to identify and regroup segments of crown into single crowns have not been implemented yet. When all the possibilities (or rules) have been
exhausted, a bitmap of individual tree crowns is generated.
From then on, individual tree crowns are treated as objects
for further analysis. With a multispectral dataset, the crowns
can be fed to a species signature generation process and
then, to the classification process to identify their species.
However, interesting forestry information can also be extracted at this point in the analysis. For example, the bitmap
of individual tree crowns can be used to analyze crown
closure, stem densities, crown size, and forest gaps.
TECHNIQUES AND METHODS
One automated tree delineation method, the valley following approach (Gougeon, 1995), relies on the presence of
shaded material between tree crowns. This generally leads
to good crown delineation in high to medium density conifer forests, depending on the spatial resolution. It is also
capable of separating hardwood crowns, with a lower success rate. However, because of its basic premise, the valley
following technique has to rely on preprocessing techniques
in more open, lower density areas. This preprocessing eliminates (masks-out) non-shaded background material and can
also be used to separate vegetated from non-vegetated areas, and within the former, forested from non-forested areas.
Individual Tree Crown Classification
With multispectral imagery, the ITCs are typically introduced into a supervised classification process. Single
species, single situation training areas are delineated on
the image and ITC-based signatures are generated for each
class. Using the ITC bitmap, the signature generation process takes care of extracting the individual tree crowns
within the training areas, generating ITC-specific signatures, and combining them into class signatures (typically,
mean ITC multispectral mean and covariance of the ITCs).
In addition to, or instead of, these typical multispectral signatures, numerous other types of signatures can be generated which take into account texture, structure, or other
image features. Context can also be considered by introducing extra feature channels such as digital elevation, or
slope and aspect, or sensor look angle, etc.
The classification is based on a maximum likelihood
decision rule. ITCs are taken one by one from the ITC
bitmap, their ITC signature is calculated using the same
features and parameters as in the class signature generation process, their likelihoods of belonging to the various
classes are calculated, and a final decision is made taking a
confidence interval into consideration. When all the trees
are classified, classification results can be displayed and
evaluated.
In a supervised classification approach, the prevalent
way to evaluate classification accuracy is to have put aside
during the training phase several representative areas (or
generate new ones) to be used as testing areas from which
to a confusion matrix is calculated. Alternative approaches
Preprocessing
With multispectral imagery, one can sometimes rely on
the multispectral information itself to create a mask that will
eliminate some areas from further processing. This is typically the case when the background material is made of
woody debris, lichen, sand, rock, soil, or senescing herbaceous material (in early spring or fall). A simple multispectral rule such as, detect pixels having a near infrared radiance smaller than its mean visible band radiance, can be
used to create an effective non-vegetation mask. Within the
vegetative areas, one can often use a texture measure to separate the forested from non-forested areas. Texture is typically the only recourse when dealing with panchromatic
imagery. Of course, when a digital base map or forest cover
is available, the major non-forested areas can be masked out
by simply selecting the appropriate polygons (e.g., roads,
lakes, swamps, railroads). When LIDAR (height) data is
available, simple thresholds can be used to separate ground
level vegetated areas from regenerating (brushy) areas, and
from forested areas, making separate analysis of regeneration and mature forest possible (Gougeon et al., 2001a).
Additional preprocessing typically takes place before the
crown delineation process. For example, the selection of most
appropriate band on which to do crown delineation or the
creation of an appropriate illumination image from a combination of several of the bands, and the subsequent smoothing of that image.
44
can involve comparing the species composition obtained
within given stands with that reported by an existing forest
inventory or with species compositions obtained from
ground transects or field plots within the stands.
local maxima operation. In such cases, unless some kind of
locally adaptive method (e.g., Wulder et al., 2000) is used,
tree counts are typically unreliable. However, species
recognition is often unaffected, as the representative pixels
in each tree crown are still among the purest pixels and
generally classify the best.
The tree top approach has more or less the same
preprocessing requirements as the ITC approach and the
removal of non-forested areas is also compulsory to get
meaningful results. In forested areas, it also benefits from
the presence of shade between tree crowns and uses a
threshold to eliminate a majority of shaded pixels. It is thus
primarily efficient in medium to high density softwoodpopulated areas. In order to deal with small widely-spaced
trees in otherwise open areas, a variation of the tree top
approach which looks for a specific shadow for every tree
crown has been developed (Gougeon and Leckie, 1999b).
Using the sun azimuth and elevation as input parameters, it
looks in the opposite direction for a shaded pixel to associate
with each local maxima. A locally adaptive approach capable
of switching between the traditional and the shadow-specific
tree top techniques has also been developed. The switching
is based on an a priori generated mask that designates areas
with high directionality at an angle commensurate with the
sun’s illumination. This locally adaptive technique has been
highly successful in eliminating false positive tree detections
from the background material of open areas and leads to
much more accurate tree counts. Such a technique could
also be applied to the ITC paradigm. Another locally
adaptive approach capable of switching between the tree
top (TT) and the individual tree crown (ITC) paradigms
based on crown size would complete this picture and make
the analysis of large diversified regions more automatic.
Individual Tree Crown Regrouping
Because forest stands are an intrinsic part of forest management and still the dominant paradigm within the forestry
community, and also, because existing geographic information systems (GIS) are limited in the quantity of polygons
they can handle, regrouping individual tree crowns into forest stands or environmental strata is a necessity. Such regrouping can be based on:
(a)
(b)
(c)
(d)
(e)
forest stands from an existing forest inventory,
newly generated forest stands obtained by conventional photo interpretation methods,
automatically generated stands based on species
content, height, density, and canopy closure,
automatically generated stands based texture parameters, or,
customized regrouping for biodiversity, wildlife
management, or the application at hand.
Regrouping techniques c, d, and e, are the subject of ongoing research. They will need to be evaluated and hopefully, judged comparable to today’ standards, and acceptable by states and provincial authorities and forest companies which are responsible for forest inventories. However,
whatever the regrouping technique used, there is potential
for precise information within these regroupings. Eventually, the demands of precision forestry will force us, and the
ever increasing power of computers allow us, to forgo these
static regroupings in favor of keeping all of the information
about the individual tree crowns themselves (e.g., position,
crown area, height, species, dominance). Regrouping would
be done on demand, for each specific application, if done at
all.
Forestry Parameters Extraction
As mentioned above, numerous forestry parameters such
as stem density, crown closure, crown sizes, forest gaps,
gap distribution, etc. can easily be extracted, even without
species classifications. However, quantitative data on density, canopy closure and crown diameters still need to be
validated, especially relative to the various spatial resolutions (10-100 cm/pixel) and media (digitized B&W or color
aerial photos, aerial or satellite images from various sensors). Nevertheless, relative information such as portrayed
by stem density, canopy closure and crown size images
shown in the typical cold to hot color scheme have been
found to have their own merits in computer assisted interpretation. These images are also used as input in one of our
stand delineation process.
At this point in time, there are very little standards relative to forest gaps or gap distribution, except for regenerating areas where stocking needs to be assessed. Obvious
applications may be in wildlife management and
biodiversity. For example, mammals (big or small) will only
stay in a given region if their preferred patterns of forested
to open areas is present. Immature, mature, and old growth
The Tree Top Approach
Another approach to establishing tree locations, counts,
density and even species, a method often used with lower
resolution images (Gougeon and Moore, 1989) or with
smaller trees, such as in young regenerating areas (Gougeon,
1997), is the tree top (TT) or local maxima approach. It
consists of detecting the most brilliant pixels (or local
maxima) in an image and hopefully, only one such pixel
within each tree crown. For conifer crowns seen close to
nadir, such pixels are often located on the sunlit side of tree
crowns near the tree top. For hardwood trees, the relationship
is not as straightforward, as it is less likely to have a one-toone correspondence between a local maxima and a single
tree crown. The relationship is more dependent on the spatial
resolution of the image and the window size used by the
45
FORESTRY APPLICATIONS
Figure 1. Individual tree crowns (ITCs) delineated and
classified on a 36 cm/pixel multispectral image.
Forest Inventory
One of the primary goals behind the development of these
individual tree crown image analysis techniques and methods is the production of detailed, precise, accurate and timely
forest inventories. The same original ITC data could be regrouped to different abstraction levels depending on specific management or operational needs. In time, such an
approach may gradually replace the photo interpretation
based processes currently used throughout Canada. Then,
the availability of information at the ITC level could alter
the basis on which forests are managed from a largely stand
based premise to a tree based one. So, how close are we to
ITC-based forest and vegetation inventories?
Work on multispectral aerial images at 31 cm/pixel
(Gougeon, 1995b) has demonstrated that under ideal circumstances at that high spatial resolution 81% of the crowns
delineated by the valley following approach were the same
as those delineated by interpreters (Figure 1). It is also known
that this performance degrades as the spatial resolution becomes lower and as situations get more complicated (e.g.,
open areas, hardwood trees). For example, an ITC analysis
of a 1m/pixel IKONOS image is not likely to produce viable tree counts. On the other hand, the presence of some
tree clusters instead of ITCs may not interfere with the production of forest inventories that are better than the interpretive ones (Figure 2).
forests have different gap patterns. The detection of snags
is important to assess the nesting bird potential of a given
area.
Of course, the addition of ITC species classification
brings us closer to a full-fledge forest inventory, with detailed species composition for every stand, especially if
combined with LIDAR height data. Depending on the LIDAR data spot density, there are possibilities to derive
heights on a stand or an ITC basis. This also provides information needed to establish volume and biomass relationships.
Figure 2. TC analysis of an IKONOS image with stand polygons from the existing forest cover map. ITC classes are: white pine (red),
other softwoods (blue), yellow birch (yellow), other hardwoods (green), and hardwood regeneration (light green). Lakes (light blue)
and swamp areas (orange) are from the digital forest cover map.
46
Interesting results have been achieved with species recognition. In (Gougeon et al., 1999a), an area populated
with five coastal coniferous species was analyzed using 60
cm/pixel aerial data. The species compositions generated
by a supervised classification were found to be only 12%
off relative to field transects through sixteen stands. The
dominant species allotment was on average 10% off, and
species that composed more than 25% of a stand, were 14%
off. However, one should keep in mind that such success
was achieved with young trees (around 20 years old) and
only five species to differentiate. Young trees are typically
easier to differentiate among themselves than mature trees
where health often interferes with species recognition
(Leckie et al., 1999d). Other work (Leckie and Gougeon,
1999a), considering more species (11-19) and including
hardwood trees, reported classification accuracies of 56%
for eight coniferous species and 65% for three hardwood
species. Compared to interpreters placed in the same situation and using powerful image enhancements, results for
coniferous species were inferior by 15 percentage points,
but superior by 6 percentage points for the hardwoods.
Regrouping ITCs into forest stands and generating stand
attributes is another subject of major interest. In some jurisdictions the inventory process is done in two phases: a)
the stand delineation phase and b), the stand attribution
phase. For these jurisdictions, one could rely on the interpreters for the first phase and use ITC image analysis for
the second. Summarizing the ITC-based information for
each polygon and storing it as polygon attributes is a
straightforward process. This could mark the end of
“simple” stand description in terms of 10%, or even 25%,
species composition breakdown (Leckie and Gillis, 1995).
Knowing species content with more precision and more
importantly, seeing their spatial distribution within stands
would help with operational decisions on important, yet
minority species within stands (Gougeon et al., 2001b).
Two main options are being explored for fully automatic
stand delineation and attribution: independent and dependent stand delineation. Independent stand delineation could
be done, as in a two phase interpretation process, by first
delineating forest stands on images using a spectral-textural based segmentation process, followed by an ITC image analysis for stand attribution. Dependent stand delineation could be done after the ITC image analysis by regrouping the delineated and identified tree crowns into
stands based on density, crown closure and species composition (Gougeon et al., 1999a). Ideally, height (from LIDAR
data?) and other significant parameters would also be considered in delineating stands.
For a majority of foresters, the automatic forest inventory picture will only be complete when volume assessments
are properly conducted. At first, the present system of stratification based on ecological features and volume assessment from stratum representative plots followed by summaries on a regional basis could remain essentially unchanged. Later, individual tree based volume assessment
Figure 3. Isolation of defoliated trees from a 36 cm/pixel
multispectral image.
could be considered. Similarly, growth models and volume
projections through time could also be based on ITC analysis.
Damage Assessments
Fire is a big component of the boreal ecosystem. Although
it contributes to the forest rejuvenation, it is often considered in terms of losses. An important issue is the amount of
residual live trees in tree patches present within a burn. Similarly, the Canadian forest is continually being damaged by
insects and diseases, some damage leading to permanent
losses, others to a reduction in growth. Mapping the extent
of these damaging agents is an ongoing concern. Although,
significant improvements over sketch mapping can be
achieved simply by using satellite imagery such as from
Landsat, high resolution images may have an important role
to play.
For damaging agents covering wide areas, an ITC-based
analysis of sample areas could help to parametrize and alleviate some of the confusion factors encountered in lower
resolution image analysis. For example, assessment of
spruce budworm damage levels from Landsat images can
get biased by the presence of unaffected pine or deciduous
trees in an otherwise spruce or fir dominated stand. The
level of defoliation of individual trees can usually be classified to confirm assessments at stand level (Leckie et al.,
1988). Figure 3 illustrates the ease of detection of completely defoliated tree.
For very localized damaging agents, such as root rot,
where the emphasis is on the detection of small infestation
pockets, high resolution imagery and ITC-based assessments could become very important. Although even on an
ITC-basis, light damage symptoms may be difficult to detect, moderate and severe damage including needle loss are
usually detectable (Leckie et al., 1999b). In any case, the
pattern of increasingly stressed trees around a center, often
47
Figure 4a. Plantations of various ages and densities from
a 31 cm/pixel multispectral image.
Figure 4b Synergy between treetops and ITCs to
evaluate stocking in regenerating areas.
marked by a gap in the canopy from fallen trees, is a characteristic of this type of infestation that can be used to detect them if one has individual tree information.
ongoing, individual tree crown analysis of high resolution
images is still largely a research endeavor. There are numerous issues to be addressed, least of which is repeatability of results under less than ideal circumstances. If ITCbased inventories are to become a mainstay for forestry applications, there is a requirement to quantify delineation
and species identification accuracy (i.e., recognition accuracy) relative to various spatial resolutions and media (airborne scanners, digital frame camera, digitized aerial photos, satellite images, etc.).
Regeneration Assessments
The treetop (or local maxima) technique, although applicable to mature trees in poor resolution images, is usually the most appropriate to analyze the small trees found in
typical regenerating plantations. Young trees of various ages
(3-10 years old) and density levels (600-4000 stems/ha) have
been analyzed with the locally adaptive treetop technique
(Gougeon and Leckie, 1999b). The switching that took place
between the shade-based and the shadow-based treetop techniques eliminated the majority of ground induced false positive hits in the widely spaced plantations (7 year old, 4 m
spacing), while preserving full detection capability for the
very small, very dense areas (3 year old, 3876 stems/ha). In
addition, the ITC technique was also successfully applied
to the plantations with bigger trees to get their average crown
sizes. A non-vegetation mask obtained by preprocessing had
been used to remove ground areas in widely spaced plantations that had the potential to create false positive crowns.
Here, this was made possible by a senescing herbaceous
ground cover during a late fall image acquisition. The best
overall results were obtained by combining the resulting
bitmaps from the two approaches, leading to single pixels
designating small trees and full crowns for bigger trees (Figure 4). Such results lead to information on stem count, density, tree spacing, stocking, ingrowth and sometime, species or health status.
In addition, research is ongoing relative to:
•
•
•
•
•
•
•
•
preprocessing techniques, specially for trees in
open areas;
locally adaptive thresholds and locally adaptive
technique switching;
delineation rules improvement, especially relative
to crown over and under splitting;
spectral signature extension to other images;
estimations of volume and biomass from crown
areas, height and density, by species;
compensations for illumination and view angles,
topography, haze and seasonal effects;
interference of health factors on species recognition; and,
synergistic effects of spectral, textural, structural,
and contextual information.
CONCLUSION
This paper presented some of the image analysis concepts, methods and tools to extract individual tree crown
based information from high spatial resolution images and
reported on a few successful applications, some of the current limitations and ongoing research. It was speculated that
ITC-based forest inventories would be more precise, accu-
ONGOING RESEARCH
Although interesting results have been achieved with a
variety of forestry applications and technology transfer is
48
rate and timely. In addition, the unprecedented level of details would allow the extraction of a variety of additional
multiresource management information such as, snag locations, forest gap sizes and distribution, highly-valued tree locations, riparian zone mapping; permit detailed damage and
regeneration assessments; and generally encourage and facilitate precision forestry.
Leckie, D.G.; Teillet, P.M.; Fedosejevs, G.; Ostaff, D.P. 1988.
Reflectance characteristics of cumulative defoliation of
balsam fir. Canadian Journal of Forest Research
18(8):1008-1016.
Leckie, D.G.; Gillis, M.D. 1995. Forest inventory in Canada
with emphasis on map production. The Forestry
Chronicle 71(1): 74-88.
REFERENCES
Leckie D.G. and F.A. Gougeon. 1999a. An assessment of
both visual and automated tree counting and species identification with high spatial resolution multispectral imagery. Pages 141-152 in Hill, D.A. and Leckie, D.G., eds.
In Proceedings Int’l Forum on Automated Interpretation
of High Spatial Resolution Digital Imagery for Forestry.
February 10-12, 1998. Victoria, B.C., Canada. Natural
Resources Canada, Canadian Forest Service, Victoria,
B.C., Canada.
Gougeon, F.A.; Moore, T. 1989. Classification individuelle
des arbres à partir d’images à haute résolution spatiale.
Pages 185-196 in Bernier, M., et al., eds. Télédétection et
gestion des ressources Vol. VI - 6e congrès de L’association
québécoise de télédétection. Sherbrooke, Québec, Canada,
May 4-6, 1988.
Gougeon, F.A. 1995. A crown-following approach to the automatic delineation of individual tree crowns in high spatial resolution aerial images. Can. J. Rem. Sens. 21(3):274284.
Leckie, D.G.; Jay, C.; Paradine, D.; Sturrock, R. 1999b. Preliminary assessment of Phellinus weirii-infected (laminated root rot) trees with high resolution CASI imagery.
Pages 187-195 in D.A. Hill and D.G. Leckie, editors.
International forum: automated interpretation of high
spatial resolution digital imagery for forestry. In Proceedings of a symposium held at Victoria, British Columbia, February 10-12, 1998. Canadian Forest Service,
Pacific Forestry Centre, Victoria, BC
Gougeon, F.A., D. Leckie, I. Scott and D. Paradine. 1999a.
Individual tree crown species recognition: the Nahmint
study. Pages 209-223 in Hill, D.A. and Leckie, D.G., eds.
In Proceedings Int’l Forum on Automated Interpretation
of High Spatial Resolution Digital Imagery for Forestry.
February 10-12, 1998. Victoria, British Columbia, Canada.
Natural Resources Canada, Canadian Forest Service,
Victoria, Canada.
Leckie, D.G., M.D. Gillis, F. Gougeon, M. Lodin, J. Wakelin,
and X. Yuan. 1999c. Computer-assisted
photointerpretation aids to forest inventory mapping:
some possible approaches. Pages 335-343 in Hill, D.A.
and Leckie, D.G., eds. In Proceedings Int’l Forum on
Automated Interpretation of High Spatial Resolution
Digital Imagery for Forestry. February 10-12, 1998.
Victoria, British Columbia, Canada. Natural Resources
Canada, Canadian Forest Service, Victoria, B.C., Canada.
Gougeon, F.A. and D. Leckie. 1999b. Forest regeneration:
individual tree crown detection techniques for density and
stocking assessment. Pages 169-177 in Hill, D.A. and
Leckie, D.G., eds. In Proceedings Int’l Forum on Automated Interpretation of High Spatial Resolution Digital
Imagery for Forestry. February 10-12, 1998. Victoria, British Columbia, Canada. Natural Resources Canada, Canadian Forest Service, Victoria, Canada.
Leckie, D., C. Burnett, T. Nelson, C. Jay, N. Walsworth, F.
Gougeon, E. Cloney. 1999d. Forest parameter extraction
through computer-based analysis of high resolution imagery. Pages 205-213. In Proceedings Fourth Int. Airborne Rem. Sens. Conf. and Exh. / 21st Can. Symp. Rem.
Sens, Vol. II. Ottawa, June 21-24, 1999.
Gougeon, F.A.; St-Onge, B.A.; Wulder, M.;Leckie, D.G.
2001a. Synergy of Airborne Laser Altimetry and Digital
Videography for Individual Tree Crown Delineation. To
appear In Proceedings 23rd Can. Symp. Rem. Sens., Quebec City, Aug. 20-24, 2001.
Gougeon, F.A.; Labrecque, P.; Guérin ,, M.; Leckie, D.;
Dawson, A. 2001b. Détection du pin blanc dans
l’Outaouais à partir d’images satellitaires à haute
résolution IKONOS. To appear In Proceedings 23rd Can.
Symp. Rem. Sens., Quebec City, Aug. 20-24, 2001.
Wulder, M.; Niemann, K.O.; Goodenough, D.G. 2000. Local maximum filtering for the extraction of tree locations
and basal area from high spatial resolution imagery. Rem.
Sens. of Env. 73:103-114.
49
Chapter 5
Taking GIS/Remote Sensing into the Field
RICK R. KERNS
THOMAS E. BURK
MARVIN E. BAUER
Abstract—Field data recorders have long been used in forestry fieldwork. In most instances their application has focused on
improved collection of attribute data for subsequent integration into an office-based information or decision support system. As
the capabilities of portable computers improve in terms of storage, processing speed, and screen space/resolution it is perhaps
time to consider extending the application of such devices. This paper reports research from a NASA-sponsored project investigating the possibilities of carrying imagery products into the field with a GIS-capable device. We discuss use of off-the-shelf
technology for this purpose highlighting applications such as regeneration surveys and corner location/boundary line establishment. Results from the use of high-resolution satellite imagery, specifically IKONOS imagery, in such applications are presented.
INTRODUCTION
illustrative case studies of a system for aiding in forest management applications.
As pressures on the forest resource increase, the need for
more accurate, timely data and information for tactical and
strategic decision making likewise increases. This is obviously the case in situations where management intensity for
a single resource is increasing, though similar circumstances
arise where management for competing resource uses is the
focus.
Foresters have long had an interest in using field data
recorders to enhance their data collection efforts. Data recorders have been combined with a GPS receiver for the
collection of spatial attributes alongside non-spatial attribute
data.
Field data recorders have not been as broadly deployed
in forestry applications as first anticipated. There are several reasons for this. Since the hardware is fairly specialized, its cost is high relative to its capability. General software applications seemingly fall short of users needs; specialized applications are expensive and not broadly available. Data recorder hardware/software systems don’t always
integrate well with desktop systems and information systems specifically. These shortcomings must be addressed if
field data recorder applications are to be successful.
Traditionally the need for spatial data products in field
applications has been met with hardcopy aerial photography and maps. Digital imagery of various types and GIS
layers have the potential to replace hardcopy products, providing the field forester with up-to-date spatial information
that can be manipulated while in the field.
In the following sections we discuss the hardware, software, and imagery components that comprise systems for
taking geo-spatial data into the field and end by providing
Hardware Solutions
To investigate the potential utility of a field GIS system,
we will focus on current off-the-shelf products capable of
displaying digital imagery and GIS vector layers. The computer to be considered is known as a pen-tablet or pen-computer. The reasons for using this kind of computer in a field
environment are: light weight, reduced size, ability to sketch
or draw, and no need for a flat stable work surface.
Another important hardware component to the field GIS/
mapping system is the GPS receiver. A GPS gives us the
ability to get a fast and accurate geographical location in
the field. Knowing where the data are provides a more realistic and accurate representation of a study area.
Pen-computers today come equipped to meet almost any
application imaginable. Specifically, improvements in speed,
weight, sunlight-readability, ruggedness and data transfer
options mark a coming of age of GIS data collection (Jonas
2000). Nevertheless, selecting a pen-computer can still be
a confusing task. To help simplify things we will characterize three different categories of pen-computers: small, medium and large devices.
GPS receivers also come in many shapes and sizes today. Traditional stand-alone units range anywhere from
backpack systems to hand-held units that are so light they
will float if dropped in the water.
Since space does not allow a complete description of all
pen-computer configurations, and configurations are constantly updated, the reader is urged to check manufacturers
web sites for the most accurate and up-to-date information
51
Table 1. Pen-computer hardware characteristic comparison. (current as of May 1, 2001)
Device
Type
Screen
Size
Screen
Resolution
(dpi)
O perating
System
M emory
P rocessor
Data
Storage
W eight
P rice
Range
(Small)
PPC
3.77”
diagonal
320X240
PocketPC
Up to 64
RAM ,
Flash ROM
133 - 206
MHz
RAM ,CF
& PC card
slots
Less
than 10
oz.
$500$800
(Small)
H PC
6.5”
diagonal
640X240
W indows
CE
Up to
3 2M B
RAM ,
RO M
133 M H z
RAM ,CF
& PC card
slots
1 - 2 lbs.
~$1000
(M ed.)
H PC
7”-8.5”
diagonal
640X480
VG A
W indows
CE
Up to
3 2M B
RAM ,
RO M
129 – 168
MHz
RAM ,CF
& PC card
slots
2 - 3 lbs.
~$1000
(M ed.)
Win9X /2
K /N T
~ 8.5”
diagonal
SVGA &
XG A
W indows
9X/2K/N T
Up to
256M B
Up to
500M H z
Celeron
Hard Disk
D rive, CF
& PC card
slots
2.5 – 4
lbs.
$2000 $30 00
(Lg.)
Win9X /2
K /N T
10” – 1 3”
diagonal
SVGA &
XG A
W indows
9X/2K/N T
Up to
256M B
Up to
400M H z PIII
Hard Disk
D rive, CF
& PC card
slots
3–6
lbs.
$3000 $50 00
Corporation (Figure 1), and
Magellan GPS315 receiver. The iPAQ meets
our processor needs with a
206 MHz Intel StrongArm
processor. This is presently
the fastest processor available for the PocketPC.
Compact Flash or PC
Cards can be used for storing large or multiple image
files and GIS data layers.
For a field application the
ability to read a display in
direct sunlight is critical.
At this time the iPAQ is the
only PocketPC with a truly
sunlight-readable display.
One of the features of
this system we really wanted to exploit is the ability to capture GPS data directly into the field GIS software. In order
to capture data directly, the GPS receiver must output a
message using the NMEA or Trimble TSIP protocols. The
GPS message contains information such as satellite position and status as well as your location. For this project we
chose the lightweight and inexpensive Magellan GPS315.
With this GPS configuration there must be a cable connection between the two devices. Serial cables from both
Compaq and Magellan are readily available; the two cables
can be connected with a standard null-modem adapter and
gender changer. The Magellan GPS315 includes features
such as: track logging, route planning, and many different
navigation screens. These features can be used at any
(Appendix A). Key characteristics of pen-computing devices are listed in Table 1.
Small Pen-Computing Platform. This category includes
the HandheldPC (HPC) and new PocketPC (PPC) devices.
This platform is characterized mainly by their small screen
size and WindowsCE operating system. One advantage of
such a small computer is portability. The main disadvantage of these devices is the relatively small screen size. GIS
and remote sensing are very visual sciences and the small
display may be an issue for some users.
Medium Pen-Computing Platform. The medium size
computers are available with the Win9x/NT or WinCE operating system, which share a similar user interface. A big
advantage of these devices is that they are still lightweight
and portable yet have bigger displays, more memory, faster
processors and more resident storage for data than do small
pen-computers. This category makes a very powerful statement by maximizing capability while minimizing size and
weight.
Large Pen-Computing Platform. Full sized pen-computers have the ability to run virtually any software that a
desktop computer runs. This capability allows for seamless
integration with the desktop computer and software. This
category of pen-computer approaches the power and flexibility of laptop computers. The main disadvantage of large
pen-computers is size and weight. Some of these devices
can be very cumbersome and hard to use in the field.
Hardware Selected For This Project. As size and weight
is always a concern with field equipment the choice was
made to employ a small hand-held computer and small handheld GPS receiver in this project. Specifically, the hardware package we put together for the field system consists
of an iPAQ PocketPC computer, from Compaq Computer
Figure 1. iPAQ PocketPC.
52
time and don’t interfere with the output that is going to the
computer. There is also the possibility to use the GPS unit
alone to map features and then later down load and convert
the track log. Using a GPS unit with this setup is subject to
the same limitations of any GPS field application.
cussed above are much improved over those for “data recorders”. Such tools offer the opportunity to develop highly
customized GIS solutions where off-the-shelf solutions are
too limiting.
As hardware is constantly improving, so are software solutions. To obtain more detailed and up-to-date information
it is highly recommended you visit the developers’ web pages
(Appendix A).
Software Selected For This Project. Only a few GIS/
mapping software packages are ported to the PocketPC platform, and ESRI’s ArcPad is the only one that operates
seamlessly with ArcView desktop GIS. We chose ArcPad
as a very streamlined GIS/mapping software. ArcPad allows us to digitize points, lines or polygons on the screen
with a stylus (Figure 2) or directly from GPS input, while
displaying a background raster image or shapefile (Figure
3). As we are interested in exploring some high-resolution
imagery options, support for MrSID image compression was
a factor in choosing ArcPad.
Another important component of this software is the ability to create custom input forms for easily attributing features while in the field (Figure 4). The custom forms are
created in ArcView’s dialog designer on the desktop and
then exported to the PocketPC device. Custom input forms
help to ensure our data quality by predefining the format of
the input or by building them around existing data dictionaries. As shown in figure 4, the ability to enter data via a
“pick list” not only speeds the process, but also reduces error through standardization.
Software Solutions
As the range of hardware available is growing rapidly, so
is the software that makes field GIS possible. Technology
is moving rapidly in the world of PDA/hand-held/pen-based
field data collection (Graham 2000). Several mapping/GIS
software packages were researched in order to characterize
their potential functionality in a field GIS system. These
packages include ESRI’s ArcPad and ArcView, PenMap by
Condor Earth, PC GPS by Corvallis Microtechnologies
(CMT), iMap by Sokkia, and SOLO CE by Tripod Data
Systems (Table 2).
Some key issues we identified for choosing a software
solution are: ease of GPS integration, straightforward user
interface, extensibility, and desktop GIS integration.
The integration of GPS into mobile GIS software seems
to be a priority with both software developers and users. All
of the software that we researched, with the exception of
ArcView, has made this form of data input a primary feature. The lack of this feature in ArcView is probably due to
its primary use as desktop software.
A straightforward user interface is essential when designing a field system. With ArcPad, the user interface is very
simple and very similar to the ArcView desktop GIS software. In the spirit of simplification, ArcView has a highly
customizable user interface. ArcView’s interface can be
stripped down to show only the controls needed and custom, user defined, controls can be added. The remaining
software options address “ease-of-use” similarly with either
a user interface that is already very simple and straightforward or an interface that is easily customized.
Extensibility can be defined by the ability to customize
the capability of the software through scripting or programming languages. A common use of extensibility, in mobile
GIS software, is the integration of input devices beyond GPS.
Custom extensions either already exist or can be written for
all software packages mentioned using various scripting and
programming languages.
Desktop integration is a priority for the software developers as well as users, as it is hard to do anything with your
data if you can’t get the data out of your mobile device. The
issue becomes “how straightforward is the integration process?”. All software researched provides at least an export
function for the common vector file formats, as well as desktop software that facilitates the data transfer process.
ArcView and ArcPad both use the shapefile format directly,
eliminating the export step. Furthermore, the attribute table
of the ESRI shapefile is stored as a .dbf database file, allowing it to be imported easily into various database and spreadsheet software.
Software development tools for the hardware devices dis-
Figure 2. Stylus-based digitizing in ArcPad. Approximate twoacre stand has been digitized (solid line). Small dashed
rectangle identifies polygon extent.
53
Table 2. Pen-computer software characteristic comparison. (current as of May 1, 2001)
Features
Raster
Vector
ArcPad
5.0.1
PenMap
CMT s PC
GPS 3.6d2
iMap
Solo CE
ArcView
3.2
CADRG, Jpeg,
.bmp, MrSID
(8-bit only)
BMP, JPG or
ESRI Tiff
World Files,
import of
GeoTIFF
GeoTIFF,
DOQ, DRG,
TFW, JPG, and
BMP
GeoTIFF,
TIFF and
JPG image
files
GeoTIFF,
TIFF and
JPG image
files
Extensive
raster
support
Shapefile only
Shapefiles,
DXF files,
Shapefiles,
AutoCAD
DXF files,
ASCII
Shapefiles,
AutoCAD
DXF files,
ASCII, MIF
Shapefiles,
AutoCAD
DXF files,
ASCII, MIF
Extensive
vector
support
Streaming
pen input
with
Minnesota
Dept. of
Natural
Resources
extension
Yes
Pen
digitizing
Streaming pen
input
Streaming
pen input
Streaming pen
input
Yes
One point at
a time,
“sticky”
mode allows
rapid point
placement.
Vertex
editing
No, should be
available with
next major
release
Yes
Yes
No
Yes
GPS
Real-Time
mapping and
navigation
built in
Real-Time
mapping and
navigation
built in
Real-Time
mapping and
navigation built
in, as well as
differential
post processing
Real-Time
Real-Time
Custom
data entry
forms
Yes, via
ArcPad Tools
for ArcView
(incl. With
ArcPad)
Yes, via Form
Generator
(included)
Yes
No
Yes
Additional
features
Scaled down
for use on
Windows CE
and Pocket PC
devices.
Customizable
GUI,
Extensions
for more than
70 field
sensors are
available
Foresters
Toolkit ($385)
Compatible
with laser
range
finders,
waypoint
navigation,
customizable
interface
Compatible
with laser
range
finders,
waypoint
navigation
Highly
customizable
GUI
None
None
ArcView
Layouts
advanced
cartographic
tools
$1,000
$1,000
$1,200
Plotting
Output
None
Simple
plotting
Professional
plotting output
via Quick Plot
or Custom Plot
Cost
$500
$2,400
$1,995
54
Real-Time
with
GPSview
software
($500), or
extension
Yes, via
dialog
designer
extension
(incl. With
ArcView)
Aerial photographs record a lot of detail. Standard techniques yield photos that resolve ground objects slightly larger
than a foot in diameter at a scale of 1:24,000. Such imagery
is used to identify and measure features including stand
height and volume. Aerial photos can be viewed stereoscopically to interpret species, size class, and percent crown
closure while in the field. When combined with a GPS receiver, hard-copy aerial photos are aids for orienteering and
sample plot location (Caylor 2000).
Landsat and SPOT provide readily available, medium
spatial resolution (20 – 60 meters depending on the sensor)
digital satellite imagery in multiple bands. This type of imagery has been used to extract forestry information for standbased applications. Information, such as change in status or
conditions of individual stands is appropriate for tactical
management. Applications include assessing the correctness of stand boundaries, and assessing the accuracy of stand
labels. Landsat and SPOT are often combined with aerial
photography for field applications (Roller 2000).
New high-resolution digital imagery provides an opportunity to develop new products for use in the field. Whether
high-resolution aerial or satellite imagery, these data are already in digital format. The imagery can usually be purchased rectified or not. With multispectral digital imagery
it is easy to create either natural color or false color images.
It is also possible to create a high-resolution Normalized
Difference Vegetation Index (NDVI) or greenness map. This
type of a background image can be very useful for identifying and mapping damage or areas under stress.
Imagery Selected For This Project. The imagery chosen for this project is the IKONOS Carterra Geo product
from Space Imaging Inc. (Figure 5). The IKONOS data
were provided to us by NASA’s Science Data Purchase program. An area of approximately six square miles, containing the University of Minnesota’s Cloquet Forestry Center,
near Cloquet, Minnesota was acquired for this project.
IKONOS imagery is well suited to meet our requirements.
Visual interpretation is handled easily, as high spatial resolution is provided at approximately 1 meter per pixel. Spectral resolution is good with IKONOS imagery. IKONOS
has 11-bit radiometric resolution capable of displaying 2,048
levels of tonal variation, in five separate bands. Temporal
resolution is also good; we were able to obtain five image
dates in the year 2000, some as quickly as two weeks from
the time the order was placed. Although satellites are tied to
their orbits, IKONOS greatly increases its temporal resolution with the ability to aim or point the satellite (Green 2000).
The IKONOS Geo multispectral product was received as
five separate files of blue, green, red, near infrared and panchromatic data. These five bands required pre-processing
in order to make the data usable. The first step in the process was to bring the four separate color bands into a one
multi-band image. This was accomplished in ERDAS Imagine image processing software using the Layer Stack model.
The second step was to merge the panchromatic image,
which is one meter per pixel, with the multispectral image,
which is four meters per pixel. This process produces a
Figure 3. GPS-based digitizing in ArcPad. Area has been
digitized by traversing it with the GPS unit active (solid line).
Cross hair identifies current position from GPS. Small dashed
rectangle identifies polygon extent.
Figure 4. Screen shot of custom input form created in
ArcView’s dialog designer.
Imagery Solutions
Different types of imagery have been taken into the field,
traditionally in hard-copy format. These types of imagery
include aerial photographs for aid in navigation and feature
location, and Landsat or SPOT satellite imagery for ground
truthing a cover type classification project.
55
Figure 5. False color image (bands 4,2,1) of Cloquet Forestry Center, (A) April 25, 2000 and (B) August 10, 2000.
Figure 5A.
Figure 5B.
pattern of stocking on the area. Stocking pattern can be
ascertained by plotting the location of stocked and
nonstocked plots on the hard-copy map of the surveyed area
(Avery and Burkhart 1994). This method of determining
stocking pattern is both time consuming and can be inaccurate.
By taking GIS to the field we believe the regeneration
survey can be greatly improved in speed and accuracy.
Before entering the field the desktop GIS is used to prepare several aspects of the regeneration survey. Using background images, the boundary of the regenerating stand is
delineated as a polygon. Based on the shape and size of the
stand, the traverse for the survey is laid out as a line coverage and the plot centers as a point coverage (Figure 7). The
attributes collected at each sample plot are built into custom
forms for the mobile GIS. The attributes collected are: stocking (stocked or nonstocked), species, and size. Once the
locations of the plot centers are determined, coordinates (x,y)
are generated and loaded into the GPS unit. Once in the
field, the GPS unit is used for navigation through the stand
to the plot centers, and the pen-computer is used for collecting data at the plot. When a plot is reached, a 1/250-acre
plot (for example) is identified and centered on the data collector. In the GIS, the plot is opened for editing by simply
tapping on the point with the stylus. At this time the position of the point is updated by the GPS while the attributes
are being entered. While navigating with the GPS gets us
close to the pre-determined plot center, updating the position after we get there gives us the actual location of the plot
center.
With the use of an NDVI or greenness map we can identify and delineate holes in the regeneration using the desktop GIS before entering the field (Figure 8). Once in the
field these specific sites can be visited during the survey,
their actual condition verified and then mapped using the
GPS.
Corner Location & Boundary Line Establishment (line
running). The task of corner location and line running has
been identified as one that could benefit greatly from new
simulated one-meter or pan-sharpened image (Figure 6).
Image Compression. As with any high-resolution imagery IKONOS data sets can be quite large. In our application the resulting image had an approximate file size of 415
megabytes following the layer stack and resolution merge.
We chose to utilize the aggressive compression characteristics of MrSID software by LizardTech, Inc. At a 20:1 compression ratio, the reduced file size (approximately 20 megabytes) is easily portable to a handheld PocketPC device.
Specific Field Applications
The real drive behind implementing new technology is
the existence of important applications that can immediately
benefit from the new technology. We felt it important to
initially focus on straightforward and easy-to-implement
tasks. Two tasks were chosen because of their extensive
application in forest resource management practice. The
first task we decided to focus on was the regeneration survey. Regeneration survey work is pervasive and obviously
a very important data collection effort. Corner location/
boundary line establishment (line running) was chosen as a
second task; this task traditionally takes an excessive amount
of time and effort.
Regeneration Surveys. Surveys are undertaken early in
the life of a stand to determine the success of regeneration.
If regeneration is not adequate, then measures can be taken
to improve the situation. There is an important spatial component to regeneration surveys that is often ignored: specifically the location and extent of any failures. With the
new technology, holes in the regeneration can be mapped
easily and attributed as to the characteristics of the possible
cause.
The stocked-quadrat method of survey is common when
interest lies in regeneration. Traditionally the surveys are
carried out using hard copy aerial photos and maps only as a
reference to locate and traverse the stand. The resulting
stocked-quadrat tally provides an estimate of the percentage
of area occupied by trees; however, it does not reveal the
56
Figure 6. Resolution merge process with IKONOS imagery.
Figure 7. Stand boundary with traverse and sample plot centers.
4-meter resolution multispectral image.
Figure 8. NDVI greenness map, developed using pre-growingseason IKONOS imagery, with “holes” delineated (brighter
pixels = healthier vegetation).
1-meter resolution panchromatic image.
Simulated 1-meter “pan sharpened” multispectral image.
57
technology. Generally, prior to a management activity, corners are located and a line between the corners is run and
marked with paint. This line identifies the boundary of the
management unit or activity. This task is traditionally accomplished with aerial photos, map, hip-chain and compass,
and is well known to take an inordinate amount of time given
its simplicity.
The real difficulty of this task comes from locating the
corner markers. These markers are usually an iron stake,
sometimes a wood post, or a tag. Research of the survey
notes will get you in the neighborhood of the corner, but
finding the actual marker can be frustrating. Usually the
first corner located is near a road. Often a tree close to the
road is tagged. The tag commonly provides a bearing and
distance to the actual corner marker. In the worst case scenario there is no marking at the road and one is virtually
searching “for a needle in a haystack”.
Once the first marker is located and flagged with ribbon,
the second corner is located. The bearing and distance to
the second corner is derived from survey notes. Navigation
to the second marker proceeds, using the compass and hipchain, with ribbon flagging along the way. This corner is
now marked and considered an approximate location. It is
prudent and recommended that a different “surveyed” marker
is then found and navigation to our “approximate” second
corner location repeated. This brings up another problem
associated with this task. Often times the survey notes that
are used to locate the markers were written a long time ago.
In some cases compass declinations weren’t recorded or perhaps recorded in error. Over short distances this will not be
a problem, but if the compass is off one degree, a one hundred-foot error can result over the distance of a mile. Once
corners are satisfactorily located, the line is run as accurately
as possible, and marked with paint.
Before entering the field the desktop GIS and survey information is used to locate and map the corner locations.
More and more survey corners have actual coordinates associated with them these days. By running the survey information through COGO (coordinate geometry) in ArcInfo or
ArcView, the actual corner coordinates are estimated and put
into a coverage. The estimated coordinates are then entered
into the GPS receiver for navigation. The background imagery, roads and streams vector layers, along with the corner
location layer are then downloaded to the pen-computer.
Using the roads and streams layer for reference and the GPS
integrated with the mobile GIS software, the field personnel
navigate to within very close proximity of the first corner
marker. Once the first corner is found and marked, its location is updated with GPS in ArcPad. While updating the
location of the corner marker, attributes are added via a custom form to identify and describe the marker. A waypoint is
then set for the next corner location. The boundary is then
marked with ribbon while navigating from one corner to the
next. There may be some location error with the GPS receiver, but the error should be consistent and not grow over
distance. With the field GIS system it would be very easy to
provide interested parties with the GPS locations of the cor-
ners and boundary. With this kind of field GIS system the
time spent marking these should be reduced significantly,
perhaps made more accurate, and definitely recorded more
thoroughly.
CONCLUSION
The components for developing effective solutions for
taking geo-spatial data into the field are available now. A
number of options in terms of hardware, software, and data
products exist. The number and quality of options will increase. As the market for many components extends far
beyond forestry or even natural resources, price-performance
ratios will likewise improve. Field personnel should be made
aware of this potential and partnered with to develop solutions that aid in the conduct of the many forest management
activities that possess a significant spatial component.
CITATIONS
Arar, Y. 2001. Pen and PC: Sony’s Winning Combo. PC
World Magazine. 19(3):58.
Avery, T.E., and H.E. Burkhart. 1994. Inventories with
sample strips or plots. p. 198-216 in Forest Measurements: Fourth edition. McGraw-Hill, Inc., New York,
NY.
Caylor, J. 2000. Aerial Photography in the Next Decade.
Journal of Forestry. 98(6):17-19.
GATC, 1999. http://www.gatc.cz/gatc/us-pen-tablet.htm
Graham, L.A. 2000. Life in the Fast Lane. GEOWorld.
13(7):30–35.
Green, K. 2000. Selecting and Interpreting High-Resolution
Images. Journal of Forestry. 98(6):37-39.
Jonas, M. and B. Hillman. 2000. Take Your Computer to the
Field. GEOWorld. 13(11):32–34.
Roller, N. and K. Bergen. 2000. Integrating Data and Information for Effective Forest Management. Journal of Forestry. 98(6):61-63.
ACKNOWLEDGEMENTS
This work was conducted under NASA grant NAG1399002 to the University of Minnesota. Special thanks to
Mr. Timothy J. Mack for discussions related to specific applications in this paper.
58
APPENDIX A
Portable Computers
Amrel Rocky rugged laptop
Compaq iPAQ 3650 pocket PC
cassiopeia PDA
Dauphin Orasis
Fujitsu pen-computer
GeneSys & Ramline
GeoAstor GAC-PC wearable PC
Hitachi Handheld PC
HP Jornada 720 handheld PC
Husky rugged handheld PC
Intermec Norand & Pen*Key
Melard Sidearm rugged handheld PC
MicroSlate pen-computer
Tadpole pen-computer
ViA wearable computer
Walkabout Hammerhead pen-computer
Web Site
http://www.amrel.com/
http://www.compaq.com/
http://www.casio.com
http://www.dauphintech.com/
http://www.fujitsupc.com/
http://www.xploretech.com/index1.html
http://www.geoastor.ch/e/home_e.htm
http://www.hitachi.com
http://www.hp.com/
http://www.itronix.co.uk/
http://www.intermec.com/products/pen.htm
http://www.melard.com/
http://www.microslate.com/
http://www.tadpole.com/
http://www.via-pc.com/
http://www.walkabout-comp.com/
Portable GIS Software
ArcPad 5.0.1
ArcView 3.2
CMT
iMap
PenMap
Solo CE
Web Site
http://www.esri.com
http://www.esri.com
http://www.cmtinc.com
http://www.sokkia.com/Products/GPSGIS.htm
http://www.condorearth.com/penmap.html
http://www.penmetrics.com/solo/html
59
Chapter 6
RTI-Real-Time Inventory
A New Approach to an Old Problem
MARK M. MILLIGAN
Abstract—RTI is a totally different concept and approach to forest inventory and precision forestry. Developed by GeoTech
Systems in cooperation with Haglof, and Landmark Applied Technologies, it combines the power of Global Positioning Systems
(GPS), Geographic Information Systems (GIS), electronic field data collection instruments, and collection/processing software to
bring about a very powerful solution for field foresters.
The way this system works is through the use of a high-precision DGPS (Real-time corrected Differential GPS), with advanced
multi-path rejection, the forester is able to navigate throughout a user-defined, and easily generated cruise grid. Once on plot
center, the forester presses a button, then collects cruise data with the same device, or optionally, with a digital caliper and digital
hypsometer.
There have been attempts in the past to achieve the same goal of GPS-assisted navigation to plot center. These used very
expensive, and bulky solutions that consisted of separate GPS receivers, batteries, antennas and data recorder and/or field computer. On top of that, these systems could easily run $25-$35K.
Now, thanks to the availability of smaller, highly accurate DGPS that excels under canopy, Windows CE and Pocket PC
platform, cheaper hardware and better software, these systems are not only less bulky, but easier to use and much more affordable.
ernment agencies and consultants are greatly concerned with
the cost and errors attributed to both spatial (positional) and
information (sampling) errors. The past two decades of technological development has addressed the information side
of this equation quite well. Data recorders, digital calipers
and other field computers have enabled many organizations
to take advantage of collecting field data only once. This
has resulted in less transfer errors (to the desktop computer),
reduction of data entry errors (due to error checking features in many programs), and faster turnaround on data processing since data entry is already accomplished in the field.
These benefits have contributed significantly to the bottom
line for those organizations that have taken advantage of
this technology. Larger, more established entities, such as
industrial timber companies and most government agencies,
have reaped the rewards. The vast majority of foresters using this technology, that our company has surveyed, indicate that, if given the choice, they much prefer the handheld
data recorders to the tally book.
However, a significant number of forestry firms, mostly
small timber & land companies, consultants and procurement foresters, have not taken advantage of electronic data
recorders and cruise software. In our company’s experience, this has to do with either initial cost issues, a lack of
understanding technology, and/or the inability to “see” the
benefits. Since most of these firms lack the resources to
have IT personnel on hand, they oftentimes must fill that
INTRODUCTION
Overview
The forest industry has undergone many changes in the
last five to ten years. There have been many mergers, buyouts
and a tremendous change in land ownership. Coupled with
the current economic situation and the status of free trade,
this has resulted in the need for more accurate and faster
cruising methods than ever before. The main limitations involved with collecting and processing this information is time
and elimination of errors.
Though the data collected is simple in nature (species,
dbh, height, product, age, form class, etc.), there can literally
be thousands of plots on one job. Organizing, collecting,
and processing this data is very time-consuming and can be
quite complex. The primary way of accomplishing this task
is to draw cruise grids (i.e. 5 chains x 10 chains with each
line at 10 chains and each plot spaced 5 chains along the
grid) on a paper map. The forester/technician uses a compass and pacing methodology, while flagging the line, to get
to each designated plot center. Once on plot center, many,
but not all organizations (many still using paper tally cards),
are using DOS-based data recorders.
Sampling Error
It should be no surprise that many timber companies, gov61
role with existing personnel. This can be a significant cost
to a small company up-front, and training & support issues
are a foremost concern in this segment of the industry.
Figure 1. Integration of several cruising tools
used in Canada.
Spatial Error
The second part of the equation in more effective timber
cruising is the error associated with spatial accuracy. Since
cruisers use a compass and their own pacing, or a string
box, for navigating to plot center, significant time, effort
and errors can be produced.
With the advent of GPS, some cruisers are using this tool
to navigate to plot centers. The problems associated with
this approach has to do with the cost of a stand alone GPS
unit, bulkiness and equipment weight issues, accuracy of
the GPS, and the lack of coordination/integration between
the GPS and the electronic date recorder.
For these reasons, the overwhelming majority of organizations have elected not to utilize GPS in this manner. Although GPS technology has assisted many thousands of foresters in deriving accurate acres and producing quality maps,
the GPS-assisted forest inventory has been a difficult solution at best. Many forestry professionals that we have trained
over the years continue to ask when GPS and forest inventory will be truly integrated.
There have been several other attempts, but these solutions still either lack true integration that is needed by the
industry, costs are too prohibitive, or there are simply too
many pieces of hardware to trudge through the woods.
ASSESSING LIMITATIONS WITH
CURRENT TECHNOLOGY
Up to now, current technology has not provided an efficient, cost-effective way in which to integrate GPS and timber cruising in one package. There have been attempts, such
as with CMT’s MC-GPS. Two Dog forest inventory software, written for both ROS and DOS operating systems,
had a quick button on their earlier version of software to
quickly exit the program and go straight to the GPS field
module, for the MC-GPS. This was not a true solution since
ROS and DOS systems are only single-tasking machines.
There is no way to run two programs simultaneously, therefore, the user has to re-initiate the program each time a switch
is made.
Another effort has been made by one of the Canadian
provinces. Figure 1 outlines their approach using a Trimble
ProX model GPS, a Husky PX5 rugged Windows computer,
a Criterion laser and a Mantax caliper. Although these
accessories can provide a powerful navigation and cruising
solution, this is just not a practical option for the industry.
Aside from the obvious weight and bulk of this field
package, the costs can be quite prohibitive. A system such
as this could easily run close to $30K.
Two Dog has made an improvement on their new version of software. There is an available option whereby the
software will accept NMEA messages to appropriate data
fields via the serial port. This will probably help some users, but is still a far cry from true integration of the two
technologies. In addition, cost and weight/bulk issues are
still a concern here.
ANALYSIS OF POSSIBLE SOLUTIONS
Win CE and Pocket PC
There have recently been two technological
developments that have made the integration of GPS and
forest inventory, a possibility. Windows CE devices and
Pocket PC’s (Figure 2) have developed to the point where
the operating systems, processors and memory availability
afford the end-user a wide array of options for running the
latest software applications. Since it is based upon similar
technology to desktop windows, it is a multi-tasking
platform. Also, there are developer’s tools that allow
integration of programs.
Costs for these systems are very competitive, and indeed,
usually less expensive than traditional data recorders. Price
primarily depends upon the degree of ruggedness and, to a
lesser degree, factors such as processor speed and memory.
Prices for non-rugged devices such as the Compaq IPAQ
and Aero run approximately $500 and $300, respectively.
Prices for more rugged devices such as the Symbol PPT
2700 and the TDS Ranger range between $1,500 to $2,500.
Though this is much more than for non-ruggedized units,
the performance and utility is usually well worth it. Even at
that level, the prices are very competitive with those of traditional data recorders, and the CE’s provide a much more
robust platform and set of capabilities.
62
Figure 2. Three of the most prominent Windows CE units for in-field use.
Symbol PPT 2700
TDS Ranger
Advances in GPS – DGPS and WAAS
Compaq IPAQ
Figure 4. The WAAS Network.
In addition, there have very recently been significant advances in GPS technology. The improvements have manifested itself in a similar manner to that of other computer
hardware. GPS has now become smaller, more powerful
(more accurate), easier to use and less expensive. In addition, receiving real-time differenFigure 3. CMT RT-GPS. tially corrected (DGPS) positions
are both more widely available
and less costly. These are certainly
positive trends for the success of
the subject integration.
One such unit is the RT-GPS
by CMT (Figure 3). Real-time
(DGPS from any 0183 RTCM
source) GPS is accomplished
through an integrated GPS/Beacon Antenna and Receiver all in
one rugged unit. It tracks very
well under canopy and produces
1-3 meter accuracy. Price range
starts under $1,000.
A very exciting development is
the expansion and early production of units using the Wide Area
Augmentation System (WAAS).
WAAS was developed and administered by the FAA primarily for the purposes of much more accurate and reliable
GPS-assisted landing, takeoff and navigation. The WAAS
(Figure 4) works through a system of ground-based reference stations working in conjunction with communication
satellites that transfer DGPS messages via the L1 band (the
same signal used by GPS satellites).
It is expected that WAAS integrated antennas/receivers,
even smaller than that of the RT-GPS, are to be available
soon at similar prices. The advantage here is that corrections are easier to maintain (very wide availability in North
America and part of Canada only), and achieving sub-meter
accuracy is an inexpensive reality. According to some of
the engineers, because of the design of the WAAS, even
uncorrected data is better than that on non-WAAS receivers.
T-Cruise WIN CE Software
The only known cruising software developed for Windows CE and Pocket PC is T-Cruise, distributed by Haglof,
Inc. In addition, versions exist for the Palm OS and Haglof’s
Mantax computer caliper.
Using the T-Cruise for Win CE software, data collection
is simple, yet flexible and efficient. Default species codes
(numbers or characters) may be used as well as default
counts. Products may be auto-assigned (based on DBH),
and multiple heights may be taken even on the same tree.
Any cruise method is supported and the data output can be
exported to other programs for further processing. There
are also other fields for stand, site index, off plot and reproduction data collection.
Data input is in a spreadsheet-like format (Figure 5).
Any individual column may be turned on or off and data
entry is accomplished through either a stylus or keypad
(Ranger).
63
version, 6.0, will not allow for either “nested” points (point
features taken while collecting either a line or area feature),
nor offset features (features mapped a distance and direction from where the user is located). The lack of these capabilities serves as an indication that professional-level GPS
data collection is still some time to come with ArcPad.
Tripod Data Systems (TDS) has been developing its
SOLO CE GPS software for years. Their focus is on providing advanced GPS data collection capabilities to the GIS
market. There are several features that make this software
attractive. Among them being the ability to map several
different features at once, interfacing with lasers, five-level
data structure (as opposed to the usual four), and easy custom form creation. Although MrSid files currently cannot
be read due to an agreement between LizardTech and ESRI,
TDS has its own file compression format that is quite impressive.
Figure 5. T-Cruise WIN CE’s data input interface.
RESULTS
In addition, the office version of T-Cruise is quite easy to
set up, yet very powerful. Templates can be set up in minutes and the parameters quickly downloaded to the handheld.
Instead of relying upon only tables for volume generation, TCruise uses profile functions, which are very
accurate for each species, and based upon equations. Therefore, giving the program a full array of power in order to
merchandize products. Top diameters can very quickly be
changed as well as DBH limits for each product.
Multi-product and grading functions have been added, as
well as a custom report generation module. The program is
based on C++, so calculations are practically instantaneous,
and it is much more stable than other available cruise programs. Many enhancements are being made monthly as this
program is quickly gaining popularity in the marketplace.
By analyzing the capabilities of the preceding hardware
and software products to meet the demand for an integrated
GPS/Forest Inventory solution, the major factors under consideration are the following: 1) Weight and size of hardware; 2) Cost; 3) Ease of use; 4) Flexibility of the system.
The products under consideration are: 1) Which Windows
CE or Pocket PC device to use; 2) Which GPS to incorporate; 3) Which GPS software to use. Finally, a determination has to be made how to package all of the components
together. For the forest inventory solution, the only real
choice for this software platform is T-Cruise.
Weight and Size
1)
Windows CE/Pocket PC GPS Data
Collection Programs
There are several of these programs available. Among
them, are ESRI’s ArcPad, TDS’s SOLO CE and Fieldworker.
For the purposes of this paper, only the first two have undergone extensive review. Both of the companies carrying these
products are also promising to soon have available a set of
developer’s tools for customizing interfaces and increasing
utility. However, as of this writing (May of 2001), no beta
versions have yet to be released.
ESRI is by far the widest known name in GIS software.
That is why their ArcPad software for Windows CE and
Pocket PC is worth a look. The current version, 5.0.1, has
several unique features such as ArcIMS capability and MrSID
image support. However, it seems the focus of ESRI in developing ArcPad is to allow for easier GIS spatial and database maintenance, and to allow developers to customize the
interface. This is a good step, and will benefit users in the
long run. However, GPS support and data collection appears to be a secondary goal. This is evidenced by the fact
that they have indicated even the new, soon-to-be-released
2)
3)
Windows CE or Pocket PC device - This issue is
being addressed by utilizing any of the available
Windows CE and Pocket PC’s appropriate for field
use. The choice will be one of personal prefer
ence, available financial resources and demands of
the work performed. The TDS Ranger has the edge
in ruggedness, while the Compaq IPAQ gets high
marks for its quality color screen.
Which GPS to incorporate - The RT-GPS gets
the nod on price be a slim margin, but the new
WAAS receivers will be hard to beat for their size
and weight.
Which GPS software – Not applicable.
Cost
1)
2)
3)
64
Windows CE or Pocket PC device - The Compaq
Aero and IPAQ are the lowest-cost units, while the
Symbol and Ranger are the next highest, in order.
Which GPS to incorporate – See 2 above.
Which GPS software – Since it is so new, ArcPad
is less expensive than TDS SOLO for now, how
ever, this is expected to change in the future.
Ease of Use
1)
2)
3)
Figure 6. RTI System with Compaq Aero and RT-GPS.
Windows CE or Pocket PC device – For the professional field user, the Ranger edges out the pack
since it has single-hand operation capability and data
entry through either the touch screen, alphanumeric
keypad, and through arrows and tabs on the key
board.
Which GPS to incorporate – Either GPS solution
is similar in this regard.
Which GPS software – A lot of time has been spent
with both of these programs. We have found that
SOLO had a faster learning curve, but that is at least
partially due to the fact that this software has been
available for some time when compared to ArcPad.
We suspect, that over time, the two programs will
be at about the same level of ease.
numbering plots. Programming is currently being performed
so that the grid generation is performed on the PDA. Stratified plot sampling is currently available, and soon random
plot generation will be offered.
Once the grid is generated, the program creates a plot
number, but will also accept a plot ID generated by another
source, and sends the data to T-Cruise. A tolerance distance
is also selected by the forester to indicate when he/she is
close to plot center. Once on plot center, the forester
switches to the T-Cruise WIN CE interface (usually via a
hot button), and collects sample data. When finished, SOLO
will send the actual LAT/LON over to T-Cruise, so that
efforts can be made to ensure accuracy of the spatial integrity
of the cruise. The symbol appears differently for those plots
that have been visited (Figure 7).
An audit function is being added to both the SOLO and
T-Cruise software. This will allow auditors to either hand
pick, or state a percentage of plots to be visited. The program will generate a randomly selected group of plots that
the auditors may visit. Once on plot center, the old plot data
may be reviewed for verification. The specifications of the
audit function are not complete at this time.
Flexibility of the System
1)
2)
3)
Windows CE or Pocket PC device – The Ranger
gets high marks here because of the I/O ports including a DB-9 serial port, IrDA port, 26-pin
multiport with serial I/O, 10-BaseT Ethernet and
audio in/out. Also because of its flexible data in
put options. Most other similar devices lack both
of the above serial ports and the ethernet capability.
Which GPS to incorporate – Flexibility in this
case would refer to the system’s ability to operate
accurately. The WAAS would get the nod, but in
the U.S. and parts of Canada only, where the WAAS
network is functional. The RT would get the go
ahead in areas where RTCM 0183 messages are
readily available.
Which GPS software – There is no doubt, as of
this writing and for at least the next 18 months,
that the SOLO CE gets the edge in this regard. Its
advanced GPS functionality and continuous soft
ware enhancements serve to meet the needs of the
forestry professional best.
Figure 7. Cruise grid with plots visited (blue square) and not
visited (grey dots).
The Real-Time Inventory (RTI) Solution
This solution consists of combining the above-recommended components. They are T-Cruise for Win CE; either
the IPAQ, Symbol or Ranger handhelds; either the RT-GPS
or the new WAAS capable receivers (or any other appropriate NMEA-compatible GPS) and the SOLO CE software.
In addition, a custom cruise vest and collapsible pole has
been developed to house the components. This design was
made to keep all wires within the vest, and keep bulk and
weight to an absolute minimum (See Figure 6).
Cruise grids are currently generated in ArcView via scripts
already widely available for that purpose. Forestry Tools for
ArcView by FORS is one such package, but there are others
that can simply be modified, especially for the purposes of
65
Future Enhancements
CONCLUSION
There are a number of users who prefer to collect cruise
data with the Haglof Mantax Caliper. In these cases, the
data can be stored either on the caliper, or in addition, backed
up via radio transmission to the PDA. This is being made
possible through a new product yet to be released by Haglof.
Similarly, we are testing the feasibility of sending tree height
data to the PDA via infrared from the Vertex Hypsometer.
There will no doubt be suggestions for improving this
product, and giving the end-user more flexibility and greater
functionality.
The RTI system holds a lot of promise for the forest industry. This is finally a product that gives true integration of
GPS and forest inventory for the field forester. The system
may be used for either GPS data collection, forest inventory,
or both, simultaneously.
The integration of GPS into cruising and forest inventory
gives the GIS manager a better solution as well. Point data
is easily integrated into the GIS since the data link is done in
the field. A point-based GIS inventory database is superior
to that of the polygon base, since changes in stand structure
and polygon boundaries are now quickly and accurately reflected in standing inventory numbers. Guesswork is now
replaced with real, and accurate data to reflect the changes
in forest structure and volumes as the GIS is updated.
66
Chapter 7
Use of Airborne Laser Terrain Mapping System for
Forest Inventory in Siberia
IGOR DANILIN
EVGENY MEDVEDEV
T. SWEDA
Abstract—The use of laser scanning method provides a number of principally new possibilities on remote sensing of forest
vegetation. The high productivity of laser sensing survey (up to 30 thousand original measurements per second) combining with
the spatial resolution and accuracy of tens centimeters allow making the effective algorithms of morphology analyses, ensuring
automatic extraction of many important information characteristics of forest cover. The analysis of stand structure integrated
with GPS data, digital aerial video and highly accurate (10-15 cm of actual linear resolution) photographic images makes it
possible to highly reliable interpretation of different types and layers of forest vegetation separating it by a tree species, density
and the other parameters. The consequent processing of a laser profiling data using developed in Russia original ALTEX software, integral calculations, the Fourier and mean free path analysis makes possible to get such important and precise information on vegetation as a timber stock, forest type, NDVI at a direct way or by mediate – on correlation with tree crown diameter,
density, crown vertical extent and tree stand height. The regression method provides high accuracy assessment of a stand
biomass when processing a laser profiling data. The methodology aspects of airborne laser sensing method use for forest survey
are considered in the article. The description of the morphology algorithms is shown. The results of a practical application of
airborne laser sensing by ALTM-1020 machine to forest inventory in Central Siberia are discussed in the paper.
During last years with appearance and availability at civil
branches of economy of satellite positioning and navigation
systems (US GPS and Russian GLONASS), satellite and
aerial laser scanning and digital video- and photography, new
possibilities appears for remote sensing of terrestrial ecosystems and forest cover with highly accurate measuring
relief and heights of ground objects at about ±10-15 cm and
positioning of their space coordinates at about ±15-20 cm
and higher accuracy [Ritchie et al. 1993; Sweda et al. 1998;
Danilin et al. 1999, 2000; Medvedev 2000].
INTRODUCTION
Laser terrain mapping and survey is important constituent part of the newest methods and technologies of
geoinformatics and digital photogrammetry and is widely
adopted at solving many tasks of ecological monitoring and
forest mapping and inventory. Laser forest terrain mapping
and survey can be done independently and in complex with
space and aerial digital video- and photography combined
with ground sampling as well.
In North America laser mapping methods using satellite
and aircraft platforms was developed and widely practiced
at geodesy, cartography and forest inventory and survey
[Aldred et al. 1985; Krabill et al.1987; Chappelle et al.1989;
Kalshoven et al. 1990; Ritchie et al.1993; Weltz et al.1994;
Blair et al.1996; Lefsky 1997; Ackermann 1999; Means et
al. 1999, 2000; Magnussen et al. 2000].
In Russia (at that time Soviet Union) laser methods, directed toward forest resources’ survey had been developed
before onboard aircraft and satellite lasers appear. There are
mainly investigations which had been done at Leningrad
Forestry Research Institute (LenNIILH) by V.I. Solodukhin
with coauthors, who demonstrated high efficiency of forest
laser profiling combined with traditional (classic) aerial photography for timber cruising purpose [Solodukhin et al. 1977,
1985; Stolyarov et al. 1987].
METHODS AND RESEARCH AREA
Aerial survey is made from Mil-8 helicopter by laser scanning set ALTM 1020 by Optech Inc., Canada. Video and
photo scenes are recorded simultaneously by Sony DCRPC110 megapixel mini DV camcorder with Carl Zeiss VarioSonar optics and KODAK DSC-EOC 1c digital camera with
3000´2000 pixel resolution CCD which provides 10 cm high
on ground linear resolution from 300 m of flight altitude at
about 300 m of scanning path width [Opten’s…2001, Danilin
et al. 2000].
The width of the patch covered in a single pass of aircraft
depends on the scan angle of the laser system and the aircraft flight height. Typically operating specifications are at
flying speeds of 200 to 250 kilometers per hour (55-70 meters
67
Figure 1. General scheme of airborne laser terrain mapping system (Laser, 2000).
per second), flying heights of 300 to 3,000 meters, scan
angles from 0 up to 20 degrees, and pulse rates of 2,000 to
25,000 pulses per second. These parameters can be selected
to yield a measurement point every few meters, with a footprint of 10 to 15 centimeters, providing enough information
to create a digital terrain (DTM) and forest vegetation model
adequate for many forestry and forest engineering applications, including the design of forestry operations, projecting and alignment of forest roads, the determination of timber stock and volumes of ground works, and the design of
harvesting schemes and structures.
The position of the aircraft at the time of each measurement is determined by phase difference kinematic Global
Positioning System (GPS). Airborne and ground based GPS
receivers Ashtech Z-12, Ashtech Z-Field Surveyor, Ashtech
Z-Surveyor are used for the laser airborne survey. Rotational
positions of the beam director are combined with aircraft
roll, pitch and heading values determined with an inertial
navigation system, and the range measurements to obtain
vectors from the aircraft to the ground points. When these
vectors are added to the aircraft locations they yield accurate coordinates of points on the surface of the terrain. Uncertainty of one-time measurement of geographical coordinates using ALTM-1020 machine and Ashtech GPS does
not exceed 0.1% of flight height (Basic...2001).
General scheme of airborne laser terrain mapping system
had been used for inventory of forest lands is shown at Figure 1.
Laser sensing data and digital video images and photographs are exmining for quality and preprocessing at once
on board of aircraft and later on ground using special software which allow to get geometric parameters and highly
accurate coordinates of separate trees and well readable
morphostructural characteristics of forest canopy along the
flight course.
All images and database are presented at three dimensional (3D) view as customers may work with digital terrain
model (DTM) and video- and photographs at more comfortable regime.
At 2000 general methodology of forest cover mapping
and stand structure interpreting by ALTM-1020 machine was
studied and developed. The research had been placed along
200 km flight and sampling transect at Turukhansk region
of Krasnoyarsk krai, Central Siberia, within Bakhta river
basin (right tributary of Yenisei river) (63-64°N, 91-92°E)
at subzone of Siberian middle taiga dominated by larch
(Larix sibirica) - spruce (Picea obovata) – Siberian pine
(Pinus sibirica) forests in some sites mixed with birch
(Betula pendula) and sphagnum bogs and wetlands.
For ground verification of aerial survey data 35 sample
plots were established along the transect representing all
dominating forest types and site environments. At the sample
plots sample trees were selected, measured and cut down by
2 cm DBH and 1 m height range for stem analysis and stand
phytomass assessment.
RESULTS AND DISCUSSION
Interactive processing of laser and digital photo data is
implemented by original Altex Obser ver software
68
Figure 2. Visualization and complex geometric measurement of forest stand laser location data at Altex Observer working
window.
bitrary selection of a lighting source.
[Altex…1999] which is intended for use on IBM-compatible computers and provides the following capabilities:
· Tuning export of data to any software for their specialized post processing such as AutoCAD, PLSCADD, POLE-CAD, ArcView, ArcInfo, MapInfo,
Erdas Imagine (Figure 2).
· Visualization of primary data of laser location survey
and results of their topological processing in user-defined scale and aspect.
As a result of processing of the laser location data we
get primary ground profile, which consist of vegetation
and topographic (ground) surface profiles (Figure 3à). Topographic surface is interpolated consequently by equalization and joining (unification) of points, where laser beam
had reached the ground penetrating through tree crowns/
leaves (Figure 3b). Eliminating topographic profile from
the primary one we get forest vegetation/tree stand profile
(Figure 3c).
The analysis of forest vegetation and tree stand structure integrated with aerial digital photographic and video
data allow to accurate interpretation of different types and
layers of forest vegetation dividing it by tree species, density and other parameters (Figure 4).
Subsequent processing of the laser profiling data by
means of integral calculations, Fourier and mean free path
analysis make it possible to get such important and precise information on vegetation as timber stock, forest type,
NDVI at a direct way or by mediate – on correlation with
tree crown diameter, density, crown vertical extent and tree
stand height (Figure5). The regression method provides
high accuracy of stand biomass interpretation when processing a laser profiling data [Usol’tcsev 1998].
· Data separation into true terrain geometry, vegetation,
manmade objects and structures with a possibility of
their individual visualization.
· Visualization of digital aerial photos of the route draped
on laser data.
· Displaying of digital or raster topographical map and
marking location of the surveyed object on it.
· Defining of space and altitude geodetic coordinates of
any objects and areas.
· Available data sets (copy of screen)
· Creation and visualization of profiles and sections (putting of corridors and different projections).
· Computer-aided carrying-out of complex geometric
measuring: calculation of distances between trees, between crowns and ground, between the trees and other
objects marked by operator etc.
· Conducting of information-searching operations and
data output with forming listing of critical vegetation
zones (burnings, cuttings, silkworm forest damaged
areas and others).
· Synthesizing of halftone shaded scene images with ar69
Figure 3. Main stages of processing of laser location data of forest vegetation profiles.
Figure 4. Integrated profile of tree stand: mixed mature (MM), young coniferous (YC) and secondary deciduous (SD) stands.
70
Figure 5. The method of laser profiling data calculation and analysis.
The laser location survey method combined with highresolution digital photography as well as on-ground and airborne GPS support allow remotely, operatively and with high
accuracy to get information on forest cover condition and
environment with a basic scope of data on surveyed object
which may be a foundation for various thematic GIS compilations of different complexity. Such basic scope of data includes:
axis orientation angles) along with DTM created by laser locator allow all photogrammetric measurements on
such photos, as well as automatic creation of digital
orthophotoplans and photomaps on their basis (Figure
6).
The scope of collected data may be expanded in number
of cases. Thus, the parallel IR survey performance appears
necessary quite often. Classic examples are detection of forest fires sites’ and measuring forest floor temperature at line
survey, or control of thermal status of burning materials like
timber residuals and wastes. Another practically important
addiction may be a multispectral scanner, which data is extremely useful in number of forest health and ecological
applications. Utilizing of all such thematic sensors naturally
fits the concept being implemented – availability of GPS
facilities on board as an absolute time source and precise
information on aircraft position during survey allows to only
synchronize data of all such sensors with unified onboard
time, after which during on-ground processing to mathematically apply aircraft evolution and perform geometric correction of thematic data. Practical implementation of such approach does not presume significant expenses as most of
state-of-the-art airborne remote sensors produce signals that
can be directly used for such synchronization, and necessary universal correction software has been developed.
Thus, the basic scope of data can be, if necessary, upgraded with one or several GIS thematic layers connected to
the unified topographic base, without increase of aerial
· Accurate detailed description of terrain relief and cre-
ation of regular Digital Terrain Model (DTM) where
relief is represented in its primary form not affected by
forest vegetation.
· Precise topologic models of natural and man-made facilities being objects of survey. Thus computer-aided
contouring of forest management compartments, cuttings, burnings, roads, waters etc. (in absolute geodetic
coordinates) and elevation thereof, coordinates of trees
and of the compartments’ boundaries, intersections
identified, relief cross section along route axis compiled, other significant objects allocated within a corridor of a required size along the survey route.
· Georeferenced digital aerial photos representing natural True Color image of survey object, with terrain resolution of 8-12 cm per pixel. Mandatory fixation of aerial
photos exterior orientation parameters during survey
(spatial coordinates of principal point, camera optical
71
Figure 6. The final result of data processing in form of photomap. Raster topography map, laser image and orthorectified photo along with
main contours and other semantic data are presented. The data was obtained with laser locator Optech ALTM-1020 and digital photo
camera KODAK EOS DSC 1a (Laser… 2000).
location also shows the advantage of higher degree productivity for large scale topography survey of forested
areas at about 100-150 êm, and on fixed-route flight
survey about 500-600 km per day.
survey works, ensuring same strict methodology and metrology requirements.
CONCLUSION
· Besides, for number of cases the laser location appears
the only possible source of information on forested
lands.
Principal advantage of presented technology versus classic approach is explained by the following:
· Only digital methods are used on all stages of collec-
· True relief measurements (ground surface) without significant loss of accuracy is possible with laser location
methods for open forest areas and even under tree canopies.
tion, archiving and processing data, both on the aircraft
and during ground based processing. As a result data
processing starts in the landing point immediately after
flight day.
· The work areas without visual texture is possible to
survey like forest openings, sands, fully snowed areas
etc.
· One of major factors is a 3-D nature of obtained data.
Application of laser locator allows to, firstly, immediately obtain 3-D image of terrain and all ground objects
and perform geometric measurements on them, and, secondly, with no technological effort achieve any image
detail resolution by choosing appropriate flight and survey regimes: altitude and speed, as well as width of
swath coverage.
· Survey of location and shape of objects of a complex
structure mostly man-made, for example forest plantations, power line towers and lines, buildings and facilities, etc.
· The state-of-the-art aerial survey methods provide almost complete exclusion of ground geodetic works from
technology cycle, as the data obtained principally at
WGS-84 no need for plan/elevation on-ground support.
· Thus, while ensuring practically unachievable by classic aerial topographic methods accuracy of DTM, laser
72
Figure 7. Installation of ALTM-1020 machine equipment to helicopter Mil-8 (Laser…2000)
forest surveys // Inf. Report PI-X-51. Technical Inf. and
Dist. Center. Petawawa National Forestry Inst. Chalk
River. Ontario, 1985. 62 p.
The only exclusion is positioning and operating base
GPS stations. At survey of extended objects, to ensure
absolute geodetic accuracy of laser location and photo
data of 15–20 cm such stations should be located with
100–150 km interval of each other. If accuracy requirements are not that strong this interval may be increased.
Operation of GPS base stations does not require any
special geodetic knowledge of operator and basically is
limited to its locating in indicated place and switching
power supply on (Laser…, 2000).
Altex Observer User Manual, 1999. / E. Medvedev and K.
Pestov. Version 3.1. ALTEX OBSERVER Opten Limited Copyright, Moscow, Russia.
Basic specifications of ALTM 1020 airborne module, 2001.
http://www.opten.ru/eng/altm/index.html Opten Limited
Copyright, Moscow, Russia.
The mostly principal and critical point is declining of the
method’s labor costs comparably to classic methods it is
cheaper at more than two times in general (Danilin et al.
2000). Aerial laser survey equipment may be installed within
one day to any light aircraft like Antonov-2 or helicopter
Mil-8 with permanent cargo/survey hatch. This makes possible to use aircrafts of local aviation companies and enterprises at any region, eliminating high expenses of long distance driving of aircraft to a survey site (Fgure 7).
Blair, J.B., and Coyle, D.B., 1996. Vegetation and topography mapping with an airborne laser altimeter using a highefficiency laser and scannable field-of-view telescope.
In: Proceedings Of the Second Int. Airborne Remote
Sensing Conf. And Exhibition, Vol. II., Environmental
Research Institute of Michigan, Ann Arbor, Michigan,
2:403-407.
Chappelle E.W.,
Williams D.L.,
Nelson R.F.,
McMurtrey J.E. Lasers may help in remote assessment
of vegetation // Laser Focus World. 1989. ¹ 6. P. 123126.
LITERATURE CITED
Ackermann, F., 1999. Airborne laser scanning – present status and future expectations. ISPRS J Photogram & Remote Sens., 54(2-3): 64-67.
Danilin, I., and E. Medvedev, 2000. Investigation of forest
cover structure by method of laser aerial surveying
(Izuchenie struktury lesnogo pokrova metodom lazernoi
Aldred A.H., Bonnor G.M. Application of airborn lasers to
73
aeros’emki). In: Lesnaya Taksaciya I Lesoustroistvo (Forest
Inventory and Forest Planning), Siberian State Technological University, Krasnoyarsk, Russia: 153-162 (in
Russian).
Sensing, Vol. XXXII Part 6W8/1: Proceedings of the ISPRS
Workshop International Cooperation and Technology
Transfer, Ljubljana, Slovenia, Feb., 2-5, 2000 http://
www.opten.ru/eng/gis/index.html: 26 pp.
Danilin, I., and Sweda, T., 1999. Laser profiling for studying changes of forest vegetation. In: Methods for Assessment of Forest Ecosystems’ State and Stability, Proceedings of Int. Workshop, Aug. 8-13, 1999, Krasnoyarsk,
Russia. V.N. Sukachev Institute of Forest, Siberian
Branch of the Russian Academy of Sciences: 51-52.
Opten’s Home Page, 2001. http://www.opten.ru/eng/home/
html: Opten Limited Copyright, Moscow, Russia.
Ritchie J.C., Evans D.L., Jacobs D. et al. Airborn laser measurements of forest and range canopies // Appl. of advanced inf. technol.: effective manag. of natural resources.
In: Proceedings of the Conf. 18-19 June 1993, Spokane,
Washington. 1993. P. 428-435.
Kalshoven J.E., Dabney Ph.W. Airborne laser polarimetry
measurements during the forest ecosystems dynamics
experiment // Proceedings. of the IGARS’90 Symp., University of Maryland, College Park, MD, May 20-24. 1990.
V. 1. P. 897-899.
Solodukhin, V.I., A.Ya. Zukov, I.N. Mazugin, 1977. Possibilities of laser aerial photography for forest profiling
(Vozmoznosti lazernoi aerofotos’emki profilei lesa),
Lesnoe Khozyaistvo (Forest Management), No. 10: 5358 (in Russian).
Krabill W.B., Martin C.F. Aircraft positioning using global
positioning system carrier phase data // Navigation Journal Inst. Navigation. 1987. V. 34. ¹. 1. P. 1-21.
Solodukhin, V.I., K.V. Shevchenko, I.N. Mazugin, and T.K.
Bokova, 1985. Space distribution of trees in correlation
with stand height, detected at laser profile (Raspredelenie
derev’ev po ploshadi v svyazi s vysotoi drevostoya,
opredelyaemoi po lazernoi profilogramme), In:
Lesoustroistvo, Taksaciya i Aerometody (Forest Planning,
Forest Inventory and Aerial Methods), Collection of Scientific Works, Leningrad Forestry Research Institute
(LenNIILH), Leningrad, Russia: 75-83 (in Russian).
Laser cartography. Complex aerial survey for geoinformation
support of power lines designing and other topography
works, 2000 / E. Medvedev http://www.opten.ru/eng/
altm/index.html Opten Limited Copyright, Moscow, Russia.
Lefsky, M.A., 1997. Application of lidar remote sensing to
the estimation of forest canopy and stand structure, Ph.D.
thesis, University of Virginia, 185 pp.
Stolyarov, D.P., and V.I. Solodukhin, 1987. About laser forest inventory (O lazernoi taksacii lesa), Lesnoi Zurnal
(Forest Journal), Russia, No. 5: 8-15 (in Russian).
Magnussen, S., and P. Boudewyn, 2000. Derivations of stand
heights from airborne laser scanner data with canopybased quantile estimators, Canadian Journal of Forest
Research, 28: 1016-1031.
Means, J.E., S.A. Acker, D.J. Harding, J.B. Blair, M.A.
Lefsky, W.B. Gohen, M.E. Harmon, and W.A. McKee,
1999. Use of large-foot-print scanning airborne lidar to
estimate forest stand characteristics in the western Cascades of Oregon, Remote Sensing of Environment, 67(3):
298-308.
Sweda T., T. Yamamoto, and Z. Shibayama, 1998. Airborne
infrared-laser altimetry of forest canopy profile for extensive and accurate assessment of timber resource and
environmental functions of forests. In: FORESEA
Miyazaki 1998 Forest Sector Analysis. Proceedings of
the Int. Symp. on Global Concerns for Forest Resource
Utilization – Sustainable Use and Management, October
5-8, 1998, Seagaia, Miyazaki, Japan. Tokyo: Japan Society of Forest Planning Press, Vol. I.: 736-745.
Means, J.E., S.A. Acker, B.J. Fitt, M. Renslow, L. Emerson,
and C.J. Hendrix, 2000. Predicting forest stand characteristics with airborne scanning lidar, Photogrammetric
Engineering & Remote Sensing, 66(11): 1367-1371.
Usol’tcsev, V.A., 1998. Forming of Data Banks of Forest
Phytomass (Formirovanie bankov dannykh o fitomasse
lesov), Ural Branch of the Russian Academy of Sciences,
Ekaterinburg, Russia: 306-441(in Russian).
Medvedev, E., 2000. Digital automatic orthophoto production with laser locator and aerial photography data. In:
International Archives of Photogrammetry and Remote
Weltz, M.A., Ritchie, J.C., and Fox, H.D., 1994. Comparison of laser and field measurements of vegetation height
and canopy cover, Water Resour. Res. 30: 1311-1319.
74
ACKNOWLEDGMENTS
ment of State. None of these organizations are responsible
for the views expressed herein. The author would particularly like to recognize the very careful and considerate review, including many detailed editorial and language suggestions, made by an anonymous reviewer, which helped to
very significantly improve the organization and content of
this paper.
Research for this article was supported by Ehime University with funds provided from the Environmental Agency of
Japan and by a grant from the International Research and
Exchanges Board (IREX) with funds provided by the Bureau of Education and Cultural Affairs (ECA), US Depart-
75
Chapter 8
Diameter Sensing Using Radio Frequency Identification for
Precision Forestry Applications
DENISE M. WILSON
SEAN HOYT
DOUG ST. JOHN
Abstract—This paper presents results for a prototype radio-frequency identification and sensing system focused on identifying trees and timber via wireless means and on sensing properties of the tree/timber during the identification process. A
prototype reader demonstrates 25 cm of reading range, using a zero-power tag in a magnetically coupled arrangement at 125 kHz
interrogation frequency. The range is resilient to variations in wood type and moisture content. Because of this resilience at the
relatively low operating frequency of 125 kHz, the system is also capable of producing measurements parasitic to the identification process that provide 7.8 mm accuracy in determining log or timber diameter (distance between reader and tag assuming
symmetrical placement of tag inside the wood sample). These results form the basis of extending RFID technology in the forest
products industry to smart chip technologies that can gather a wide variety of information about trees and wood in-situ and in
real-time.
INTRODUCTION
moisture, and biochemical content in the path between the
RFID tag and the reading (interrogating) device. Fortunately,
these unique problems associated with using RFID in forestry applications are not intractable. In this work, we present
initial experimental results that demonstrate range and robustness applications for forest and timber products applications.
Use of a wireless link between a powered interrogator
and a low or zero power reader (tag) is not inherently restricted to purposes of identification and tracking. RFID
methods can also be used to potentially measure biochemical, moisture, density, and other information in-situ under
no-power conditions. Transmission of the sensed information requires power, but can be mediated by the electrical
power supplied by the interrogator (reader) rather than the
tag. Smart chip extensions of RFID technology, that provide information beyond identification of objects and events,
are an important next step in advancing wireless information technology. Our work here demonstrates the potential
of smart-chip technology in the context of sensing, under
zero tag power conditions, the diameter of a log during the
RFID process.
Technology can provide invaluable mechanisms for improving the efficiency of the forest products industry at many
stages including seedling, harvest, timber processing, and
reforestation. Genetic engineering has supplied stronger seedling stock that has been supplemented by improved methods
for nourishing and optimizing the health of trees during the
growth cycle. Information technology, including a variety
of remote sensing methods, has offered the ability to monitor tree health while also gauging the impact of harvest on
the environment. The first and most obvious in-situ information that can benefit both forest management and timber
processing is the identification of each tree and subsequent
log by its history, genetic stock, pruning sequence, harvest
date, planting location and other relevant information. Radio frequency identification (RFID) methods have been developed to service a variety of applications that require touchfree identification of valuable resources. The primary advantage of radio frequency identification over conventional
bar-code identification methods is the resistance of an embedded RFID tag to environmental wear and tear and vandalism. An RFID tag can be stored, battery-free (the RFID
interrogator provides all necessary power to the system), in a
tree for years between planting and harvest or it can be embedded at harvest to facilitate robust and reproducible identification of the log and its piece parts during the timber production process. Identification of trees at this level carries
with it some unique application requirements including the
need for relatively long range (25 to 40 cm) and small tag
sizes coupled with resilience to large variations in wood,
BACKGROUND
Radio Frequency Identification using passive transponders (RFID) has become increasingly popular for monitoring and tracking a wide variety of objects, including biological organisms. RFID transponders, both electricallycoupled and magnetically-coupled, are available through a
wide variety of source companies. Over 150 companies are
currently involved in the production of RFID transponders,
77
tags, supporting infrastructure and systems. Over 124 publications have been generated in the technical literature in
the 1990’s on RFID targeted at a wide range of topics associated with optimizing RFID for particular classes of applications. This work extends the current use of RFID as a
fixed interrogation of passive transponders into systems that
are configured to sense parameters of interest in the forestry
environment. Only three efforts published in the technical
literature are reported to use RFID for sensing applications.
Neuzil et al report an integrated circuit embedded in an
RFID transponder tag that is capable of transmitting output
from generic capacitive microsensors, but has not yet been
interfaced to such sensors. Nikiforaki et al have used RFID
transponders to measure detailed fluid mixing properties;
each transponder has its own ID and is used to map, in time
and space, the properties of the fluid during mixing. Finally, Hamel et al. combine strain transducers capable of
passive detection of the peak strain on composite structures
(aircraft, dams, etc.) to RFID transponders and interrogation systems. Fully integrated sensing of environmental parameters, embedded in RFID constructs and systems, to the
best of our knowledge, has not yet been explored in the research community.
Other means of non-destructive interrogation of wood
properties via electromagnetic, sonic, or ultrasonic methods
also show a limited presence in the technical literature.
Kamaguchi et al. employs a later impact vibration method
to determine defects in the wood and moisture content.
Numerical methods that analyze interaction of L- band microwave on tree trunk are introduced by Tetuko et al. . Their
results are applied to discriminate tree trunk diameter within
Synthetic Aperture Radar images.
RFID, although not an economical replacement for barcodes in all cases, has particular use in applications where
destruction of a visual, external bar-code is an issue over the
life span of the object to be identified. Trees, which are
harvested up to 70 years after a seedling is planted in a managed forest, are one such example of an application, where
the greater cost of RFID is justified by the desired life cycle
of the tag inside the tree. The transponder tag itself can be
either passive (no-battery) or active (on-board battery); passive tags are particularly attractive for the precision forestry
effort, as they require no battery and therefore no maintenance during their potential 70 year use period . RFID has
obvious applications for the timber industry, especially as
major retail organizations, such as Home Depot and Lowes
have recently mandated that all lumber sold in their stores
be certified as to its quality and origination in a properly
and formally managed forest. This research seeks to address the appropriate use of RFID and associated smart-chip
sensing technologies for optimizing the value of forest and
timber, from seedling to processed timber. This effort demonstrates, at a proof-of-concept level, the ability to meet range
requirements for RFID in the forest products industry at a
(low) frequency that is resilient to variations in wood type
and moisture content. Using this resilience to environmental variation, our results also demonstrate the feasibility of
measuring log or tree diameter as an integral part of the identification process.
A radio frequency identification system consists of a
reader, which both transmits and receives a magnetic or electromagnetic field and a tag which, upon receiving a wireless signal from the reader, transmits identification information stored on the tag back to the reader. The percentage
of the transmitted field captured by the tag is dependent on
the size and geometry of the tag antenna and the orientation
of the tag relative to the reader antenna. When the tag is
small, increasing the size of the transmitted magnetic field
cannot be simply accomplished by increasing the size of the
reader antennas, since it is the percentage of the transmitted
field that the tag receives in its antenna that determines the
ultimate power delivered to the tag for responding to interrogation by the reader.
SYSTEM DESCRIPTION
The radio frequency identification system demonstrated
and used in this work consists of the following basic components:
•
•
•
•
•
•
Driving circuit: provides current to the antenna
circuit to be transformed into the desired magnetic
field.
Antenna circuit: transmits a magnetic field to the
tag
Level adjustment circuit: reduces received magnetic field to voltage levels appropriate for decoding
Decoding circuit: decodes the received encoded
signal into a serial digital signal
Microcontroller: converts the series digital signal
to a form that can be transferred to a datalogger or
other computing device and provides control signals (including clock) to the driving circuit.
Tag: consists of an antenna and silicon chip with
on-board power supply and transmitting memory
storage capability.
The reader is a modification of a system provided by
Microchip in application notes. Discrete components in the
driving circuit have been replaced with an integrated current buffer (Burr-Brown BUF634) for better matching in the
driving transistors and an integrated voltage booster
(NE5538) to supply more voltage to the antenna circuit while
allowing the entire reader to run on two 9 volt batteries.
Decoded RFID signals from the battery-less tag are interpreted by the microcontroller as a complete tag identification sequence and converted to standard RS232 protocol for
transmission to a variety of possible microcomputers, including dataloggers, personal digital assistants and PCs.
The two output signals of the RFID reader circuit are the
data stream encoded as an RS232 signal produced by the
78
Figure 1:Experimental Setup - The experimental setup is held perpendicular to the tree to control the orientation of antenna to tag. The
antenna slides to and from the tag. Vanalog is read from the digital multimeter at increments of 1 cm from 11 to 20 cm.
microcontroller and the voltage Vanalog. The difference between the maximum and minimum values of Vanalog represent the difference in signal strength between a digital “0”
and “1”. This difference is expected to vary with the distance between tag and reader (which affects the strength of
the shunting of the magnetic field by the tag as received by
the reader).
Figure 2: Viable Tags for use in Forest Products -The (a) first
tag is in the shape of a nail and is suitable for insertion into a
live tree before harvest; the (b) second, circular tag is suitable
for insertion into a log (a nail is inserted through the center hole)
after harvest and demonstrates greater range because of
increased antenna area.
(a)
(b)
EXPERIMENTAL SETUP
Maximum read ranges have been evaluated with the use
of the RFID reader and the experimental setup of Figure 1.
The tag used is circular and approximately 3.5 cm in diameter, obtained from Intersoft. It is hammered into each species of wood under evaluation. To evaluate the impact of
distance, moisture and species on signal strength, a single
sample of each of three species of wood (12" in diameter; 2’
in length, no defects) are implanted with the 3.5 cm diameter tags at a distance of 10 cm from the bark. A tag packaged into the shape of a nail and this tag are considered
viable options for insertion into trees before and after harvest respectively (Figure 2 ). The species included in the
experiment were Species 1: Northwest Douglas-Fir
(Pseudotsuga Menziesii), Species 2: Western Red Cedar
(Thuja Plicata), and Species 3: Red Alder (Alnus Rubra)
with wood densities of 28.1 lbs/ft, 21.9 lbs/ft, and 23.1 lbs/
ft, respectively.
Electromagnetic theory dictates that the signal strength
will vary with orientation of the reader relative to the tag. To
reduce this variation, and to keep the setup as consistent as
possible, the orientation between antenna to tag is held fixed
by employing an apparatus affixed to the side of each tree,
perpendicular to the bark (Figure ). This apparatus consists
of a wood channel fitted for the wand antenna to slide toward and away from the tree. During each experiment, the
wand antenna is placed in the channel and moved from 11
to 20 cm away from the embedded tag in increments of 1
cm. Starting voltage with no tag present and voltages at each
increment are recorded. Moisture content is also measured
by extracting core samples at a fixed distance from the end
of the tree along a fixed radius. The core is weighed immediately upon removal, oven dried at 101 degrees Celsius
(215°F) for 12 hours, then weighed again. The difference in
core weight divided by the original weight provides the percentage of moisture at that radius. Experiments are conducted
every other day to allow the tree samples to release moisture.
Moisture content variation within the species samples is
considered. To evaluate variation in the input parameter
(moisture content), data are collected at the following six
locations:
•
79
Three equidistant points along the perimeter of
the sample
Figure 3: Output Voltage vs. Diameter for Wood Species 1- This figure demonstrates the second order polynomial
nature of the relationship between diameter (distance between reader and tag) and the signal level of Vanalog. The polynomial
of best fit is shown. Worst-case RMS error in diameter measurement for all 3 wood species is of 7.8 mm in a range from 1120 cm range of diameter measurements
•
Three points along the radius of the sample at the
same axial location as the perimeter data points
tag. Electromagnetic theory dictates that the signal strength
will vary with orientation of the reader relative to the tag,
introducing potential errors to this diameter measurement
scheme. However, the circuitry can be modified to scan the
field in different orientations for maximum signal strength
at Vanalog, thereby providing a more robust indication of diameter than if a measurement were extracted at a random
orientation between reader and tag.
Variation of Vanalog used to verify range reproducibility is
plotted in for wood species 1 as moisture content decreases
over time. These results do demonstrate a variation in the
relationship between signal strength and diameter across
different experiments. This relationship is not monotonic
(i.e. signal strength consistently decreases or increases with
moisture content), suggesting that the variation in experimental measurements is not due to moisture content but to
some other parameter in the experiment (an uncontrollable
factor). Drift in circuit operation has been observed as a
function of the drive circuits in the antenna; future circuit
modifications must compensate for (temperature) drift in
these drive circuits in order to reduce the error in diameter
measurement (up to a maximum of 1 cm or 10% in this set
of experiments). Overall error in diameter measurement due
to this drift has an RMS value of 7.8 mm.
As expected (Figure 3 ), range varies little as the moisture content of wood in which the tag is embedded. Scatter
plots in Figure 4 display a random relationship between
moisture content and reader voltage for the three species of
wood. Since only six points (sparse data) are available for
every species of wood to evaluate the relationship between
Topological variation in moisture content are significant,
up to 9% in some cases. Due to the fact that core samples
are extracted within the same area, moisture content variation is minimal within each experiment.
Although wood, moisture, and the wood/air interface attenuates the transmission of electromagnetic waves between
RFID interrogator and tag, these effects are expected to be
negligible at the small (125kHz) frequencies used in this
demonstration system. High frequency systems (MHz and
GHz) provide greater range but are more substantially impacted by attenuation and are therefore, not appropriate for
combined ID and diameter sensing.
EXPERIMENTAL RESULTS
Maximum read ranges are demonstrated at 25 cm at a
drive current of 40 mA, well within an acceptable life time
for two 9 V, 150 mAh batteries. The drive circuits are not
operating at their absolute limits (250-300mA), implying
that further range improvements can be achieved with the
use of a more efficient matched antenna. The voltage Vanalog
has proven, in the experimental results, to be an effective
measure of diameter of the wood. Vanalog represents the field
strength in the presence of an interaction (communication)
between the reader and the tag. This voltage varies with the
distance between tag and reader as the field received by the
tag decreases as the cube of the distance between reader and
80
Figure 4:Output Voltage (Vanalog) vs. Moisture Content for Three Species of Wood - Voltage data for distances of 11-20 cm are taken for
three species of wood. Each scatter plot demonstrates how volt­age responds to a decrease in moisture content. The correlation coefficients
are listed below the scatter plots. The closer the number is to +1 or -1, the stronger the relationship between moisture content and voltage
is. A value of 0 implies no relationship. In each case, the x-axis is moisture content and the y-axis is Vanalog (diameter).
moisture content and signal level, the correlation coefficients
in Figure , in some cases, represent a relationship between
moisture content and the output signal, Vanalog that is not necessarily valid. The predominance of correlation coefficients
of values less than 0.5 and greater than -0.5 confirm that,
even in spite of sparse data limitations, output signal level
is independent of moisture content across wood species. In
fact, when taken as a whole and normalized, the entire experimental data set across moisture content, diameter, and
wood species type demonstrates a correlation coefficient of
approximately -0.346, a very weak correlation between the
controlled variables (moisture, wood species) and signal
strength Vanalog. Immunity to wood moisture and variation is
expected at 125kHz due to a combination of the inherent
resistance to noise exhibited by digital systems and due to
the fact that a 125kHz wave is much more able to turn corners and propagate consistently across different media than
waves at higher frequencies (despite the decreased resolution of these lower frequency waves).
at 125KHz. Ranges of up to 25 cm have been demonstrated
and are expected to increase when the driving capacity of
the reader circuits is optimized through future circuit modifications. Ranges are also expected to increase in future
work as the circuits are streamlined and smaller changes in
the received voltage at the reader can be identified with more
precision. Determination of log diameter has also been demonstrated using a parasitic measurement from the RFID process with a resolution of 7.8 mm, limited by reader circuit
drift. These results are the first step toward practical Radio
Frequency smart chip sensing systems for identifying a wide
variety of tree characteristics in-situ and in real-time for the
forest management and timber products industries.
LITERATURE CITED
Frontline Solutions Website: http://
www.frontlinemagazine.com/rfidonline/listings.htx
CONCLUSIONS
Pavel Neuzil, Oskar Krenek, Michael F. Serry, and Jordan
G. Maclay, “Measurement and wireless transmission of
embedded capacitive microsensor’s output using
SigmaDelta conversion and radio frequency identifica-
This paper has presented successful experimental results
for radio identification of wireless, zero-power tags in wood
81
tion (RFID) technology,” In Proceedings of SPIE 3044,
Smart Structures and Materials 1997: San Diego, CA,
March 4-6, 1997, pp. 178- 185.
Radio Frequency Identification — RFID, a basic primer,
AIM (Automatic Identification Manufacturers) Global
Network, http: //www.aimglobal.org/technologies/rfid/
resources/papers/rfid_basics_primer.htm.
. Nikeofraki, K.C. Lee, M. Yianneskis, “Circulation time
measurements with miniature radio frequency identification devices,” In Institution of Chemical Engineers
Symposium Series Fluid Mixing: Bradford, UK, July 7August 8, 1999, pp. 47-57.
A. Kamaguchi, T. Nakan, T. Nakai, A. Tamura, “Measurement of heartwood moisture content of Sugi by the lateral impact vibration method: Comparison with data
obtained from the moisture meter method and measurement of longitudinal distribution,” In Journal of the Japan Wood Research Society, v47, n3, 2001, pp 235-241.
Michael J. Hamel, Christopher P. Townsend, Steven W.
Arms, “Micropower peak strain detection systems of
remote interrogation,” In Proceedings of SPIE 3990,
Smart Electronics and MEMs: Newport Beach, CA, pp.
104-109.
Josephat S.S. Tetuko, R. Tateishi, K.Wikantika,”Method to
estimate tree trunk diameter and its application to discriminate Java-Indonesian tropical forest characteristics,”
In 2000 Interantional Geoscience and Remote Sensing
Symposium: Honolulu, HI, Jul 24 - Jul 28 2000, pp 405407.
82
Chapter 9
Cooperative Use of Advanced Scanning Technology for
Low-Volume Hardwood Processors
LUIS G. OCCEÑA
TIMOTHY J. RAYNER
DANIEL L. SCHMOLDT
A. LYNN ABBOTT
Abstract—Of the several hundreds of hardwood lumber sawmills across the country, the majority are small- to medium-sized
facilities operated as small businesses in rural communities. Trends of increased log costs and limited availability are forcing
wood processors to become more efficient in their operations. Still, small mills are less able to adopt new, more efficient technologies due to initial cost, payback period, and modifications to operations. Based on the current marketing structure in the hardwood industry, the prevalent links between small- to medium-sized landowners and similar sawmills are log concentration yards.
This paper examines the utility and logistics of locating industrial CT scanners in log concentration yards. If installed, logs can be
scanned, automatically graded, and potentially optimized for sawmill breakdown (processing decisions) at the yard. By knowing
the correct grade of each log, a mill can properly manage its log inventory. In addition, by knowing the internal structure of each
log, a mill can optimize the sawing of that log. The advantage of this proposed system for the small mill is that it is affordable and,
with little or no infrastructure changes, they can buy and utilize value-added logs that have been CT-scanned upstream.
INTRODUCTION
of the small- and medium-sized hardwood mills from participating in the propagation of technology. Small mills are
under increasing financial pressure and face a difficult time
ahead if they cannot access at least some of the advanced
technology.
These small hardwood mills are numerous and important. Mills under 25 million board feet (MMBF) per year
make up about 70% of hardwood mills (Figure 1). It is clear
from the graph that small hardwood producers are a key contributor to the industry as they represent a significant share
of the market. Nevertheless, fewer than 20% of the hardwood mills surveyed in 1997 by the newsletter Hardwood
Weekly Review report the use of any new technology
(Luppold and Baumgras 2000). In a recent survey of 424
sawmills, 63% reported no type of scanning or optimizing
equipment in the sawmill operation (Bowe et al. 2001). The
amount of under-utilization of advanced technology and the
corresponding pressure on land use are significant.
Wood Processors
With reduced timber harvesting on public lands, more
pressure is being placed on individual landowners to supply
an increasing demand for hardwood products. That pressure is also transferred to the sawmills that have to fill that
demand with a relatively poor-quality log supply from cutover lands. Historically, the highest value yields come from
older trees, but current and future supplies of older trees,
such as old-growth and long-rotation timber, are very limited (Luppold and Baumgras 2000). There are two approaches to this paradox. Either landowners must significantly change the way they manage their forests to produce
more high-value logs (which is being done in some cases),
or sawmills must turn to technology to better utilize existing, and anticipated, small-diameter, low-quality raw material.
Extracting high-grade lumber from diminishing timber
resources has been an increasing challenge for most hardwood sawmills. Most large softwood mills and many large
hardwood mills have implemented the latest sawing and optimization technology to increase lumber yield and value.
Lumber yield over the past years has steadily increased with
the advent of optimization technology. However, the capital cost of much of this advanced technology prohibits many
Sawmill Operations
In contrast to softwood lumber, which is valued in terms
of volume and mechanical strength, the value of hardwood
lumber is based more heavily on appearance-related criteria. For this reason, hardwoods account for most wood used
in the manufacture of high-value furniture, cabinets, flooring, millwork, and molding, along with hardwood exports
83
Figure 1. Hardwood lumber production is concentrated in small
(<25 MMBF) mills.
Figure 2. Tomography produces cross-sectional “slices” of a
log that capture density-related internal features.
889
399
307
276
258
159
Export
yds.
50
Veneer
1-10
10-25
25-50
50-100
<1
77
21
100-200
1000
900
800
700
600
500
400
300
200
100
0
200
# of mills
Number of mills vs. production
Annual Lumber Production (MMBF)
sawyer’s capability to produce high-value boards. Developing nondestructive sensing and analysis methods that can
accurately detect and characterize interior defects is critical
to future efficiency improvements for sawmills (Occeña
1991). This type of sawing would be information-augmented
in the sense that internal defect information is leveraged to
obtain better value and volume yields from the log (Occeña
et al. 1997). Increased information from scanning removes
much of the guess-work that hinders sawyers’ skill, and it
also enables log sorting/processing decisions prior to breakdown.
dependent on the quantity, type, and location of defects, thus
each log must be sawn to minimize (subject to board sizing
constraints) the number, size, and severity of defects in the
resulting boards.
Conversion of hardwood logs into lumber involves a number of steps. While there are variations of twin line or twinband headrigs, the following description is typical of a singlesaw headrig. Logs entering the mill are debarked. Following this operation, they go to the headrig where a sawyer
moves the log repeatedly past a saw to remove flitches
(unedged and untrimmed boards) one at a time. The sawyer
chooses a sawing strategy by visually examining the exterior of a log, and periodically picks a better grade face by
rotating the sawing face as the log interior is exposed
(Malcolm 1961). This type of sawing is information-limited in the sense that the sawyer only has knowledge of external indicators of internal features (e.g. knots) (Occeña et
al. 1997). Flitches go through subsequent operations of edging and trimming, where defects and bark at the edges and
ends are removed to increase the board’s grade, and consequently its commercial value. The cant may either enter a
resaw operation where additional boards are sawn, or may
be sold intact for use in low-value products, such as railroad
ties or pallet parts. During initial log breakdown, profitcritical decisions are made by the sawyer that can significantly affect downstream processing operations. This observation suggests that targeting sawlog breakdown improvements can drastically increase lumber value recovery, as well
as volume recovery, for downstream operations.
The sawyer uses log shape, external indicators of internal
defects, and knowledge of lumber grades to make sawing
decisions. While sawyers are highly skilled in this task, studies (Tsolakides 1969, Richards et al. 1980, Wagner et al.
1990, Steele et al. 1994) have shown that the lumber value
of logs can be improved anywhere from 10% to 21% by carefully selecting the proper sawing strategy. However, the
current level of information available to sawyers during the
log breakdown operation is inadequate for enhancing the
Internal Log Scanning
Because most defects of interest are internal, a nondestructive sensing technique is needed that can provide a 3-D
view of a log’s interior. Several different sensing methods
have been tried, including nuclear magnetic resonance
(Chang et al. 1987), ultrasound (Han and Birkeland 1992),
and x-rays (Benson-Cooper et al. 1982, McMillin 1982,
Cown and Clement 1983, Onoe et al. 1984, Taylor et al.
1984, Burgess 1985). Due to its efficiency, resolution, and
widespread application in medicine, x-ray computed tomography (CT) has received extensive testing for roundwood
applications (McMillin 1982, Taylor et al. 1984, Funt and
Bryant 1987, Roder 1989, Zhu et al. 1991a, Som et al. 1992).
An x-ray CT scanner produces image “slices” that capture
many details of a log’s internal structure (Figure 2). Because x-ray attenuation is linearly related to wood density
(Shadbolt 1988) and many wood features (including defects)
exhibit density differences (Hopkins et al. 1982), many lumber-quality defects (e.g., knots, voids, and decay) are visible
in CT images.
CT technology is quite capable of providing sawmills with
valuable log processing information. The critical issue is
how to introduce internal log scanning into an industry whose
members have valid concerns about initial cost, payback period, and integrating new technology into existing mill
84
Figure 3. The current prototype log scanner includes
specialized conveyors to move logs through the scanning ring.
Figure 4. After all CT “slices” are generated, longitudinal
“virtual” boards can be created.
operations. The next section introduces industrial CT, describes preliminary prototype tests conducted in a softwood
mill, and covers CT data handling issues. Then, we present
our vision of how private landowners and small processors
can share CT technology, and we lay out a series of step to
accomplish this goal. Finally, we offer some observations
on why this approach to technology transfer is particularly
beneficial.
INDUSTRIAL COMPUTED
TOMOGRAPHY
m3/year) in Austria (Schmoldt et al. 2000b). The sawmill
specializes in producing window frames from alpine softwood species, mainly Norway spruce (Picea abies, L.).
Some tests were also performed for European larch (Larix
decidua, Miller), another common species of this region.
All dimensions and qualities were chosen from the normal
dimension and quality product lines of the mill.
Two sets of logs (over 100 logs total in this study) with
comparable diameters were randomly placed into a scan
group and a control group. For the scan group, logs were
CT scanned and reconstructed images were presented to the
sawyers. Using a tool developed by InVision Technologies,
sawyers were able to see simulated board faces (virtual cuts)
for different cutting positions (Figure 4). Selection of the
cutting positions for log breakdown was done as usual, except that the sawyer could view deeper cuts without the risk
of making expensive mistakes. For this application, a thicker
board has a higher value compared to several thinner boards.
But, without CT assistance there is significant risk in cutting too deep into the log to capture a thicker board. The
quality of the hidden face of a thick board may be worse
then expected. In such a case, one or more high quality thin
boards would have been a better choice. No additional computer-based optimizations were performed. Sawing patterns
were then manually marked on the log ends to serve as a
guide for the physical sawing of the logs at the headrig.
The control group was processed according to the normal operation of the mill. After the primary breakdown, all
boards from both groups were blind-graded by expert grad-
Background
CT scanning entails rotating an X-ray source around an
object. X-ray scans are collected from 360 degrees around
an object to produce a tomographic slice of that object. The
slice is a detailed density map of the object’s internal structure. CT was introduced in 1972 for medical imaging. Since
that time, CT has become a cornerstone diagnostic tool for
hospitals. Although medical CT systems would appear to be
easy to adapt to such applications, it is, unfortunately, the
case that current medical CT systems have been engineered
for infrequent, short duration use. This is incompatible with
industrial sawmill needs (Schmoldt et al. 2000b). In 1994,
InVision Technologies, Inc. produced the CTX 5000 scanner that became the first industrial CT scanner for the detection of small amounts of explosives in passenger luggage for
aviation security. CTX machines can now be found at airports throughout the world. Due to similarities in application environments (Schmoldt et al. 2000b), the industrial CT
engine from the aviation security application can be translated to log scanning for sawmills. InVision has developed
and tested a log-scanning prototype (Figure 3).
Log Scanning Mill Test
The first long duration study of a CT scanner in a sawmill
was performed in a medium-sized softwood mill (~100,000
85
Figure 5. Slat yields by grade illustrate a shift to higher quality slats between the control and scan groups of logs.
by viewing a sequence of 2-D CT images, or even “virtual”
boards. Furthermore, generating CT images produces tremendous amounts of data. For example, depending on resolution and frequency of scans, scanning a single 4 m log
may result in 20-800MB or more of image data. Storing
and handling that amount of data between scanning and
processing for many logs would be cumbersome at best.
Fortunately, CT data contain a large amount of redundancy,
which can be exploited to condense the data into a form
that is more manageable and usable.
Only those internal features of a log that are important
for subsequent processing need to be identified (log shape
comes free as tomographs are captured). These features
are the defect areas within a log. Each density-related defect is relatively contiguous and each such defect type is
fairly homogeneous with respect to density. Consequently,
over the past 20 years researchers have begun to develop
automated methods to interpret CT imagery (Hopkins et
al. 1982, McMillin 1982, Funt and Bryant 1987, Zhu et al.
1991b, Schmoldt et al. 1993, Li et al. 1996, Zhu et al. 1996,
Schmoldt et al. 1997, Schmoldt et al. 1998). Once different internal log defects can be automatically detected, then
those views can be integrated into a 3-D rendering of the
log. Defect information can also be used in generating automated log breakdown decisions. One defect detection
algorithm using artificial neural networks (Schmoldt et al.
2000a) achieved accuracy above 96% for single-species
classifiers, 90-97% for two-species classifiers, and 91-92%
for three-species classifiers. It is able to label an entire CT
image (containing 64K pixels) in 1-2 seconds (depending
on processor speed).
Eventually, CT data will be used to automatically arrive
at log breakdown decisions. One early computer model
designed to deal with this defect specific approach was
PDIM (Pattern Directed Inference Model), which generates a log breakdown pattern specific to the internal defect
configuration found inside the log (Occeña 1992). It accomplishes this by enveloping defects in a defect hull and
analyzing a composite end view that represents an aggregation of the defects’ distribution through the log. Automated decision making was driven by the shape of the hull,
and density numbers that reflected the defect concentration
ers (without knowing to which group boards belonged).
Using volumes and prices, the value yield for each log was
calculated. As one example out of the study, the results of
grading 30 high quality spruce logs with diameters of 51 cm
and larger showed an increase in value per m3 of 6.3% for
the scanned group (8.8% if the best and worst logs in each
group were removed).
Normally, most of the high quality boards in that mill are
cut as intermediate products for the window frame industry.
They are processed further to produce slats in a secondary
breakdown. If the classification of the primary board was
correct, a high percentage of the board can be used for slats.
So this secondary stage of processing is an excellent indicator of classification accuracy for the CT system compared to
experienced sawyers, and might be a better demonstration
of yield improvements by the CT system.
Slats where produced from boards with a thickness of
153 mm. With this dimension, it is very difficult to correctly
estimate the quality of the board on the back face. After
grading the slats, the control-group slats yielded 31% (in
value) for grade A (the highest value grade), 49% for grade
B, and 20% in grades C and D (the lowest value grades). On
the other hand, the CT optimized boards resulted in 71%
grade A and only 1% in grades C and D (Figure 5). This
improvement cannot normally be achieved without knowing the quality of the board in advance.
CT Data Application
While the previous mill test illustrates the potential value
gains that internal scanning can achieve, it relied on sawyers
carefully examining full CT imagery. In viewing 2-D “virtual” boards, they were able to accurately determine how
thick boards should be without risking much-reduced slat
quality. In general, however, this interactive use of CT imagery is frequently incompatible with mill operations and
not the most effective for log breakdown. In most hardwood
mill operations, log breakdown decisions must be made
quickly, and sawyers benefit greatly from a complete 3-D
view of a log, where both log shape and internal defects are
clearly visible, i.e. a “glass log” view. This cannot be achieved
86
tion. Because scan information will allow simulated breakdown of the log into lumber (which can be computer graded),
it should also be possible to generate a lumber grade distribution and lumber value for each log. When a sawmill takes
delivery of a log, it could have an accurate log grade, a description of the lumber contained in the log, and information about the internal structure of the log. Sawmills can
purchase whatever level of detailed information they wish
to use, and feasibly could do so on an individual log basis.
The nature of the internal information can be manifested in
one of two ways:
along the length of the log. Designed to be a generative
process-planning model, PDIM generated sawing instructions that could be used to direct a numerically controlled
sawing headrig and log carriage. This and other similar
models have the potential for effectively automating log
breakdown decisions.
CONCENTRATION YARD SCANNING
Log Yards
Log brokering operations (concentration yards) provide
a convenient intermediary between loggers/land owners and
sawmills/log buyers. In particular, these yards form an important link between the small- to medium-sized landowners and similar sawmills. Harvested logs are brought to these
concentration yards either by independent loggers or by broker-contracted crews. In most instances, logs are bucked,
i.e., log lengths of 2.5-5m (8-16 ft) are produced from treelength stems on the log deck at the harvest site. This greatly
simplifies transport, but leaves an important processing decision to variably trained—and incorrectly rewarded—logging operators. Most of these bucking operators have a volume, but no value, incentive for their bucking efforts (Pickens
1996). From concentration yards, logs are then merchandised to various client sawmills based on a bill of materials
(Bush et al. 1992). Prior to sale, the merchandiser grades
each log (based on external characteristics) and estimates
board-foot volume. Logs are then sold in batches to appropriate mills based on species and anticipated product. The
highest quality logs go to veneer mills, or are exported.
Lower quality logs go to various sawmills for lumber production.
•
•
Saw line markings on the log ends can indicate the
location of the best opening face, or the location of
maximal clear wood content.
A computer file can contain reconstructed 3D images, and/or log breakdown optimization data with
lumber yield.
Information about log grade and lumber content for each
log can help a mill properly manage its log inventory. By
knowing the internal structure of each log, in addition, a
mill can optimize the sawing of that log and also improve
downstream edging and trimming operations.
Another immediate advantage of such a system for the
small mill is that with little or no infrastructure changes,
they can buy and utilize logs that have been CT-scanned
upstream at the concentration yard. If a sawmill chooses to
use the grade and the location of the best opening face or
clear wood (as marked on the log ends), then there will be
no setup cost. On the other hand, if a sawmill would like to
customize certain logs—e.g., the intermediate grade logs
which value varies greatly depending on breakdown pattern—by also procuring the prescribed log breakdown optimization, then it will require a small setup investment in a
PC-based computer system, a barcode reader, and a visualization/imaging system. The latter investment will be required to register the log with the stored scan data and download the breakdown optimization. Furthermore, it should
be possible in the future to tailor optimization patterns to
individual mill operations and product lines.
Valued-Added Scanning
Locating a log scanner at a concentration yard possesses
several theoretical advantages. First, it allows many members of the hardwood industry to take advantage of advanced
technology. Second, it generates important log information
early in the processing stream, so that it can be used throughout mill operations. Third, it greatly increases the objectivity of log valuation; less guesswork is involved in log pricing. Precise pricing can be reflected in prices paid to landowners, can provide incentives for loggers to buck for log
value (not volume), and can ensure mills that they are getting the log quality that their operations need. The information generated adds value to each log—value that is retained
in the local economy. While the price of the log might increase by the cost of the scanning operation, it will be more
than offset by the benefits mentioned above accruing from
the information (Hodges et al. 1990).
With a concentration yard log scanner, logs can be
scanned, automatically graded, and optimized for sawmill
breakdown. Current hardwood log grading rules estimate
clear wood volume by examining external characters of the
log faces. Detailed internal information should provide better estimates of log grade and even permit development of
an alternative set of log grades based on internal informa-
Implementation
A CT engine for log scanning already exists in the form
of InVision’s baggage scanner. It has already been prototyped
as a log scanner and tested successfully in two sawmills (one
hardwood mill in the western U.S. and one softwood mill in
Austria). Considerable additional work needs to be done,
however, before it is ready to scan hardwood logs in an industrial setting, and before it can operate as envisioned above.
Four of the important next steps are outlined in the following sections.
Optimization and Defect Identification Software
The goal of this step is to develop and optimize defect
87
Integration with Log Handling
identification software based on CT-derived volumetric images. Log breakdown software already exists that uses external dimension information, from commercial log shape
scanners for example. It needs to be extended to optimize
log breakdown based on full 3-D internal information derived from CT. This development will rely on the defect
identification software reported earlier, as well as CT image
processing and explosives/narcotic detection experience
gained in baggage inspection. Because breakdown optimization requires knowledge of lumber value (which includes
lumber grade), existing lumber grading software will be included (Klinkhachorn et al. 1988). Optimal edging/trimming software will also be added to the set of software tools,
so that this important downstream process can be integrated
into upstream decision making. These tasks, and several
others, are listed below:
Access to, and use of, internal log defect information will
require the design and development of an automated log and
scan-data tracking system. This system will tag the log,
mark the opening face on the log ends, store the scan-data
and/or optimized breakdown instructions, associate the log
with the scan-data/optimized breakdown instructions, and
transmit or enable the retrieval of stored scan-data/breakdown instructions at the client’s sawmill. A log/data tracking scheme, a database architecture that can store scan data,
as well as traditional data formats, and efficient data communications (including business-to-business data exchange)
are necessary for easy access to the scan-data/breakdown
instructions associated with each log at the concentration
yard. These tasks can be summarized as follows.
Task 1. Modify existing external dimension log optimization software to utilize internal CT information.
Task 2. Develop defect identification software based
on CT images.
Task 3. Develop software to map defect size and type
onto a wire frame representation of a board.
Task 4. Develop protocols to send the board representation to board edging/trimming software.
Task 5. Develop a user interface to allow users to
modify optimization parameters.
Task 1.
Develop a barcode or radio frequency tagging
system to identify and track each log.
Task 2. Develop an automated system to mark log ends
with opening face positions and to optically
read those markings.
Task 3. Design and develop database and data warehousing systems to store and retrieve associated scan-data and/or optimized breakdown instructions.
Task 4. Test, implement, and maintain above-mentioned data and log handling systems.
Increase Acquisition Speed of CT Engine
Data Collection and In-House Testing
Increased log throughput requires improved software and
hardware for continuous scanning. Continuous scanning of
logs will require the use of a fast data streaming and storage
system, as the current InVision CT prototypes’ computer
architecture cannot handle the vast amount of data generated by volume scanning. To scan a log using the current
computer architecture the scanner must stop acquiring data
for a short period of time (a few hundred milliseconds), move
the acquired data from one memory location to another, then
start scanning again. This data acquisition stopping and starting causes small gaps (0.5-1 cm) during continuous data
collection. To ensure that all defects are fully imaged, these
gaps must be eliminated. Continuous scanning and data storage should be possible after completion of the following
tasks.
In-house data collection can be used to thoroughly test
the system prior to field-testing. These in-house tests can
serve as a final proofing of the system prior to field deployment where any software and hardware bugs can be worked
out efficiently and quickly. To perform meaningful tests,
several hardwood log specimens from the eastern U.S. should
be used. Given their importance to the eastern hardwood
industry, the species red oak (Quercus rubra, L.), yellowpoplar (Liriodendron tulipifera, L.), black cherry (Prunus
serotina, Ehrh.), and sugar maple (Acer saccharum, L.) are
preferred. For each species, several logs would be selected
in each of the three log grades. These samples would provide examples covering a variety of wood densities, wood
structure, and wood quality. This variability will aid testing
both the scanning engine (hardware/software) and the application software (sawing optimization, edging/trimming).
Task 1. Specify a high-speed data buffer and computer
system capable of streaming a continuous flow
of CT log images and storing them prior to
processing.
Task 2. Convert the present discrete slice CT system
to incorporate the high-speed buffer and computer system.
Task 3. Develop the software necessary to acquire continuous CT log images.
Task 1. Obtain a representative sample of different
hardwood logs that will be used for the inhouse data collection.
Task 2. Setup CT system ready for log screening together with full data collection and storage.
Task 3. Test and record data from each log.
88
DISCUSSION
rather than a few, large mills—transportation costs can be
kept lower and not add to the increasing cost of hardwood
lumber. In this scenario, log concentration yards will no
longer be just marketing points for raw material, but will act
as technology distribution centers for the industry.
Unlike most softwood sawmills, hardwood sawmills are
not completely automated at the headrig. Because most
hardwood sawmills produce appearance grade lumber, they
require a sawyer to evaluate the outside of the log to predict
the best opening face and the optimum sawing pattern. Small
sawmills have been able to stay competitive in this environment because log breakdown is a learned skill that can be
developed over years of training and experience. A good
sawyer specializing in a specific species or product mix can
be responsible for significant value increases in that sawmill, whatever the size of the mill might be.
The last frontier in automated hardwood sawmill optimization is the ability to see inside a log before sawing, which
would reduce the sawyer’s role in determining value yield
from a log. The ability to see inside a log is on the horizon.
Unfortunately for small sawmills, they will face the same
problem as they have historically, i.e., a CT scanner is a
capital-intensive product in which small sawmills will likely
not be able to invest.
Small mills are in further danger because one of their key
competitive advantages will be removed with the introduction of CT, in that the sawyer’s learned skills and experience will become less important. Sawmills using CT technology will be able to predict product mix, product grade,
and product value more accurately before ever sawing a log.
Using CT technology has already been established in several studies to be more accurate than the sawyer. CT brings
a wide range of benefits, from reduced log inventories at the
front end of a mill, to edging and trimming based on grade
at the back end of the mill.
Our approach to technology transfer will allows mills to
share the cost of technology and to also avoid mill layout
changes. Therefore, small- and medium-size sawmills can
benefit from the advantages of internal log scanning at little
cost and without negative operational impact on their mills.
By including small mills in the benefit stream of CT, the
70% of hardwood mills that are not currently impacted by
technology will have equal opportunity. This will lead to
far more logs being optimized. Better optimization and inventory control will allow more trees to stay in the forest
longer, and allow landowners more flexibility when it comes
to land management.
Technology can either alienate people/groups or it can
democratize them, leveling the playing field for all. The
keys to enabling the latter are access and acceptance. Installing internal log scanning operations in log concentration yards will provide technology access for a wider clientele, while the economic benefits will hopefully engender
their acceptance. Many hardwood mills are located in rural
areas. By providing those mills access to technology that
would otherwise be beyond their reach, rural economic stability will be enhanced. The information attached to each
log at scanning will add value to that log and will allow
rural areas to benefit from that value addition. Furthermore,
with many, small mills located close to the raw material—
LITERATURE CITED
Benson-Cooper, D.M., R.L. Knowles, F.J. Thompson, and
D.J. Cown. 1982. Computed tomographic scanning for
the detection of defects within logs. Forest Research Institute, New Zealand Forest Service, Rotorua NZ, Bull.
No. 8. 9 p.
Bowe, S.A., R.L. Smith, and P.A. Araman. 2001. A national
profile of the US Hardwood Sawmill Industry. Forest
Products Journal 51(10):25-31.
Burgess, A.E. 1985. Potential applications of medical imaging techniques to wood products. In 1st International
Conference on Scanning Technology in Sawmilling,
Szymani, R. (ed.). Forest Industries/World Wood, San
Francisco CA.
Bush, R.J., P.A. Araman, and J. Muench Jr. 1992. A comparison of market needs to the species and quality
compostion of the U.S. Hardwood resource. P. 275-277.
In Wood Product Demand and the Environment. Forest
Products Society, Madison WI.
Chang, S.J., P.C. Wang, and J.R. Olson. 1987. Nuclear magnetic resonance imaging of hardwood logs. In 2nd International Conference on Scanning Technology in
Sawmilling, Szymani, R. (ed.). Forest Industries/World
Wood, San Francisco CA.
Cown, D.J., and B.C. Clement. 1983. A wood densitometer
using direct scanning with x-rays. Wood Science and
Technology 17(2): 91-99.
Funt, B.V., and E.C. Bryant. 1987. Detection of internal log
defects by automatic interpretation of computer tomography images. Forest Products Journal 37(1): 56-62.
Han, W., and R. Birkeland. 1992. Ultrasonic scanning of
logs. Industrial Metrology 2(3/4): 253-282.
Hodges, D.G., W. C. Anderson, C. W. McMillin. 1990. The
economic potential of CT scanners for hardwood sawmills. Forest Products Journal 40(3):65-69.
Hopkins, F., I.L. Morgan, H. Ellinger, and R. Klinksiek.
1982. Tomographic image analysis. Materials Evaluation 40(20): 1226-1228.
Klinkhachorn, P., J.P. Franklin, C.W. McMillin, R.W.
Conners, and H.A. Huber. 1988. Automated computer
grading of hardwood lumber. Forest Products Journal
38(3): 67-69.
89
Li, P., A.L. Abbott, and D.L. Schmoldt. 1996. Automated
analysis of ct images for the inspection of hardwood logs.
In Proceedings of the 1996 IEEE International Conference on Neural Networks. Institute for Electrical and
Electronics Engineers, Inc., Piscataway NJ.
Schmoldt, D.L., J. He, and A.L. Abbott. 2000a. Automated
labeling of log features in ct imagery of hardwood species. Wood and Fiber Science 32(3): 287-300
Schmoldt, D.L., P. Li, and A.L. Abbott. 1997. Machine vision using artificial neural networks and 3d pixel neighborhoods. Computers and Electronics in Agriculture
16(3): 255-271.
Luppold, W., and J. Baumgras. 2000. The changing structure of the hardwood lumber industry with implications
on technology adoption. P. 89-94. In Proceedings of the
Twenty-Eighth Annual Hardwood Symposium, Meyer,
D.A. (ed.). National Hardwood Lumber Association,
Memphis TN.
Schmoldt, D.L., E. Scheinmann, A. Rinnhofer, and L.G.
Occeña. 2000b. Internal log scanning: Research to reality. P. 103-114 in Proceedings of the Twenty-Eighth Annual Hardwood Symposium, Meyer, D.A. (ed.). National
Hardwood Lumber Association, Memphis TN.
Malcolm, F.B. 1961. Effect of defect placement and taper
setout on lumber grade yields when sawing hardwood
logs. Report #2221. Madison WI: U.S. Department of
Agriculture, Forest Service, Forest Products Lab, 17p.
Schmoldt, D.L., D.P. Zhu, and R.W. Conners. 1993. Nondestructive evaluation of hardwood logs using automated
interpretation of ct images. P. 2257-2264 in Review of
progress in quantitative nondestructive evaluation, vol.
12, Thompson, D.O. and D.E. Chimenti (eds.). Plenum
Press, New York.
McMillin, C.W. 1982. Applications of automatic image
analysis to wood science. Wood Science 14(3): 97-105.
Occeña, L.G. 1991. Computer integrated manufacturing issues related to the hardwood log sawmill. Journal of Forest Engineering 3(1): 39-45.
Shadbolt, P.A. 1988. Some aspects of non-destructive testing using computerised tomography. M.S. Thesis. Department of Applied Physics, Chisholm Institute of Technology.
Occeña, L.G. 1992. Hardwood log breakdown decision automation. Wood and Fiber Science 24(2): 181-188.
Som, S., P. Wells, and J. Davis. 1992. Automated feature
extraction of wood from tomographic images. in Second
International Conference on Automation, Robotics and
Computer Vision.
Occeña, L.G., D.L. Schmoldt, and S. Thawornwong. 1997.
Using internal defect information for log breakdown. P.
63-68 in ScanPro: Advanced Technology for Sawmilling,
Dennig, J. (ed.). Miller-Freeman, San Francisco.
Steele, P.H., T.E.G. Harless, F.G. Wagner, L. Kumar, F.W.
Taylor. 1994. Increased lumber value from optimum orientation of internal defects with respect to sawing patterns in hardwood sawing. Forest Products Journal
44(3):69-72.
Taylor, F.W., J. F. G. Wagner, C.W. McMillin, I.L. Morgan,
and F.F. Hopkins. 1984. Locating knots by industrial tomography - a feasibility study. Forest Products Journa l
34(5): 42-46.
Onoe, M., J.W. Tsao, H. Yamada, H. Nakamura, J. Kogura,
H. Kawamura, and M. Yoshimatsu. 1984. Computed tomography for measuring the annual rings of a live tree.
Nuclear Instruments and Methods in Physics Research
221(1): 213-220.
Pickens, J.B. 1996. Computerized hardwood log bucking.
in Proceedings of the 1996 Hardwood Research Symposium, Meyer, D. (ed.). National Hardwood Lumber Association, Memphis TN.
Tsolakides, J.A. 1969. A simulation model for log yield study.
Forest Products Journal 19(7): 21-26.
Richards, D.B., W.K. Adkins, H. Hallock, and E.H. Bulgrin.
1980. Lumber value from computerized simulation of
hardwood log sawing. USDA For. Serv. Res. Pap. FPL356. 10 p.
Wagner, F.G., T.E.G. Harless, P.H. Steele, F.W. Taylor, V.
Yadama, and C.W. McMillin. 1990. Potential benefits of
internal-log scanning. P. 77-88 in Proceedings of Process Control/Production Management of Wood Products:
Technology for the 90’s. The University of Georgia, Athens GA.
Roder, F. 1989. High speed ct scanning of logs. In 3rd International Conference on Scanning Technology in
Sawmilling, Szymani, R. (ed.). Forest Industries/World
Wood, San Francisco CA.
Zhu, D., R.W. Conners, F.M. Lamb, D.L. Schmoldt, and P.A.
Araman. 1991a. A computer vision system for locating
and identifying internal log defects using ct imagery. in
4th Internation Conference of Scanning Technology in
Sawmilling, Szymani, R. (ed.). Forest Industries/World
Wood, San Francisco CA.
Schmoldt, D.L., J. He, and A.L. Abbott. 1998. A comparison of several artificial neural network classifiers for ct
images of hardwood logs. P. 34-43 in Photonics West
1998, vol. 3306. SPIE.
90
Zhu, D., R.W. Conners, D.L. Schmoldt, and P.A. Araman.
1991b. Ct image sequence analysis for object recognition — a rule-based 3-d computer vision system. P. 173178 in 1991 IEEE International Conference on Systems,
Man, and Cybernetics.
Zhu, D., R.W. Conners, D.L. Schmoldt, and P.A. Araman.
1996. A prototype vision system for analyzing ct imagery of hardwood logs. IEEE Transactions on Systems,
Man, and Cybernetics 26(4): 522-532.
91
Chapter 10
Applications of an Automated Stem Measurer for
Precision Forestry
NEIL CLARK
Abstract—Accurate stem measurements are required for the determination of many silvicultural prescriptions, i.e., what are
we going to do with a stand of trees. This would only be amplified in a precision forestry context. Many methods have been
proposed for optimal ways to evaluate stems for a variety of characteristics. These methods usually involve the acquisition of total
stem measurements, which are expensive and difficult to collect. A video rangefinder instrument is presented here as a move
toward efficient collection of this total stem data used on a per stem or precision basis for fertilization, thinning, harvesting, and
merchandising decision-making.
INTRODUCTION
heads-up display that indicates the exact location of the laser pulse and the ranging information. Video data is output
to a portable video cassette recorder via a standard video
cable. The recorder has an IEEE-1394 “i.LINK” interface
which would allow additional information (i.e., the range
and orientation data) to be written synchronously with the
video to the tape, though the camera is not equipped to do
this at this time. Currently, the range and orientation data
are output to a memory card, which is later read into the
processing program “Tree Measurement System” (TMS) as
a separate file.
The procedure to extract the desired information from the
data currently requires the use of a desktop computer. The
TMS program synchronizes the range and orientation data
from the memory card with the video data from the videotape. Currently, manual input is required to extract frames
and determine stem edges, etc. The frame extraction and
image processing components are being developed to allow
automation of information output (e.g., point & click diameter/height/volume measurements).
A previous study (Clark et al. 2001) shows that this instrument performs comparably to standard optical methods
for determining stem volume and height. This investigation
compared diameter, height, and volume measurements for
20 hardwood and 20 softwood trees using optical calipers
and aluminum height poles, conventional caliper and tape
measurements on the felled tree, and the TMS instrument.
The TMS instrument did not perform as well for individual
diameters, however investigations are underway to improve
performance.
Field data collection time was reduced with the TMS instrument. One person using the TMS instrument captured
the required data in approximately the same length of time
as 4 crews of 4 people using the optical calipers and alumi-
The goal of precision forestry is to enhance value,
assumedly where the benefits gained outweigh the costs of
management. This is a complicated task as the many public
and private entities which own, manage, and regulate forest
operations often have different management objectives and
constraints. A common denominator among all of these entities is the need for accurate and thorough information on
which to base, prioritize, assess, and defend management
decisions. The increase in knowledge about the impacts of
forest operations and the increasing demand for forest products are intensifying the level at which we must know and
manage this valuable resource. Forest operations from fertilization to thinning to harvesting and merchandising all rely
on accurate measurement of individual tree stems.
Tree heights, upper-stem diameters, and lengths and form
of bole sections are some of the many variables needed for
the more advanced models and assessment procedures available today. The acquisition of many individual tree measurements beyond DBH (diameter at breast height) is a time
consuming endeavor. For this reason new instruments are
needed to provide this detailed information in a very rapid
and accurate manner. Here an instrument for collecting dimensional information very quickly is presented and its potential application for precision forestry is shown. It is only
a prototype instrument at this point and more work is needed
to make it ready for regular field use, but it provides a very
important link to the applications involved in precision forestry.
The TMS Instrument
An instrument has been developed incorporating a standard format video camera, a 3-axis magnetometer (to measure instrument orientation), and a laser-rangefinder (table
1, figure 1). The camera is positioned to record through the
93
height poles, and one crew of 4 people using the felled tree
methods. The time is only given as an approximation as the
measurements were taken on the same trees on the same
days, so bottlenecks did occur using some methods. For
instance, the felled tree method crew had to be last for obvious reasons.
arrives, these particular
Figure 1. TMS instrument
stems can be gathered for
on monopod with portable
delivery.
DV videocassette recorder.
There have been a number of programs written for
predicting local timber
product yield and optimizing the utilization of each
stem (Olsen et al. 1991).
(Use
of
BUCK®
tradenames for informational purposes only and
does not constitute endorsement by the USDA Forest
Service,
http://
www.forestyield.com/) is an
example of a program that
generates local timber product yield tables based on
user-defined stem inputs and
product specifications. HWBUCK (Pickens et al. 1993)
is a computer simulation
program designed to be used
as a training tool to increase
value recovery and optimal
bucking of hardwood logs.
There are other similar programs that have been developed for a variety of species and markets that offer these
optimization functions (e.g., MARVL (Deadman and
Goulding 1979, Firth et al. 2000) available for Pinus radiata D. Don). These models require detailed stem data in
order to generate beneficial output. The TMS instrument
can provide this detailed data.
POTENTIAL APPLICATIONS FOR
PRECISION FORESTRY
Precision forestry will reduce management units from the
stand level to small groups of trees or even individual stems.
Directed treatments on these smaller units will aid in meeting multiple objectives and in better utilization. There have
been many forest-related analysis tools developed for computers in the last few years for inventory, visualization, modeling, decision support, and simulation.
Volume and Merchandising
Many studies have shown areas where potential improvements can be made for volume estimation. Wiant et al. (1992)
state that errors as great as 30% can occur if inappropriate
volume tables or functions are used. Goulding (1979) has
shown that when using widely spaced upper stem diameter
measurements volume estimate errors can still exceed 8%.
Importance sampling (Gregoire et al. 1986 ) and its cousins
centroid (Wood et al. 1990) and critical height (Van Deusen
1990) have been underutilized due to difficulty of field implementation. The height-accumulation method (Grosenbaugh
1954) has been found to be reliable, yet time consuming by
some based on the need to obtain multiple upper-stem measurements (Ferguson and Baldwin 1995). The TMS instrument can efficiently collect data for the entire stem allowing
the implementation of any or all of these methods.
As long as bias is not present, individual stem errors have
minimal impacts in a large-area inventory, when many stems
growing in varying environments are grouped together. In a
precision forestry context, however, errors of individual
stems are a great concern. Moreover, the parameter of interest may not be only total volume, but volume by product
class. Figure 2 demonstrates our limited ability to adequately
model a stem with existing profile equations. Generally,
stem form is omitted from stand inventories. At most, an
individual stem may be categorized into a particular product class. The efficient data collection made possible by the
TMS instrument will allow better methods of analysis to be
realized. It may also extend the boundaries for new types of
evaluation, especially in the area of stem form, product yield,
and merchantability to multiple markets.
Stem Mapping
Stem mapping is useful for relocating a particular tree
for remeasurement or spatial analysis. Other precision forestry developments such as Radio Frequency Identification
(Ringstad et al. 2001) would likely require stem location as
an attribute. Obtaining a GPS point at each stem in a stand
would be very time intensive. Stem mapping using a tape
and compass is quite time consuming and frustrating, especially in thick understory. Sonic devices (Doyle 1994) as
well as laser rangefinders (Peet et al. 1997) have been used
with varying levels of success to attempt to speed up the
process of stem mapping. Although any rangefinder would
be adequate for this task, the issue involved in this entire
precision forestry endeavor is efficiency. The TMS instrument can capture the distance and azimuth digitally which
can then be automatically tied to a coordinate position and
individual tree record.
Stem Form / Products
Current technologies and resource demands are creating
the need for more detailed data about individual trees.
Goulding (2000) presents an idea of a pre-harvest inventory
where stems are evaluated for product yield while still standing. The information along with the locations of these stems
are available so that when an order for a certain product
Modeling
Not only are the data collected with the TMS instrument
useful for evaluation of what currently exists, these data can
94
Table 1. Technical Specifications of the TMS instrument.
Dimensions
Weight
21.5 x 11.5 x 19 cm
2.1 kg
Distance (no reflector)
Azimuth
Range 2 – 610 m
Range 0.0º – 359.0º
Accuracy
Inclination
CCD Camera
Range ±50º *
480 x 720 RGB color
Accuracy
*
±15.3 cm
Accuracy
±1.5º RMS
±0.4º
Mounted at a 30º incline to cover a range from -20º to +80º respective to horizontal
Figure 2. The model approximations (left) that are currently used are inadequate to meet the demands for merchandising optimization. Data collected with the video system will provide the stem data needed to optimize product output.
95
also be inputs for predictive modeling. A survey of growth
and yield modeling over the last 17 years (Baldwin & Cao
1999) has revealed that the development and sophistication
of these models has closely followed that of computers and
software. And while few of these models have been packaged for convenient delivery to users, there has been a progressive step to incorporate some of these models into decision support and expert systems. Surveys of some current systems are presented in Reynolds et al. (1999) and
Rauscher (1999).
These predictive models also extend to biomass (Means
et al. 1994), fire (Cook 2001), and silvicultural modeling
(Dean 1999). Public awareness and integration into decision-making processes have lead to the creation of landscape visualization software. Programs such as the Landscape Management System (McCarter 1997) and Envision
(McGaughey 2001) have been developed to assist in stand
and landscape-level and forest ecosystem analysis and planning. Multi-resource inventories (Lund 1998) are now
being realized in some situations where multiple entities
cooperate in data collection efforts to share the labor while
meeting the requirements of each group. Nearly all of these
models require data from individual stems in the form of
location, density, crown ratio, volume, stand structure, etc.
An instrument such as the one presented in this paper will
be necessary to efficiently collect the massive amount of
required data inputs.
dendrometer. Developing this visual context and verification for the TMS instrument may be difficult. These obstacles need to be addressed before the instrument is truly
deemed field-ready.
Data Distribution
It is great to have large amounts of data and a suite of
analysis tools available, but if they are not organized and
accessible, they won’t be used (Martin et al. 2000). A big
effort in this area is web-based data distribution for custom
user analysis. This may present more problems in assuring
that metadata are also reported and warning the user of the
validity of the data for their analysis. This issue trickles
down through the entire sampling design. An issue to consider with the TMS instrument is the lowest level of data or
information to retain and make accessible to the overall system. In the usual case, individual diameters or height measurements are the lowest level of individual stem data recorded. These data are processed and aggregated into informational units (e.g., volume of pine sawtimber by county)
deemed to be the finest level of detail desired or statistically
viable to present to users. The lowest level of data collected
with the TMS instrument is a matrix of brightness values,
time (intrinsic with data streaming), range, and orientation.
This massive amount of data requires storage and an organized mode of access. For each project, a data model considering the lowest level of data (frames, range and orientation data, or a compressed mosaic of the entire stem) or information (derived diameters and heights) and its organizational structure would need to be determined to avoid being
overloaded with data.
DISCUSSION
Though there are some really exciting possibilities in
the integration of all of these tools and technologies for
more timely and cost-efficient analysis, there are some potential challenges as well. Any time that new methods are
developed and introduced there are always some new obstacles that are often not anticipated. This is generally the
cause of the queasy feeling that the production manager
gets when attempting a “better” way to accomplish the task
at hand. One solution is to stick with the tried-and-true.
There is a higher confidence of success, yet the return is
not likely to satisfy the expectations of the investor. Here
are a few issues that can be foreseen with this camera system.
Tackling the Precision Issue
Sampling has been developed and used to estimate a parameter that is impossible to measure directly. Usually there
is a compromise between sampling and measurement error.
Foresters have been accustomed to being very concerned
about measurement error for a number of sample elements
(i.e., DBH) and tolerating a huge model error when predicting the population parameter (i.e., volume). What digital
devices lose to analogue instruments in measurement precision, they make up for in number of measurements collected.
So an assessment should be made to evaluate the performance of more precise element measurements used with a
less precise model, versus a greater number of less precise
observations used with a more precise model.
Field Issues
Currently the TMS instrument is ruggedized, however
the video cassette recorder used to store the camera data
may not withstand much abuse. Once resolved, on-board
processing and data reduction can eliminate the need for
this external storage. The need for portable electric power
is also inconvenient. The present configuration allows 1.5
hours of use per recharge for the instrument. The video
cassette recorder has a separate rechargeable battery that
has 1 to 3 times that duration. And finally, the camera
records incident radiation within the visible range so light
conditions and occlusions can have unpredictable effects
on the results. A forester will usually confirm that visual
contact is sufficient when using some other optical
Automation
The great advantage of the system being digital from start
to finish is capability for the automation of results. Just by
pulling a trigger and scanning a tree, the volume can be estimated and made available for analysis. This will eliminate
blunder errors caused by transcription. Verification is an
issue with automated processes. Unless the system is robust to the point of being foolproof, a verification step should
be introduced to safeguard against collection blunder errors.
If this can be accomplished, the time savings presented by
96
rapid collection and automation of information extraction
should greatly facilitate the usefulness of improved models
and analysis tools.
Firth, J., R. Brownlie, and W. Carson. 2000. Accurate stem
measurements key to new image based system. New
Zealand Journal of Forestry. 45(2):25-29.
CONCLUSION
Goulding, C.J. 1979. Cubic Spline Curves and Calculation
of Volume of Sectionally Measured Trees. N.Z. J. For.
Sci. 9(1):89-99.
The future of forest management is going to increase in
complexity as the land base decreases, values diversify, and
regulations increase. Fortunately, many useful tools, such
as merchandising optimization programs and decision support systems, are becoming available to assist managers in
dealing with this complexity. In order for these tools to be
useful in a precision forestry context, large quantities of
detailed data will be required at rapid rates. The TMS system, presented here, can currently be used to rapidly acquire
heights and diameters for an entire tree bole. Furthermore,
the imaging nature of the collected data allows more intensive variables such as stem form and some defect information to be assessed. With some additional work to make
this instrument field-ready, adding some verification procedures, and smoothing the data flow, these data can be used
in many applications for precision forestry.
Goulding, C.J. 2000. The forest as a warehouse. P. 276282 In Hansen, Mark; Burk, Thomas, eds. Integrated tools
for natural resources inventories in the 21st century: an
international conference on the inventory and monitoring of forested ecosystems; 1998 August 16-19; Boise,
ID. Gen. Tech. Rep. NCRS-212. St. Paul, MN: U.S. Department of Agriculture, Forest Service, North Central
Research Station.
Gregoire, T.G., Valentine, H.T. and Furnival, G.M. 1986.
Estimation of bole volume by importance sampling. Can.
J. For. Res. 16:554-557.
Grosenbaugh, L.R. 1954. New tree-measurement concepts:
height accumulation, giant tree, taper and shape. USDA
Forest Service Southern Forest Experiment Station, New
Orleans, LA. Occasional Paper 134. 32 p.
LITERATURE CITED
Lund, H. Gyde. ed. 1998. IUFRO Guidelines for Designing
Multiple Resource Inventories. IUFRO World Service
Vol. 8. Vienna, Austria: International Union of Forestry
Research Organizations. 216 p. Baldwin, V.C., Jr., and Q.V. Cao. 1999. Modeling Forest
Timber Productivity in the South: Where Are We Today? P. 487-496. In Tenth Biennial Southern Silvicultural Research Conference, Shreveport, LA, February 16,
1999.
McCarter, J.B. 1997. Integrating forest inventory, growth and
yield, and computer visualization into a landscape management system. P. 159-167 in Teck, R., M. Moeur, and
J. Adams (comps.), In Proceedings of the Forest Vegetation Simulator conference. Gen. Tech. Rep. INT-GTR373. Ogden, UT. USDA Forest Service, Intermountain
Research Station.
Clark, N., S. Zarnoch, A. Clark III and G. Reams. 2001.
Comparison of standing volume estimates using optical
dendrometers. (In Press) In 2nd Annual FIA Symposium, Salt Lake City, UT, November 2000.
Cook, W. 2001. http://www.fire.org/
McGaughey, R.J. 2001. http://forsys.cfr.washington.edu/
svs.html
Deadman, M.W., and C.J. Goulding. 1979. A method for
assessment of recoverable volume by log types. New
Zealand Journal of Forestry Science. 8(2):225-239.
Martin, F.C., T. Baggett, T. Wolfe, and R. Mita. 2000. Using sampling theory as the basis for a conceptual data
model. P. 464-470. In Hansen, Mark; Burk, Thomas,
eds. Integrated tools for natural resources inventories in
the 21st century: an international conference on the inventory and monitoring of forested ecosystems; 1998
August 16-19; Boise, ID. Gen. Tech. Rep. NCRS-212.
St. Paul, MN: U.S. Department of Agriculture, Forest
Service, North Central Research Station.
Dean, T.J. 1999. Using live-crown ratio to control wood
quality: an example of quantitative silviculture. P. 511514. In Tenth Biennial Southern Silvicultural Research
Conference. Shreveport, LA, February, 16, 1999.
Doyle, T.W. 1994. Sonic distance device improves field sampling efficiency and accuracy. National Biological Survey Research Information Bulletin 79. 3 pp.
Means, Joseph E., Heather A. Hansen, Greg J. Koerper, Paul
B. Alaback, Mark W. Klopsch. 1994. Software for computing plant biomass—BIOPAK users guide. Gen. Tech.
Rep. PNW-GTR-340. Portland, OR: U.S. Department of
Agriculture, Forest Service, Pacific Northwest Research
Station. 180 p.
Ferguson, R.B., and V.C. Baldwin, Jr. 1995. A comparison
of height-accumulation and volume-equation methods for
estimating tree and stand volumes. USDA Forest Service Southern Forest Experiment Station, New Orleans,
LA. Research Note SO-378.
97
Olsen, E., S. Pilkerton, and J. Garland. 1991. Evaluating
timber sale bids using optimum bucking technology. Applied Engineering in Agriculture. 7(1):131-136.
Ringstad, M., T. Mohr, J. Tryall 2001. RFID Tagging of
Seedlings. http://www.cfr.washington.edu/research.pfc/
presentations/rfid_seedlings/
Peet, F.G., D.J. Morrison, and K.W. Pellow. 1997. Using a
hand-held electronic laser-based survey instrument for
stem mapping. Can. J. For. Res. 27:2104-2108.
Van Deusen, P.C. 1990. Critical height versus importance
sampling for log volume: does critical height prevail?
For. Sci. 36(4):930-938.
Rauscher, H.M. 1999. Ecosystem management decision
support for federal forests in the United States: a review.
Forest Ecology and Management. 114: 173-197.
Wiant, Jr., H.V., G.B. Wood and T.G. Gregoire. 1992. Practical guide for estimating the volume of a standing sample
tree using either importance or centroid sampling. For.
Ecol. Man. 49:333-339.
Pickens, J.B., G.W. Lyon, A. Lee, and W.E. Frayer. 1993.
HW-BUCK Game Improves Hardwood Bucking Skills.
Journal of Forestry. 91(8):42-45.
Wood, G.B., H.V. Wiant, Jr., R.J. Loy and J.A Miles. 1990.
Centroid sampling: a variant of importance sampling for
estimating the volume of sample trees of radiata pine.
For. Ecol. Man. 36:233-243.
Reynolds, K., J. Bjork, R. Hershey, D. Schmoldt, J. Payne,
S. King, L. DeCola, M. Twery and P. Cunningham. 1999.
Chapter 28: 687-721. In Decision Support for Ecosystem Management. Proceedings, Ecological Stewardship
Workshop.
98
Chapter 11
Using Ultrasound to Detect Defects in Trees:
Current Knowledge and Future Needs
THEODOR D. LEININGER
DANIEL L. SCHMOLDT
FRANK H. TAINTER
Abstract—Ultrasonic decay detectors (UDDs) have been available commercially for several years and can be used successfully to detect decay in live hardwood and conifer trees. Recently, a UDD has shown promise in detecting bacterial wetwood in
red oaks in the southern United States and in a Chilean hardwood species. Two improvements to the UDD tested would make it
more useful to the broader forestry community. Current UDDs only measure ultrasound signal time of flight (i.e., velocity) from
the transmitter to the receiver across the diameter of a tree. This measurement is insufficient to distinguish wood decay from a
void, or either of those from the cell wall degradation caused by bacterial wetwood. Further, in order to position ultrasonic
transducers in contact with the wood of a live tree for good signal propagation, a “wad” punch is used to create 5-cm diameter
holes in the bark. This process takes time and causes wounds that serve as entry points for pathogens and insects. An overview
is presented of the current effort to develop a UDD that records time-domain and frequency-domain waveforms that can be
positively linked to individual types of defects, and that uses smaller, pointed transducers to minimize tree wounds. Preliminary
results support current knowledge regarding ultrasound interactions with sound and unsound wood, and suggests that further
experimentation can lead to a new generation of UDDs.
INTRODUCTION
Significance of the problem
It is estimated that for every 100 million board feet of
timber harvested every year in the United States, heart decay fungi destroy about 30 million board feet of timber volume. Heart decay is thought to cause more than twice as
much timber volume loss as all other hardwood and conifer
diseases combined (Tainter and Baker 1996). Boyce (1961)
in his book on forest pathology stated that there are so many
decays caused by such a large number of wood-destroying
fungi that he could only briefly discuss the most important
ones. The greatest number of these fungi can decay dead
wood while a smaller number of fungi can extensively decay heartwood in living trees.
Practically all hardwoods and some conifers are susceptible to bacterial wetwood infections (Tainter and Baker
1996). Bacterial wetwood has a large economic affect on
the hardwood lumber and veneer industry causing problems
with color, drying, and machining and reducing overall quality and value (Ward 1982). In 1990, the value of hardwood
lumber at point of production in the eastern United States
was more than $3.5 billion (Murdoch 1992). Nearly 25% of
hardwood sawtimber in eastern forests is in the red oak species group. It has been estimated that bacterial wetwood
causes annual losses due to drying defects in oak lumber of
about 500 million board feet (Ross et al. 1995). Clearly,
with this much economic value at stake, the development of
Foresters need to identify and assess the extent of damage from wood decay fungi and bacteria in the wood of living trees because damage by these organisms affects subsequent processing for wood products. From a wood utilization perspective, it is also important to distinguish between
pathogen-caused wood degradation and macroscopic flaws
(e.g., splits). A forester who is aware of the types and extent
of pathogen-caused damage in a forest stand is better
equipped to prescribe silvicultural and other management
practices. Similarly, arborists faced with deciding whether
or not to remove a landscape tree would benefit from knowing the severity of heart decay or other types of lower stem
damage. The challenge to researchers and developers is to
provide foresters, arborists, and other forest health practitioners with an easy-to-use, field portable instrument that
will aid them in detecting different types of internal damage
in trees, while minimizing the need to wound or damage
subject trees in the process. An effort is underway to develop and field test a prototypical instrument that uses ultrasound to detect abnormalities in wood structure. Preliminary research findings are presented along with plans for
future research and development.
DISEASES AFFECTING HEARTWOOD
99
an instrument that can detect wood decay and bacterial
wetwood would greatly benefit forest managers of private
and public lands. Governmental and private timber sales
nationwide would benefit from more accurate volume estimates taking into account volume lost to diseases. It would
also be difficult to estimate the importance of this kind of
instrument to the arboriculture industry in terms of economic
value and job performance.
a tree only to a height of about eight feet above the ground.
Using an ultrasound decay detector in a fire-damaged stand
of bottomland hardwoods in the Delta National Forest near
Rolling Fork, MS, Leininger (unpublished data) verified the
presence of heartwood decay at least three feet higher on
the bole of a mature green ash (Fraxinus pennsylvanica
Marsh.) than was evident using the sounding technique. The
extent of the rot column was verified by felling the tree.
Heart Decay vs. Bacterial Wetwood
Examination Tools
Wood decay fungi cause heart rots in conifers and hardwoods by secreting extracellular enzymes, which break down
the structure of host cell walls. When decay is advanced,
the breakdown in cell structure usually causes wood tissue
to be discolored, soft, and crumbly. In the most severe cases,
decayed wood may have the consistency of pudding. In terms
of ultrasound wave conduction, the effect of wood in this
condition is similar to that of a void through which little
energy is transmitted.
Bacteria can cause a disease condition in both conifers
and hardwoods known as wetwood, or bacterial wetwood.
Unlike the total, or near total, destruction of cell wall integrity that occurs in the decay process, the cells in wetwoodaffected heartwood remain intact structurally but may be
separated from neighboring cells. This situation occurs because wetwood bacteria secrete extracellular enzymes that
degrade the middle lamellae and pit membranes of cell walls
while leaving the majority of the cell wall intact (Tainter
and Baker 1996). Wetwood-infected lumber that is dried
under a normal kiln-drying schedule typically develops defects in the wood such as ‘honeycombing’, the radial separation of cells along the rays, and ‘shake’, due to cell separation tangentially along the rings. The slipping or pulling
apart of neighboring cells from each other due to the enzymatic breakdown of cell wall structure causes these conditions.
A number of other methods that employ various instruments have been used to detect wood decay in the absence
of visible indicators; Dolwin et al. (1999) provided a review
of these. Portable hand or electric drills or steel rods can be
used to probe the extent of superficial decays and, to some
extent, the severity of decay based on resistance of the wood
to penetration by the probe, and on the color and consistency of wood shavings. An increment borer can be used to
extract a core of wood extending from the bark into the pith;
the core can be examined for signs of decayed, discolored,
water-soaked, and malodorous tissue, all of which are indicative of disease. Micro-drills, resistographs, fractometers,
and compression meters have been used with varying degrees of success to detect wood quality in live trees and all
have the same common drawback; they require use of a either a drill or an increment borer, both of which leave wounds
that can be invaded by pathogens and insects (Dolwin et al.
1999). Filer (1970) experimented with the transmission of
gamma radiation through stems of standing trees to detect
decay; Miller (1988) built a portable x-ray tomography device for utility pole inspection. Shigo and Shigo (1974)
pioneered the use of differences in electrical resistance between decayed and sound wood as a means of detecting decay. Gamma radiation equipment proved to be too bulky
for field use, and electrical resistance in trees is too variable
to provide a reliable measure of wood structural integrity
and requires wounding the tree by drilling holes for the electrical resistance probes.
DETECTING WOOD DECAY IN
LIVING TREES
Sonic Devices
Traditional Methods
McCracken and Vann (1983) were among the first to
present data showing that signal attenuation of sound waves
(100 Hz and 1 kHz) from vibrations pulsed through tree stems
was different for stems with heart rot than for those without. Using a commercial ultrasound instrument (James Vmeter), those investigators also found a direct relationship
between the diameters of healthy trees and the time of flight
(TOF) of ultrasound waves at 54 kHz and 150 kHz. Transit
times were unaffected by variations in moisture content or
specific gravity for eastern cottonwood (Populus deltoides
Bartr.), green ash, and willow oak (Quercus phellos L.).
However, wood decay increased apparent ultrasound TOF
in all three species, a finding that is the basis for ultrasound
detection of decay in trees. McCracken and Vann also observed that good contact between the tree and the
In the absence of visible indicators of disease such as:
open wounds; swellings; darkened, stained bark; or fruiting
bodies, it is not easy to determine the existence or extent of
a disease in a tree. Even when visible indicators are present,
foresters and arborists must estimate the internal radial and
longitudinal extent of an infection within a tree bole. This
estimation is often done using knowledge gained as a result
of having harvested similarly diseased trees or, alternatively,
by striking a tree, with a mallet or the back of a hatchet,
along the bole and around its circumference while listening
for a change in pitch to indicate the transition from sound
wood to decayed wood. This second method, called ‘sounding’, tends to underestimate the extent of decay because it is
not refined enough to distinguish between sound wood and
incipient decay, and because the average person can ‘sound’
100
Beall et al. 1994; Tiitta et al. 1999), but they did not demonstrate the ability of these parameters to detect bacterial
wetwood and other defects in living trees, especially hardwoods.
Using the ADD, heart decay is defined by a propagation time (in msec) that is three-fourths or more of the tree
diameter in millimeters. Sound wood is defined by a propagation time that is one-half or less the tree diameter in
millimeters. Based on this relationship, a propagation time
of 400 msec, or less, for an 800 mm diameter tree indicates sound wood. Whereas a propagation time of 600
msec, or more, indicates heart rot. A wave propagation
time within the 200 msec range between the 400- and 600msec thresholds could indicate incipient decay according
to the manufacturer of the ADD, and thus warrants further
examination. Xu et al. (2000) found that wave propagation times falling between those two thresholds in red oaks
in Mississippi and South Carolina were indicative of bacterial wetwood infections. In the following section, we
review past approaches to identifying wetwood using both
chemical/water relations and ultrasonic measurements.
signal transducers was
Figure 1. The Arborsonic
very important. They reDecay Detector can be readily
ported that reducing the
transported throughout the
bark to a thin layer greatly
forest as its portability
reduced TOF variability
permits quick and simple tree
inspection.
and signal attenuation, but
it also caused damaging
wounds to trees. These
two issues; the need to reduce signal attenuation
across bark and wood in
various states of health or
decay, and the related issue
of needing to minimize or
eliminate tree wounds during measurements, are two
important challenges to
successfully using this
technology.
Since the early 1990s,
ultrasound decay detectors
(UDDs) have been available to the arboriculture
community.
The
Arborsonic Decay Detector (ADD, Fujikura Europe
Limited, Wiltshire, England) is one implementation of this
type of device, which is compact and portable (Figure 1).
The ADD produces an ultrasound signal of 77 kHz, which
according to the manufacturer’s operational guide traverses
a tree of any species at a relatively constant speed of 2000
ms-1, up to a maximum diameter of 1.4 meters (Anon. 1995).
Dolwin et al. (1999) reported variation in the actual signal
velocity from about 1600 to 2000 ms-1, and that signal attenuation occurs in trees greater than 1 m diameter. The
ADD was designed specifically to detect heart decay in trees.
Because sound is an elastic wave, any disruption of the
wood structure (e.g., decay, bacterial infection, or other flaw)
that is different from sound wood will affect ultrasonic signal propagation. This phenomenon can be seen by comparing the signal strength of a waveform received for wood
containing a defect to the greater signal strength of the waveform for sound wood (Figure 2). While most types of wood
defects affect ultrasonic signal strength (Kabir et al. 2000),
the current ADD only measures signal time of flight. The
ADD developers reasoned that any decay present in the signal path would either cause the signal to take a longer path
to the receiver using good wood for propagation (greater
time of flight), or cause the signal to weaken to the point
that it could not be detected above a preset threshold (a
“timed-out” condition). The problem with this approach is
that there is no way to determine if a “time-out” condition
was caused by disease, a void, or if there was bad transducer-to-wood contact. Recognizing this problem, Beall and
others used ultrasound time and frequency domain parameters to detect decay and other defects in softwood utility
poles and round wood sections of softwood (Beall 1996;
DETECTING WETWOOD IN
LIVING TREES
Chemical Methods
Wetwood bacteria produce acetic acid and other fatty
acids that have a distinct rancid, vinegar-like odor (Schink
and Ward 1984). The easiest and most reliable way to
discern bacterial wetwood outside the laboratory is by sniffing to detect these odors at their source; for example, increment cores, log ends, or milled lumber from freshly cut
trees. This is problematic, however, as different individuals vary in their ability to detect odors.
Increment coring, which is inconvenient and damages
the tree, is the only way to detect bacterial wetwood in
standing trees while in the field. An effort was made to
develop an enzyme-linked immunoassay to detect wetwood
bacteria, but a working system is still years away
(McElreath et al. 1995, 1997a and b, 1998). Field and
laboratory studies have revealed certain characteristics
about the chemical, physical, mechanical, and drying properties of bacterial wetwood in red oaks. Methane concentration was the best indicator of wetwood in living red
oak trees sampled in Mississippi, South Carolina, and
Florida, whereas other associations with wetwood—
greater concentrations of acetic acid, and total K+, and
lesser concentrations of nonstructural carbohydrates—depended on the severity of wetwood rather than occurrence
alone (Xu et al. 2001a). In a related study, Xu et al. (2001b)
found that wetwood-affected red oaks in Mississippi and
South Carolina were characterized by greater moisture
content, abnormally high radial and tangential shrinkage,
and lower tension strength perpendicular to the grain compared to wood from healthy red oaks. In addition, moisture content of increment cores was a good indicator of
wetwood in the heartwood of sampled red oaks.
101
Figure 2. Normalized signal strength (volts) is plotted against time (µsec) for an ultrasonic signal propagating through wood with or
without a defect. The time of flight is relatively unaffected by the defect, but there is substantial energy loss.
1.50E+00
1.00E+00
With Defect
Without Defect
Volt (V)
5.00E-01
0.00E+00
-5.00E-01
-1.00E+00
-1.50E+00
0.00E+00
1.00E+02
2.00E+02
3.00E+02
4.00E+02
5.00E+02
6.00E+02
Time (µs)
Figure 3. Graph of linear regression lines of ADD readings for trees in Mississippi that were healthy ( #), had bacterial wetwood (+), or
had decayed heartwood (· ).
Y=5.89428X
(Wetwood)
Y=6.69876X
(Decayed)
Y=5.57547X
(Healthy)
102
of lines for wetwood-infected and healthy trees (P=0.0607),
although the regression lines are close to being significantly
different. When heartwood-decayed trees are included with
wetwood trees, there is a significant (P=0.0001) difference
between slopes of regression lines of this combined group
and healthy trees.
There is no significant difference between any of the regression lines relating ultrasound propagation times to bole
diameters for healthy, wetwood-infected, and heartwooddecayed trees in South Carolina (Figure 4). However, when
heartwood-decayed trees are combined with wetwood trees,
there is a significant difference (P=0.0241) between that
combined group and the healthy trees.
For the ADD readings of Chilean tepa logs, there was a
significant difference between the slopes of the two regression lines (P=0.0022), one representing unstained logs, the
other representing logs with butterfly stain (Figure 5). Significance between the slope coefficients of the two lines was
tested using the dummy variable approach presented by
Gujarati (1995). The relationship between propagation time
readings and log diameter was very strong for unstained logs
(R2=0.97) and for logs with stain (R2=0.97). Because both
lines were forced through the origin (0,0), the lines are close
together for small diameters and diverge as log diameters
increase. The significant difference between the slopes of
these two lines indicates that the ADD shows promise for
detecting butterfly stain in tepa logs (or trees), especially
for larger diameters.
Development of an ultrasonic decay detector that can
distinguish between incipient decay, voids, and bacterial
wetwood, in addition to detecting advanced decay, would
be a big improvement over existing devices. Two other
major drawbacks of the ADD are, first, that two, 5-cm
diameter wounds to the xylem are required for each
ultrasound reading; and, secondly the ADD is engineered
to detect a signal with a threshold amplitude, thus providing
limited electronic information even though more signal
information may be available.
Figure 4. Graph of linear regression lines of ADD readings for
trees in South Carolina that were healthy ( #), had bacterial
wetwood (+), or had decayed heartwood (·).
However, the potential of these variables to be wetwood
indicators depends on the severity of the wetwood infection
and involves destructive sampling. Increment cores to the
heartwood might have to be removed from several places
on the butt log of a large oak to adequately sample for
wetwood. This procedure is time consuming, injures the
tree, produces entry points for other pathogens, and reduces
the dollar value of the log once it is harvested.
ADD Inspection
Xu et al. (2000) and Tainter et al. (1999) used the ADD
to detect bacterial wetwood infections in red oaks (Quercus
spp.) and the Chilean hardwood tepa (Laureliopsis
philippiana [Looser] Schodde), respectively. In the red oak
study, ultrasound detection of wetwood and heart decay was
compared. Healthy, heart-decayed, and wetwood-infected
willow and nuttall (Quercus nuttallii E.J. Palmer) oaks were
examined in Mississippi, while in South Carolina, northern
red oaks (Q. rubra L.), southern red oaks (Q. falcata Michx.),
black oaks (Q. velutina Lam.), and scarlet oaks (Q. coccinea
Munchh.) were tested. A complete description of the experimental approach and methods is given in the original
report (Xu et al. 2000).
Regression lines relating ultrasound propagation times
to bole diameters for healthy, wetwood-infected, and heartwood-decayed sample trees in Mississippi are graphed in
Figure 3. The slopes of the lines of heartwood-decayed trees
differ from those for wetwood trees (P=0.0001) and healthy
trees (P=0.0001). There is no difference between the slopes
IMPROVING ULTRASOUND
DETECTION OF INTERNAL DAMAGE
Results to date with the ADD are encouraging. However, the use of TOF-diameter relationships (effectively velocity) lacks discriminating ability (substantial overlap exists in the data points) and seems to be somewhat species
dependent, as is evident when multiple species are included
in the analyses. With the current stage of ADD technology,
a particular TOF value for a tree with a particular diameter
cannot reliably predict which of several internal defects are
present. We have concluded that ultrasound velocity measurements alone are unlikely to provide sufficient information to overcome these problems.
Related work on inspection of pallet parts suggests that
we should be able to distinguish different types of internal
tree damage using ultrasonic measurements. Using freshcut (green) pallet deckboards of northern red oak and yel103
of the specimen simulated the void sample. Because pallet
parts are cut from green cants, the moisture content is similar to that for live trees. Ultrasonic measurements were
made in a pitch-catch arrangement from face to face, primarily a radial direction—similar to diameter measurements
on standing trees. Measurement regions of the stringers
were selected such that ultrasound transmissions traversed
both sound and unsound wood, similar to transmission
through good sapwood and defected heartwood in a standing tree. Aside from destructively sampling trees for defects of interest and conducting ultrasonic tests on those
specimens, these “simulated tree” tests closely approximate
the essential characteristics of actual measurements on standing trees without any bark-related attenuation problems.
An industrial prototype ultrasonic scanning system (Kabir
et al. 2001) was used to take measurements on the samples.
All measurements were carried out at 120 kHz transmitting
frequency and received signals were sampled at 500 kHz.
The transmit voltage and receiver gain were 130 V and -4
to -1 dB respectively. Several measurements were taken to
ensure that the captured signals were representative of the
internal condition being sampled.
Ultrasonic time-domain waveforms appear in Figure 6
for the four samples tested. Compared to sound wood, the
loss of internal wood integrity for decay, wetwood, and void
samples is evident in their greatly reduced signal strength.
Because signal strength as a function of time aggregates
the contributions from all frequencies, it provides limited
information. By applying a Fourier transformation (FFT)
from the time domain into the frequency domain, we can
obtain a more detailed picture of how wood characteristics
influence ultrasound propagation. Figure 7 displays the FFT
magnitude graphs for our samples. For each spectra, energy values peak at the transmit frequency of 120 kHz.
Again, as in the time-domain graphs, the sound wood sample
has the highest peak energy. The wetwood sample has an
intermediate peak energy value, which distinguishes it from
decay and void. The decay spectrum depicts greater signal
attenuation than the hole spectrum in frequencies above the
peak frequency; this is consistent with results on decayed
wood by other researchers (e.g., Halabe et al. 1995). There
are a variety of ultrasonic parameters that can be calculated
to quantitatively capture what we see in these graphs, and
that can then be used to distinguish one wood type from
another.
These tests are very preliminary and only make use of a
single test specimen in each case. Nevertheless, the results
are consistent with our knowledge of wood structure and
ultrasonic propagation in that medium and with prior wood
scanning results. If these ultrasound signatures can be reproduced consistently in other samples and other species,
they can form the basis for a new generation of UDD.
Figure 5. Graph showing the relationship between regression
lines of healthy (o) and butterfly-stained (·) tepa logs
low-poplar (Liriodendron tulipifera, L.), Kabir et al. (2001)
characterized a variety of defect types with ultrasound, including decay and voids. Several TOF, energy, and pulse
length variables were measured. While no single ultrasound
variable appears able to discriminate between all defect types,
some subset of those variables taken together should be able
to classify different defect types. Measurements in those
experiments, however, were taken directly on, or adjacent
to, particular defect types. This situation differs from heartwood defects in standing trees where internal degradation
is surrounded by clear wood regions in most cases. Furthermore, those pallet part experiments did not include wetwood
as a defect type. Consequently, a preliminary set of tests
(using larger pallet parts) was performed to illustrate how
ultrasonics might be used to distinguish between sound
wood, decay, wetwood, and voids in standing trees.
Pallet Stringer Tests
Three red oak pallet stringers and one red oak pallet
deckboard that included sound wood, advanced decay, a void,
or wetwood were collected from a local pallet manufacturer.
For the wetwood sample, a stringer was not available, so a
deckboard part was used. Based on the intensity of odor,
this wetwood sample was classified as moderate infection.
Pallet stringers tested were approximately 3.8 cm (1.5 inches)
thick; the deckboard was 1.6 cm (0.625 inch) thick. Drilling a 1.25 cm hole (0.5 inch) longitudinally into the center
Nondestructive Transducer Contact
Because tree bark provides a sound propagation interface (reflecting substantial energy) and uneven contact
104
Figure 6. Time-domain waveforms plot energy (volts) versus
time (µsec) for (a) sound wood, (b) advanced decay, (c) a hole,
and (d) wetwood. After reaching peak energy, signal strength
gradually decreases until the sampling window of 256 µsec is
exceeded.
a
Amplitude (V
Amplitude (V
1.00
Figure 7. Fourier transformation (FFT) graphs for the four
samples plot signal amplitude (magnitude of the FFT) versus
frequency (kHz): (a) sound wood, (b) advanced decay, (c) a
hole, and (d) wetwood. Energy is at a maximum near the
transmitted frequency of 120 kHz.
0.50
0.00
-0.50
-1.00
0
50
100
150
200
250
300
18
15
12
9
6
3
0
0
1.00
b
0.50
0.00
-0.50
0
50
100
150
200
250
300
Amplitude (V
Amplitude (V
b
0
c
0.50
0.00
-0.50
100
100
200
300
c
0
Amplitude (V
Amplitude (V
d
0.50
0.00
-0.50
50
100
150
300
100
200
300
Frequency (kHz)
1.00
0
200
18
15
12
9
6
3
0
Time(µs)
-1.00
300
Frequency (kHz)
1.00
0
200
18
15
12
9
6
3
0
Time(µs)
-1.00
100
Frequency (kHz)
Amplitude (V
Amplitude (V
Time(µs)
-1.00
a
200
250
300
18
15
12
9
6
3
0
d
0
Time(µs)
100
200
Frequency (kHz)
105
300
occur through transducers that are the most “tree friendly”
method available. Once a working prototype is developed,
meaningful statistics relating sample trees to the larger forest resource should be developed to provide the broadest
use for the new device. The final goal is to bring this improved technology into the marketplace as a field-portable,
affordable unit.
with transducers, there is often great signal attenuation. But
removal of bark and xylem is highly undesirable, as noted
above. Therefore, bark must be penetrated in some way to
achieve good signal transmission. For the ultrasound frequencies that we are interested in transmitting, relatively
large (2.5-4 cm diameter) cylindrical transducers are most
efficient. However, market needs have not driven any search
for alternative transducer designs. Recently, though, wood
scanning efforts by Perceptron, Inc. (Ultrasound Technology Group) have led to the development of a more conical
transducer with some bark penetration capability. These
transducers have a contacting face of about 1 cm in diameter, yet they operate in the desirable frequency range for
wood inspection. It is possible that these transducers could
be held in place by hand—or mounted on some sort of handheld device (perhaps using a scissoring caliper action)—
that would supply sufficient pressure for bark penetration
with negligible bark and xylem damage. Most thick-barked
trees have deep fissures in their bark that could provide adequately wide access points for such transducers. In addition to tests that determine ultrasound relationships for internal tree damage, we would need to verify the reliability
of this transducer-wood contacting mechanism across of
variety of species and bark characteristics.
LITERATURE CITED
Anonymous. 1995. Arborsonic Decay Detector operational
guide. Fujikura Europe Limited, Wiltshire, England. 14
p.
Beall, F.C. 1996. The use of acousto-ultrasonics for NDE
analysis of biodeterioration in wooden structural elements. P. 409-413. In Proceedings of the 3rd Conference
on Nondestructive Evaluation of Civil Structures and Materials, Schuller, M. and D. Woodham (eds.). Express
Press, Boulder, CO.
Beall, F.C., J.M. Biernacki, and R.L. Lemaster. 1994. The
use of acousto-ultrasonics to detect biodeterioration in
utility poles. J. Acoustic Emission 12(1/2):55-64.
CONCLUSIONS
Boyce, J.S. 1961. Forest pathology, third edition. McGrawHill Co., New York 572 p.
The operating principles inherent in the current generation of UDDs can only be marginally effective. While TOF
measurements are useful for interrogating standing tree quality, they are conceptually flawed and provide only ambiguous readings. When no signal is received across the stem, it
is impossible to distinguish between bad transducer contact
with a tree and internal damage. Furthermore, different types
of internal damage cannot be readily discriminated. Foresters and arborists need more diagnostic equipment to perform their jobs effectively.
Both past ultrasonic experiments on wood and preliminary tests reported here suggest that a more sophisticated
and robust device can be built. Furthermore, our preliminary tests suggest that it might be experimentally expedient
and advantageous to conduct additional laboratory tests using small samples as we have done here. After the ultrasonic relationships are better understood using laboratory
experiments, then more elaborate, destructive field tests
could be conducted on standing and felled trees. These fullscale tests would help fine tune the relationships identified
in the laboratory and would also address less idealized conditions, e.g., simultaneous occurrence of several types of
internal damage, transducer-to-wood contact and bark penetration.
Further research and development goals to improve upon
existing UDD technology should include designing the electronics to analyze important waveform patterns indicative
of advanced decay, incipient decay, bacterial wetwood, and
voids. Ultrasound signal transmission and reception should
Dolwin, J.A., D. Lonsdale, and J. Barnett. 1999. Detection
of decay in trees. Arboricultural J. 23(2):139-149.
Filer, T.H., Jr. 1970. Gamma radiation detects defects in trees
and logs. Phytopathology 62:756.
Gujarati, D.N. 1995. Basic Econometrics, second edition.
McGraw-Hill, New York. 838 p.
Halabe, U.B., H.V.S. GangaRao, and V.R. Hota. 1995. Nondestructive evaluation of wood using ultrasonic frequency
domain analysis. P. 1653-1660 in Review of Progress in
Quantitative Nondestructive Evaluation, vol. 14, Thompson, D.O. and D.E. Chimenti (eds.). Plenum Press, New
York.
Kabir, M.F., D.L. Schmoldt, and M.E. Schafer. 2000. Detection of defects in red oak deckboards by ultrasonic
scanning. P. 89-96 In Proceedings of the 4th International
Conference on Image Processing and Scanning of Wood,
Kline, D.E. and A.L. Abbott (eds.). USDA Forest Service, Southern Research Station, Asheville NC.
Kabir, M.F., D.L. Schmoldt, and M.E. Schafer. Time domain ultrasonic signal characterization for defects in
pallet deckboards. Wood and Fiber Science. In Press.
McCracken, F.I. and S.R. Vann. 1983. Sound can detect
106
Serv. Res. Pap. NE-294. 11 p.
decay in standing hardwood trees. USDA For. Serv. Res.
Pap. SO-195. 6 p.
Tainter, F.H. and F.A. Baker. 1996. Principles of forest pathology. John Wiley and Sons, Inc. New York. 805 p.
McElreath, S. D., T.D. Leininger, and F.H. Tainter. 1997a.
Development of a field test for detection of wetwood in
oak trees. P. 482-485. In Proc. of the SNA Research Conference, Forty-Second Annual Report 1997, James, B.L.
(ed.), Vol. 42. Southern Nurseryman=s Association, July
13-August 1, 1997, Atlanta, GA.
Tainter, F.H., T.D. Leininger, and J.G. Williams. 1999. Use
of the Arborsonic Decay Detector to detect butterfly stain
in Chilean tepa. Interciencia 24:201-204.
McElreath, S. D., T. D. Leininger, and F. H. Tainter. 1998.
Development of an ELISA for detection of an Erwinia
sp. in tree sap. Phytopathology 88(9):S60.
Tiitta, M., J.M. Biernacki, and F.C. Beall. 1999. Using spatial averaging and relative measurement techniques to
improve acousto-ultrasonic decay detection efficiency.
P. 185-194. s In Proceedings of the 11th Symposium on
Nondestructive Testing of Wood. Madison, WI.
McElreath, S. D., E. G. Porter, F. H. Tainter and T. D.
Leininger. 1995. Bacteria associated with wetwood in
southern oaks. Phytopathology 85(10):1196.
Ward, J.C. 1982. Bacterial oak and how to dry it. South.
Lumberman 243:8 10.
Xu, Z., Leininger, T.D., Lee, A.W., and Tainter, F.H. 2001a.
Chemical properties associated with bacterial wetwood
in red oaks. Wood and Fiber Sci. 33(1):76-83.
McElreath, S. D., F. H. Tainter, G. P. Birrenkott, L. K. Fulton,
H. E. Farris and T. D. Leininger. 1997b. Polyclonal antibodies to wetwood bacteria from egg yolk of immunized
hens. Phytopathology 87(6):S113-114.
Xu, Z., Leininger, T. D., Lee, A. W., and Tainter, F. H. 2001b.
Physical, mechanical and drying properties associated
with bacterial wetwood in red oaks. For. Pro. J. 51(3):7984.
Miller, W.H. 1988. Design and implementation of a wooden
pole inspection device based upon computerized axial
tomography. Nuclear Instruments and Methods in Physics Research 270(2/3): 590-597.
Xu, Z., Leininger, T.D., F.H. Tainter, and J.G. Williams. 2000.
Examination of the Arborsonic Decay Detector for detecting bacterial wetwood in red oaks. South. J. Appl.
For. 24:6-10.
Murdoch, C.W. 1992. Detection system to identify wetwood
in standing living trees and in cut logs and boards. Report prepared for the Technology Transfer Information
Center, National Agricultural Library, U. S. Dept. Agric.
and Nat. Inst. of Standards and Technology, U. S. Dept.
Commerce. ISSN 1964 3451. 16 p.
ACKNOWLEDGMENTS
The authors wish to recognize research contributions by
Zicai Xu and Jeffrey McLemore, Clemson University; field
assistance by Dexter Bland, Curt McCasland, and Nathan
Schiff, USDA Forest Service, Southern Research Station;
ultrasonic testing by M. Firoz Kabir, Virginia Tech; and administrative and field assistance by Ralph Pearce and Larry
Moore, USDA Forest Service, Delta National Forest and
Mark Monroe, Tony Parks, and John Hodges of AndersonTully Company. The use of trade or corporation names in
this publication is for reader information and does not imply endorsement by the U.S. Department of Agriculture of
any product or service to the exclusion of others that may be
suitable.
Ross, R.J., J.J. Fuller, and J.R. Dramm. 1995. Nondestructive evaluation of green defect-prone red oak lumber: a
pilot study. For. Prod. J. 45:51-52.
Schink, B. and J.C. Ward. 1984. Microaerobic and anaerobic bacterial activities involved in formation of wetwood
and discolored wood. International Assoc. Wood Anatomists Bull. 5(2):105-109.
Shigo, A.L. and A. Shigo. 1974. Detection of discoloration
and decay in living trees and utility poles. USDA For.
107
Chapter 12
An Automated Log Grading System
Based on Computed Tomography
TIM RAYNER
WALTER GARMS
ELAN SCHEINMAN
Abstract—An automated system for grading logs is an important element of forest resource management. Log grading ensures that each log is suitable for its intended purpose, reducing waste and aiding in conservation. Furthermore, log grading
enhances market opportunities for producers of logs by giving better insight into the product. Automation of the log grading
process is desirable because of the possibility for enhanced accuracy, efficiency, and consistency. InVision Technologies, Inc.
(InVision), has developed an automated log grading system based on computed tomography (CT). CT uses x-rays to produce
high-resolution cross-sectional images of the internal structure of a log. The CT images reveal the grain pattern, knots, and other
defects that vary in density from the surrounding wood and that are larger than the scanning resolution. The development of
InVision’s automated grading system was accomplished by examining field data from a sawmill where an industrial CT-based
log-scanning system has been installed. On a set of test logs, the automated CT grade agreed 80% of the time with the manually
assigned grade normally used at the mill. The automated CT grading process has improved accuracy over manual grading methods, since more information is available. The logs are not only graded based on external characteristics, but also on the detailed
CT images of their internal structure.
INTRODUCTION
information is made available. Further, the CT scan provides
high quality detailed images of the internal structure of the
log, allowing an indisputable grade can be attached to each
log. The grade can even be verified, for example, by viewing the CT images.
InVision has developed a system based on CT to automatically grade logs. Much of the development was done at
a sawmill in Austria where Invision has installed an industrial CT-based log-scanning system. The sawmill produces
finger lamellas for the appearance grade industry. A set of
test logs was graded using the mill’s usual manual grading
system that assigns grades based on observations of the external characteristics of the logs. CT scans were taken of
each log. Six candidate grading metrics under consideration
by InVision were applied to the CT image data to compute
the automated grade for each log. The best performing grading metric achieved an accuracy of 80%; in other words, the
automated grade agreed 80% of the time with the manually
assigned grade.
An automated system for grading logs is an important element of forest resource management. Log grading ensures
that each log is suitable for its intended purpose, reducing
waste and aiding in conservation. Furthermore, log grading
enhances market opportunities for producers of logs by giving better insight into the product. Automation of the log
grading process is desirable because of the possibility for
enhanced accuracy, efficiency, and consistency.
At present, sawmills and wood-product manufacturers
typically purchase logs from log dealers and distributors.
Since the value of a log is dependent on the nature and extent of defects in the log, log purchasers employ experts to
evaluate logs prior to purchase. The conventional evaluation methods require the physical inspection of the exterior
of the log in an effort to predict cracks, voids, decay, knots,
and other internal defects. Based on this inspection, the log
is then graded and its dollar value estimated.
One method of automating the grading of logs is by using
the imaging technology x-ray computed tomography (CT).
CT uses x-rays to produce high-resolution cross-sectional
images of the internal structure of a log. The CT images
reveal the grain pattern, knots, and other defects that vary in
density from the surrounding wood and that are larger than
the scanning resolution. The automated grading process has
an advantage over manual grading methods since internal
BACKGROUND ON X-RAY BASED CT
InVision manufactures of a family of CT scanners specifically designed for use in harsh industrial environments.
The CTX™ scanner is based on third generation CT that
creates geometrically accurate, cross-sectional plane images
through an object. The CT scanner acquires numerous x-ray
109
Figure 1. The basic concept of CT, where the image is produced by rotating an x-ray source and detector array around an object.
wood and that are larger than the scanning resolution are
visible. The image has a pixel resolution of 1.6mm X 1.6mm
and 16-bit contrast resolution.
Cross-sectional images can be taken at regular intervals
as the log moves through the scanner. The multiple images
enable the size, type, and placement of internal defects to be
determined. By stacking a series of CT cross-sectional images together in a three dimensional representation, the CTX
can simulate a transparent log, showing only internal defects.
Another way of arranging the CT images is to create a
virtual board face. The virtual board face can be a useful for
verifying the grade assigned to the log. The virtual board
face is computed by combining successive cross-sectional
images of the log.
Cross-sectional images taken from an alder (Alnus
Glutinosa) log at 10mm intervals were used to construct the
virtual board face in Figure 3. A photograph of the actual
cut board face is also shown in Figure 3, demonstrating the
accuracy of the virtual board face. False color is added to
the image to make it appear more lifelike; the color scale
assigns darker colors to the areas of higher density.
The sawmill operator can view the virtual board face im-
attenuation images as an x-ray camera is rotated around the
object; these images are then used to compute the final CT
cross-sectional images. A diagram that illustrates the scanning process is shown in Figure 1.
CT constructs the cross-sectional images based on computed material densities. The image contrast in a CT image
is determined by the different densities. Each density is
given a CT number. Water, at a density of 1.0 g cm-3, has a
CT number of 1000. Natural products such as wood have
densities in the range 0.4 to 1.0 g cm-3, which correspond to
CT numbers between 400 and 1000, respectively.
InVision’s automated log grading system is comprised
of the CTXTM scanner and a heavy-duty log handling system. A log is placed in the log handling system, which translates the log through the plane of the CT scanner for acquisition of the cross-sectional images.
An example of two CT cross-sectional images of a single
spruce (Picea Abies) log can be seen in Figure 2. Figure 2
clearly demonstrates the image resolution that can be obtained with an x-ray based CT image. Each image was acquired in about 1 second. The annular rings, and defects,
such as knots and occlusions can be clearly seen. Defects
that have a density that is different from the surrounding
Figure 2. Two CT cross-sectional images through a spruce (Picea Abies) log. Defects such as knots can be seen as well as occlusions.
110
Figure 3. An (Alnus Glutinosa) virtual board face (bottom)
constructed from a complete series of CT cross-sectional
images. A photograph of the corresponding actual board face
is shown on top.
Figure 4. The transformation of the 3D log image data into a
2D-image file.
φ
z
φ
360º
φ max
φ min
0
0
Z min
Z max
L
Z
Creation of a Defect Map from the CT Data
ages to place cuts in the log that maximize the grade and
volume of the product.
The automated grading metrics require information about
defects in a log in order to compute the grade of the log. A
defect map is computed from the CT data for this purpose.
Defects such as knots and splits are identified and localized.
The first step, in the creation of a defect map, is to transform the three-dimensional CT image data set into a twodimensional data set. The transformation reduces the complexity and computational overhead. The resulting two-dimensional data set is known as a “radial image” because it
is made up of the average density across a radial slice of the
log starting at the pith center and ending near the bark. In
order to reduce errors associated with mislabeling bark as a
defect, the bark itself is not included in the calculation of
average density. The two-dimensional map of a log is shown
in Figure 4. In the two-dimensional plot, the Z-axis is the
log length, and F is the angle as the log “tube” is unwound
and flattened.
THE DEVELOPMENT OF INVISION’S
AUTOMATED CT LOG GRADING
METRICS
Overview
There are a number of publications that describe the criteria for the assignment of log grades (see literature cited).
Invision faced the task of developing a computerized grading metric that translates CT image data of a log into a grade
for the log. Six candidate grading metrics were derived empirically. In order to test the candidate metrics for accuracy,
InVision utilized its CT log-scanning system at a sawmill in
Austria that processes Norway Spruce (Picea Abies) logs.
To test the six candidate metrics, a set of logs was graded
using the mill’s usual manual grading scheme that is based
on external information. CT images of the logs were then
acquired, and the six candidate log grading metrics were
applied to the data; the computed grades were compared to
the mill’s manually assigned grade to select the metrics with
the greatest accuracy.
The ultimate goal of the automated grading system is to
have the automated grade reflect the value of the final product, and to track the final product value better than the externally based grades currently in use. To make the comparison of the automated grades with the final value of a log, the
set of logs was sawn into boards; the boards were graded,
and board yield and final dollar values were calculated for
each log.
The sawmill produces finger lamellas for the appearance
grade industry. Grades are assigned to logs based on the
goal of producing as many defect free lamellas as possible
of certain required lengths.
There are four possible grade (see literature cited) assignments at the mill, A, AB, C, and D; the logs are graded based
on external characteristics. For the study, only higher quality logs of grades A and AB are considered, since greater
value can be extracted from these logs. The grading criteria
for grades A and AB are shown in Table 1.
Figure 5 . A knot bound by a “pie slice”, defined by radius r,
angle minimum and maximum and length Z.
Z
φ
Z max
φ min
φ max
Z min
111
• Defect’s location
- Z min, Z max
- φ min, φ max
-r
r
• Defect’s type
- Knot
- Void
• Defect’s size
(angle & length)
- “Big”
- “small”
Table 1. Log grading criteria.
Grading
Requirements for
Requirements for
Parameters
A39+ Quality Logs
AB46 Quality logs
Diameter
>39 cm
46 cm
Knots
1-2 knots
Knots arranged in a row, top diameter 40+ cm: regularly
arranged sound knots (separated by at least 50 – 60 cm)
Compression
Not tolerated
(reaction) wood
Maximum of 25% of the cross-section of the log (if no
other defects occur)
Resin pockets
Not tolerated
Maximum 1-2 (if no other defects occur)
Spiral grain
Not tolerated
Not tolerated
Discoloration
Not tolerated
Not tolerated
Fungus and
Not tolerated
Not tolerated
Not tolerated
Straight direction of grain, narrow ringed
insect attack
Grain
top diameter 40+ cm: at least 15-20 cm narrow ringed
Brown rot
Not tolerated
Top diameter 45+ cm: maximum 10-15% of the crosssection of the log (if no other defects occur)
Figure 6. A 2D segmented image of a spruce log showing the defects as black regions.
112
Radial slices containing defects are identified through
application of a numerical threshold to the densities. Since
densities are averaged across the radial slice, further computation is required to determine the extent of a defect,
once it is identified. Additional density profiles across the
radius of the log are computed. The defect can then be
bound within a “pie slice” polygon having a radius from
the center of the log, an angular spread, and a length, as
illustrated in Figure 5.
The defect regions are then grouped into sets of adjoining points. Those groups that do not have enough contiguous pixels to be classified as a significant defect are
ignored. The defect map is now complete showing information about the number, size, and position of defects.
Figure 6 shows the completed defect map of a CT scanned
log.
It is from this two-dimensional defect map that the grade
of the log is computed.
Figure 7. Distribution of the computed metric values for the
Score metric.
4.5
4
3.5
Frequency
3
2.5
Log Grade A
Log Grade AB
2
1.5
1
0.5
0
100
150
200
250
300
350
400
450
500
550
Computed Metric Values
Setting Threshold Values
Before the automated grades can be assigned to the individual logs, threshold values must be chosen for the six
metrics. In this study, the computed grades can be either
A39+ or AB46. If the computed metric value for a log is
less than the chosen threshold for that metric, the log graded
AB46; if the computed metric value is greater that the chosen threshold value the log is graded A39+.
The choice of threshold value is crucial to the automated
grading of the logs. The selection of the threshold values
for the six metrics was accomplished through statistical
analysis of the computed metric values in Tables 3 and 4 for
the set of 34 test logs.
Description of the Six Candidate Automated
Grade Metrics
The six candidate grade metrics were derived empirically, and compute grades based on the information in the
two-dimensional defect map. Since the sawmill where the
testing occurred produces lamellas, the grades also took
into consideration the number of lamellas that can be positioned while maximizing defect free length. A detailed description of the six candidate grade metrics is shown in
Table 2
.
Statistical Analysis of the Metric Values and
Threshold Determination
Acquisition of the CT Log Data
Plots were made, from the data in Tables 3 and 4, to show
the distribution of the computed metric values for each metric. In Figure 7, the distribution of the values computed for
the Score metric is shown; the distribution of the set of manually graded A39+ logs is shown in green, and the distribution of the manually graded AB46 logs is shown in brown.
Note that the two distributions overlap.
For each threshold value, estimates can be computed for
the probability of a correct log grade determination and the
probability of an incorrect log grade determination; here,
the term correct means that the automated grade assignment
agrees with the manual grade assignment. The area of overlap of the distributions in Figure 7 contributes to the probability of incorrectly grading a log, since the computed metric values in that area belong to both the AB46 and A39+
manually assigned grades.
A receiver operator characteristic (ROC) curve enables
the probabilities of correct and incorrect grading to be displayed for all the possible threshold values. The ROC curve
for a particular grading metric is computed from the corresponding distribution plots. The ROC curve for the automated grade metric Score is shown in Figure 8. The ROC
curve is a parametric plot of the probability of a correct log
grade assignment versus the probability of an incorrect log
grade assignment, as the threshold value varies. The thresh
A set of 34 spruce logs, graded with the mill’s manual
grading scheme was selected; five of the logs were of
manual grade A39+, and 29 of the logs were of manual
grade AB46. A log of grade A39+ is a log of grade A that is
greater than 39 cm in diameter. A log of grade AB46 is a
log of grade AB that is 46 cm in diameter. Each log was
scanned and cross-sectional image data was recorded at
1.5 cm intervals. The CT log data was manually removed
from the CT scanning system for processing.
Values were computed for each log by each of the six
candidate grading metrics according to the formulas in
Table 2. Table 3 gives the values computed for the five
logs of manual grade A39+, and Table 4 gives the values
computed for the 29 logs of manual grade AB46.
Data Analysis
The accuracy of the six candidate metrics is first evaluated by determining how well their automated CT grades
correspond to the grades assigned manually at the sawmill. The accuracy is then evaluated by studying the correspondence of the automated CT grades with the dollar
value of the final products.
113
Table 2. Empirically derived automated metrics to apply to CT data.
Metric
Description
A
Number of lamellas that contain a defect free length of 200 cm
B
Number of lamellas that contain a defect free length of 75 cm
Total B
Total number of defect free lengths that are 75 cm long
Score
Estimated average value over all the defect free of lengths
Score = ( Value A + Value B ) / N,
Where, for the set of A grade lamellas that contain a defect free length of 200 cm
Value A = $300 + Ó i=1,NA ( LAi – 200 cm ) * $1.5/ cm,
NA = number of defect free lengths of A grade,
LAi = the length of the ith defect free length of A grade,
And, for the set of B grade lamellas that contain a defect free length of 75 cm
Value
B
= $75 + Ó i=1,NB (LBi – 75 cm ) * $1.0 /cm,
NB = number of defect free lengths of B grade,
LBi = the length of the ith defect free length of B grade,
And,
N = NA + NB, is the total number of defect free lengths
Biased Score
Biased Score was developed to give weight to the longer length A grade lamellas.
Logs are favored in which the knots are arranged in a row; this is a closer approximation to external
log grading rules than the Score metric which estimates the valuation of the lamellas.
Biased Score = ( 5 * NA + NB) / N,
Where NA, NB, and N are defined in the Score section of this table
Defect Number
Number of knot defects bigger than 50 cm4 in diameter
Defect size threshold is 8 pixels, a pixel is 1.5 cm in length, and occupies 10 arc, for a 50cm diameter log, therefore the approximate defect
threshold size is 523mm or about 1 square inch.
4
.
Table 3. Values computed by the automated metrics for the set of A39+ logs
Log Number
Score
Biased S
A
B
Total B
Ndef
1
436.9
86
85
91
144
59
2
441.9
80.7
83
69
113
52
3
471
79.5
83
62
115
51
4
433.2
77.8
73
102
171
52
5
417.4
71
69
81
134
49
114
Figure 9. Roc curves for the Score and Biased Score metrics.
When a probability 80% correct determination is desired, the
corresponding probability of an incorrect log grade
determination is 28.6% for Score and 35.7% for Biased Score.
100.0%
100.0%
90.0%
90.0%
80.0%
80.0%
70.0%
70.0%
P(Correct)
P(Correct)
Figure 8. ROC curve computed for the Score metric. P(correct)
is the probability the automated grade agrees with the manual
grade assignment, and P(incorrect) is the probability the
automated grade does not agree with the manual grade
assignment.
60.0%
50.0%
40.0%
60.0%
50.0%
40.0%
30.0%
30.0%
20.0%
20.0%
10.0%
10.0%
0.0%
0.0%
20.0%
40.0%
60.0%
80.0%
Biased S
S
0.0%
0.0%
100.0%
20.0%
40.0%
60.0%
80.0%
100.0%
P(Incorrect)
P(Incorrect)
scanner gives improved accuracy over manual grading when
using final product dollar values for comparison.
A subset of the logs scanned with the CT were sawn into
boards and graded. All logs scanned by the CT were cut into
boards, but information only exists for the first ten AB46
logs. The boards generated from the logs were graded using
the metrics shown in Table 6. The price per cubic meter is
given in U.S. dollars. The yield column is the expected yield
through the board-processing step, which varies from board
type to board type.
The ten AB46 logs generated several types of boards of
several grades. The total dollar value of each log was calculated. In Table 7, the total dollar value for each log is shown
along with the previously computed Score and Biased Score
values.
The grades for the ten AB46 logs were examined, using
the thresholds of the previous section, 430 for Score, and 77
for Biased Score. Logs 2, 4, and 10, for Score, and logs 2,
and 10, for Biased Score, which were all manually graded
AB46, have been assigned the higher grade of A39+ by the
automated grading system. This higher grade valuation is
supported by the corresponding higher total dollar values of
the logs. The mean dollar values are shown in Table 8.
hold parameter itself is not shown on the ROC curve. The
ROC curve is useful in the selection of an optimal threshold
value that maximizes the number of correct log grade assignments and minimizes the number of incorrect ones. This
optimal threshold value lies near the “elbow” of the ROC
curve.
Based on the ROC curves of the six candidate automated
grading metrics, the grade metrics Score and Biased Score
were determined to be the most likely to grade logs correctly.
The Score and Biased Score automated grade metrics were
selected for further analysis. In Figure 9, the ROC curves
for the Score metric and Biased Score metric are shown.
Accuracy of the automated grading when
compared to manual grading
If an 80% probability of correct grade determination is
desired, then the corresponding threshold value for the Score
metric is 430, and for the Biased Score metric it is 77. Using these thresholds for the Score and Biased Score metrics,
the grades listed in Table 5 were assigned to the set of test
logs.
For the 34 logs in the test set, the percentage of logs where
the automated grade agreed with the manual grade was 75%
for Score and 82% for Biased Score. This level of accuracy
has been demonstrated elsewhere with x-ray based automatic
log grading systems (see literature cited).
Figure 10. The total board value in dollars plotted against the
CT derived grade metric Score and Biased Score.
600
Accuracy of automated grading when compared to final product dollar value
CT Derived Grade Metric
500
In order to prove that CT automated grading is more accurate than manual grading, the data analysis was extended
through the breakdown process of the logs. A number of
logs were tracked during the breakdown process and the
boards they generated were graded and valued. Using this
information and the thresholds of the previous section, it is
possible to show that automated grading using the CT log-
400
Biased
Score
300
Linear (Score)
Linear (Biased)
200
100
0
80
90
100
110
$ per cubic meter
115
120
130
Table 4. Values computed by the automated metrics for the set of AB46 logs.
Log Number
Score
Biased S
A
B
Total B
Ndef
1
304.5
41.5
30
99
201
84
2
463.1
79.2
81
70
124
45
3
425.2
71.7
71
75
138
63
4
434.2
76
75
81
142
41
5
358.7
64
61
79
123
48
6
309.2
47
35
107
206
72
7
329.2
50.2
39
106
201
85
8
356.5
60.3
53
97
162
54
9
310.9
53.8
44
103
184
68
10
498.5
97.8
102
77
104
37
11
295.5
50.7
43
89
146
67
12
326.4
47.7
37
101
188
54
13
196.8
28.3
13
105
182
10
14
412.2
71.7
70
80
139
47
15
255.1
32.3
17
109
213
71
16
524
100.3
114
32
38
11
17
329.8
55.3
47
97
175
81
18
471
97.7
105
61
92
29
19
280.3
50
41
95
147
84
20
451.5
74.3
76
66
119
43
21
270.7
40.2
29
96
171
83
22
265.9
34.2
18
115
220
87
23
411
73.5
70
91
155
60
24
281.8
45
35
95
173
93
25
441.1
91.5
93
84
113
44
26
336
59.7
54
88
149
62
27
327.1
53.3
46
90
188
77
28
385.4
68.5
65
86
155
62
29
350.2
53.7
47
87
157
57
116
Table 5. The results for the automatic grading of logs using the two automatic grading metrics Score and (Biased
Score).
Grade According to the X-ray scanner
Grade According to the visual grading
A39+
AB46
A39+
4 (4)
1 (1)
AB46
7 (5)
22 (24)
Table 6. Board grade metrics, price, and expected yield.
Grade
Grade_ID
Price in USD per m3
Yield
115 mm 0-I
115-0-I
321
68.00%
124 mm 0-I
124-0-I
340
67.00%
143 mm 0-I
143-0-I
340
65.00%
153 mm 0-I
153-0-I
340
64.00%
30 mm 0-I
30-0-I
334
100.00%
50 mm 0-I
50-0-I
308
100.00%
95 mm 0-I
95-0-I
289
70.00%
B
B
160
100.00%
D
D
109
71.00%
F
F
231
67.20%
FC
FC
179
67.20%
Kaufmann
KA
118
100.00%
KW
KW
118
79.00%
LC
LC
179
67.20%
RC
RC
141
79.00%
V
V
308
100.00%
117
CONCLUSION
Table 7. Computed Score and Biased Score metric
values with calculated total board value in United
States dollars.
Log
Number
Score
Biased S
1
304.5
41.5
$95.32
2
463.1
79.2
$115.08
3
425.2
71.7
$94.64
4
434.2
76
$110.00
5
358.7
64
$103.05
6
309.2
47
$89.80
7
329.2
50.2
$114.04
8
356.5
60.3
$121.53
9
310.9
53.8
$97.63
10
498.5
97.8
$126.52
The results indicate that a CT derived automated log grading system is successful at determining grade. The automated metric, Biased Score, which is an amalgamation of
the number of each type of lamella produced at the test site
sawmill, is the most accurate, having an 82% probability of
assigning a grade to a log that agrees with the manual grade.
Other CT-derived metrics that focused on a single lamella
type or number of defects had limited success in correctly
determining the grade.
The 80% accuracy rate for the automated log grading
metrics is comparable to automatic grading algorithms found
in the literature. The automated log grading method was
further validated with final product dollar value information that correlates log grade with the total value of the breakdown board products; log grades tended to correspond to a
higher total dollar value of the breakdown products. The
data showed that automated CT grade metrics can be more
accurate than the usual manual grading methods that rely
solely on external information.
Total Board
Value
LITERATURE CITED
Explosive Detection System EDS Certification is part of
the US Department of Transportation FAA 14 CFR 108.
Forest Service Guidebook for log grading.
Table 8. Mean total board values as a result of using
the automated CT grade metrics Score and Biased
Score.
CT grade metric
A39+ mean
Simplified Guidelines to Hardwood Lumber Grading
(Walton R. Smith – US Department of Agriculture –
Forest Service).
Max Pristovnik, Personal communication concerning the
sawmill log grades, 2000.
AB46 mean
total board value
total board value
Score
$117.20
$102.29
Biased Score
$120.80
$103.25
Johan Oja, Lars Wallbacks, Stig Grundberg, Erik Hagerdal
and Anders Gronlund, “Automatic grading of saw logs
using an industrial logscanner”, Proceedings of the 4th
International Conference on Image Processing and Scanning of Wood.
The total board values shown in Table 8 for the A39+ are
based on the assumption that the mean total board value for
A39+ logs is around $120. Based on this assumption, the
results indicate that Score and Biased Score grade the logs
more accurately than the manual grading by the sawmill.
The correlation of the two CT derived grade metrics,
Score, and Biased Score, with the total dollar value is shown
in Figure 10. The correlation is expected to be linear but
fitting the data to a linear relation gives an average linear
correlation coefficient of 0.6, which is not high. However,
the graph shows a definite trend. This figure is poor, due in
part, to the small sample set.
ACKNOWLEDGMENTS
The principal investigator would like to express his thanks
and gratitude to the United States Department of Agriculture for providing the funding for this work, to Dr. Charles
Cleland for his support in his role as the program point of
contact, and to Debbie Ogden in her role as administrative
point of contact. This work would not have been possible
without the help of several of my colleagues, including
Walter Garms, Maximilian Pristovnik, Sondre Skatter, Elan
Scheinman, Pierfrancesco Landolfi, and François Mesqui.
118
Chapter 13
Evaluating the Positioning Performance of GPS Surveying
Under Different Forest Conditions in Japan
SECA GANDASECA,
TETSUHIKO YOSHIMURA
HISASHI HASEGAWA
Abstract—The objective of this study is to clarify the performance of GPS in forested conditions after SA was turned off and
relationships between SNR and spatial conditions surrounding the GPS receiver. Moreover, we tried to increase the antenna
height to mitigate the negative effects of tree canopies and other obstacles on the accuracy of GPS measurements. As a result, we
found that position errors of precision and accuracy were increased by the effect of obstacles in the plantation, on the forest road
and in the natural forest compared to those at the landing, and that DGPS improved both precision and accuracy. It was also found
that position errors of precision and accuracy using DGPS were improved when the antenna height was raised from 1.0 m to 4.2
m. The signal quality observed at the landing was the highest while that observed in the plantation forest was the lowest because
of high density of tree canopies and stems. On the forest road, signal quality was decreased by multipath, in which GPS signals
were reflected by cut slopes or trees.
INTRODUCTION
METHODS
The Global Positioning System (GPS) is a satellite-based
navigation system designed and operated by the U.S. Department of Defense (DoD) for military and civilian use. The
GPS provides users with accurate information about their
position and velocity, as well as the time, anywhere in the
world. It is widely known that for US national security there
were difficulties in getting accurate position data from GPS
data because of Selective Availability (SA), which degraded
the signal available to nonqualified GPS receivers and introduced a position error of about 100 m by dithering the time
and ephemerides data provided in the navigation message.
Therefore, differential GPS (DGPS) has been used to improve positioning or navigation accuracy by determining the
positioning error at a known location and subsequently incorporating a corrective factor into the position calculations
of another receiver operating simultaneously. However, SA
was turned off a few minutes past midnight EDT after the
end of May 1, 2000. As a result, the position error of autonomous GPS was reduced to about 10 m, and the possibilities of GPS in forestry and forest management have increased.
The objective of this study is to clarify the performance
of GPS in forested conditions after SA was turned off. Moreover, we tried to increase the antenna height to mitigate the
negative effects of tree canopies and other obstacles on the
accuracy of GPS measurements. In addition, it is known
that one of the most important predictors of the accuracy of
GPS measurements is SNR (Signal to Noise Ratio) and, therefore, we tried to clarify relationships between SNR and conditions surrounding the GPS receiver.
The base station (Trimble 4600LS) was set up on the
mountain ridge of Kyoto University Forest in Wakayama.
Then, the rover (Trimble Pathfinder ProXR) was set up at
four points: at a completely open point, referred to as the
landing, in a plantation forest, on a forest road, and in a natural
forest. At the landing, the sky is almost completely open
without obstacles such as trees or slopes. In the plantation
forest, Japanese cedar (Cryptomeria japonica) was planted
in 1958, and the average tree height was 17.0 m. The width
of the forest road was 5.5 m, and the height of the cut slope
was 5.2 m. The rover was set up 1 m away from the bottom
of the cut slope. The natural forest is a secondary one, with
many species such as Fagus, Carpinus, Parabenzoin, Acer
and so on. The GPS measurements using the rover were
done at each point for 20 minutes with different antenna
heights of the GPS receiver. These measurements with an
antenna height of 1 m and 4.2 m were conducted on January
10 and 9, 2001, respectively. Table 1 shows the receiver
settings used in this research. We started each GPS measurement four minutes earlier on January 10 than on January
9 since the same distribution of GPS satellites appears four
minutes earlier day by day. The observed data was differentially corrected using the data of the local base station.
Moreover, four sets of GPS 3+ (Garmin), Oncore Active GPS Antenna (Motorola), notebook computers
(TOSHIBA) were set at the four points and NMEA data that
included the azimuth, elevation and SNR of each visible satellite were logged in order to clarify relationships between
119
Table 1. Receiver settings.
along the x and y axes, respectively, and calculated by the
following equations:
n
PDOP mask
SNR mask
Elevation mask (º)
Logging interval (sec)
ProXR
(Rover)
4600LS
(Base station)
20
0
15
5
12
0
15
15
σ x2 =
∑ ( xk − x)2
k =1
n −1
n
σy =
2
∑( y
k =1
(2)
k
− y)2
(3)
n −1
In Eqs. (2)-(3), x and y are the sample mean of the error
along the x and y axes, respectively when calculating precision. When position errors of accuracy were calculated, we
substituted xtrue and ytrue that indicated the true location along
the x and y axes respectively for x and y in Eqs. (2)-(3).
The true locations of the four observation points were determined by using dual-frequency GPS receivers prior to this
research.
SNR and spatial conditions surrounding the GPS receivers.
This observation was done at the four points simultaneously
from 12:15:07 to 15:07:10 and from 15:21:40 to 17:00:22
on January 11, 2001 (Japanese Standard Time). After that, a
fish-eye photo was taken at each point using a digital camera (NIKON E950) to get digital images of the spatial conditions surrounding the four observation points. Finally, SNR
distribution was overlaid with the fish-eye photos.
RESULTS AND DISCUSSION
Precision and accuracy
Precision
Position errors were calculated in terms of precision and
accuracy. Accuracy refers to the closeness of the observations to the true value, and precision refers to the closeness
of repeated observations to the sample mean (Leick, 1995).
Precision and accuracy are briefly explained in Figures 1a,
1b and 1c. The centers of these figures are the true location.
Points shown in these figures mean actual measured points
with GPS. Figure 1a shows high precision and low accuracy and, conversely, Figure 1b shows low precision and
high accuracy. Figure 1c shows that precision and accuracy
are high.
In this study, position errors of precision were calculated and compared in 2drms (twice the distance root mean
square). The 2drms value is commonly taken as the 95%
limit for the magnitude of the horizontal error (Kaplan, 1996),
and the calculation of 2drms is shown in the following equation:
Position errors of precision with respect to autonomous
GPS and DGPS are shown in Figures 2 and 3, respectively.
According to these figures, position errors of precision were
increased by the effect of obstacles in the plantation, on the
forest road and in the natural forest. In the natural forest,
position errors were not so high since there were no leaves
on trees in January. On the forest road, position errors were
rather high, probably because of multipath, in which the GPS
signal is reflected by trees or cut slopes and it does not reach
the GPS receiver directly.
Figure 2 shows the results of autonomous GPS measurements at the four points with an antenna height of 1 m
and 4m. There was no clear difference in position errors
with an antenna height of 1 m and 4.2 m. Figure 3 shows
the results of post-processed differential corrections. As
shown in this figure, the difference of position errors with
an antenna height of 1 m and 4.2 m was clear. This was
because DGPS reduced the atmospheric effects. These results were likely caused by ever changing conditions of ionosphere and troposphere that produce unstable travel time of
GPS signals. In summary, position errors of precision were
improved when DGPS was used, and an antenna height of
the GPS receiver was set at 4.2 m.
We also analyzed the relationships between position
errors and the independent variables, that is, positioning
points and antenna height using ANOVA. In this analysis,
the data that was differentially corrected with the local base
station was used. As shown in Table 2, the factors of positioning points and antenna height are significant at the 1 %
and 5 % level, respectively. As a result, it was proved that
positioning points and antenna height had significant effects
on position errors of precision.
2drms = 2 σ x + σ y
2
where
σx
and
σy
2
(1)
are the standard deviation of the error
120
Figure 2. Precision errors of autonomous GPS.
Note: (1) indicates the first observation; (2) indicates the second
observation.
Figure 4. Accuracy errors of autonomous GPS.
Note: See the note in Figure 2.
Figure 5. Accuracy errors of DGPS.
Note: See the note in Figure 2.
Figure 3. Precision errors of DGPS.
Note: See the note in Figure 2
Accuracy
SNR
Position errors of accuracy at with respect to autonomous GPS and DGPS are shown in Figures 4 and 5, respectively. Position errors of accuracy were increased by the
effect of obstacles in the plantation, on the forest road and
in the natural forest as shown in these figures.
Figure 4 shows the results of autonomous GPS measurements at the four points with an antenna height of 1 m
and 4m and that the difference of position errors with an
antenna height of 1 m and 4.2 m is not clear. Figure 5 shows
the results of post-processed differential corrections. According to this figure, the difference of position errors with
an antenna height of 1 m and 4.2 m is clear. As a result,
position errors of accuracy were improved when DGPS was
used, and an antenna height of the GPS receiver was set at
4.2 m.
Then, the relationships between position errors and the
independent variables, that is, positioning points and antenna height were analyzed using ANOVA. Table 3 shows
that the factors of positioning points and antenna height are
significant at the 1 % and 5 % level, respectively. Therefore, it was proved that positioning points and antenna height
had effects on position errors of accuracy.
The SNR indicates the quality of GPS signals, and SNR
is higher when signal quality is better. Figures 6, 7, 8 and 9
show SNR and spatial conditions surrounding the GPS receiver at the landing, in the plantation forest, on the forest
road and in the natural forest, respectively. In these figures,
the north is shown at the top of each figure and, for identification, signal quality is shown with colors such as green,
yellow and red.
According to Figure 6, signal quality at the landing is
very good, as shown by the dominant existence of green
color from 15° to 90° of elevation. Figure 7 shows signal
quality in the plantation forest, which is lower than that in
the natural forest, as shown by the dominant existence of
yellow and red colors. As shown in Figure 8, signal quality
on the forest road is high, as shown by the dominant existence of green color from 30° to 90° of elevation. However,
the area on the left side (in the west) is covered by red color
because the existence of the cut slope which might cause
multipath. Figure 9 shows signal quality in the natural forest, which is lower than that at the landing, as shown by the
dominant existence of yellow color. The areas of green color
are also seen from 30° to 90° of elevation because there are
no leaves on the trees in January.
121
Figure 6. SNR at the landing.
Figure 8. SNR on the forest road.
Note: Legend is the same as in Figure 6.
Figure 7. SNR in the plantation forest.
Note: Legend is the same as in Figure 6.
Figure 9. SNR in the natural forest.
Note: Legend is the same as in Figure 6.
122
CONCLUSIONS
caused multipath and even interruption of GPS signals. This
result corresponds with our experiences, in which GPS data
along forest roads are often not so accurate. Although position errors of both precision and accuracy in the plantation/
natural forests as well as on the forest road were lower than
those observed in the open area, the use of an antenna height
of 4.2 m is a practical way to reduce position errors when
using DGPS.
Position errors of precision and accuracy were increased
by the effect of obstacles in the plantation, on the forest road
and in the natural forest compared to those observed at the
landing. It was also found that there were relationships between position errors and positioning points and between
position errors and antenna height according to the analysis
using ANOVA. It was also found that DGPS improved both
precision and accuracy. In addition, position errors of precision and accuracy using DGPS were improved when an
antenna height of the GPS receiver was set at 4.2 m. As a
result of the SNR analysis using fish-eye photos, SNR observed at the landing was higher than that observed in the
plantation and natural forests or on the forest road. The
signal quality in the plantation forest was the lowest because
of high density of tree canopies and stems. The signal quality on the forest road is rather high, however, cut slopes
LITERATURE CITED
Kaplan, E.D. 1996. Understanding GPS: Principles and Applications. Artech House Publishers, Boston, 629 p.
Leick, A. 1995. GPS Satellite Surveying Second Edition.
Wiley, New York, 560 p.
Table 2. Summary of ANOVA for precision.
Factors and interactions
Degrees of freedom
Mean squares
F values
Positioning points
3
17.42
8.10**
Antenna height
1
21.63
10.05*
Positioning points × Height
3
3.35
Errors
8
2.15
Sum
15
Note: **, Significant at the level 1% level; *, Significant at the 5% level.
Table 3. Summary of ANOVA for accuracy.
Factors and interactions
Degrees of freedom
Mean squares
F values
Positioning points
3
25.41
9.27**
Antenna height
1
16.01
5.84*
Positioning points × Height
3
2.07
Errors
8
2.74
Sum
15
Note: **, Significant at the level 1% level; *, Significant at the 5% level.
123
Chapter 14
Forest Road Earthwork Calculations for Linear Road
Segments Using a High Resolution Digital Terrain Model
Generated from LIDAR Data
ELIZABETH DODSON COULTER
WOODAM CHUNG
ABDULLAH AKAY
JOHN SESSIONS
Abstract—A computer-assisted model was developed to compute earthwork for linear road segments using a high-resolution
digital terrain model (DTM). A high resolution DTM (1m x 1m) was created using LIDAR data from the Capitol Forest, Washington, USA for this study. This high resolution LIDAR DTM enables the computer-assisted model to estimate earthwork accurately
and quickly and may eliminate the need for additional ground surveys. An example of earthwork calculations for four linear road
segments using a LIDAR DTM is presented.
INTRODUCTION
roadway is evaluated. Third, the other cut and/or fill slope
is evaluated. Here, a road segment is defined as the straight
line between two design points.
The major cost component in the construction of a forest
road is earthwork. The existing methods (conventional methods) for estimating earthwork volumes for straight roadway
selections are the average end-area method and the prismoidal method (Hickerson 1964). The average end-area method
averages the earthwork calculated at two successive road
cross-sections. The prismoidal method is based on an assumption that the ground profile between roadway stations
is linear. The average end-area method tends to overestimate the volume while the prismoidal method provides a
more precise volume calculation for linear profiles (Easa
1992).
During the last few decades there has been increasing interest in computer-aided analysis of road design as it provides quick evaluation of alternatives in a more systematic
manner. Recently, computer-aided road design systems employing digital terrain models (DTM) have become more attractive with the advent of high resolution DTM data. One
method for gathering high-resolution elevation data is with
the use of airborne laser mapping technology such as LIDAR (light detection and ranging). LIDAR elevation data
have been found to be accurate to within 15 cm (Ahmed et
al. 2000). These data allows the creation of an accurate,
high-resolution DTM, enabling a computer-assisted model
to estimate earthwork accurately and quickly.
Roadway
The equation for the centerline of the road can be found with:
y2 − y1
x2 − x1
(1)
b = y1 − mx1
(2)
m=
where (x1, y1) and (x2, y2) are the beginning and ending points
of the road segment, respectively, m is the slope of the
centerline road equation, and b is the y-intercept for the
centerline road equation (see Figure 1).
For each potential cell along the roadway, the perpendicular distance from the centerline is calculated.
1
x
m
(3)
(c − b ) m
m2 + 1
(4)
ycl = mxcl + b
(5)
c= y+
xcl =
d cl =
Design of a Single Road Segment
(x − xcl )2 + ( y − ycl )2
(6)
where c is the y-intercept for the line perpendicular to the
road centerline, (x, y) are the coordinates for the cell being
looked at, (xcl, ycl) are the x and y values at the intersection of
the perpendicular and the road centerline, and dcl is the dis-
The design of a single road segment is completed in three
steps. First, elevations for cells within the roadway are calculated. Second, the cut and/or fill slope on one side of the
125
Figure 1. A road segment presented in an x-y coordinate system.
A grid cell in cut area
Y
A grid cell
on road surface
dre
Road edges
Centerline
with slope (m)
(x, y)
dcl
RW
2
(xcl, ycl)
b
db
dre
X
A grid cell in fill area
tance from (x, y) to the centerline (in meters).
If dcl is less than half the road width from the centerline,
the elevation (z) of the cell (x, y) is set to the elevation of the
centerline (zcl) Eq. (7).
z = z cl = z1 +
d end ( z 2 − z1 )
L
equations (Eq. (9) or (10)). If the elevation of the cell (z) is
above the elevation of the roadbed (zrd) at a point perpendicular to the cell, the new elevation is calculated as in Eq.
(11). If the elevation of the cell is below the elevation of the
roadbed, Eq. (12) is used to calculate the design elevation
for the cell.
(7)
where zcl is the elevation of (xcl, ycl) in meters, z1, z2 the elevation of the start- and end-point of the road segment, respectively, dend is the distance from (xcl, ycl) to (x1, y1) in
meters, and L is the length of the road segment in meters.
The assumption is made that the road grade is constant between (x1, y1) and (x2, y2). All potential cells along the roadway are checked in this manner and assigned a design elevation, if appropriate. A flowchart of this process is presented in Figure 2.
(8)
y = mx + (b − db )
(9)
y = mx + (b + db )
(10)
(11)
z = zrd − d re ∗ fillslope
(12)
where dre is the distance to the road edge, cutslope is the
slope of the cut slope (vertical distance in meters divided by
horizontal distance in meters) and fillslope is the slope of
the fill slope (vertical distance in meters divided by horizontal distance in meters).
Earthwork
Cut and Fill Slopes
Equations for the road edges can be calculated using Eq.
(8), (9), and (10).
RW
2
db =
cos(tan −1 (m ))
z = zrd + d re ∗ cutslope
The convenience of using a 1-meter by 1-meter DTM is
that the sum of differences between the designed road surface elevation and the ground elevation is equal to the volume of earthwork (Eq. 13). Earthwork calculations are
shown graphically in Figure 3. Fill volumes should be corrected by an appropriate compaction factor.
Earthwork =
∑ (z
n
di
− z si )
i =1
(13)
where Earthwork is the volume of earthwork in m3, zdi the
elevation (in meters) of the ith cell within the designed road,
and zsi the elevation (in meters) of the ith cell on the DTM
surface.
where db is the difference in y-intercept between the
centerline road equation and the road edge equation, and
RW is the road width in meters.
For each cell, the distance from the road edge is calculated as in Eq. (3)-(6) using either the right or left road edge
Example of Road Design
The data used to develop this model were from a LIDAR
data set collected by Aerotec (Bessemer, AL) under con126
Figure 2. Flow chart for segment design process.
Calculate linear equations for centerline
and road edges
Select next cell in row/column and draw a
perpendicular line from the cell to the
centerline
No
Does this cell fall
within the road width?
Fall in cut area?
Yes
Yes
Designed elevation (Z) = Zrd
Z = Zrd + dre * cutslope
Calculate elevation difference
(∆Z = Zground - Z)
Z = Zrd - dre * fillslope
Yes
No
∆Z > 0
Add ∆Z to fill volume
Add ∆Z to cut volume
No
Done with this row/column?
Yes
Select the next row/column
No
Done with all cells in the road?
Yes
End
127
No
Figure 3. A road cross-section showing how to calculate earthwork volume using an 1m x 1m DTM.
Actual ground surface
A designed
road template
1m x 1m
∆Zcut
Ground elevation on
DTM
Design elevation
Cut volume = Σ(∆Zcut)
Fill volume = Σ(∆Zfill)
Table 1. Road design results.
Road Segment
Length (m)
Road Grade
Cut Volume
(m3)
Fill Volume
(m3)
Number of
Cells in Road
Number
of Cells in
Cut and
Fill Area
1
55.54
4.12%
7.38
58.19
269
147
2
46.14
4.21%
10.1
83.26
222
126
3
48.7
3.83%
67.93
0.40
240
116
4
35.13
0.76%
25.95
32.27
178
78
Total
185.51
111.36
174.12
909
467
128
At this point, this method only applies to straight road
segments and does not deal with horizontal or vertical curves.
This model also does not take insloping, outsloping, curve
widening, or turnouts into consideration. However, these
modifications would be relatively easy to implement.
This method is able to look at every cell within the road
template and determine the design elevation and compare
that elevation to the surface elevation. Although it is felt
that this is a more accurate method of determining earthwork
than traditional road design methods, this assumption has
not been field tested. A field test would include a comparison of time and accuracy of computerized road design cut
and fill volumes with earthwork volumes generated using
traditional methods. These should be validated with actual
volumes measured during road construction activities.
Figure 4. Shaded elevation map created from LIDAR data.
CONCLUSION
A method has been developed utilizing high-resolution
DTM data to calculate earthwork for a forest road. This
method assigns design elevations to all cells within the road
template and calculates cut and fill volumes from the difference between design and surface elevations. This method
is believed to be more accurate and efficient than traditional
road design methods, however field trials would be necessary to validate this assumption.
tract to the USDA Forest Service as part of the joint WADNR/USFS PNW second-growth harvesting options study
in western Washington. The software program TerraScan
was used to convert the random points of the LIDAR data to
a DTM. This is done by first filtering out all non-ground
points. Second, a triangulated irregular network (TIN) is
created, from which the software pulls off elevations at regular intervals, in this case every 1 meter, to create a DTM.
The DTM used in this example is shown in Figure 4 as a
shaded elevation map.
A short road was designed consisting of four linear road
segments. Road segment descriptions are shown in Table 1.
LITERATURE CITED
Ahamed, K.M., S.E. Reutebuch, T.A. Curtis. 2000. Accuracy of high-resolution airborne laser data with varying
forest vegetation cover. 2nd International Conference on
Earth Observation and Environmental Information, 1114 November 2000, Cairo, Egypt.
DISCUSSION
Easa, Said M. 1992. Estimating earthwork volumes of curved
roadways: Mathematical model. Journal of Transportation Engineering. 119(6):834-849.
This method may significantly decrease road design time
and effort, both in the field and in the office. An engineer in
the field equipped with a GPS unit would need to make fewer
field measurements in order to complete road design
earthwork calculations meaning a greater number of alternatives could be examined. If an area is covered by a highresolution DTM, a field engineer would need only a series of
road center points in order to calculate cut and fill volumes.
When the time came to mark the road location on the ground
for construction, the engineer would have final design elevations available at every point within the road template.
Hickerson, Thomas F. 1964. Route location and design.
McGraw-Hill, United States. 634 p.
ACKNOWLEDGEMENTS
The authors wish to thank Joe Means, Department of
Forest Science, Oregon State University for assistance with
LIDAR processing and Stephen Reutebuch of the USFS
PNW Research Station, Seattle, WA for permission to use
LIDAR data.
129
Chapter 15
Tightly Coupled Inertial/GPS System for Precision Forestry
Surveys Under Canopy: Test Results
JOEL GILLET
BRUNO M. SCHERZINGER
ERIK LITHOPOULOS
Abstract—This paper describes the Position and Orientation System for Land Survey (POS/LS) that Applanix is developing
for survey applications. The POS/LS is derived from the Applanix POS product family, and will include Applanix’s next generation tightly coupled inertial/GPS integration. Differential GPS (DGPS) and Real-Time Kinematic GPS (RTK) are widely used
today as sources of accurate position data for various survey applications in precision farming and forestry surveying. This paper
describes the problems incurred in forestry due to DGPS outages under canopy and how tightly coupled inertial/GPS systems can
reduce or eliminate these problems. From the position of individual trees, to the location of a stream’s headwaters and all other
attributes that affect the forest, ground position can now be a primary key for all required forest attributes. This document presents
an overview of the technical implementation of the POS/LS, and evaluates some preliminary test results.
INTRODUCTION
stacle to the creation of such a database so far was the impracticality, lack of precision, or environmental impact of
positioning instruments under canopy.
The POS/LS will be a light, portable instrument, fitting
in a backpack or on a walking staff for example, and providing a continuous metric-level position in three dimensions
for the forester. Its practicality will be comparable to that of
a GPS receiver in open areas (GPS does not work under
canopy).
With such a device it will be easier to create a data sheet
for every tree studied to include genetic heritage, ecosystem, age, (etc...) based on its location, and cross-reference it
to maps and aerial photographs.
It will enable selective harvesting based upon continuous
monitoring to determine the optimum time for cutting or the
optimal path to be followed in the forest to minimize the
effects of soil compaction, some of the basis of precision
forestry.
The combination of radio-frequency identification (RFID)
which is the passive identification of objects (trees in this
case) with precise three dimensional location will change
forest practices. The lineage of the tree can now be traced
from the forest to the lumber dealer thereby creating a more
sustainable forest management.
The combination of radio-frequency identification (RFID)
which is the passive identification of objects (trees in this
case) with their precise three dimensional location will
change forest practices. The lineage of the tree can now be
traced from the forest to the lumber dealer thereby creating a
more sustainable forest management.
This paper presents the Position and Orientation System
for Land Survey (POS/LS) that Applanix is developing for
survey applications in hostile GPS environments such as forested areas. Originally designed for oil exploration survey,
the POS/LS will provide continuous position information in
a variety of survey conditions that range from full access to
Global Positioning System (GPS) satellites signal to complete blockage of GPS signals for possibly the entire survey.
This attribute is called robust positioning, and is a consequence of integration of available aiding data into an aided
inertial navigation system, the core of the POS/LS.
Until now, precise forestry surveying, such as boundary
was done using conventional survey instruments and methods such as theodolites and electronic total stations. As part
of post-processing the surveyor performed elaborate geometric computations in order to obtain the coordinates, elevation profiles and plans of every point or line surveyed. The
instruments required access to line-of-sight survey lines.
Survey crews had to create straight cut-lines into forested
areas to obtain the line-of-sight corridors, which added expenses to the survey operation for the slasher crews and in
some jurisdictions stumpage fees for the cut trees. This
method therefore has a low productivity and a high environmental impact, as well as high cost.
At the same time, there is a growing need to establish a
geographically-referenced database of information on the different species of plants and animals, soils, waterlines, properties and other attributes of the forest habitat. The only ob131
Applanix proposes this new instrument, which will bring
GPS-type productivity and ease of use to all areas, even to
the densest of forests. We will present here the basic
principals of aided inertial technology and some preliminary
tests results in the development of a prototype instrument.
Figure 1. Loosely coupled inertial/GPS architecture.
INERTIAL NAVIGATION
An inertial navigation system (INS) contains two core
components: an inertial measurement unit (IMU) and a navigation processor (NP). The IMU contains three accelerometers and three gyros, whose respective input axes form an
orthogonal triad, plus digitization and digital interface electronics. The accelerometers measure the specific force that
the IMU experiences, comprising accelerations and gravity
with respect to an inertial reference. The gyros measure the
angular rate that the IMU experiences, comprising its angular rate with respect to the earth plus the earth’s angular rate
with respect to the inertial reference. The NP receives the
inertial data and performs two functions. First it performs
an alignment, during which it establishes an initial position
and orientation using the local gravity vector as the vertical
reference and North component of the earth rate vector as
the heading reference. Having established a navigation frame
of reference that is locally level and having a known heading with respect to North, the NP then transitions to its freeinertial navigation mode. It solves Newton’s equations of
motion in the navigation frame on the earth from the measured specific force and angular rate data to generate a current position, velocity and orientation solution at a specified sampling rate.
The key advantage of an INS is that, once aligned, it
navigates autonomously of external signals or
communications. The key disadvantage of an INS is that
being essentially a dead-reckoning system, its position error
grows with time due to alignment errors and inertial sensor
errors. A medium accuracy INS contains ring-laser gyros
(RLG) with less than 0.01 degrees/hour bias and pendulous
servo accelerometers with less than 50 micro-g’s bias
(following factory INS calibration), and exhibits a typical
position error rate of less than one nautical mile per hour
or 0.5 meters per second. This free-inertial drift rate is
acceptable for aircraft navigation but not usable in a survey
instrument.
loop INS error regulation architecture is well known in the
navigation community, and has been the basic method of
aided inertial navigation design for the last 30 years [1]. The
dominant aiding sensor in recent years is GPS. It provides a
position solution whose errors are noisy but stable, whereas
the INS provides position that is smooth but prone to drift.
The GPS position solution can drop out due to antenna shading, whereas the INS solution is uninterrupted. The INS and
GPS navigation solutions are complementary, in that a deficiency of one sensor is a strength of the other. A GPS-aided
INS provides a navigation solution that inherits the best characteristics of both sensors.
Figure 1 shows a loosely coupled inertial/GPS integration. In a loosely coupled integration, the Kalman filter processes the GPS position and velocity solution to aid the inertial navigator. In this case, the GPS receiver is a self-contained navigation subsystem that is capable of self-contained
positioning so long as it can receive signals from four or
more satellites. When the receiver tracks fewer than four satellites, it cannot provide position and velocity fixes to the
Kalman filter. In this case the inertial navigator operates unaided, and is subject to drift imposed by the residual errors
in its alignment and its inertial sensors following correction
by the Kalman filter’s error state that it extrapolates from the
last time of valid GPS data. The receiver may continue to
output the pseudorange and phase observables from up to
three satellites, which a loosely coupled integration is not
able to use.
Figure 2. shows a tightly coupled inertial/GPS integration. This implies the Kalman filter processes the GPS
pseudorange, phase and Doppler observables. In this case,
the GPS receiver is strictly a sensor of the GPS observables.
The navigation functions in the GPS receiver, namely posi-
AIDED INERTIAL NAVIGATION
Aided inertial navigation is a method to regulate the INS
errors, in particular the position error drift, and to align the
INS or improve its alignment while moving. Other navigation sensors that measure position, velocity and/or orientation in various combinations and formats provide the INS
aiding data. A Kalman filter performs the integration of the
INS and aiding navigation data, as is shown in Figure 1.
The Kalman filter estimates the INS and aiding sensor errors based on the navigation data presented to it, and then
corrects the INS based on the estimated errors. This closed132
on the order of a GPS receiver. It shall allow the surveyor to
carry the unit indefinitely with no significant fatigue. It shall
have low power consumption, so that it can operate all day
on a small set of batteries.
Figure 2. Tightly coupled inertial/GPS architecture.
3.
It shall be rugged, waterproof and dustproof. It shall
be able to endure abuses such as shock from being dropped,
vibrations from field vehicle engines, and exposure to rain.
4.
Its operational temperature and humidity range shall
allow operation in all possible environmental conditions,
from the Tropical rain forests, to northern Canada.
5.
It shall provide continuous survey data, automatically accommodating changes from open ground to forest,
from fields to cities, without interruption.
6.
It shall provide a comprehensive quality assurance
(QA) and quality control (QC) of the computed data, so that
the surveyor has a reliable indication of the quality of the
computed position at all times.
7.
The unit shall contain a reprocessing function that
performs a smoothing operation on the recorded navigation
data between two or more specified position fixes.
tion and clock offset fixing and possibly RTK, are not used.
The key advantage of this configuration is that it makes use
of all GPS data available at all times. The tightly coupled
integration makes effective use of GPS observables from
one or more visible satellites to control the navigation errors. This is especially important in areas of marginal GPS
coverage such as forests where fewer than four satellites
may be visible.
Also shown in Figure 2 is a second source of aiding data,
the zero velocity update (ZUPT). This data comprises a
known zero velocity when the INS is known to be stationary. It allows the aided INS to zero or “ground” its velocity
error and thereby improve the calibration of velocity error
sources such as the alignment and the inertial sensor errors.
ZUPTs are possible in a land navigation system, but not in
an aircraft or on a ship. ZUPTs performed periodically will
provide a lower position error rate than a free-inertial INS is
capable of. ZUPTs are implemented when GPS is not available. ZUPTs typically are required every 1 to 2 minutes and
last 15-30 seconds to achieve minimum position error.
8.
The unit shall be easy to use, requiring no special
knowledge or skill. It shall provide control and display of
status and navigation data to the operator via a hand-held
control and display unit (CDU).
Such an instrument will multiply survey productivity four
or five times, and thereby significantly decrease survey costs.
It will also provide a dramatically lower environmental impact than the traditional methods.
After considering these requirements, Applanix
Corporation has decided to develop and make available for
sale a survey instrument called the POS/LS that fulfills these
requirements. POS/LS will be developed in several phases
in order to adapt more readily to ongoing market
requirements. An open architecture will allow several
versions of the product to be proposed to different markets
according to precision and legislation requirements.
The POS/LS will operate in one of two modes: alignment and navigation. Following power-up, the instrument
enters the alignment mode, during which it is required to be
stationary for 5-10 minutes unless a GPS signal is available.
The purpose of this mode is to establish a level and oriented
navigation frame for the subsequent navigation modes.
Following alignment, the POS/LS transitions to a Zero
velocity UpdaTe (ZUPT) -aided or GPS-aided inertial navigation mode, whose signal processing architecture is shown
in Figure 2. A ZUPT is basically a period when the instrument is still, for example when put on the ground. If four or
more GPS satellites are visible to provide a full 3D-position
solution, then the POS/LS will provide position fixes with
A NEW SURVEY INSTRUMENT
By studying the drawbacks of past and current survey
systems, Applanix has drafted the requirements for the nextgeneration survey and navigation system in wooded areas.
The key requirements are the following:
1.
The unit shall be borne by a single surveyor. This
implies a handheld unit, a backpack format or some variation thereof.
2.
The overall unit shall have small size and weight,
133
the same accuracy as the GPS without ZUPTs. If fewer than
four satellites are visible, then the POS/LS continues to use
the available GPS observables and thereby control the position error drift, but not in all dimensions. When the POS/LS
has determined that its position error has begun to grow
with time, it begins to request periodic ZUPTs from the operator. If no GPS data are available, then the frequency of
ZUPT requests is expected to be every 1-5 minutes. If the
operator stops to do something (eg. plant a stake, collect
data, mark a tree), he can put the POS/LS on the ground so
that it will automatically perform a ZUPT. If the operator
stops to plant a stake or measure a tree, every 2 minutes,
then the POS/LS will likely not need to request a ZUPT.
When only aided by ZUPTs, the POS/LS will exhibit a
position error growth that is a function of distance. In absence of GPS signal, the real-time accuracy of a survey will
depend on the length of the traverses between two position
control points or GPS position fixes. If a real-time position
accuracy of less than 3 meters and a post-processed accuracy of less than a meter are sufficient, then the POS/LS
operator will be able to go all day long without control. He
will tie to a position fix at the end of the day. If a better
position accuracy is needed, the operator can survey one or
more control traverses with the POS/LS to provide more
control points on the lines. Control traverses are traverses
between position fixes that intersect the survey lines at the
control points. The POS/LS position solution on a control
traverse between two position fixes will have the accuracy
required for control. This accuracy can be improved with
re-processing. No alternative method of establishing control is needed.
The POS/LS will provide a multi-layer real-time error
evaluation and reliability assessment of the error evaluation
in the form of a QA/QC function. This feature will keep the
operator informed on the quality of the POS/LS position
and will warn him if the position error begins to drift outside of an operator defined accuracy specification.
The POS/LS will include an in-field primary reprocessing function that will automatically re-compute positions as
soon as the operator reaches a control tie point. These reprocessed positions will be sufficiently accurate in most cases
and can be delivered to the client on arrival to base camp
every evening. A full “final” post-processing will also be
offered along with a more complete QA/QC report and
graphic representation of errors, statistics, and maps of the
day’s work. Data handling time for pre-loading and post
processing will be reduced to less than an hour a day in the
office compared to full days of work for the current survey
supervisors.
micro-g. The INS is capable of 2 nautical miles/hour freeinertial navigation after a full ground alignment. Such an
inertial system does not represent the state-of-the-art performance that the POS/LS will incorporate, however it was
used because it was available for the first level of development. It also presented a design challenge to extract reasonable positioning performance from this lower quality inertial instrument.
The INS was connected to a data acquisition system that
time-stamped and recorded the IMU data coming out of the
INS. The equipment also included a roving L1/L2 GPS receiver and data acquisition computer that moved with the
INS and a base L1/L2 receiver. The GPS receivers were used
only to compute an accurate reference trajectory against
which the experimental POS/LS position solution can be
compared.
The reference trajectory was computed with Applanix’s
POSPAC post-processing software package [2]. POSPAC
contains a GPS processing component (POSGPS) that computes a kinematic GPS position solution with 2-5 centimeter accuracy, and a GPS-aided inertial navigation component (POSProc) as shown in Figure 1 that computes a postprocessed full 6 degree-of-freedom navigation solution with
the same accuracy. The real-time POS/LS solution was computed using an experimental version of RTSIM, Applanix’s
real-time POS simulator. RTSIM reproduces the real-time
embedded software in Applanix’s POS products on a PCcompatible computer. The POS/LS version of RTSIM used
in these tests included the following components specific to
a land surveyor:
•
The Auto-ZUPT function automatically detects when
the IMU is stationary, and performs ZUPTs so long as
the IMU remains stationary.
•
The Position Fix function accepts an operator-entered
position fix and corrects the POS/LS position
•
The reprocessing module implements a smoother/
corrector that reprocesses specified segments of the
POS/LS navigation solution to improve its accuracy
The results from two test programs are reported here. The
first test program comprised a series of van tests near
Applanix’s building in Richmond Hill, Ontario. The van carried INS, roving GPS receiver and data acquisition computers. The base receiver was located on Applanix’s building.
The van drove different trajectories including straight-line,
zig-zag and circular trajectories.
TEST RESULTS
“Straight-Line” Controlled Test
The following is a description of the results of experimental tests of a prototype land survey instrument. The IMU
used in these tests was a 20 year old RLG INS that was
designed for aircraft navigation. The RLGs and accelerometers exhibit respective biases of 0.01 degrees/hour and 200
This test followed a straight trajectory shown in Figure 3
with heading approximately Northwest for 1 hour and 3
minutes after the 20-minute initialization. The ZUPTs occur every 60 seconds and last for 15 seconds. The length of
the trajectory is 3.2 kilometers. The dynamic environment
134
Line 10 Test
Figure 3. Straight-line test trajectory (m).
Line 10 comprised 14 stakes approximately 80 meters
apart for a total of 1.1 kilometers. The Argo vehicle started
at one end and drove along the side of the road from one
stake to the next at a walking pace as set by one of the authors walking in front of the ATV (see Figure 6). It stopped
either at a stake or between two stakes for a 30 second ZUPT.
The time between ZUPTs was approximately 30 seconds.
Figure 7 shows a plan view of the trajectory in North and
East meters from the starting point. The distance is approximately 1.1 kilometers.
The ATV underwent significant dynamics because of the
characteristics of the vehicle, the uneven terrain on which it
drove and the severe mud that limited its maneuverability.
The INS was subjected to typical dynamics that a real land
survey would generate.
Figure 4. Straight-line real-time position errors (m).
in the van was fairly benign, hence the results are quite good
for the IMU being used. They do however demonstrate error characteristics that are typical for a straight-line survey
between position fixes.
Figure 4 shows the simulated real-time horizontal and
vertical position errors. Both errors have a growth trend that
is linear with time, and after approximately one hour both
grow to 1.5 meters. This performance can be classified as
1.5 meter in one hour or 0.6 meters per kilometer in the
horizontal and vertical directions. This is a typical error characteristic of an inertial surveyor on a straight-line trajectory.
Figure 5 shows the reprocessed horizontal and vertical
position errors. The horizontal position error has a maximum 0.35 meters with no linear growth trend. The vertical
position error has been reduced to less than 0.2 meters except for an occasional outlier. Reprocessing thus has improved the position accuracy by a factor of 5 or better.
Field Test
Applanix and Enviro-Tech Surveys Limited (Calgary
Alberta) jointly performed a field test of the experimental
setup in a typical environment west of Drayton Valley, Alberta
on 14 June 2000. Enviro-Tech provided a previously surveyed site and the Argo all-terrain vehicle (ATV) that carried the test equipment. The survey site contained three survey lines with stakes at approximately 80-meter intervals.
Line 10 ran along a road and provided fairly good GPS coverage. The other two lines 91 and 81 ran perpendicular to
the road along previously cut survey lines through the forest. No reliable GPS coverage was available along these lines.
The results from the Line 10 test are reported here because a
good GPS reference solution is available for this test.
135
Figure 5. Straight-line reprocessed position errors (m).
Figure 6. Field test survey in progress.
Figure 7. Line 10 trajectory plan view (m).
Top View: Latitude
200
0
-200
-400
-600
-800
Figure 8 shows the simulated real-time horizontal and
vertical position errors during the survey. Both errors grow
at an approximate rate of 2 meters per kilometer, which is
significantly larger than the rate during the van tests. This is
a consequence of the significantly higher dynamics and the
limited capability of this particular INS in these dynamics.
Figure 9 show the re-processed horizontal and vertical
position errors. The maximum horizontal and vertical errors
are respectively 0.8 meters and 0.4 meters, a significant improvement in accuracy from the real-time errors.
-1000
-1200
-8
-4
-2
0
2
4
6
Longitude
tages this instrument will provide are:
·
significantly increased productivity compared to
traditional and GPS survey methods,
·
significant reduction in environmental impact of land
surveying,
overall reduction of the cost of conducting a land
survey.
·
CONCLUSIONS
The POS/LS described in this paper will present foresters with a new way to perform their work. The key advan-
· POS/LS is an enabling technology for precision forestry
136
Figure 8. Real-time position errors on Line 10 survey (m).
Figure 9. Re-processed position errors on Line 10 survey (in
meters).
2D radial position differences (meters)
2D radial position differences (meters)
3.5
1.0
3.0
0.8
2.5
2.0
0.6
1.5
0.4
1.0
0.5
0.2
0.0
349500 350000 350500 351000 351500 352000 352500 353000
0.0
349500 350000 350500 351000 351500 352000 352500 353000
Time (s)
Time (s)
Down position difference (meters)
Down position difference (meters)
0.5
0.6
0.0
0.4
-0.5
0.2
-1.0
0.0
-1.5
-2.0
-0.2
-2.5
349500 350000 350500 351000 351500 352000 352500 353000
-0.4
349500 350000 350500 351000 351500 352000 352500 353000
Time (s)
Time (s)
The POS/LS will use Applanix’s aided inertial navigation technology found in Applanix’s current POS product
family. It will be a second generation GPS-aided inertial land
survey instrument that will incorporate state-of-the-art inertial sensors and new processing features such as reprocessing in a package that has the same size and weight as current generation GPS backpacks.
This paper has described the results of a requirement
analysis of the POS/LS and preliminary design and performance evaluation using available hardware. The test results
were intended to examine the signal processing functions
that the POS/LS will have and to demonstrate the performance characteristics of an inertial land surveyor.
The POS/LS will automatically use GPS data when available and provide position accuracy comparable to the GPS
position accuracy. It will perform automatic ZUPTs when it
is stationary, and request ZUPTs from the operator when it
needs to in order to maintain a specified position error rate
during marginal or no GPS coverage.
REFERENCES
George Siouris, Aerospace Avionics Systems, A Modern Synthesis, Academic Press 1993.
B. Scherzinger, A Position and Orientation Post-Processing
Software Package for Inertial/GPS Integration
(POSProc), In Proceedings of the International Symposium on Kinematic Systems in Geodesy, Geomatics and
Navigation (KIS97), Banff, Canada, 03-06 June 1997.
137
ACRONYM GLOSSARY
ATV
All Terrain Vehicle
GPS
Global Positioning System
IMU
Inertial Measurement Unit
INS
Inertial Navigation System
NED
North-East-Down
POS
Position and Orientation System
QA
Quality Assurance
QC
Quality Control
RTSIM
ZUPT
Real Time SIMulator
Zero velocity UPDate
138
Chapter 16
RealTime Harvester: The Future of Logging
BRIAN H. HOLLEY
Abstract—This paper will report on the application of current GPS-assisted technology integrated with timber harvesting. An
ongoing problem in the forest industry involves navigating through harvesting operations. The purposes of using GPS systems
with harvesting machines include steering clear of unintentional timber trespass, avoiding known hazards and buffer areas, and
determining optimum skidding distances. The searching and verification involved in these activities can result in lost time and
production. The consequences of failing to adhere to these constraints can result in greater costs and lost time, and possible public
relations and legal problems.
The integration of GPS and GIS technology can better prepare loggers, foresters, and landowners prior to and during timber
harvesting activities. With demands on the logger and forester continually on the rise, RealTime Harvester (RTH) will increase
efficiency and confidence in timber harvesting operations. This will also result in less downtime and lower rates of intrusion into
areas that are out-of-bounds. Issues to be addressed include data transfer, logger and forester education and training, data logging,
and future developments.
Via Differential GPS (DGPS), rugged computer technology, and user feedback we are monitoring user satisfaction, concerns,
accuracy of systems, and ease of use. We have received much interest from the forest community with very favorable responses.
INTRODUCTION
(SMZ’s), wells, and sensitive areas are just a few of the features that might be of importance to the logger. The logger
will know when he approaches any of these features, as it
will be displayed on the screen inside the harvesting machine. Making the logger aware of his surroundings could
make the difference between identifying a property line before a trespass occurs. Being aware of important features in
the field and where the logger is at any given time in relation to those features could mean the difference between a
successful harvest and a disaster that could have been prevented.
Many foresters are currently using GPS systems on a daily
basis for timber cruising, acreage determinations, and other
applications requiring precise locations. Enabling the logger to utilize data that has already been collected by the field
forester only makes sense. However, before the logger can
utilize this system, we must train the foresters on how to
deliver, or transfer, the information to the logger.
As a forester, it is hard sometimes to envision the forest
through the logger’s eyes. Many of the problems that face
foresters do not face the logger and vice-versa. Identifying
these differences and solving the problems that arise from
them does not always have to be difficult. With technology
at our hands and powerful tools to assist in our efforts, our
only limitation is our imagination.
It has been said that some loggers are not often businessmen. First, they are loggers and whatever follows is part of
logging. This situation is changing rapidly as requirements
upon loggers grow heavier. Burdens are increasing from all
angles: financially, environmentally, educationally, and technically. With the current depressed state of the forest
economy, loggers must find every edge to help make them
profitable, efficient, consistent, and reputable. Development
of the RealTime Harvester (RTH) system was designed to
help loggers achieve each of these goals. RTH is an integrated Global Positioning System (GPS) and navigation solution that is designed to aid the logger during harvesting
operations. RTH utilizes the flexibility of ArcPad (Environmental Systems Research Institute, ESRI) to provide the
solution for the mapping and navigation needs of RTH.
RTH is a system that consists of a rugged computer integrated with a Differential GPS (DGPS) receiver. This system will display the logger’s realtime location in relation to
features in or near the stand of timber he is harvesting. Property lines, harvest boundaries, streamside management zones
DISCUSSION
Forester Responsibilities
Before the logger can use RTH, either the forester or the
logger must collect data about the harvest area. In most
cases, the forester will be responsible for collecting the data.
Thus, for this discussion, we will assume the forester will
be collecting the data.
There are several different mapping solutions used in the
forest industry. The two GPS/GIS systems most commonly
139
easily transferred via communication ports or infrared ports.
The forester must then take the information to the field to
transfer the data to the logger. The transfer in the field will
occur via infrared communication to avoid having to use
cables. After in-the-field data transfer has occurred, the logger can now begin using the RTH system for harvest navigation.
Depending on the GIS system the forester is using, there
could be additional training on the ArcPad software. A
manual for step-by-step instructions will be created to prevent confusion and simplify the transfer process for the loggers and foresters.
Figure1: Forester Data Collection.
Financial Benefits
The financial burden facing logging operations is one that
does not recognize rainy days or mill quotas. It is a challenge that many foresters do not consider when creating
harvest plans or stopping harvest production. However, this
burden can be eased if managed correctly. The RTH benefits the logger and the procurement forester in several ways.
In many situations loggers are held legally responsible for
timber trespass unless contractually agreed upon by the timber company and the logger prior to harvesting. Many if
not all loggers are required to carry insurance for that specific reason. We feel RTH will give logging insurance agents
incentive to credit loggers that have RTH installed on their
machines because of their increased awareness in the field.
Even without insurance credits, the value of having a device
that displays approaching property lines and sensitive areas
could be significant. As the old adage states, an ounce of
prevention is worth a pound of cure.
As a forester, it is often difficult to find where the harvest
boundary should be drawn between each loading dock. By
being able to draw (Heads-Up-Digitize) on a GPS map and
give that map to the logger, a couple of problems are eliminated. It will save the forester from having to walk the harvest line and flag or mark the harvest line, which is a line
that really means nothing more to the logger than stop harvesting for this particular loading dock. It has no boundary
importance. This will also be more accurate than having to
run a compass in the field and hope your line is straight or
you are at the right point. The increased accuracy will also
benefit the logger. If the logger does not have to harvest
timber that would be better suited for another loading dock,
skid distances will be optimized. In turn, the value of RTH
has increased because of the logger’s efficiency in the woods
and the time it is saving the forester and timber company.
used in forestry are the systems provided by Corvallis
MicroTechnology (CMT) and Trimble Navigation. Both of
these systems are capable of giving accurate information to
the logger. However, there are several steps that need to be
taken.
First and foremost, the data collected must be collected
accurately. The legitimacy of RTH depends on working with
reliable data. After the data collection has been completed,
the forester must download and differentially correct his GPS
data, unless he is using a DGPS receiver. This is important
due to the fact that there is still a certain amount of error in
GPS signals. There are also foresters who still do not correct their GPS data, which could be risky if the logger is
depending on accurate data. It is also the responsibility of
the forester to locate and map the areas that are of interest to
the landowner, logger, or himself. When the file is given to
the logger, it must be complete and accurate.
After the data are collected and corrected, we must convert the data into shapefiles (the file format used by ArcPad
and other ESRI products) and transfer the data to the logger.
Both CMT and Trimble have the capability for this conversion. The most straightforward approach to transferring files
is to transfer the files from the forester’s computer to a
handheld Windows® CE device, such as a Compaq iPAQ,
Hewlett Packard Jornada, etc. These units are compact,
mobile and transfer data easily. After synchronizing the
handheld device with the forester’s computer, files can be
Figure 2: Example RTH Navigating Map
Environmental Benefits
With the ever-increasing environmental demands being
placed on loggers, something will be needed to help loggers
prevent environmental mistakes and make them more aware
of the impact they are having on the earth. Knowing where
SMZ’s are located could help loggers maintain the integrity
of the watersheds. Being able to locate the SMZ’s without
140
Figure 3: RTH Approaching an SMZ.
Figure 4: Logger Navigating Using RTH.
entering the zones will prevent unnecessary soil compaction
and will also help the logger plan for felling before reaching
the line. Instead of felling a tree toward a zone and having to
enter the zone to remove the tree, RTH will allow the logger
to know where the zone is and fell the trees in such a manner
that the tree can be removed without entering the zone. Not
only will this prevent unnecessary soil compaction but will
prevent the logger from having to physically remove any
limbs or debris from the stream or water body, which could
be time consuming and costly.
With the ability to locate sensitive areas, the logger can
plan his harvest with the stand in mind. Knowing how each
water body is situated and the flow of the streams, the logger
can plan the timber harvest and skid trails. This will help the
logger to minimize skid trails and minimize the impact during the harvest.
One of the benefits of RTH is its versatility. Not every
region of the United States has the same environmental challenges facing them. We must learn from the past and make
strides to better our conservation practices. The western
United States has seen many battles over the environment
pertaining to wildlife and water quality. While not having
the same wildlife species or water sources, the southeastern
United States is beginning to see the same struggles that
plagued the West Coast. The RTH system will let the logger
use the information the forester has identified on the ground
and RTH can also help loggers justify where they have been
with their equipment by recording GPS data as they are logging. This information will let the forester know where the
logging equipment has traveled in relation to the areas previously identified. If the logger has traveled across sensitive
areas, the forester will be able to identify those areas and
repair any damage caused by the logging operation.
Other areas to be protected by the forester may not only
be environmentally sensitive areas, but historical areas as
well. I was training a forester on a GPS system recently and
while out in the stand, we found a grave headstone. This is
just one example of historical areas that need to be preserved.
By identifying this area with a GPS system, we have the
data necessary to inform the logger to ensure its preservation. There are many more historical sites that need to be
noticed, marked and preserved, RTH can ensure this preservation from future logging operations.
Educational Challenges
It is widely known that many in the forest industry are
among the last to move toward existing and new technology. Within the U.S. forest industry, loggers generally lag
behind other professions on the technology curve. To ensure the survival of logging in the future, we must act now
to educate those that are technology averse. RTH is a system that would allow loggers to utilize the latest in technology and begin working on the same technology level as foresters.
The RTH system will need to be utilized by loggers that
will be able to access GPS data files. Whether collecting
the data themselves or using the data provided by the forester, the logger must establish a GPS file of each tract or
stand. This type of technology can be administered in two
ways, individual or group training. We initially plan on educating loggers on an individual basis, which is effective due
to the amount of personalized time that can be given to each
logger, but is not cost effective. The training will include
use of the RTH system and integration with the existing
GPS data of the forester. The end goal is to provide a system to the logger that he can operate efficiently and that he
perceives as an asset instead of a headache. The only way
to achieve this goal is effective training.
Loggers are currently being educated on a large scale on
better business techniques, effects of logging on the environment, and safety issues. We feel this type of collective
training can also be effective with RTH, if administered properly. The largest problem with training a collective group of
loggers is the time involved and less individual attention,
however it is more cost effective. This would mean that a
large meeting place away from all of the loggers’ operations
would have to be arranged. Some might have to take downtime to attend training and may include overnight travel,
which would be countered effective as to our end goal.
Other educational challenges involve the foresters themselves. Although there are many foresters who understand
141
technology and GIS systems, there are still those that are
technology averse. Correct GPS data collection, conversion and transfer will need to be taught in some cases.
ester will be able to email the file to the logger. The logger,
using a wireless Internet connection will be able to download the information as he sits in his harvesting machine
and begin to utilize the data within minutes. Likewise, the
benefit to the logger is the ability to provide tract data to the
forester in a short period of time. The largest obstacle at
this time is the limited range of cellular phone signals. Most
major interstates are covered fairly well but not all tracts of
timber are located near major interstates. As coverage improves so will the likelihood of the upgrade. The cost and
speed of a wireless Internet connection is another factor to
consider and should improve in the future as technology
improves. The ability to save the forester time translates
into a higher value for the logger involved, thus should mean
an increase in logging rates to compensate for the cost of
the system.
The use of DGPS to assist loggers in harvesting operations is not a new concept. There has been much research
and successful implementation of DGPS in harvesting operations in the Scandinavian countries partly due to the different logger-mill relationship than exists in the United
States. As technology increases, so will the application for
RTH. We can see the application taking the place of painting lines in the future. Instead of a forester painting SMZ
lines, the logger will be able to navigate to these lines. We
understand there are a number of issues to be resolved before this can happen, such as GPS satellite coverage and
GPS accuracy under canopy. We do feel, however, that these
will not be issues five years from now. If this is the case and
RTH can save the forester from having to paint SMZ lines,
large forest companies will be able to benefit tremendously
just on the cost savings from tree marking paint. As a check
behind the logger, the forester can ask the logger to provide
him with a harvest log file, which will be the file created by
the logger that will show his travel paths as he harvests the
stands. The forester will then be able to overlay the harvest
log on the original SMZ GPS line created by the forester
and compare to see if there has been an encroachment.
These are just a few enhancements for RTH in the future.
With the increase implementation of RTH, we hope to gain
more feedback to continuously develop a system that will
increase the value of loggers.
With the time constraints on foresters getting larger each
day, loggers can help ease the burden with efficient time
management in the field. RTH will allow the logger to harvest the stand with fewer questions and less confusion. This
is a value added service the logger will provide which will
increase his value to the company for which he is harvesting. Fewer mistakes, less confusion, quick, precise information and more efficient production are the aims of RTH.
Future Enhancements
RTH can help loggers in many ways as it stands. There
will be enhancements in the future to make the system fluid
and easy to use. As more loggers use the system we will
gain more feedback from the logging industry on what could
be done to improve the system. The following enhancements could be modified in the future to meet the request of
the logging industry.
Very soon, RTH will be able to collect GPS data as the
logger is harvesting the timber. The data could then be given
to the forester at the end of each week or at the end of the
job. This will allow the forester to view where the harvesting equipment has been on the tract and help the forester
and logger in several different ways. The forester will be
able to identify if the harvester has entered areas that should
not have been entered, such as SMZ’s. If the harvester has
entered these areas, the forester will know exactly where so
that he or she will be able to assess the damage, if any. Without RTH, the forester must walk every protected area to see
if the logger has encroached on the areas. This will also
save the forester time and energy in assessing harvested areas.
Collecting data with RTH does not have to be limited to
felling machines. RTH can be installed on skidders as well.
Many studies can be implemented by collecting data with
skidding machines. By collecting data with skidders, latitude/longitude points can be assigned to skid trails. This is
important because of the soil compaction that occurs with
normal harvesting operations. By knowing exactly where
the compaction has occurred, the growth rates of future
stands can be monitored to study the impact of soil compaction during harvesting operations.
Knowing where the skidder has traveled will also ensure
the forester that all of the felled timber has been removed. If
there is an area that has not been traveled by the skidder, the
forester can walk the area to make sure the skidder operator
has not missed felled timber. With the data collected by
RTH, identifying these areas will be much easier.
Another enhancement for RTH will be showing where
the equipment has traveled. As the harvesters or skidders
travel across an area, the area will be shaded a different color
to help the operators identify quickly where they have or
haven’t been. One enhancement that will save time for foresters and loggers is wireless data transfer. Once the file
has been downloaded and corrected in the office, the for-
142
Chapter 17
Vehicle Management System for Forest
Environmental Conservation
KAZUHIRO ARUGA
JUANG RATA MATANGARAN
KOJI NAKAMURA
RIN SAKURAI
MASAHIRO IWAOKA
TOSHIO NITAMI
HIDEO SAKAI
HIROSHI KOBAYASHI
INTRODUCTION
system, the harvester conducts harvesting operation according to the plan, a harvester’s computer obtains operational
information such as volumes, species, and locations of harvested trees in a harvesting operation, the harvester’s computer transfers operational information to the host computer,
the host computer plans a logging operation taking into account operational efficiency and the number of vehicle passes
allowed on the trail, the host computer transfers the plan of
the logging operation to the forwarder, and the forwarder
conducts the logging operation according to the plan.
Introduction of large forestry machines to Japan has improved operational efficiency and reduced industrial accidents. However, large forestry machines have had an impact
on forest environment, causing soil disturbance and residual
stand damages. Especially, reduction of tree growth response
is a serious problem in tropical forests, where soil compacted
by large forestry machines requires a long time to recover
(Matangaran, J. R. and Kobayashi, H).
To reduce the impact of soil disturbance, it is necessary
to restrict the trail areas and limit the number of vehicle passes
at the same trail. For proper vehicle management, it is important to plan forest operations by using information about
forest properties and vehicle specifications, to conduct forest operations according to plan, and to evaluate the forest
environment after the operation. The paper describes first
the vehicle management system and then discusses the number of vehicle passes that should be allowed on a given trail
so that the tree growth response is not affected.
Location and Operational Information
As is well known, most Timberjack harvesters are
equipped with the Timberjack 3000 measuring and control
system. The system allows optimized processing of a trunk
based on price and priority lists. The system also allows easy
monitoring of the harvested volumes and species, and twoway real-time data transfer is possible between the harvester
and the forest company (http://www.timberjack.com/).
It is also known that Ponsse harvesters are equipped with
the advanced harvester data system Ponsse Opti. The system is easy to use because it is based on a normal PC using
the Windows operating system. The Windows operating system enables data transmission and map applications during
logging. The GPS system and maps of the stand marked for
cutting make it possible for the operator to see, on a large
color screen, the location of his machine, the borders of the
stand marked for cutting, and any natural preservation areas
(http://www.ponsse.com/).
If these systems are used in Japan, there are two problems, namely, the accuracy of GPS and the communication
system. Positional error when using GPS is between 5 to 10
m in open areas because the Selective Availability of GPS
was cancelled on May 2nd, 2000. When a vehicle runs on a
THE VEHICLE MANAGEMENT SYSTEM
We have studied the arrangement of strip roads and logging trails, which is important for improving operational efficiency, by means of the simulation with GIS (Kobayashi,
H., Sakurai, R., Cho, K., Sakai, H., Iwaoka, M. and Nitami,
T). This simulation model will present a vehicle management system by introducing factors relating to both operational efficiency and the number of vehicle passes allowed
on a trail (Nitami, T. and Kobayashi, H). This system will
consist of a host computer in an office, local computers processing location information from GPS and operational information in machines, and a network connecting the office
and the machines. If we assume a harvester and a forwarder
143
Figure 1. GPS data transfer examination. Copyright©1999 ALPS Mapping Co., LTD.
stump that is 2 m away from another stand, GPS may mistakenly assume that vehicle is positioned farther from this
neighbor stand for such a level of GPS error. Therefore,
GPS cannot be trusted to guide vehicles on logging trails.
However, the error of DGPS calculated from the GPS data
of smaller DOP is 2 m even in the forest (Hasegawa, H.,
Yoshimura, T., Yamate, N., Sakai, S. and Fukuda, M). The
Agency of the Japan Costal Guard started to broadcast differential correction information from April 1999 all over
Japan. In this way, it was easy to measure the position of
GPS users using real-time DGPS.
panies such as Hitachi Construction Machinery Co., Ltd. have
started to equip construction machines used as base-machines
of forestry machines with satellite communication systems.
If such machines become widespread, location and operational information will be transferable anywhere in a forest.
DETERMINING THE NUMBER OF
VEHICLE PASSES ALLOWED ON A
TRAIL
Seedling Growth Response in Compacted Soil
According to logging operation rules in Indonesia, forest
companies are obligated to plant some local tree species in
all skidding trails. This obligation ensures the sustainability
of the forest for the next cycle and environmental protection.
However, many forest companies neglect their obligation,
and the level of soil hardness in the skidding trail becomes
dangerous for the sustainability of the forest. If the critical
limit of seedling growth in compacted soil can be determined,
the skidding intensity in logging operation can also be
planned.
The research was conducted to examine the growth response of seedlings of Shorea selanica at certain levels of
compacted soil (bulk densities 0.9, 1.1, 1.3, 1.4 g/cm3 with 9
replications). This species is a member of the
dipterocarpaceae family. The soil type was podsol soil. The
soil contained about 70 % silt and clay. The plastic pot was
filled with compacted soil, and the Shorea selanica seedlings were planted in the upper layer of soil. In each pot, only
one seedling germinated (Figure 2). After the seedlings had
grown for 6 months in the pot, the depth of root penetration
and height growth were measured.
Duncan’s multiple range test was performed to find out
the level of bulk density which caused the critical limit of
Communication System
As cellular phones are very popular in Japan, they could
be instrumental in spreading the vehicle management system. We examined transferring GPS data by cellular phones
in Tokyo University Forest in Chichibu. A car equipped with
a GPS receiver (Trimble AgGPS124) traveled from Tokyo
University Forest office to the end of the Irikawa Forest
Road along R140 and the Irikawa Forest Road. The distance between the office and the end of the Irikawa Forest
Road is 25 km. GPS data obtained from the GPS receiver
was downloaded to a computer in the car and transmitted
from the computer to a computer in the office through a
cellular phone. It was demonstrated that GPS data can be
transmitted anywhere except for sections such as tunnels
and the forest road along a stream (Figure 1).
The maximum data transfer speed of a cellular phone is
9,600 bps. This may be slow for transferring operational
information and GPS data. The speed of the next-generation cellular phones launched last year is 38,400 bps. Therefore, it will be possible to transfer operational information
by using next-generation cellular phones. Furthermore, com144
Figure 2. Shorea selanica seedling in compacted soil.
Figure 3. Relationship between rut depth and cone
index (n=10).
] Q
' &
, ( *
3
P
F
]
K
W
S
H
G
W
X
5
5
&
R Q H L Q G H [ R I X Q G L V W X U E H G V R L O &
, N J I F P
Figure 4. Relationship between cone index and rut depth.
Table 1. Duncan’s multiple range test for root
penetration and seedling height.
Bulk density (g/cm3)
0.90
Root penetration (cm) 13.3
**
1.10
1.30
11.1 **
7.10
**
3.70
15.6
13.1
**
10.4
1.40
Seedling height (cm)
18.6 **
P
F
]
K
W
S
H
G
W
X
5
Note: ** significant difference at 99 % confidence limit
\ [ 5
seedling growth response. From Table 1, it can be seen that
the average root penetration response in each level of bulk
density was significantly different at 99 % of confidence
level. The average of seedling height response did not indicate the difference at the 99 % level for 1.1 and 1.3 g/cm3 of
soil hardness. This means that the height growth was not
significantly influenced by the level of bulk density at 1.1
and 1.3 g/cm3. On the other hand, the bulk density, 1.3 g/
cm3 and 1.4 g/cm3, showed a difference concerning seedling height response. The growth response was significantly
reduced at the level of 1.4 g/cm3. Therefore, bulk density up
to 1.3 g/cm3 was determined to be the tolerable limit of soil
hardness for height growth.
R Q H L Q G H [ D IW H U W U D F W R U S D V V H V &
, N J I F P
Here, EGPtracks is the Effective Ground Pressure of the tracked
machine, CIb is the cone index of the undisturbed condition
before a tractor passes (surface to 50 cm depth), D is a standard fixed wheel diameter that would give a valid comparison between machines, 1.5 m, n is the number of tractor
passes on the trail, and z is rut depth.
Figure 4 expresses the relation between cone index after
logging and rut depth, and Figure 5 explains the relation
between cone index and bulk density. Figures 4, 5, and 6
can be used to apply a restriction in the number of tractor
passes on a trail. For example, bulk density 1.3 g/cm3 equals
cone index 8.1 kgf/cm2 (Figure 5), and the cone index value
8.1 kgf/cm2 equals to 11.3 cm rut depth (Figure 4). If the
cone index of undisturbed soil is 1.5 kgf/cm2 and the rut
depth 11.3 cm, a tractor can be allowed on a trail 6 times.
Rut Estimation
Once the critical limit of the bulk density that impairs on
the seedling growth is known, the next problem is how to
decide the number of tractor passes allowed on a trail in
conjunction with the level of bulk density, cone index, and
rut depth. There is a relationship between soil hardness and
rut depth. Rut depth may be the most convenient way to
measure the level of soil hardness in the field. Rut depth (z)
can be predicted from the cone index of the condition of
undisturbed soil (Wronski, E. B. and Humpreys, N). In the
research conducted in Tokyo University Forest in Hokkaido,
the relationship could be expressed by the following formula (Figure 3):
z = 3.22n0.50D(CIb/EGP tracks)-2.60
&
CONCLUSIONS
In this study, the vehicle management system was explained, and the method used to decide the limit number of
vehicle passes allowed on a trail was discussed. From the
above research, the decreasing tendency of growth response
and the percentage of seedling growth reduction in compacted soil were shown, and the critical limit of seedling
(1)
145
growth response can be determined. In addition, a theoretical formula could be used to estimate the ruts by using the
cone index value and the ground pressure of the tractor. The
bulk density, 1.3 g/cm3, can be considered as the critical
value for Shorea selanica seedling growth response in compacted soil. If the critical limit of seedling growth response
is 1.3 g/cm3, 6 times is the maximum number of times a
tractor should be allowed on a trail. More than 6 times or
more than 11.3 cm rut depth means that the area must be
replanted for the sustainability of the forest. By introducing
the number of vehicle passes allowed on a trail, we can develop a vehicle management system for forest environmental conservation in conjunction with GIS, GPS, and communication systems. We must decide the critical values for
Cryptomeria japonica, Chamaecyparis obtusa, and so on
in order to use this system in Japan.
Figure 5. Relationship between bulk density and cone index.
P
F
I
J
N
,
&
[
H
G
LQ
H
Q
R
&
\ [ 5
% X ON G H Q V L W\ % ' J F P
Figure 6. Relationship between rut depth and cone index
(n=2, 4, 6, 8, 10).
LITERATURE CITED
Hasegawa, H., Yoshimura, T., Yamate, N., Sakai, S. and
Fukuda, M. (1998) A study on the accuracy and the
method of surveying with differential GPS in mountainous areas. J. Jpn. For. Eng. Soc. 13(2) : 89-98. (in Japanese with English summary)
P
F
]
K
W
S
H
G
W
X
5
Kobayashi, H., Sakurai, R., Cho, K., Sakai, H., Iwaoka, M.
and Nitami, T. (1999) Thinning plan using a technique
of GIS. J. Jpn. For. Eng. Soc. 14(3) : 193-198. (in Japanese with English summary)
&
R Q H LQ G H [ R I X Q G L V W X U E H G V R L O &
,
N J I F P
Matangaran, J. R. and Kobayashi, H. (1999) The effect of
tractor logging on forest soil compaction and growth of
Shorea selanica seedlings in Indonesia. J. For. Res. 4(1)
: 13-15.
Nitami, T. and Kobayashi, H. (2000) Environmental damage control by information system for harvesting vehicles.
In Proceedings of IUFRO world Congress 2000 : 77-78.
Wronski, E. B. and Humpreys, N. (1994) A method for
evaluating the cumulative impact of ground-based logging system on soils. J. For. Eng. 5(2) : 9-20.
146
Chapter 18
Using A Laser Rangefinder To Assist Harvest Planning
MICHAEL WING
LOREN KELLOGG
Abstract—Harvest planning includes locating and mapping unit boundaries, landings, skid trails, skyline corridors, and special management areas. Locating and mapping tasks associated with harvest planning have traditionally been performed using
manual survey techniques with varying degrees of precision, accuracy, and efficiency. This paper examines the use of a laser
rangefinder in conjunction with a digital compass to digitally capture and map data associated with harvest planning. The benefits
and limitations of this digital approach over traditional data collection techniques in terms of assisting harvest planning and layout
activities are discussed. The potential benefits of global positioning system (GPS) technology in working with data from the laser
rangefinder and digital compass equipment are also briefly discussed.
INTRODUCTION
METHODS
Traditional harvest planning techniques include locating
and mapping unit boundaries, landings, skid trails, skyline
corridors, roads, and special management areas. Locating
and mapping tasks associated with harvest planning have
been typically performed through manual survey techniques
with varying degrees of precision, accuracy, and efficiency
(Kellogg et al. 1997). Inaccuracies in location and mapping activities can add to the cost of harvest operations. Tasks
associated with harvest planning have become more demanding as the focus of forest management, particularly in the
national forests, has shifted from timber production to include other goals that can be broadly defined as ecosystem
values (Swanson and Franklin 1992). This focus shift has
required harvesting operations to adopt practices that include
uneven aged management techniques such as patch cutting,
selective thinning, and multiple entries (Kellogg et al. 1996).
Uneven aged management techniques place greater demands
on harvesting operations in terms of the types and accuracy
of forest data that must be collected for harvest planning
and layout. One solution to meeting this challenge is to use
technologies that can collect and analyze forest data more
efficiently and accurately than through traditional means.
This paper examines the use of a laser rangefinder in conjunction with a digital compass to digitally capture and map
data associated with harvest planning. The laser rangefinder
allows field planners to capture locations and gradients of
objects up to several hundred feet away while the digital
compass calculates an azimuth towards an object. The benefits and limitations of this digital approach over traditional
data collection techniques in terms of assisting harvest planning and layout activities are discussed. The potential contributions of global positioning systems (GPS) to this data
collection approach are also briefly discussed.
We tested a laser rangefinder coupled with a digital compass. Both of the instruments were manufactured by Laser
Technology Inc. The laser rangefinder is capable of detecting distances and gradients between the instrument and an
object within visible distance. The digital compass works
in tandem with the laser rangefinder and simultaneously computes an azimuth and distance between the instrument and
the object. These devices are portable and can either be carried by hand or mounted on a staff to provide stability and
support to both the user and instrument.
The operation of the laser rangefinder is relatively straightforward once the user understands several mechanics of the
instrument. The laser rangefinder has two lenses on its front
or forward-looking panel. One lens emits short pulses of
infrared light while the other lens is designed to receive
emitted pulses that reflect back from a targeted object. Distances and gradients are measured by calculating the amount
of time it takes for a signal to reflect off of an object and
return to the receiving lens. Although signals are easier to
detect from objects with smooth, reflective surfaces, signals
can also be reflected from objects with coarse surfaces. By
itself, the instrument weighs one kg, fits into the palm of a
hand, and is approximately 15 cm long (Figure 1). A sighting device is located on top of the instrument and a red dot is
centered in the scope to assist the user with aiming the device. A LCD screen on the front panel of the instrument
provides a visual menu system. The user navigates through
the menu system to select the type of measurement desired.
Options include object height, inclination and three measures of distance: horizontal, slope, and vertical. The user
takes a measurement by using the red dot in the sighting
scope to center an object in the instrument’s aiming path
and pressing a button on the side of the instrument; measurements are immediately displayed on the LCD
147
Figure 3. Laser rangefinder
strument. As the laser
and digital compass tandem
rangefinder and compass are
mounted on staff.
turned in different azimuths,
an audible tone informs the
user of proximity to the target azimuth; as the user gets
closer, the tone increases in
pitch until the target azimuth
is found. This signals the
user to take a measurement
and can help maintain a bearing when multiple measurements are required following
a consistent heading.
Although the laser
rangefinder and digital compass combination provides a
visual output of measurements on a LCD screen, digital data
recorders are available in a variety of formats ranging from
HP 48 calculators, to small hand-held collection units, to
mini-computers. These devices can be hooked directly to
the laser rangefinder or compass to collect and store measurements from either or both units. These accessories have
the potential to make field data collection more efficient and
to make data transfer to a GIS or mapping system more expedient. One possible disadvantage is that the digital collection approach means that another piece of equipment must
be carried and operated by the user, and additional costs are
imposed by equipment and accessory purchases.
We tested the laser rangefinder and digital compass combination in two typical field forestry applications to determine whether these instruments might be useful for harvest
planning and layout. Our first test was to use the instrument
combination to create a traverse around a harvest planning
unit. The second test was to run a terrain profile along a
skyline corridor route. Both of our tests were conducted in
Oregon’s Coast Range in predominantly 55-75 year old Douglas-fir (Pseudotsuga menziesii) stands covering slopes ranging from two to 50 percent. The objective of these tests was
to provide baseline information for assessing the potential
uses and disadvantages of the digital equipment in a set of
realistic harvest planning activities.
Figure 1. Laser rangefinder.
screen. The resolution of distances collected by the instrument is within 0.01 m and inclination accuracy is 0.1 degree. The maximum distance of objects that may be sighted
is listed at 575 m (Adams and Zukowski 1997).
The digital compass weighs approximately 1.25 kg, is 31
cm tall, and 3 cm by 5 cm wide (Figure 2). The instrument
is designed to calculate azimuth or direction between the
user and an object. The reported accuracy of digital compass measurements is 0.3 deFigure 2. Digital
grees and the resolution is 0.1
compass.
degrees (Mannik et al. 1998).
The digital compass can be
physically and digitally connected to the laser rangefinder
through a mounting bracket
and cable (Figure 3). After the
digital connection has been established between the two instruments, magnetic declination for the data collection area
must be entered into the instrument so that astronomic north
is used in calculations. In addition, calibration of the compass is necessary to ensure that
magnetic variations are accounted for in azimuth measurements. These calibration
procedures require about 15
minutes of time and are relatively easy to undertake. Azimuth between the user and an object within visible distance
are calculated simultaneously as distance measurements are
recorded by the laser rangefinder. A digital display of measurements is provided on the compass LCD screen.
One of the unique features of the digital compass is its
audible aiming assistance. For applications when a consistent azimuth is desired in measurements, such as creating a
terrain profile, a target azimuth can be entered into the in-
RESULTS
Our first test was to create a traverse around a harvest
planning unit. Foresters typically run a traverse to locate
stand boundaries, leave areas, riparian boundaries, or any
other measurement operation in which a boundary or zone
must be delineated. We used two field personnel and had
them record data manually in a field notebook. One person
operated the rangefinder and was required to sight and take
measurements of objects along the traverse line. Whenever
possible, the reflective surface of the helmet worn by the
other person was used to sight and take measurements along
the unit boundary. Dense vegetative understory, such as vine
maple (Acer circinatum) made getting reflective signals
148
challenging and in some
Figure 4. Traverse area and
cases impossible. When
points collected by laser
this occurred, the boles of
rangefinder and digital
trees that rose above the
compass.
understory were used to
record distances. The
flexibility of the laser
rangefinder to record both
horizontal distances in
addition to slope distances proved advantageous in this regard.
When
the
laser
rangefinder and compass
returned a distance and
bearing, these measurements were verbally communicated to the second person. The second person was
required to record the measurements in a field book and to
keep descriptive notes of the objects that were used to create the traverse. The second person was also required to
delineate break points along the traverse where changes in
the course or slope of the unit boundary required taking
length and gradient measurements. The perimeter of the
area traversed was approximately 1,291 m (Figure 4) and
was covered in three hours.
Our second test was to run two terrain profiles so that
elevation change along a cross section could be derived.
Terrain profiles are typically created by field foresters using
a clinometer, tape, and compass. Terrain profiles may be
desirable in assessing skyline corridors, skid trails, road locations, and other harvesting operations in which slope is a
consideration. Our first profile began just uphill from a
potential logging landing. The profile ran downhill, ended
at a creek bed for a total of 169 m (Figure 5), and required
approximately one-half hour to complete. Our second profile was taken using an actual cable logging corridor that
ran from the creek bed to the landing site for a total of 104
m (Figure 6) and took about twenty minutes to complete.
For both profiles, we input a target azimuth into the digital
compass and used the audible tone option described above
to maintain a consistent course.
Figure 5. Terrain profile.
Figure 6. Terrain profile of cable yarding corridor.
DISCUSSION
We found the laser finder and digital compass combination to be an effective set of tools in helping us to locate and
create spatial data for a traverse and set of profiles. This
digital approach to collecting data offers advantages over
traditional data collections means in that distance and direction information can be quickly and accurately calculated.
We found that we could generally take distance measurements quickly and did not have to worry about controlling
sag, tension, or for uneven ground surfaces as would be required with a tape. The ability of the laser rangefinder to
almost instantly return horizontal or slope distances also
provided us with flexibility in returning the type of measurement we desired. The digital compass quickly calculated azimuths at the same time as distances and were easy
to read from the LCD display. Although we did not test the
accuracy of measurements returned from the laser
rangefinder and digital compass, if measurements are within
the manufacturer’s accuracy specifications (listed above),
they would be more accurate than the traditional measurement tools such as a tape, clinometer, or compass. These
traditional tools have long been used to collect traverse and
profile information. One of the challenges in their use is
ensuring that accurate and precise measurements are being
read and recorded from all instruments. During inclement
weather or in demanding conditions, such as those found on
steep, densely vegetated forest landscapes, the ability of field
foresters to consistently discern instrument measurements
may be challenged. In addition, the skill level of field foresters in using these instruments will influence the quality
and reliability of measurements. One advantage, however,
of manual measurement techniques is that tools such as a
steel or cloth tape are more durable than digital tools and
may require less maintenance and care.
One of the primary disadvantages of the laser rangefinder
was the difficulty we encountered in penetrating the under-
149
story. Leaves and even small branches between the user and
the object reflected the laser signal and returned distance measurements that were short of our true objective. In some cases,
we could shorten the length of distance measurements and
attempt to move through vegetation and avoid blocked paths.
In other cases, we were able to move the vegetation from the
path of the laser. Another way of handling this hindrance was
to sight tree boles above the understory and use these as
traverse points. Unfortunately, trees weren’t always conveniently placed to solve all difficulties with line of sight obstacles.
Other strategies for working with obstacles in the line of
sight that we did not test include the filter and gate options on
the laser rangefinder. The filter option setting will configure
the laser rangefinder so that only signals from reflective surfaces are returned. The gate option sets a distance range for
objects so that distances are only returned from those objects
that fall within a user-specified minimum and maximum distance.
Data collected by the laser rangefinder and digital compass can also be incorporated into other mapped products,
such as digital orthophotos or digital raster graphs (DRGs)
provided that a control point has been established that links at
least one vertex of the traverse or profile to a known location.
This is an application where GPS might prove useful. The
application of GPS technology has been limited in Pacific
Northwest forests due to the limitations of receiving continuous sufficient satellite signals and risks of multipath errors
that can result from dense vegetation and dissected terrain
(Van Sickle 1996). With GPS mission planning that identifies the times during which GPS receivers have the best opportunity to receive satellite signals, users can usually collect
several data points to establish ground coordinates for a
traverse or profile. Once ground coordinates have been established and transferred, users can view their field data products in conjunction with other mapping products through a
geographic information system (GIS). Mapping and terrain
analysis conducted using a GIS can assist operation planners
in locating and scheduling harvest activities.
Although the maximum distance of the rangefinder is listed
as 575 m, we found the practical maximum distance to be
around 75 m during our forestry applications. We determined
this by gauging our ability to repeatedly collect measurements
across a range of distances. The practical maximum distance
assumes that the rangefinder is mounted on a staff or some
other form of support that stabilizes the instrument. The practical maximum distance is around 40 m without instrument
stabilization. These distances could be extended if a larger
and more cooperative reflective surface is used; in our applications, the reflective portion of a hardhat provided a moderately useful surface to work with. A more active solution might
be to use reflective tape mounted on a flat surface.
the most important of these criteria are utility, ease of use,
and cost. Our initial tests have provided encouragement
that the laser rangefinder and digital compass combination
can provide some advantages over manual data collection
techniques. The equipment was not difficult to learn to use
and the relatively small size and weight of the equipment
did not impose significant burdens on field staff. When the
equipment was mounted on a staff, extra stability was provided to the field operator and the practical range of the
instrument was extended. The cost of the equipment was
also not prohibitive. The retail cost of the laser rangefinder
is about $2,800 and the digital compass sells for about
$1,500. These prices are well within the price range of most
organizations that are involved in forest operations.
Our future work will involve evaluating measurements
taken by the rangefinder to those collected by traditional
measurement tools. We plan to make comparisons across
several different forest traverses and profiles so that results
of both approaches can be contrasted in terms of the time
required to collect the data and the quality of the final data
product.
REFERENCES
Adams, J. and L. Zukowski. 1997. Impulse manual. Laser
Technology Inc., Englewood, CO.
Kellogg, L.D., P. Bettinger, and R. M. Edwards. 1996. A
comparison of logging planning, felling, and skyline yarding costs between clearcutting and five group-selection
harvesting methods. Western Journal of Applied Forestry
11(3):90-96.
Kellogg, L.D., G.V. Milota and B. Stringham. 1997. Logging planning and layout costs for thinning: Experience
from the Willamette Young Stand Project. Oregon State
University, Forest Research Laboratory, Research Contribution 20.
Mannik, L., Adams, J. and S. Colburn. 1998. MapStar Electronic Compass Module User’s Manual. Laser Technology Inc., Englewood, CO.
Swanson, F.J. and J.F. Franklin. 1992. New forestry principles from ecosystem analysis of Pacific Northwest forests. Ecological Applications 2:262-274.
Van Sickle, J. 1996. GPS for land surveyors. Ann Arbor
Press, Chelsea Michigan.
ACKNOWLEDGMENTS
CONCLUSIONS
The authors wish to acknowledge the support of the Center for Wood Product Utilization and the assistance of Ben
Spong, OSU Forest Engineering graduate assistant.
Digital tools must meet several criteria if they are to be adopted
and used to improve the efficiency of forest operations. Among
150
Chapter 19
Using GPS to Evaluate Productivity and Performance of
Forest Machine Systems
STEVEN E. TAYLOR
TIMOTHY P. MCDONALD
MATTHEW W. VEAL
TON E. GRIFT
Abstract—This paper reviews recent research and operational applications of using GPS as a tool to help monitor the locations, travel patterns, performance, and productivity of forest machines. The accuracy of dynamic GPS data collected on forest
machines under different levels of forest canopy is reviewed first. Then, the paper focuses on the use of GPS for monitoring forest
harvesting and site preparation equipment. Finally, the paper discusses future trends in precision forestry for intensive forest
operations.
INTRODUCTION
racy of travel patterns and velocities of a tractor operating
in open sky and forest canopy conditions. They reported
that GPS successfully tracked machines under open sky conditions; however, under forest canopies there was a major
decrease in accuracy.
Veal et al. (2002) further quantified accuracy of GPS position data collected on wheeled skidders. Two different
commercially available GPS receivers (12 channel Trimble
ProXR and six-channel GeoExplorer II) were used to track
wheeled skidders under different canopy conditions at two
different vehicle speeds (5.4 kph and 9.1 kph). Three different courses were established in a loblolly pine plantation
under different forest canopy density conditions: open
canopy (0% crown cover in a recent clearcut), light canopy
(57% crown cover in a heavily thinned area), and heavy
canopy (85% crown cover in a lightly thinned area). While
the skidders traversed each course, locations of the tires were
marked. Traditional optical surveying techniques were used
subsequently to determine the actual tire track locations.
These actual vehicle locations were compared to the GPS
machine paths to determine errors.
Maps from data collected by both receivers showed general travel patterns of the skidders; however, as canopy density increased, more discontinuities and irregularities were
observed in GPS maps, especially for the GeoExplorer II.
These discontinuities were attributed to multipath effects
and to receivers switching satellite constellations. Figure 1
shows mean position errors in GPS data collected by both
receivers under different canopy conditions. Position accuracy showed a decreasing trend as the canopy changed from
open to heavy. For example, mean 3D position errors for
the ProXR were 1.26 m, 1.77 m, and 3.76 m, for the open,
light, and heavy canopy conditions, respectively. It is important to note that some of these errors are similar to the
The Global Positioning System (GPS) is being used in an
ever-increasing array of applications for managing forests
and our natural resources. When used on mobile forest harvesting machines, data collected from GPS and additional
external sensors can improve forest engineering design and
management decisions based on machine performance data
as a function of terrain and timber stand variables. Applications employing GPS capabilities are being developed for
use in site preparation, planting, and managing intensive
culture plantations. Many of these applications stem from
the successful integration of GPS into “precision agriculture”, which can be defined as managing crop inputs, such
as fertilizer, herbicide, etc. on a site-specific basis to reduce
waste, increase profits, and maintain the quality of the environment. Developments in GPS technology and precision
agriculture are readily adapted to problems in forest operations, particularly in intensive forest production systems. This
paper will review recent work at Auburn using GPS as a tool
to help researchers and practitioners measure performance
and productivity of forest machines. Also, future trends for
precision forestry in intensive forest management will be
discussed.
Accuracy Considerations in GPS Machine
Tracking
Before discussing applications for GPS machine tracking, it is important to understand the accuracy of dynamic
GPS data collected in forest conditions. Most previous studies on GPS accuracy under a forest canopy were concerned
with static positions. Spruce et al. (1993) used a typical
mapping-grade GPS receiver and measured relative accu151
Background
Figure 1. Mean dynamic GPS position errors for different
canopy conditions. From Veal et al. (2002).
Among those using GPS-based systems to monitor forest machine systems, McMahon (1996, 1997) used GPS to
evaluate site disturbance of tree-length harvesting systems
by mapping paths of harvesting machines. His system then
made a mathematical transformation of the data to calculate
the amount of area disturbed by the machine. The result
was a map that showed the number of times that a machine
passed over a given point during the harvesting operation.
The GPS-based machine location point data were given a
width corresponding to the machine width plus two times a
mean position error. The position error data were based on
comparing GPS point data collected over time at a known
point against location data collected at the point using survey-grade GPS. The mapping system allowed the researcher
to detect areas where multiple skidder passes occurred and
then to predict areas of greater soil compaction due to the
logging operation.
Thompson et al. (1998) used GPS to map the movements
of tracked skidders in southeastern Australia. They produced
maps of skid trail networks and measured traffic intensities
(in terms of number of passes) over the network. Also, they
estimated the time that the machine was performing each of
six different machine cycle elements. For their study, they
concluded that the methods were sufficiently accurate to
indicate traffic intensities and machine productivity information. Although it was not always possible to to determine what the machine was doing at all times, the system
performed well in defining travel out and travel in time elements. Traffic intensity data generated by the GPS-based
methods were reliable and compared well with field observations. However, they noted that using the off-the-shelf
GPS unit was very labor intensive. They recommended that
for similar work in the future, it would be helpful to have
more of a “black box” that would automatically start and
stop when the machine was used, and that would record data
over a longer, unattended period.
Reutebuch et al. (1999) tried to determine time study data
using GPS receivers on a feller buncher, a hydraulic shovel,
and a tracked skidder. Machines were monitored so cycle
distances and times could be calculated. Due to apparent
errors in the GPS positions (occasionally over 100 m), travel
distances could not be calculated accurately. In addition to
studying harvesting systems, the commercially-available
“Silvitracs” GPS system was used by Stjernberg (1997) to
map movements of site preparation equipment. This system provides information on the travel patterns of both mechanical and chemical site preparation equipment. The system also provides data on area covered during the operation
and gross productivity information.
Canopy Effects
6
5
Mean 4
Position
3
Error
2
(m)
1
Heavy
Light
0
ProXR
ProXR
2D
Geo 2D
3D
Geo 3D
Open
Canopy
Receiver Type
width of the machine. Tests conducted before and after deactivation of Selective Availability showed little differences
in differentially-corrected GPS data. Finally, the machine
speeds tested did not significantly affect accuracy of GPS
positions for either receiver type.
These results indicate that researchers or practitioners
need to be cautious when relying on GPS to track forest
machines in heavy forest canopies. For general knowledge
on where machines travel, many mapping-grade GPS receivers, when using differential correction, will probably be
sufficient. However, for more detailed studies of travel patterns or environmental impacts at specific points (e.g. soil
compaction studies), typical mapping grade receivers may
not have the level of accuracy needed. For detailed studies
that require sub-meter or sub-centimeter accuracy, more sophisticated hardware and firmware will be needed.
USING GPS TO MONITOR FOREST
MACHINE SYSTEMS
Several recent research efforts have been aimed at learning more about performance, productivity, and site impacts
from forest harvesting machines and from site preparation
equipment by using GPS to monitor machine movements.
Traditional methods of studying machine productivity and
site impacts required researchers to work in close proximity
to machines to observe and videotape activities and travel
paths of the machines and people. These labor-intensive
methods pose safety problems for personnel involved in the
study. To alleviate these problems, GPS receivers can be
mounted on each machine of interest to determine travel
paths and velocities of machines during operations. Additional sensors and data acquisition equipment also can be
installed on machines to record information on machine functions or performance. Using these types of data allows researchers to quantify and model productivity or potential
site impacts of individual machines or the entire machine
system.
Estimating Harvesting Impacts from GPS
Machine Tracking
In work at Auburn, McDonald et al. (1998a) developed a
method to use GPS tracking data to determine the area impacted by a machine as it traveled over a site. The method,
which was similar to one presented by McMahon (1997),
152
properties by measuring changes in the properties at select
points that corresponded to estimated traffic intensities
within the harvest tract. They found that bulk density and
cone index responded to increased traffic intensities and
achieved peak values after a limited number of passes. This
ability to compare detailed data on traffic intensities with
soil strength properties would not have been possible without integration of GPS into the impact assessment process.
Figure 2. Harvest traffic intensities (number of vehicle passes
indicated by colors) monitored by GPS during a clear-cut
harvest of a loblolly pine plantation. From Carter et al. (2000a).
Number of
Passes
Harvesting Machine Tracking – Automated
Time Study
>11
6-10
5
The machine tracking work begun by McMahon (1997) and
McDonald et al. (1998a, 1998b, 1998c) was extended to
facilitate time and productivity study of harvesting machines
by McDonald (1999) and McDonald et al. (2000a, 2000b).
The system to develop time study data solely from GPS position information was implemented using two components:
1) a feature extraction sub-system to identify characteristics
of a machine path, given some site-level information, independent of the type of machine being tracked, and 2) an
event processor that applied machine-specific knowledge
to combine characteristic movements and sub-events into
operational functions. The intention was to develop a system that incorporated no domain-specific knowledge and
was therefore useful to analyze the functional performance
of any type of machine where movement and position were
important factors in its operation.
McDonald et al. (2000a) conducted further research using GPS for unattended time study of grapple skidders.
During field operations, a time study was conducted by researchers using traditional methods. The GPS data were
reduced to movement-defined events, then movement events
were combined into machine functions, and elemental times
(travel loaded/empty, delimbing, positioning and grappling)
were determined. For gross time study measurements, the
data acquisition system performed well, recognizing over
90 percent of the time elements. The average difference
between total cycle time estimated from the GPS data and
the manual time measurements was less than 3.5 percent.
Skid distances determined by the GPS-based machine functions were significantly higher than those measured on the
ground. Some of this distance discrepancy was attributed
to additional movements recorded by GPS during grappling
and delimbing that were not measured by typical manual
techniques.
Elemental time study was also possible, but correspondence with manually-determined elemental times was not
as precise. Travel empty and travel loaded times were close
to observed clock times, but grappling times were subject to
some large errors (in 25 percent of the cycles, grapple time
was overestimated by nearly 100 percent).
These methods also have been applied to automating time
study of wheeled feller bunchers (McDonald et al., 2000b).
In addition to collecting GPS position data, a field computer in the machine monitored the states of two switches
that indicated feller buncher activity: 1) cutting a tree as
indicated by micro switches on the foot pedals that controlled
4
3
2
1
used pairs of x,y position data to represent sampled locations of a machine, then assumed that machinery moved linearly between adjacent location samples. These x,y pairs
were transformed into a map showing how many times the
machine passed a given location. Final output of the transformation was a raster map, with cells in the raster having a
value equal to the number of times the object, or machine,
passed over a particular location in a rectangular region.
The model was tested initially by using data collected
from a rubber-tired skidder working in part of a clearcut
harvest. Several features of the harvesting operation were
discernable from the mapped paths: the deck or landing, the
delimbing area, the main skid trail, and the return skid trails.
They noted that the receiver type made a significant difference in the apparent accuracy of the maps. Although they
did not conduct any detailed position error determination,
they concluded that overall the calculated travel patterns
matched the true machine movements closely enough for
stand-level assessments.
In a later study, McDonald et al. (1998b) and Carter et al.
(2000a) used the same methods to map the travel paths of
feller bunchers and skidders over an entire harvest tract. The
output from this study was a traffic map of cumulative totals of traffic intensities and their distribution in the tract.
Figure 2 shows the traffic intensity map resulting from this
study. They found that 25 percent of the stand received no
traffic, 25 percent received more than five tire passes, and
50 percent received one to five tire passes. When visual
disturbance assessment methods were compared with GPS
estimated traffic intensities, the visual methods overestimated
the presence of heavily trafficked areas. They noted that the
GPS-based method was superior to the traditional methods
because it was less time consuming and presumably more
accurate.
Carter et al. (1999, 2000a, 2000b) presented detailed results of the soil physical responses measured during the study
introduced by McDonald et al. (1998b). They assessed the
impact of traffic intensity on spatial variability of soil physical
153
Figure 3. Map of feller buncher movements during harvest
operations. From McDonald et al. (2000b).
Figure 4. Results from tracking a site preparation sprayer while
broadcast spraying herbicide.
Spray on path
Spray off path
Sprayed area
Unsprayed area
Missed areas
Group of
Standing Trees
N
%
100
the felling head grabbing arm, and 2) felling head tipping as
indicated by a set of magnetic switches mounted on the felling head linkage. Figure 3 shows a map of feller buncher
movements across a study plot as well as the locations of
tree cut and head dump events as indicated by the data acquisition system.
The system performed well in a gross time study, and for
individual felling cycles the automated system agreed well
with traditional time study methods. With more accurate
information on the location of cut trees and with an additional system to measure tree size, it will be possible to
measure yield across the site. Current work at Auburn is
developing a tree diameter sensor that can be mounted on
wheeled feller bunchers. Also when the locations of the
bunches are known, it may be possible to optimize skidder
performance by routing skidders to the nearest bunch.
0
100
200 Meters
grated GPS capabilities, engineers will be able to develop
custom data acquisition systems to collect much more performance data as a function of machine location.
One of the most exciting areas of precision forestry, however, is in precision planning and implementation of forest
operations. For example, today we are mapping tracts that
need to be harvested and then planning optimal road layouts, deck locations, and harvest prescriptions. However,
during future harvests, we should be able to map the locations and sizes of felled trees (i.e., yield maps) and then send
these data to skidding or forwarding machines to help optimize performance and minimize site impacts. From these
data, we have the potential to develop reforestation plans
that minimize site impacts, soil erosion, machine costs, etc.
While these operations take place, we can use GPS to guide
site preparation plows, sprayers, or tree planters. For sprayers or fertilizer spreaders, we have existing technology to
control pumps and spray nozzles to prevent overspray and
to apply appropriate amounts of inputs. For mechanical site
preparation, GPS can help operators steer machines (or GPS
systems can steer them) so that beds or subsoiled rows are
placed on contours to minimize erosion. With new communications technology, we will soon be able to monitor forest
operations remotely and be more responsive to changes in
market or weather conditions. This can result in further integration of forest production, procurement, and product
manufacturing, which in turn may facilitate more stable inventories and reduce external pressures on the wood supply.
Site Preparation Machine Tracking – Time
Study
Additional research has focused on defining the productivity of mechanized site preparation equipment, such as plowing and bedding and herbicide spraying operations. One
example set of data is presented here for a broadcast site
preparation sprayer. While the sprayer was operating, a GPS
receiver recorded travel paths of the machine and times were
recorded when the spray nozzles were turned on or off. The
resulting map, in Figure 4, shows areas missed or
oversprayed. Productivity data from this type of activity
can be used to plan operating procedures or to design machine features to make the operation more productive.
SUMMARY
TRENDS IN MACHINE MONITORING
AND CONTROL
Using GPS as a tool to help monitor performance or productivity of forest machines is becoming more widespread.
Several studies focused on using off-the-shelf GPS hard ware
to track forest machines and determine site impacts and
machine productivity. This research successfully developed
the methodology to determine number of machine passes
over the terrain and to determine machine functions based
The studies highlighted previously show just a few uses
of GPS and data acquisition equipment as tools to help
monitor the performance of forest machines. As data acquisition equipment manufacturers continue to offer more inte154
on GPS travel patterns and additional sensors. At the same
time, researchers quantified the position accuracy from using current GPS hardware to track machines moving through
the forest canopy. In some cases where detailed soil samples
or other site specific data are needed, more sophisticated
GPS equipment may be required. However, for general
machine tracking for productivity studies, typical mapping
grade GPS receivers are sufficient.
Council on Forest Engineering (COFE), 21st Ann. Mtg.
Portland, OR, 20-23 July 1998. Corvallis, OR.
McDonald, T.P., E.A. Carter, S.E. Taylor, and J.L. Torbert.
1998c. Relationship between site disturbance and forest harvesting equipment traffic. p. 85-92. In: Whiffen,
H. J-H. and W.C. Hubbard (eds.). Proc. 2nd South. For.
GIS Conf., Athens, Ga. 28-29 October 1998. University
of Georgia, Athens, GA.
McDonald, T.P., S.E. Taylor, R.B. Rummer. 2000a. Deriving forest harvesting machine productivity from GPS positional data. American Society of Agricultural Engineers
(ASAE) Technical Paper No. 00-5011. ASAE, St. Joseph, MI.
LITERATURE CITED
Carter, E.A., T.P. McDonald and J.L. Torbert. 1999. Application of GPS technology to monitor traffic intensity and
soil impacts in a forest harvest operation. p.609-613. In:
J.D. Haywood (ed.), Proc. 10th Bienn. South. Silv. Res.
Conf., Shreveport, LA. 16-18 February 1999. Gen. Tech.
Rpt. SRS-30, USDA- FS, SRS, Asheville, NC.
McDonald, T.P., R.B. Rummer, and S.E. Taylor. 2000b. Automating time study of feller bunchers. In Proceedings
of the Council on Forest Engineering. 23rd Annual Meeting.
Carter, E.A., T.P. McDonald, and J.L. Torbert. 2000a. Harvest traffic monitoring and soil physical response in a
pine plantation. In Proceedings of the 5th International
Conference on Precision Agriculture and Other Resource
Management. University of Minnesota, Minneapolis,
MN. 10 p.
McMahon, S.D. 1996. Measuring machine travel with a GPS.
Technical Note TN-24. Logging Industry Research
Organisation (LIRO). Rotorua, New Zealand. 2 p.
McMahon, S.D. 1997. Unearthing soil compaction: the hidden effect of logging machines. GPS World 8(3):40-45.
Carter, E.A., T.P. McDonald, and J.L. Torbert. 2000b. Assessment of soil strength variability in a harvested loblolly
pine plantation in the Piedmont region of Alabama,
United States. N.Z. J. For. Sci. 30(1/2):237-249.
Reutebuch, S.E., J.L. Fridley, and L.R. Robinson. 1999. Integrating real-time forestry machine activity with GPS
positional data. ASAE Technical Paper No. 99-5037.
ASAE, St. Joseph, MI. 18 p.
McDonald, T.P. 1999. Time study of harvesting equipment
using GPS-derived positional data. In Forestry Engineering for Tomorrow. In Proceedings of the International
Conference on Forestry Engineering. Edinburgh University, Edinburgh, Scotland. Institution of Agricultural
Engineers, Bedford, UK. 8 p.
Spruce, M.D., S.E. Taylor, J.H. Wilhoit, and B.J. Stokes.
1993. Using GPS to track forest machines. ASAE Technical Paper No. 93-7504. ASAE, St. Joseph, MI. 23 p.
Stjernberg, E.I. 1997. The Truckbase Silvitracs GPS monitoring system. Forest Engineering Research Institute of
Canada (FERIC) Field Note No: Silviculture-99. FERIC,
Vancouver, BC, Canada. 2 p.
McDonald, T.P., R.B. Rummer, S.E. Taylor, and J.D.
Roberson. 1998a. Using GPS to evaluate traffic patterns
for forest harvesting equipment. p. 429-435. In Proceedings of the First International Conference on Geospatial
Information in Agriculture and Forestry, Lake Buena
Vista, FL. 1 – 3 June 1998. ERIM International, Inc.,
Ann Arbor, MI.
Thompson, J.D., G. Delbridge, R.J. McCormack, and J.
Coleman. 1998. Tracking forest harvesting machines with
GPS in southeastern Australia. ASAE Technical Paper
No. 98-7018. ASAE, St. Joseph, MI. 11 p.
McDonald, T., E. Carter, S. Taylor, and J. Torbert. 1998b.
Traffic patterns and site disturbance. In Harvesting
logisitics: from woods to markets, Proceedings of the
Veal, M.W., S.E. Taylor, T.P. McDonald, D.K. McLemore,
and M.R. Dunn. 2002. Accuracy of tracking forest machines with GPS. Transactions of ASAE. In Press.
155
Chapter 20
Global Demands Drive Advances in Data Management and
Hierarchical Decision Support Systems in
Northern British Columbia
JOHN NELSON
DAVE HARRISON
Abstract—Recent trends in the environmental awareness of consumers have significantly altered corporate policy and strategic direction for forest product firms in British Columbia. Regardless of where we practice forestry, we are accountable to the
global community. This paper describes a joint research project between Canadian Forest Products Ltd. and the University of
British Columbia that will develop a hierarchical based decision support system in northeastern British Columbia. The system
will build on recent resource inventories and the advanced database management system developed by Canadian Forest Products
and it is intended to forecast scientifically credible future forest conditions. Components of the decision support system include
stand-level modelling, forest-level modelling, wildlife habitat modelling and visualization modelling. Because of the large
geographic scale of the operation (650,000 ha) and the broad range of economic, environmental and social goals, data management and modelling challenges are abundant. Examples of these challenges are provided and focal areas for immediate research
are identified.
and Tumbler Ridge (Figure 1). Canfor Corporation is a leading Canadian integrated forest products company with the
majority of its woodlands operations and manufacturing facilities in British Columbia and Alberta. The company is a
producer of lumber, plywood, kraft pulp and paper plus
remanufactured lumber products, hardboard panelling and a
range of specialized wood products. The Tree Farm Licence
is composed of five blocks that extend from the Peace River
lowlands across the foothills into the Hart Ranges of the
Rocky Mountains. The total area of the TFL is approximately
643,500 ha. The biogeoclimatic units include Boreal White
and Black Spruce (BWBS), Sub-Boreal Spruce (SBS) and
Engelmann Spruce Subalpine Fir (ESSF) and Alpine Tundra (AT). Forests are dominated by aspen and white spruce
in the lower elevations, with lodgepole pine occurring on
drier sites. Engelmann spruce and subalpine fir are the main
tree species at upper elevations.
The BWBS has frequent stand replacing fire events, some
of which are quite extensive (>1000 ha). Fire frequency generally decreases with increasing elevation; however, upper
elevations in the eastern areas have a shorter fire return interval. Mammal species are quite diverse, including mule
and white-tail deer, grizzly and black bear, moose, elk, mountain goat, caribou, wolverine and fisher.
There is a relatively short history of industrial forest harvesting in the area compared to other parts of the province.
There are still a number of relatively intact valleys within
the management unit. However, mineral and gas exploration, coal mining, and development of pipeline and trans-
INTRODUCTION
The ability to make credible projections of future forest
conditions is a necessary step towards the goal of product
certification and sustainable forest management. Resource
management decisions must be based on credible data and
these data must be structured to support applications within
a hierarchy of temporal and spatial scales. A three-year research project entitled “Exploring, Forecasting and Visualizing the Sustainability of Alternative Ecosystem Management Scenarios” will provide the tools to manage spatial and
temporal data at a scale required for landscape unit planning. The project will focus on the projection of future forest conditions according to economic, ecological and social
indicators of sustainability. This initiative brings together a
University of British Columbia research team with expertise
in modelling, forest ecology, stand-level management, forest-level management, wildlife, visualization and the social
sciences. Research partners and sponsors are Canadian Forest Products Ltd (Canfor), the Canadian Forest Service (CFS),
the Natural Sciences and Engineering Research Council of
Canada (NSERC) and the Social Sciences and Humanities
Research Council of Canada (SSHRC).
The Study Site
The study area is Canfor’s Tree Farm Licence # 48, located in northeastern BC, on the eastern side of the Rocky
Mountains near the communities of Hudson Hope, Chetwynd
157
Figure 1. Location of the study area in northeastern
British Columbia.
Figure 2. Typical landscape view in the Tree Farm Licence.
An important component of the certification process and
sustainable forest management is the ability to forecast longterm impacts of management decisions on environmental,
economic and social indicators. These forecasts need to
capture the full range of values, and be based on credible
science. Forecasting future forest conditions requires an
innovative decision support system linked to numerous resource inventories via a comprehensive data management
system.
Our objectives in this paper are to describe the data management and decision support systems that we are developing for Canfor’s operations, and to highlight the challenges
we face when modelling complex systems over vast geographic areas In the first section of the paper we describe
the resource inventories and the data management system
developed by Canfor. The second section describes the decision support system being developed by the UBC research
team. This section contains a summary of the models used
in the decision support system and their respective linkages.
The third section of the paper identifies challenges that we
face with modelling multiple objectives over large geographic areas and over long time horizons, in the context of
hierarchical planning systems. Finally, conclusions are
drawn and areas needing further research are identified.
portation corridors have resulted in significant alteration of
some areas. Figure 2 shows a typical landscape view in the
Tree Farm Licence.
Global Demands and the Need for Data and
Models
To remain a global supplier of forest products and technologies, forest companies will have to grow in step and in
scale with their customers. As well, a different type of supplier – customer relationship will have to be developed and
supported. Price may no longer be the sole requirement to
achieve market share. Instead, alternate tactics will have to
be pursued to create value by aligning company policy and
strategic direction with the increased scope of customer demand. The push towards sustainable forest products demanded by Canfor’s large retail customers reflects the new
realities of the market where suppliers must now react quickly
to customers’ expressed needs. As such, Canfor is committed to responsible stewardship of the forestlands entrusted
to its care. In keeping with this commitment, Canfor acknowledges the importance of the independent recognition
of its stewardship practices and has made it a priority to
certify its forestlands to viable and accepted certification
standards.
INVENTORIES AND DATA
MANAGEMENT
Canadian Forest Products Data Management
System
Our Forest Management Information Systems (Genus)
allow us to create maps, to analyze and report on inventory
data for strategic, tactical and operational planning projects
and to monitor the forests within the Tree Farm Licence.
Genus applies integrated hardware technologies with stable
data platforms and software such as Geographic Information Systems (GIS) database applications. When delivered
in concert through our Wide Area Network (WAN), the integrated approach provides field personnel and resource
managers immediate access to spatially explicit data in support of operational decisions.
158
is then further analyzed and classified into physical operability classes.
Recent Inventories of the Chetwynd Division
Canfor has embarked on an aggressive inventory program
covering a Vegetation Resource Inventory (VRI), Terrestrial
Ecosystem Mapping (TEM), Fish and Fish Habitat Inventories, Wildlife Habitat Inventories, a Visual Landscape Inventory, and Growth and Yield projects. Summaries of the
current resource inventories follow.
Biogeoclimatic Ecological Classification
As part of the Terrestrial Ecosystem Mapping/Terrain
Mapping described above, detailed plot data were collected
to refine the Biogeoclimatic Ecological Classification. The
new lines were based on field data collected in the 1998 and
1999 seasons.
Vegetation Resources Inventory
Figure 3. Grizzly bear is one of 12 species being modeled
on the TFL
The Vegetation Resources Inventory uses three processes:
1. photo-interpreted estimates (Phase I),
2. ground-sample measurements (Phase II), and
3. statistical analysis and adjustment of the original estimates.
The inventory conforms to Resource Inventory Committee (RIC) standards. The VRI has been updated to account
for all harvesting activities and silviculture surveys completed up to March 2000. Canfor’s spatial block tracking,
silviculture and road management suite of applications (Genus) was used as the source for the update and done through
automated GIS routines. The main objectives of the VRI
Phase II sampling are:
Wildlife and Wildlife Habitat
1. to adjust the Phase I estimates to provide
statistically valid timber volumes in the Tree
Farm Licence, and
2. to provide baseline ecology and coarsewoody debris data to support other
projects, including predictive ecosystem
mapping, site index ecological correlations,
monitoring, and forest certification.
Since 1996, Canfor has undertaken a series of measures
to address wildlife and wildlife habitat. These measures include wildlife habitat modeling, wildlife inventories, habitat monitoring and wildlife research.
Wildlife habitat modeling began in 1997. The species
chosen for habitat modeling were selected relative to their
provincial or federally listed status (e.g. Grizzly Bear, Figure 3). In 1998, additional field data on wildlife habitat were
collected and used along with data from recently completed
wildlife inventories to refine existing models. Other species will be modeled as guilds that respond similarly to habitat structure (e.g., shrub nesters, secondary cavity nesters).
A document summarizing the likely response of all listed
species to forest practices is being prepared. The habitat
models will be applied to the entire Tree Farm Licence when
a Predictive Ecosystem Model is available. The models will
then be combined with timber forecasts to predict wildlife
habitat availability over time.
Terrestrial Ecosystem Mapping
Terrestrial Ecosystem Mapping (TEM) has been completed for approximately one third of the TFL area. The
scope of the TEM has been modified to include a Landslide
Inventory. The results were then used to calibrate the Stability Index MAPing (SINMAP) model to produce a terrain
hazard map for the entire estate. The output will provide the
basis for deriving physical and economic operability.
The scope of the TEM project was also modified to include a field program designed to adjust the Biogeoclimatic
Ecological Classes. The work will form the basis of a Predictive Ecosystem Mapping (PEM) project to be completed
in 2001.
Stream Inventories
Since 1995, Canfor has been conducting 1:20,000 reconnaissance level fish and fish habitat surveys. Initially, work
was limited to the cutblock or small watershed level. As the
process became defined, the scale of the program was successfully increased to address larger watersheds. Fish and
fish habitat inventories have been completed on approximately 85% of the estate. The bulk of the watershed level
work will be completed for the TFL by 2001. Fish invento-
Physical Operability
Using the terrain components of the Terrestrial Ecosystem Mapping work, terrain mapping and Landslide Inventory Terrain Stability Classes were derived using the Stability Index MAPing (SINMAP) model. The SINMAP output
159
ries at the cutblock level will continue to be required on an
operational basis.
Figure 4. Models used to forecast future forest conditions in the
decision support system.
Visual Landscape Inventory
In 1999/2000, Canfor completed an update of the Visual
Landscape Inventory for the Tree Farm Licence. The inventory was done at a scale of 1:50,000 using the provincial
Visual Landscape Inventory Procedures and Standards
Manual.
ATLAS (forest-level model)
SIMFOR (habitat model)
Cultural Heritage
VISUALIZATION:
stand & landscape level
Canfor obtained GIS coverages for the Archaeological
Overview Assessment and Archaeological Site Information
for the Dawson Creek Forest District from the Ministry of
Small Business Tourism and Culture in June 1999. The data
is maintained under a Confidentiality Agreement with this
Ministry. Currently there are 20 known heritage sites within
the Tree Farm Licence.
FORECAST (stand-level model)
FORECAST
Growth and Yield
FORECAST is a stand-level, ecosystem management
simulation model designed for evaluating long-term site productivity and value trade-offs associated with stand-level
management. It uses stand-level biophysical data in conjunction with standard growth and yield data to simulate tree,
understory and soil conditions for specific management activities. Output includes biomass production, timber volumes, slash, forest floor and coarse woody debris mass,
snags, nutrient inventories and cycling, energy and carbon
budgets, plus stand level economics.
Field sample plots provide periodic and permanent data
for growth and yield estimates. Stand projection models
based on these data help predict the total increase in volume
over a given period (growth), as well as the accumulated
volume of a stand at any given age (yield).
DECISION SUPPORT SYSTEM
Description of the Research Team and the
Decision Support System
The research team consists of UBC experts in forest-level
modelling (John Nelson), forest ecology and stand-level
modelling (Hamish Kimmins, Brad Seely, Clive Welham),
wildlife habitat (Fred Bunnell, Ralph Wells), visualization
(Stephen Sheppard), social issues and recreation (Michael
Meitner) and economics (David Haley). The project team
also includes several research technicians and a number of
graduate students. In addition, scientists at the Canadian
Forest Service will collaborate on data systems and criterion and indicators of sustainable forest management.
To forecast future forest conditions, a decision support
system has been developed that includes four primary models. These are: 1) the stand-level model FORECAST, 2) the
forest-level model ATLAS, 3) the wildlife habitat model
SIMFOR, and 4) a visualization model applied at both the
stand- and forest-levels. The FORECAST model is used to
supply growth and yield libraries to the ATLAS model, standlevel habitat attributes to SIMFOR and stand structure information to the visualization model. Future landscapes are
generated with the ATLAS model that are subsequently used
in SIMFOR to assess habitat and in the visualization model
to assess the visual impacts of harvest schedules. Figure 4
illustrates the linkages of the four models within the decision support system. A description of model function and
output follows.
ATLAS
ATLAS is a forest-level harvest simulation model. The
model is spatially explicit with respect to forest polygons
and road networks. ATLAS is designed to schedule harvests
according to a range of spatial and temporal objectives. These
include policies related to harvest flows, opening size, riparian buffers, seral stage distributions and patch size distributions. Silviculture systems, rotation ages and stand
growth and yield are assigned to each polygon. At each time
period, the model reports the status of every polygon in the
forest estate. These periodic inventories can be quickly displayed with a map viewer to assess harvest patterns and/or
be exported to other models. Road construction, length of
active road, and other indicators of road network activity
related to the harvest schedule are also reported.
SIMFOR
SIMFOR is a decision support tool designed to assist
forest managers to evaluate the consequences of forest management scenarios on landscape pattern and wildlife habitat. SIMFOR calculates potential landscape patterns and
wildlife habitat conditions from spatially explicit harvest
schedulers such as ATLAS. Its projections indicate general
trends in indicators of forest ecosystem structure and func160
policy/regulation, economics, markets, harvest technology,
natural disturbance, knowledge of ecosystem functions and
biology of species). Validation is especially challenging on
public forest land where many of the social and political
variables are more uncertainly relative to private ownership.
tion over space and through time. SIMFOR’s projections
of landscape pattern created by the interaction of silvicultural treatments with forest growth and succession are primarily expressed as maps and summary statistics of area in
each seral stage, patch size class, and amount of edge of
different types. The habitat projections are based on the abundance of vegetation and structural characteristics, such as
snag density, shrub density, or downed wood. Information
on stand structure can be generated empirically, from inventory data and from process based models such as FORECAST.
SOCIAL AND ECONOMIC ANALYSIS
Linked to the decision support system are social and economic analyses. Canfor formed a Public Advisory Committee to provide input to the development of local values, goals,
indicators and objectives based on the Canadian Council of
Forest Ministers Criteria and Critical Elements. The research
team works with the Public Advisory Committee to ensure
that their values, objectives and indicators are incorporated
into the models, and that all scenarios show how these change
over time. Visualization plays an important role in communicating scenario output to the Public Advisory Committee,
especially for people that are unaccustomed to standard
model output in the form of planar maps. Surveys of the
community will also be undertaken to capture other perceived
forest values and assess how strongly these values are held.
This information may lead to further revisions to the forecasting models.
Profitability issues for Canfor and economic impacts on
communities and stakeholders are also an important component of the project. Key model outputs need to be linked
to employment and income multipliers to determine the number of jobs and regional income generated through time.
Estimates of economic rent for the management scenarios
will be used as an indicator of the return to the public landowners.
VISUALIZATION
The visualization model includes quantifiable factors such
as topography, stand height, vegetation pattern, ground conditions, plus distance and view angle. Visualization software (currently World Construction Set) is used to portray
stand and landscape level conditions resulting from management scenarios (e.g. ATLAS harvest schedules). While
various visualizations systems are in use today, some of
which have been available for many years, there is a need to
develop a more direct and transparent link between the visualization “end-product” and the modelling inputs (e.g.
harvest rotation, silvicultural technique, stand dynamics).
The requirement of such a program are that it must:
• readily accept data from multiple sources, including common GIS data formats,
• permit near real-time visualization of altered landscape conditions as model inputs are changed,
• allow visualization of both stand-level and landscape level scenes, with seamless movement from
one to the other,
CHALLENGES OF SCALE
• allow visualization of temporal variation in con-
In multiple-objective forest planning, the standard procedure is to create resultant polygons by GIS overlays (forest cover, habitat, riparian, visual, etc.). This procedure creates numerous resultant polygons, and often 30-50% of these
are slivers under 0.1ha. Even after sliver removal, the number of resultant polygons on a large estate like the Tree Farm
Licence can exceed 1 million. This level of detail is appropriate for watersheds and compartments that require special
analysis to address complex resource trade-offs. However,
for strategic planning over the entire estate (timber supply,
seral stages and patterns) this level of detail is excessive and
burdensome. One of our challenges is to identify the appropriate level of spatial detail that is necessary to capture key
trends and answer relevant questions at each level of the
planning hierarchy.
A good example of an activity that requires hierarchical
decision-making is partial cutting (e.g. shelterwood and selection systems). At the forest-level, these can be modelled
as a simple entry-removal-growth sequence on each polygon. The specifics of which parts of the polygon are cut
during a harvest entry are not important. However, to as
sess the visual impacts of these systems, it is necessary
ditions, using time lapse techniques, and
• achieve a reasonable level of realism, sufficient
to permit qualitative judgments of aesthetic quality.
The results of the visualization model will be presented
to the stakeholder/client groups using large screen projection formats. Viewer responses will be documented as further input to the next round of modelling, and to help validate or improve the visualization program.
The FORECAST, ATLAS and SIMFOR models have
been verified (i.e. they produce correct results according to
their respective algorithms and assumptions) and they have
been used in forest planning projects for about a decade.
Because of the long-time horizons in forestry, they have not
been validated (i.e. their forecasts actually implemented and/
or observed). FORECAST has been “validated” against
other stand-level models, and for each project it is calibrated
against existing stands. Forest-level models such as ATLAS
and SIMFOR may never be validated because the original
assumptions change over short-time horizons (e.g. forest
161
Figure 5. Small polygons (left) were created by splitting the resultant polygons (right). These small polygons are used to simulate
patch cutting in a visually sensitive area. Source: Nelson, J.D. 2001. Timber supply analysis report of Lemon Landscape unit
management scenarios. Arrow Innovative Forest Practices Agreement, University of British Columbia, Vancouver, BC. 24p.
Figure 6. Visualization of patch cuts.
162
to know precisely where the harvests took place, plus the
age structure and stand attributes within the polygon at each
time step. For the forest-level model to generate this information, we need to split the polygons into smaller units that
represent individual harvest entries. Figures 5 shows an example of smaller polygons created to simulate patch cuts in
a visually sensitive area, and Figure 6 shows an image of the
area rendered with visualization software. However, if we
were to use these small polygons in a strategic timber supply
analysis of the entire 650,000 ha estate, data management
and model processing times quickly become unmanageable.
Clearly, a hierarchical planning system is needed that links
the strategic analysis at the forest estate level to specific resource analyses at the watershed and compartment levels.
A similar case is made for wildlife habitat. Over the large
forest estate, our interest is in strategic thresholds, trends
and patterns in seral stages. At this broad scale, we are interested in how management scenarios differ from historic disturbance patterns and what impacts this may have on wildlife and biodiversity. Within specific watersheds/landscapes,
detailed habitat indicators such as the distribution of snags
and understory abundance may be necessary to assess impacts on listed species with localized distributions.
Another example of where linkages are needed in the planning hierarchy is access management. This is especially important because of conflicting objectives over public access
to remote areas and sensitive habitats. Recreation users generally prefer more access while trappers and guide outfitters
prefer less. Coordination of public access with harvest operations, stand tending and forest protection, plus the uncertain demands of oil and gas exploration is challenging. In
the short-term, within compartments, detailed road network
analysis is necessary to determine which roads should be
constructed, maintained, or decommissioned. These detailed
access management plans need to be consistent with the strategic opening and closing of compartments over the entire
Tree Farm Licence. At this scale, detailed road networks are
unnecessary for answering strategic questions related to access management.
The cost of obtaining good information is also a solid
reason behind hierarchical planning. To make informed decisions at the tactical and operational levels requires expensive and time consuming field surveys and inventories that
are simply not practical at the strategic level which covers a
very large forest estate. In this case, we invest our scarce
resources in smaller compartments where the needs are the
greatest and the benefits of good information are maximized.
The above examples highlight the need for a hierarchical
approach in designing the decision support system. Establishing targets that link each level is reasonably straightforward. For example, harvest targets for a specific compartment can be extracted from the strategic timber supply analysis of the entire estate. The more difficult task is manipulating spatial data between these hierarchical levels. We have
found considerable utility in polygon splitting and aggregation routines that can quickly manipulate spatial data to meet
the relevant questions at each level. We also see strong potential for heuristic methods that minimize the number of
resultant polygons created during the GIS overlay process.
CONCLUSIONS
Regardless of where we practice forestry, we are accountable to the global community. Until recently, northeastern
British Columbia was generally considered remote and insignificant in the global consciousness. Today, company
policy and strategic direction must be aligned with the increased scope of customer demand. Consumer awareness
is resulting in firms seeking certification of their products
and operations. In Canfor’s case, this began with significant investments in resource inventories and development
of an advanced database management system that is now
available through a wide area network. The objective our
the joint research project described in this paper is to develop a decision support system that combines Canfor’s inventories and database management system with scientifically credible forecasting models designed by the UBC team.
The project is relatively unique because of the large geographic scale of the Tree Farm Licence. This presents both
data and modelling challenges because detail needed in some
compartments is both unmanageable and unnecessary at the
estate level. To help resolve these issues, a hierarchical approach is being used to guide the design of the decision
support system. An important goal of the project is to be
able to identify the minimum data requirements necessary
to answer relevant questions at each level in the planning
hierarchy. The hierarchical approach also provides flexibility to add models (e.g. hydrology) to the system.
Global demands are clearly driving local data acquisition
and advances in data management and decision support systems. We envision the time when a forest product salesperson can access these systems from a kiosk in the USA or
Europe to answer specific questions about the product and/
or sustainable forest management. This project is one of
the first steps towards this vision.
ACKNOWLEDGEMENTS
We gratefully acknowledge the contributions of Don
Rosen, Warren Jukes and Andrew de Vries of the Canfor
Chetwynd Division; Hamish Kimmins, Brad Seely, Clive
Welham, Stephen Sheppard, Michael Meitner, Fred Bunnell,
Ralph Wells and David Haley from the University of British
Columbia, and the financial support of Canadian Forest Products Ltd., the Natural Sciences and Engineering Research
Council of Canada, the Social Sciences and Humanities Research Council of Canada, and the Canadian Forest Service.
163
Chapter 21
Precision Log-Making to Maximise Value Recovery from
Plantation Forests
KEVIN BOSTON
Abstract—Poor log making is the leading cause of value loss. To improve value recovery, Carter Holt Harvey Forests has
adopted the IFR TimberTech System. The system is designed to increase value recovery by using a log optimisation tool in
production log making. The TimberTech System yields additional value from the stem and log data collected during harvesting.
This data can be used for marketing decisions as well to continually improve the value recovery.
INTRODUCTION
TimberTech System to improve value recovery. The system has three components:
New Zealand’s forest industry has made value recovery
as a cornerstone in its competitive strategy. A value recovery
program is one that emphasises the three elements of the
profit equation: price, volume, and cost. The causes of value
lost appear throughout harvesting operations from felling to
unloading the logs at the mill. Some of the factors include:
• A logger tool attached to a calliper,
• A stocks tool, and
• A database management system.
Once the grades have been defined in the database management system, the process begins by creating the cutting instructions. These contain the grades, lengths, and
priority values assigned to each crew, usually for a fortnightly period. The cutting instructions are entered into
the database and the files are transferred into directories to
be downloaded by the crews. Using the stock tools, the
cutting instructions are downloaded from the database and
finally transferred to the logger tools. The logger tool is
attached to a calliper and automatically records the length
and diameter starting from the butt end of the log. The logmaker will describe the stem by recording its attributes
such as branch size, roundness, pith location, and defects.
After the stem has been fully described, the log-maker uses
the individual stem log optimisation algorithm that is part
of the software contained on the logger tool. The algorithm uses a linear taper function, the stem description,
and the desired grades, length, and relative values from
the cutting instruction to maximise the value recovered
from each stem. It assumes that the markets are unlimited
for all grade-length combinations on the cutting instructions. The logger tool saves both the stem descriptions and
the recommended logs produced from the optimisation
algorithm. This data are transferred back to the stock tools
and is uploaded to the database every second day. The full
process description is shown in Figure 1. A similar version has been developed for mechanised systems.
In the Central North Island operations in New Zealand,
Carter Holt Harvey harvests approximately 50 000 tonnes
per week of pine and eucalyptus produced by approximately
• Stump heights – higher operating costs, loss in
value and volume
• Felling damage – higher operating costs, loss in
value and volume
• Extraction damage – higher operating costs, loss
in value and volume
• Log-making – loss in value
• Loading-unloading damage – loss in value and volume
Poor log-making results in the single largest cause for
value lost. Initial studies in New Zealand have shown a loss
of 26% of the total value (Murphy and Twaddle 1986).
Mechanised log-making operation in a loblolly pine forest
in Florida (USA) showed that 43% of the total potential
value from poor log making practices. The cause in this
operation was poor length measurements (Boston in review).
Poor log-making results in the single largest cause for
value lost. Initial studies in New Zealand have shown a loss
of 26% of the total value (Murphy and Twaddle 1986).
Mechanised log-making operation in a loblolly pine forest
in Florida (USA) showed that 43% of the total potential
value from poor log making practices. The cause in this
operation was poor length measurements (Boston in review).
The Interpine Calliper System
Carter Holt Harvey Forests had adopted the use of IFR
165
Figure 1: Process flow through the system.
Database
System
Stock Tool
Cutting
Instruct. and
Stem and log
data
Cutting
Instruc. and
Stem and log
data
50 harvesting crews. The central North Island serves a number of domestic mills, (Carter Holt Harvey and others companies), and is the main source for export volume for the
company. Logging uses both ground and cable harvesting
systems with log making occurring at the extraction skids
or full stems are hauled to centralised superskids that can
receive logs from multiple extraction points. There are restrictions on the number of grade-lengths combinations that
can be made on each landing due to safety and the requirement that full loads are completed before the logs deteriorate due to sap staining. This time frame can vary from 3
weeks in winter to 2 days during summer. The process has
three potential causes for value loss. They are:
Logger
Tool
The main use for the database is to evaluate and monitor
the three potential sources of value lost in log-making including incorrect grades, incorrect relative value, and improper log making practices. By comparing the value from
applying the two cutting instructions to a similar set of stem
data that will be encountered by the harvesting operation.
One can select the cutting instruction that produces the highest value.
The second source of value recovery is the evaluation of
the relative value. It is through the relative value that the
market considerations are included in the unconstrained log
optimisation algorithm. To properly assign relative values
to 50 crews that are each cutting between 12 and 20 cuts to
satisfy 15 markets is a difficult task that Carter Holt Harvey
Fibre Solutions has not mastered. Currently, we can only
simulate the impact of changing relative values on the production for key grades. The example in figure 2 shows the
production of three grades, two structural and one internode
grade, one that must have at least 1.3 meters section of wood
between branch whorls.
This analysis shows the impact of increasing relative value
on production of the internode, primarily at the expense of
the SLM grade (large diameter medium knot sawlog) and
SLS (large diameter, small knot sawlog) grade (Figure 2).
These analytical tools highlight some of the capabilities
of the TimberTech database. It is the auditing function that
is the most valuable aspect of the database. Some reports
that have been developed include:
• Not assigning the correct grade-length to the cutting instruction.
• Assigning the incorrect relative value to each
grade, too low will result in markets not being
satisfied; too high will result in a surplus that implies additional costs for storing or downgrading
the log.
• The manufacturing losses that exists with manual
log-making such as failing to properly recognise
log attributes, improper use of the tool (ie too few
measurements or poor calibration).
The tool is not a panacea for log making. It will not
replace a well-trained log-maker, but it can complement a
well-trained log-maker especially as the cutting instructions
can have as many as 25 length-grade combinations (sorts)
on a central processing operations.
One of the most valuable aspects of the TimberTech System has been the capture of the stem and log data. It has
numerous applications; one is to evaluate marketing decisions. For example, a customer would like to know the impact of changing from an unrestricted large-end diameter
(LED) to a constrained LED of 50 cm. The what-if analysis
capabilities within the TimberTech system, allows one to
change log descriptions and evaluate the strategy based on
real data, not just intuitive beliefs. The results from this
analysis show that the volume will be reduced by 7% from
the unconstrained to the constrained LED (Table 1).
• Evaluation of felling practices and the incidence
of slabbing damage,
• Consistent use of log attributes,
• Consistent use in log-making, and
• Comparison of pulp percentages.
For example, Table 2 shows stem descriptions collected
from a recent harvesting operation. The incidence of slabbing and stump-pull damage, (codes L and Y), on four stems.
Slabbing damage resulted in a 1.25 metre in loss in log 1, a
1.65 metre loss in log 2, 0.57 metre loss in log 3, and 1.12
metre loss in log 4. It is important to note that the I and P
attributes (Internode and veneer) with size 0 indicate that
166
Table 1: Results from What-if Analysis showing the impact of a 50 cm LED.
Figure 2: Change in output of SLN, SLS, or SLM with increasing SLN relative value.
Volume m3
Change in volume output with varying relative values for
SLN
79
89
99
109
SLN relative value
SLS
SLN
167
SLM
Table 2: The incidence of felling damage from 4 logs due to slabbing damage.
Record No.
Starting distance
from end of log
Ending distance
from end of log
Attribute
Size
2,880,663
2,880,663
2,880,663
0.00
1.25
6.08
1.25 L
6.08 I
31.92 P
0.00
0.00
6.00
2,880,678
2,880,678
2,880,678
2,880,678
0.00
1.65
7.67
9.30
1.65
7.67
9.30
32.52
L
P
C
P
0.00
6.00
0.00
6.00
2,880,679
2,880,679
2,880,679
2,880,679
2,880,679
0.00
0.57
5.43
16.20
19.57
0.57
5.43
16.20
19.57
23.98
L
I
P
R
P
0.00
0.00
6.00
0.00
6.00
2,881,006
2,881,006
2,881,006
2,881,006
2,881,006
0.00
1.12
5.56
24.89
29.30
1.12
5.56
24.89
29.30
30.08
Y
I
V
P
S
0.00
0.00
6.00
5.00
5.00
Table 3: Report showing the inconsistency between log-makers on same crew.
Attribute Codes Report
Crew 1
Crew 2
B
GH
2.72
4
JB
2.34
1
KH
1.21
1
RK
1.36
168
C
1
D
I
17
11
15
25
8
9
7
7
Table 4: Different value recovered from the two logmakers.
Grade
Volume( M3)
Count
Log maker
SLM
SLN
SLS
UBR
UHR
UKB
GH
108.71
51.59
26.92
155.21
68.10
18.97
185
67
30
596
115
138
Total Pulp
Log maker
SLM
SLN
SLS
UBR
UHR
UKB
Total Pulp
4). Log-maker GH is producing a much smaller percentage
of internode volume and small-knot structural wood, but a
higher percentage of the medium knot size, plus almost 1.5
times the pulp. Log-maker GH was scheduled with the highest priority to receive training to improve his log-making
skills.
The ability to track the performance of individual logmakers is important when trying to change the culture of the
crews from one of maximising production to one of
maximising value. Fibre Solutions can quickly show where
value is being lost and with proper training and supervision
can work with the logging contractors to improve their performance and our profitability.
Percent of
Total Volume
14.46
6.86
3.58
20.65
9.06
2.52
32.23
JB
77.45
121.23
65.31
83.68
47.03
25.99
THE FUTURE
132
166
73
299
86
201
10.76
16.84
9.08
11.63
6.54
3.61
There are a number of modifications to the system that
Carter Holt Harvey Fibre Solutions would like to implement.
One would be a more complex taper function. The linear
taper function causes too many errors in the log making and
we believe as Olsen et al. (1991) that a better taper function
can improve the value recovered with the tool. Rejects or
missed opportunities are a result of the difference between
the estimated diameter from the taper function and the actual diameter.
21.78
these losses occurred on pruned stems, and eliminated them
from the high value 5.0-metre length to the lower value 4.4metre length.
The auditing reports can show the consistency in the use
of log attributes by the individual log makers on each crew.
One should expect the use of attributes to describe the stems
to be similar for log-makers working on the same landing
from the same quality of forests, but this is not often the
case.
Table 3 shows that log-maker JB and GH in crew 1 are
not using the “I” attribute, (internode) with the same frequency during a two week period with GH using the I attribute less than half as often as JB. While the results from
Crew 2 show a much more consistent use of the log attributes by the two log-makers. This can be translated into
different value recovered from the two log-makers (Table
LITERATURE REVIEW
Boston, K. In review. Value recovery from mechanised logmaking operations in the southeastern United States.
Southern Journal of Applied Forestry.
Murphy, G., and A. Twaddle. 1986. Techniques for the assessment and control of log value recovery in the New
Zealand forest harvesting Industry. In Improving
productivity through forest engineering. Proceedings of
the Council on Forest Engineering.
Olsen, E, J. Garland, and J. Sessions. 1989. Value Loss from
Measurement Error in the Computer-Aided Bucking at
the Stump. American Society of Agricultural Engineers
5:283-285.
169
Chapter 22
A Look to Future Precision Tree Length Stem Analysis and
Processing
PHIL ARAMAN
Abstract—USDA Forest Service and university researchers are developing machine vision and optimization software systems. Some of these efforts and additional efforts are being coordinated with industry efforts on the equipment side. The results
will be precision tree stem analysis of standing timber and optimal bucking of felled timber. This paper will present an overview
of current and future efforts aimed at achieving the above. We will also present our ideas on future industrial scanning and
optimization software systems and their capabilities.
171
Chapter 23
Using Data-Driven Visual Simulations to Support
Forest Operations Planning
ROBERT J. MCGAUGHEY
Abstract—Foresters and engineers charged with selecting stands for treatment, developing silvicultural prescriptions, and
designing forest operations to meet specific goals often find it difficult to comprehend the complex spatial interactions that occur
across landscapes. Visual simulations based on accurate terrain models and detailed vegetation data provide a visual environment
that can significantly improve the design and implementation of forest operations. Currently visual simulations are used to present
the final results of planning efforts. The use of visual simulations to support operational planning, implementation, and monitoring is relatively new. Visual simulation can provide feedback during the design of silvicultural prescriptions and unit boundaries;
facilitate operator training using equipment simulators; supply in-cab displays to provide information to equipment operators; and
assist in monitoring compliance with designated travel paths, cut/leave tree specifications, and forest practice regulations. Current
computers have sufficient computing power and graphics capability to make data-driven visual simulations a valuable part of
forest operations planning.
INTRODUCTION
technical and non-technical audiences. More sophisticated
visual simulations can be used to assess public preference
for a variety of management activities. Simulations can be
highly abstract, such as a display from a geographic information system (GIS) showing a perspective view of polygons draped over a terrain model, or very realistic, such as a
scene depicting vegetation using individual plant icons derived from actual photographs.
This paper will discuss using visual simulations to provide support for forest operations planning, design, implementation, and monitoring. Data requirements for simulations will be discussed relative to the size and scale of the
area depicted in simulations. Examples will illustrate using
visual simulations to depict large-area planning activities,
communicate stand structure changes resulting from forest
operations, and support on-going operations.
The ability to design, understand, and communicate forest conditions and the effect of silvicultural treatments on
the forest environment is essential for the successful management of forest lands. Foresters and engineers charged with
selecting stands for treatment, developing silvicultural prescriptions, and designing forest operations to meet specific
goals often find it difficult to comprehend the complex spatial interactions that occur across landscapes. Traditional
work methods involving fieldwork, maps, aerial photographs,
and computer analysis provide enough information to assess
individual treatments but may not provide adequate information to fully evaluate the cumulative visual impacts of
several treatments or the impact of treatments implemented
over time. Computer-based visual simulations are a recognized tool for assessing the potential visual impact of landuse decisions and management activities. Data-driven visualization methods that rely on accurate terrain models, detailed tree and plant data, and have the ability to roam freely
through the scene, provide a visual environment that can significantly improve the design and implementation of forest
operations. Such simulations can provide visual feedback to
forestry professionals and can help communicate stand and
landscape conditions and how management activities, harvest operations, natural disturbances, and growth over time
change these conditions.
Visualization methods can provide a variety of display
and communication products (Wilson and McGaughey 2000).
Simple images can communicate the simulated spatial arrangement and complexity of forest types, management activities, natural disturbances, and human development to both
Land Use and Forest Operations Planning
Land use and forest operations planning can be divided
into three levels:
•
•
•
strategic,
tactical, and
operational.
Strategic planning involves large land areas (7500 hectares and larger), long time frames (several decades), and
utilizes low spatial resolution. Strategic plans define overall
land use allocations, product outputs, landscape attributes,
and various targets used to guide the development of tactical plans. They have some spatial resolution but, due to
173
their broad scope, leave many of the spatial details to be
resolved during development of more site-specific plans.
Tactical planning efforts focus on smaller land areas (2000
to 7500 hectares), often basins or sub-basins, and span 10 to
30 years. Outputs from tactical planning include a schedule
of road construction and harvest operations and indications
of how well overall targets established in strategic plans will
be met. Tactical planning requires enough spatial detail to
relate treatment units to one another and to existing and proposed transportation systems, stream networks, identified
hazards, and areas of special concern such as habitat for
threatened and endangered species.
Operational planning involves design of specific silvicultural treatments, harvest systems, and monitoring protocols. Operational plans generally involve less than 500 hectares and rely on detailed spatial information collected in the
field and managed using a GIS. The amount of spatial detail
incorporated into operational plans varies depending on the
organizational requirements for these plans, the quality and
quantity of data available, and regulatory requirements applicable to the planning area.
Visual simulation is seldom used in conjunction with strategic planning since strategic plans lack sufficient spatial
detail and the necessary data describing vegetation conditions. During tactical planning, visual simulations are commonly produced to illustrate general management strategies
or to depict road building and harvest activities through time.
The data required to produce such simulations are generally
available during the tactical planning process. Simulations
depicting subsets of the tactical planning area to illustrate
the location and general harvest pattern for a specific treatment unit can be created. However the summarized data typically used to support tactical planning do not provide sufficient detail to show specific treatment effects or operational
features. Visual simulations are perhaps most useful during
operational planning activities where they can provide important feedback to designers. Simulations depicting detailed
stand structure can help designers assess such things as risk
to individuals working on the ground, suitable retention levels and patterns to minimize damage to residual trees during
partial cut operations, and equipment suitability given terrain and stand conditions. To be truly useful, visual simulations developed to support operational planning must be
based on high-resolution, accurate terrain models, stand
polygons, and detailed inventory data that describes the vegetation present in each stand polygon.
tures) and then assemble the component models to create a
scene depicting a forest stand or landscape. In its simplest
form, this technique can be used to generate perspective
views showing typical GIS data themes such as roads,
streams, and stand boundaries draped over the ground surface. More complex applications use detailed ground surface textures and models of individual trees to represent
vegetation cover over large land areas.
Digital photo retouching methods rely on computer programs to modify images to represent changes to stand and
landscape conditions. Digital photo retouching produces fullcolor, photographic-quality visual representations that can
depict current and future conditions (Orland 1988). Retouching does not rely directly on data describing vegetation characteristics. Instead it uses a library of images representing
different forest conditions to replace portions of an original
image. Because this method is not driven by stand data, it is
often considered an artistic technique that can be significantly influenced by the artist’s choice of images to represent treatment effects.
Of the two methods, geometric modeling provides a link
between data describing stand and landscape conditions and
features in a simulation. In general, there is a one-to-one or
one-to-many relationship between data elements and objects
in the final image. Geometric modeling produces images
that are generally less realistic than those produced using
photo retouching techniques. However, when photographic
icons are used to represent individual trees and other objects, geometric modeling can produce images that contain
many of the details normally associated with photographs.
Most applications of visual simulation do not place individual plants in the same location as those elements would
appear in a photograph simply because such positional data
is not generally available.
Data Requirements for Geometric Modeling
Producing accurate, data-driven visual simulations of
forested areas requires large amounts of data. There are four
main elements needed to produce visual simulations using
geometric modeling methods:
•
•
•
•
terrain,
surface features,
cultural features, and
vegetation.
Terrain is represented using either a regular grid or a triangular irregular network (TIN) of elevations. Surface features include those things that can be represented by simply
texturing or coloring the terrain such as streams, lakes, and
areas of exposed soil. In some advanced simulation systems,
combining textures and surface deformations produces very
realistic surface features such as the earthwork and surface
associated with a road. Cultural features include structures
or objects such as roads, bridges, power transmission lines,
and buildings. Vegetation includes all plants that will be rep-
Visual Simulation Techniques
Visual simulation methods range from simple sketches
to complete virtual realities. Two computer simulation methods are suitable for producing visual representations of forest operations: geometric modeling and digital photo retouching (for a more detailed description of these and other visual
simulation methods see McGaughey 1998).
Geometric modeling methods build models of individual
components (ground surface, trees, other plants, and struc174
resented as separate objects in simulations. In some systems, vegetation growing close to the ground such as grass
or understory herbs and shrubs are represented using surface texturing rather than modeling individual plants.
Data used to represent terrain is readily available from
several sources. The USGS produces 30-meter gridded digital elevation models (DEM) for most forested areas in the
United States. Ten-meter models are in development and
are available for many areas. Both the 30- and 10-meter models are developed from the 1:24,000 scale contour map sheets.
Similar DEM products are available in other parts of the
world. These models, developed from aerial photography
generally represent overall landform well. However, they do
not capture small features and hence appear artificially
smooth when compared to the actual ground surface. Many
forestry organizations have more accurate contour maps produced for their lands. These map products may provide a
better ground surface for visual simulations but, most likely,
will not show small topographic features.
Surface features are generally represented using texture
images sampled from photographs. Textures are used to color
lines and fill polygons that delineate streams, lakes, and rocky
areas. Some applications model rock outcrops, cut and fill
areas associated with road construction, or other deformations of the ground surface in addition to applying textures
to represent these features. One of the simplest methods used
to produce visual simulations uses a rectified, geo-referenced
image depicting vegetation patterns to completely cover large
portions of the terrain. Such simulations show overall vegetation patterns but do not show height differences between
vegetation in adjacent stands. In general such draping applications are limited to producing simulations depicting large
land areas.
Cultural features that are linear and close to the ground
such as roads and fences are often represented using simple
lines or polygons rendered as surface features. As mentioned
above, sophisticated systems may model roads by modifying the terrain surface to reflect the road surface, ditches,
and cut and fill slopes. Objects such as bridges and building
are represented using object models created using 3D modeling packages. Similarly, detailed models of towers and
wires may be used to depict features like power transmission lines. Such objects and models are located by specifying their position and orientation.
Data describing forest vegetation typically includes two
components: stand polygons and tree attributes. Stand polygons, most often delineated using aerial photographs, represent areas with reasonably homogeneous vegetation cover.
For large areas, stand polygons may be obtained from satellite imagery. When identifying unique stands, there is typically a minimum polygon size that will be represented. Selection of the minimum polygon size depends on the vegetation characteristics of an area and the interpreters desire
and need to capture fine-scale variations in vegetation. Vegetation characteristics are specified for each polygon using
either general plant community descriptions or specific data
such as that obtained from forest inventories. As a general
rule, the closer a viewer is to a stand, the more vegetation
detail they can discern. The amount of data needed to adequately describe vegetation will be discussed in the next
section. Vegetation attribute data is generally organized into
a series of stand tables, one for each major species present
in the stand. Necessary parameters include stem diameter,
total tree height, height to the base of the live crown, and
the number of trees represented by the record in the stand
table. Visualization applications use these data to generate a
population of trees for each stand polygon. For most visual
simulations, individual tree locations are not needed. However, when producing displays intended to support on-going operations, tree locations may be needed.
Project Considerations
Two of the most important considerations when developing visual simulations are the amount of land area depicted in a simulation and the proximity of the viewer to
features in the simulated landscape. The amount of land area
depicted dictates the quality and resolution of data needed
to describe topography, stand conditions, and vegetation
characteristics. The viewer’s proximity to features in the
simulated landscape influences the amount of detail that must
be used when rendering specific terrain features and individual plants.
Efforts to characterize a project as either large or small
are not straightforward. The simplest measure of project size,
the extent of the area represented in a simulation, does not
provide a consistent measure of project size. Visible area,
determined by the viewpoint location, camera geometry, and
topography provides a better indication of project size. The
boundaries between project size categories are not distinct
since rendering applications can easily display only a small
subset of a large project to create views more commonly
associated with small projects.
For large projects viewed from a distance, a low-resolution digital terrain model and generalized stand polygons
attributed with the average tree size will allow simulations
of overall landscape conditions. Individual plants can be
rendered with a minimum of detail since the viewer cannot
readily discern individual plants in the final scene. Smaller
projects viewed from a distance may require medium-resolution data for both terrain and vegetation polygons. Viewers can generally see subtle differences in species composition and tree size between adjacent stands and may be able
to discern individual plants. However, because the viewer is
located some distance from the simulated area, the amount
of detail used when rendering individual plants can be kept
to a minimum. For projects that are viewed from a short
distance, high-resolution terrain and vegetation data are
needed to accurately represent topography and vegetation
characteristics. Individual plants, especially those in the foreground, must be rendered using high levels of detail.
Three project sizes, defined by visible area, are useful for
demonstrating visualization methods and discussing the
quantity and quality of data needed to produce visual
175
Figure 1. Visual simulation depicting 1,000 hectares.
Figure 2. Visual simulation depicting 100 hectares.
simulations: 1000, 100, and 10 hectares. The 1000- and 100hectare areas correspond roughly to landscape- and standscale projects. The 10-hectare visible area represents a
within-stand view, that is, viewpoints can be placed so it
appears that the viewer is able to walk through the stand.
While these arbitrarily sizes work well to illustrate visual
simulation concepts, it is important to remember that most
datasets will be used to produce images depicting several
visible areas. Figures 1, 2, and 3 show example simulations
that depict 1000, 100, and 10 hectares. These figures were
produced using a single dataset that describes about 2700
hectares.
Projects with 1000-hectare visible areas (landscape-scale
projects) can use readily available USGS 30-meter DEMs
and stand polygons larger than 5 hectares to map vegetation. Such projects effectively communicate overall harvest
unit location, size, and shape; general vegetation patterns;
and changes to overall aesthetic quality resulting from harvest activities or large-scale natural disturbances. Projects
with 100-hectare visible areas (stand-scale projects) can also
use 30-meter DEM data but might use 10-meter resolution
data to provide more terrain detail. The same stand polygons can be used but more detailed tree attributes are needed
because more detail can be discerned in the simulations.
These projects typically communicate treatment unit features such as stand structure changes, skid trail and landing
locations, and small-scale changes to the spatial pattern of
vegetation such as those produced by group selection harvests. Projects with 10-hectare visible areas (in-stand views
or the foreground portion of stand- or landscape-scale project
images) must use more accurate terrain models to depict
small topographic features. These projects rely on detailed
stand data and might even use maps of individual tree locations where absolute spatial accuracy is desired. Such
projects typically communicate detailed changes to stand
structure or specific spatial modifications that result from
harvesting activities or other disturbances.
To discuss the effect of project size on the amount of
detail needed when rendering individual scene components,
consider an image or tree icon used to represent an indi-
vidual tree. Low-resolution tree images are adequate when
rendering large visible areas with few trees in the foreground.
In figure 1, a single 33-meter tall tree is represented near the
middle of the scene using an image with a height equal to
2.6 percent of the vertical dimension of the final scene. For
figures 1, 2, and 3 the original images were 1536 pixels
wide and 1024 pixels high so 2.6 percent of the vertical dimension is about 27 pixels. In figure 2, the same tree represents 6.4 percent of the vertical dimension (66 pixels). In
figure 3, the tree represents 56 percent of the vertical dimension (573 pixels). As a general rule, to represent a reasonable level of realism in scenes depicting large visible areas, you do not need highly detailed tree images. In addition, you probably do not need data describing all vegetation to produce simulations that accurately communicate
stand conditions. For stands with closed canopies, it is only
necessary to depict overstory trees in simulations since almost no understory vegetation will be visible. For stands
with more open canopies, ground surface textures can be
applied to communicate the presence or lack of understory
vegetation. As visible area decreases, simulations must use
more detailed tree images, depict more vegetation especially
smaller understory trees and other plants, and use higher
resolution terrain data to maintain high levels of realism.
Using this more data intensive approach allows simulations
to communicate more subtle features such as understory
conditions, local topographic features, and ground surface
disturbances.
Visual Simulation to Support Operational
Planning and Implementation
Using visual simulations to support operational planning,
implementation, and monitoring is relatively new. During
the past decade, visual simulations have been used to show
the overall shape of harvest units and their appearance within
a landscape context. However, simulations are generally
produced to illustrate final design alternatives not to assist
during design of those alternatives. Current computer technology, particularly graphics hardware, has made it possible
176
Figure 3. Visual simulation depicting 10 hectares.
to more fully incorporate visual simulation capabilities
into design and implementation processes. Possible applications include:
•
•
•
•
efficient solutions that better accomplish their goals. For
example, visual simulations that depict the view from a
potential landing location including the tower, tailholds,
and a representation of the load path would help the designer assess the sensitivity of their design to minor topographic variation. When designing harvest unit shapes,
software to facilitate boundary delineation and analysis
while viewing perspectives would eliminate the trial and
error methodology where shapes are designed in plan view
and then displayed in perspective to evaluate their visual
impact. Analysis and design environments that display
shaded renderings of the ground surface with the location
of sensitive areas and vegetation information would help
planners design designated travel paths for ground-based
equipment and locate landings to avoid traversing unstable
areas.
Modern harvesting systems including single grip harvesters, steep slope feller-bunchers, and cut-to-length systems require extensive training to operate economically and
safely. The high initial cost for the equipment makes onthe-job training very costly given the low productivity associated with a new operator and potential for damage to
the equipment. By using simulators to learn machine functions, operators raise their level of expertise and their ability to produce high quality raw material in a safe, efficient,
and environmentally sound way. Two companies currently
produce equipment simulators to help train new operators.
Systems from Valmet (http://www.partekforest.com) and
Timberjack (http://www.timberjack.com/products/virtual)
both rely on actual control consoles taken from the equip-
visual feedback during the operational design process,
operator training,
in-cab displays to provide information to
harvest equipment operators,
display of equipment travel paths and cut/
leave tree locations to monitor compliance
with forest practice regulations.
Operational planning generally relies on data collected
on-site using simple survey methods. Data typically include ground profiles, location of streams and sensitive
areas, existing roads, and preliminary grade lines for new
road construction. Much of these data could come from a
GIS. However, commonly available products such as
USGS 30-meter DEMs and USGS hydrologic maps are
not considered accurate enough to support detailed analysis needed to design forest operations. Today, more accurate data products are becoming available. The USGS is
currently producing DEMs with 10-meter resolution and
many forestry companies regularly contract to have more
accurate contour maps produced for their lands. Using
these new data products, it becomes feasible to design
forest operations without extensive fieldwork and be reasonably confident that the designs can be implemented
with only minor changes. Incorporating visual simulations
into existing design tools would help users develop more
177
avoid sensitive areas and travel only on designated travel
paths. Displays that show stand structure by displaying individual trees could help operators select cut or leave trees to
meet stand spacing and species composition requirements.
Displaying additional information such as the location of
other equipment and personnel could help operators avoid
unnecessary operational delays while waiting for equipment
and personnel to clear hazard areas.
Current research is exploring the use of radio frequency
identification (RFID) tags implanted in trees. With tags in
place, information about an individual tree can be recalled
from a database and displayed as the machine approaches
the tree or grasps the tree in a felling head. Such information
might be used with on-board computers to assist with bucking optimization or might trigger alarms as the operator approaches a designated leave tree without requiring individual
tree locations. When actual tree locations are known, overhead views showing tree locations and attributes along with
the location of the machine could provide real-time feedback needed to make spacing decisions. Additional features
of such displays might include representations of designated
skid trails, locations of known hazard trees or other trees
requiring special precautions or attention, a record of the
machines path through the stand, and the location of cut trees
and log piles created during the operation. All of these display elements would make it easier for an operator to plan
their progression through a stand.
The ability to accurately record the path of a machine
through the woods can provide information to the operator
as well as someone charged with evaluating the operation.
Travel path data can provide a record used to measure compliance with designated travel path locations or quantify the
proportion of the site impacted by harvest activity. Precise
skid trail locations can be recorded for use in future operations where it may be desirable to use the same travel paths
to avoid additional compaction damage to the site.
Figure 4. Visual simulation showing a typical graphic display
from a harvest simulator that includes standing and down trees
and a portion of the machine.
ment being simulated and a highly realistic display of forest
conditions. Figure 4 shows a simulation similar to those seen
by the operator. These displays include trees, snags, ground
features, and other objects designed to make the operator
feel that they are in a “real” environment. The simulators
behave much like sophisticated computer games. Operators
can drive the machine through the simulated environment,
cut trees, and interact with the same computer system and
controls used on the actual equipment to buck trees into
merchantable products. To provide feedback to the operator
and to evaluate their performance, information such as residual stem damage, deviation from designated travel paths,
and the number and type of logs produced is recorded. Both
companies cite cost savings over real-world training, improved operator performance, and compliance with harvesting constraints as benefits of simulator training. Additional
applications include teaching operators to control residual
tree spacing thus eliminating the need to mark cut/leave trees
throughout an entire stand.
In-cab displays are currently used to provide feedback to
equipment operators using numeric displays to show equipment parameters or information related to product optimization. Map-like displays provide guidance and position information using GPS or other positioning sensing technologies.
Such displays do not typically use visual simulations. Two
evolving technologies may make it possible to provide more
operational feedback using simple visual simulations:
•
•
CONCLUSIONS
If one of the goals of precision forestry is to take
advantage of detailed stand and topographic data to adapt
forestry activities to variability within sites, then visual
simulations can be used to facilitate better understanding of
that variability and incorporate it into design activities.
Unfortunately, the amount of data required to produce
simulations that accurately depict the variability typical of
forested environments is overwhelming. Traditional
approaches to forest inventory use a stand as their basic unit.
For general, large-area visualization projects this approach
is sufficient. However, for site-specific visual simulations
including on-board displays to assist equipment operators,
the stand unit is too coarse. For these types of applications,
more detailed data is needed. This can include individual
tree location and related tree attributes. Such detail is not
currently available and it cannot currently be obtained
economically. New technologies such as LIDAR (light
detection and ranging) show promise for providing tree
real-time equipment tracking using global positioning systems (GPS) and inertial measurement units (IMU) and
radio frequency identification (RFID) tags implanted in trees.
Accurate position sensing technologies that can operate
under forest canopies will allow better links to forestry databases. Displays showing a planimetric view of a harvest unit
with sensitive areas or other no-travel zones highlighted and
the real-time position of equipment would help operators
178
through morphological analysis of a LIDAR-based
canopy surface model. In Proceedings of the 1st International Precision Forestry Symposium; 2001 June 18-19;
Seattle, WA. Seattle, WA: University of Washington.
locations and some attributes as well as very accurate models of the ground surface (Andersen 2001 this volume). Previous work to extract tree locations from airborne video also
show promise for open stands.
Computer graphics technology is no longer the limiting
factor when considering the use of visual simulations to
support operational planning and implementation. The limitations today are the availability of accurate, spatially explicit tree and terrain data that can be used to produce sitespecific visual simulations needed for detailed planning and
operational support.
McGaughey, Robert J. 1998. Techniques for visualizing the
appearance of forestry operations. J.For. 96(6):9-14.
Orland, Brian. 1988. Video-imaging: a powerful tool for visualization and analysis. Landscape Architecture. 78:7888.
LITERATURE CITED
Wilson, Jeremy S. and R.J. McGaughey. 2000. Presenting
landscape-scale forest information: what is sufficient and
what is appropriate? J.For. 98(12):21-27.
Andersen, Hans-Erik.; Reutebuch, Stephen E.; Schreuder,
Gerard F. 2001. Automated individual tree measurement
179
Chapter 24
Information Needs for Increasing Log Transport Efficiency
TIMOTHY P. MCDONALD
STEVEN E. TAYLOR
ROBERT B. RUMMER
JORGE VALENZUELA
Abstract—Three methods of dispatching trucks to loggers were tested using a log transport simulation model: random
allocation, fixed assignment of trucks to loggers, and dispatch based on knowledge of the current status of trucks and loggers
within the system. This ‘informed’ dispatch algorithm attempted to minimize the difference in time between when a logger
would use up available empty trailers and when a new empty would arrive. Simulations modeled a situation in which trucking
capacity was limited relative to logger production capacity. The fixed assignment of trucks to loggers produced the highest
amount of delivered wood, followed closely by the informed dispatch method. The difference seemed to be a result of the fixed
assignment method favoring loggers close to mills over those further removed, while the informed method dispatched trucks
more evenly among loggers regardless of transport distance. Results also demonstrated the need to balance trucking capacity
with logging production to maximize the amount of delivered wood.
INTRODUCTION
implying that elasticity in the overall supply logistics should
be maintained in trucking capacity.
Loggers, however, do not typically have the option of
varying trucking capacity to match distance to consumption
points. Although data are scarce concerning transport capacity, but it would make sense to assume that loggers retain
enough trucking capacity to handle an average haul distance.
Greene and others (2001), for example, report an average of
2 drivers plus 2 contract haulers per crew in Georgia. Loggers live with reduced utilization of in-woods equipment at
stands far removed from the mill, and with idle trucks when
working at stands close to the mill. If this scenario is correct,
pooling trucking capacity among a group of loggers should
increase log transport efficiency. Pooled transport trucks
could be rationally dispatched to the logger at which they
are most needed and eliminate the tradeoff between utilization of trucking and logging equipment caused by haul distance variations. Daniels (1994) discussed the development
of a truck dispatching system that was credited with increasing overall efficiency, resulting in cost savings to the company implementing the system. Shen and Sessions (1989)
presented a truck scheduling system that minimized costs of
transporting wood from several loggers to a single destination.
Pooling of transport capacity is also an example of how
real-time information is of potential benefit to the wood consuming industry. To work effectively, the transport network
would have to be managed to balance utilization of all loggers being served, and to maximize the amount of wood
About 80 percent of pulpwood delivered to US mills in
1996 (nearly 200 MM tons) arrived by truck (APA 1997).
Transportation of wood fiber accounts for about 25 to 50
percent of delivered costs, the variation depending primarily
on haul distance. Optimizing the transport efficiency of these
large quantities of raw material should benefit wood consuming industries. Applying information technology to log
transport has the potential to increase its efficiency, as its
application has in many other industries.
The wood transport network is somewhat different from
most other logistics systems, where a commodity is delivered from a central location to scattered distribution points.
The opposite is true for wood delivery: a commodity is collected from independent contracting agents that are widely
distributed across the landscape. These contractors are normally individually responsible for delivery of their product
to consuming mills. Each logger maintains, or contracts for,
a fleet of tractors to haul loads of wood. The number of trucks
used is typically fixed at a value high enough to keep a steady
flow of empty trailers at the logger’s deck for the in-woods
crew to load. Too few delivered empty trailers, and the logging deck will become clogged and the utilization rate of
expensive in-woods equipment will decrease. When trucks
are in oversupply, the logger is paying for transport capacity
that is unused. The optimal number of trucks should be related to at least two controllable factors: distance to the consuming mill, and production rate of the logging system. Of
these two factors, production of the logging system should
always be maximized to achieve optimal financial returns,
181
related to Euclidean distance, and an assumed (fixed) average travel speed for all trucks (45 miles per hour). Actual
travel time was triangularly distributed, with range equal
to 40 percent of the mean travel time.
Trucks cycled between loggers and mills. The dispatcher
made logger assignments at the time the truck exited a mill.
Trucks added an empty trailer to the logger’s processing
queue upon arrival, then either picked up a loaded trailer,
or waited for the next loaded trailer if none were available.
The load was hauled to the receiving mill and the cycle
repeated. There was no attempt made to optimize or prioritize which loaded trailer to haul to a mill among those
available at the logger. Trucks simply grabbed the first available in the queue and hauled it to the mill corresponding to
the product type of the load.
The dispatch agent waited for assignment requests from
trucks, then used one of three different algorithms to select a destination logger:
moved. Implementing the dispatch system will require feedback on the disposition of trucks and status of loggers, and
perhaps information on queue lengths at receiving mills. All
this data would have to be collected and analyzed in real
time in order to respond to dynamic changes in system status. Although there is no true spatial precision involved in
such a dispatching system, it would require a higher level of
information resolution and temporal necessity than currently
employed, placing this problem into the category of ‘precision forestry’.
The objective of this study was to evaluate the utility of
real-time information in enhancing effectiveness of truck
dispatch to loggers. ‘Effectiveness’, in this case, was assumed to be the ability of one dispatch method to move more
wood in the same period of time compared to another with
the same number of vehicles. A transport network simulation was used to compare dispatch methods. Methods evaluated were (a) fixed allocation of trucks to individual loggers, (b) dispatch to randomly selected loggers, and (c) an
‘informed’ approach that attempted to use data regarding
transport network status to dispatch trucks to the ‘best’ destination.
1. Randomly assigned. Probability of assignment
among loggers was the same.
2. Fixed (static) assignment. A truck was assigned
an ‘owner’ logger at startup. Loggers received a
number of trucks that was directly proportional to
their weighted distance to all destination mills.
Weighting factors corresponded to the probability a load would be sent to an individual mill.
METHODS
A simulation of a transport network was constructed that
included 3 destinations, 5 loggers, 30 trucks, and a dispatch
agent. The model was built using the AnyLogic 4.0 simulation package from XJ Systems, Inc. The destinations simulated wood consuming mills and consisted of in- and outbound scales with a crane in between. Between the crane
and scales, trucks were subjected to asynchronous delays
that simulated travel and unbinding functions inside the mill
gates. Processing times were based on information obtained
from local mills, and each was assumed to be a triangularly
distributed random variable. Truck arrivals at the mill gates
included those explicitly modeled as part of the simulation
(i.e. the 30 mentioned previously), as well as other trucks
representing the rest of the procurement system. Inter-arrival times of these additional trucks were exponentially distributed with mean inter-arrival times assigned using total
daily wood consumption data from local pulp and saw mills.
Loggers were modeled simply as loaders that processed
empty trailers. Loading times were assumed to be triangularly distributed with a fixed mean and range of variation.
Processing queues for each logger could be charged with an
initial number of empty trailers at the start of the simulation. Loaded trailers were assigned a weight that was a random variable uniformly distributed between 25 and 29 tons.
The trailer was also randomly assigned a product code that
corresponded to a destination mill. The logger was assigned
product sort information at the beginning of the simulation.
Each product was assigned a probability of occurrence that
represented the percent of that product included in the stand
being harvested. Distances to each mill were also assigned
at startup. Each mill had a fixed x-y location, as did each
logger. Mean travel times between the logger and mills were
3. ‘Informed’ (dynamic) assignment. This method
calculated the time in the future that all loggers
would exhaust the available supply of trailers either in transit or waiting at the deck. The truck
was then assigned to the logger such that the difference between the arrival time and the time trailer
supply was exhausted would be minimized. A further check was applied such that the ‘best’ assignment would be dropped in favor of the logger that
was furthest ‘behind’, i.e. the logger that had been
waiting for a trailer the longest, if that waiting
time was large.
Overall, the simulation was designed to model a transportation network that was truck-limited. It was assumed
that situation would favor a truck assignment that was informed, rather than fixed or random. The scenario modeled assumed all loggers produced loaded trailers at the
same rate, varying only in the distance to the mills. Model
parameters were assigned such that, over long periods of
time, trucks were available to haul about 75 percent of the
trailers that loggers were capable of producing.
Model parameters used in the simulations are shown in
Table 1. A travel index value, Ti from Equation [1] for
logger i, was calculated as the sum of the products of the
transit time means and associated probability of occurrence
for each mill destination, where M was the number of mills.
182
Table 1. Model parameters used in the simulations.
a
V
0.1
0.0
0.0
0.0
0.05
1
2
3
4
5
a
Mean Travel Time, tij
(min)
V
C
P
68
91
57
114
156
149
132
112
137
78
27
42
149
109
142
Product Probabilities, sij
Logger
C
0.2
0.1
0.4
0.0
0.15
P
0.7
0.9
0.6
1.0
0.8
Travel
Index, Ti
# of Trucks
Assigned
65
150
127
42
137
4
8
7
3
8
– V,C, and P designate the three mill destinations (nominally a veneer, chipnsaw, and pulp mill)
Table 2. Mean simulation output values by truck assignment method
Model Result
Assignment Type
Informed
Random
28.5
Tons Hauled
57772a
59689b
60490c
Truck Cycles
2141a
2213b
2243c
a
a
25.9
M
∑t ⋅
j=0
ij
b
25.0b
– Values with the same superscript are statistically equal, P = 0.05.
Numbers of trucks assigned in the static allocation simulations were based on the proportion of the travel index for
an individual logger relative to the total for all loggers.
Ti =
Fixed
% Logger Idle
Time
sij
tests, trucking capacity was about 25 percent below that
needed to match the output capacity of the loggers. This
effectively capped the advantage that could be gained by
any allocation method unless some structural change to the
transport network was implemented, e.g. changing the number of loggers, or their production rates. In the informed
assignment, there was a check used in the assignment algorithm that made sure no logger was ignored. Without this
check, the algorithm consistently stopped dispatches to one
logger altogether, and wound up moving about 4.8 percent
more wood than when serving all 5 loggers. Similarly, the
fixed assignment strategy allocated trucks based on an expected turn time, but the assignment was not exact because
only integer numbers of trucks could be allocated to each
logger. Rounding the number of trucks upward for loggers
closer to the mill gave them an advantage in truck response
time over those further away, which netted more wood to
the mill by trading off idle time at remotely placed loggers
for work time at closer loggers. The informed dispatch
method balanced idle time among all loggers more effectively. Standard deviation and range of logger idle time for
the informed assignment was 2.3 percent and 7.6 per cent,
respectively. For the fixed assignment, the same values were
3.8 percent and 8.7 percent.
[1]
Simulations were run for a total of 300 model hours, 5
replications for each assignment type. Output variables measured included vehicle and logger utilization, plus total number of truck cycles and tons moved of each product type.
RESULTS AND DISCUSSION
Simulation results showed no advantage gained using
dynamic dispatch information. Although more wood was
transported than using random assignment, matching trucks
to the expected emptying time of loader queues moved 1.3
percent less wood than the static assignment of trucks to
loggers. Overall, there was 1 more turn per truck over the
300-hour simulation using static assignment.
This finding was counter to our original hypothesis, but
other simulation results indicated that the advantage for the
static assignment was gained by slightly favoring loggers
closest to the consuming mills. In the scenario used for these
183
The simulations underscored the nearly linear relationship between number of trucks needed to haul for a logger
and travel distance to mills. This indicated that pooled allocation of trucks among a group of loggers would likely haul
the same amount of wood with fewer trucks, lowering delivered costs. Other structural changes in the transport network, e.g. use of surge yards to distribute truck arrivals at
mills more evenly, could also provide transport efficiency
advantages. Dynamic allocation of trucks, at least for the
conditions tested in this study, did not improve transport
efficiency, but did provide other benefits. The algorithm was
very effective at equalizing logger productivity by balancing empty trailer deliveries, an important consideration for
loggers if they were considering turning their trucking systems over to a central dispatching system.
The overall improvement in tons hauled when dropping
1 logger was also seen in the static allocation of trucks, but
to a lesser extent (2.6 percent improvement). This result
emphasized the significance of matching trucking capacity
to logging systems to maximize delivered wood amounts.
Although the static allocation of trucks was superior to a
dynamic approach in these simulations, this does not necessarily imply that real-time information flow has no use in
truck dispatching. These results were based on steady-state
conditions. Random perturbations in the system would perhaps occur often enough that a static allocation of trucks,
even repeated when the transport network changed, would
not be best, or even feasible. Future work will be done to
examine the relative importance of random perturbations on
wood delivery schemes.
Any simulation is only a model of reality and there were
a number of idealizations incorporated in this study. Shift
lengths of drivers, for example, were not considered, neither were dynamic fluctuations in arrival times at mills. Truck
arrival intervals change significantly over the course of a
day, but this was not considered in the model. In fact, no
mill information was used in the truck dispatch algorithm at
all, but could be incorporated and perhaps provide additional
efficiencies over a static allocation. This would most likely
be of benefit if there were alternative destinations available
for a single product class. Distributions of various processing times used in the simulation were selected for simplicity
rather than based on solid information. Measuring these input parameters might improve the overall realism of the simulations.
A single method of dynamically allocating trucks was
tested in this study. There are almost assuredly better methods than the approach taken here, but further research will
be needed to define what the important state variables and
decision criteria would be to improve dispatch efficiency.
LITERATURE CITED
APA. 1997. Annual pulpwood statistics summary report.
Forest Resources Association, 600 Jefferson Plaza, Suite
350, Rockville, MD. APA Report 97-A-12. 20 p.
Daniels, T. 1994. Transport systems today and tomorrow. P.
202-208 In Proceedings of the 17th Annual Meeting of
the Council on Forest Engineering; Advanced Technology in Forest Operations: Applied Ecology in Action,
Sessions and Kellogg (eds.). Oregon State University:
Corvallis, OR.
Greene, W.D., B.D. Jackson, and J.D. Culpepper. 2001.
Georgia’s logging businesses, 1987 to 1997. Forest Products Journal. 51(1):25-28.
Shen, Z., and J. Sessions. 1989. Log truck scheduling by
network programming. Forest Products Journal.
39(10):47-50.
184
Chapter 25
Hierarchical Planning : Pathway to the Future?
JOHN SESSIONS
PETE BETTINGER
Abstract—Hierarchical approaches to forest management planning involve developing courses of action at two or more scales
or levels: perhaps at a strategic level and a tactical or operational level. Some have argued that this system of planning is
appropriate given the level of data required to develop a course of action for all levels simultaneously, and the complexity of the
resulting planning problem. We see the value in this approach, in that it simplifies the analysis at each level. However, as better,
and more precise data are developed, we can also see that the collapse of planning from hierarchical levels into a single process
may also be appropriate, depending on the planning goals, the planning environment, and the associated risks of using just one, or
two or more hierarchies.
INTRODUCTION
level to the tactical planning level. Accessibility constraints
were typically dealt with by limiting the percent of acres
available in any time period within a given strata.
Barber et al (1996) summarized the approach: “the primary function of this level (strategic planning) is to analyze
forest-wide issues, concerns and opportunites, allocate lands,
adopt standards and guidelines, establish production levels
for outputs and describe environmental effects. The explicit
representation of spatially continuous areas is not relevant
or needed at this level of analysis. It can even be counterproductive. Linear models cannot really solve spatial problems.
Placing them in the model only gives the illusion of site specific schedules”. With respect to the tactical planning level
Barber et al (1996) list the main functions as 1) spatial disaggregation of the strategic plan, 2) adjusting or rescheduling the proration of the strategic plan if local thresholds are
exceeded, and 3) analyzing connected actions and cumulative effects on a tactical planning area that may extend beyond the boundaries of the tactical planning area.
If serious conflicts are identified at the tactical planning
level, then adjustment in outputs can be made at the tactical
level or the strategic plan can be revisited. Church et al (1995)
have used a goal programming formulation to try to minimize deviations from strategic plan goals during the disaggregation stage. The Woodstock and Stanley harvest scheduling system (REMSOFT 2001) is another example where
strategic plans are developed and Monte Carlo simulation is
used to disaggregate strategic plan goals.
The First International Precision Forestry Symposium is
an appropriate setting to review the use of improved information in forest planning. One hypothesis of the symposium
is that precision forestry will support the development of
precise forest plans that can be implemented accurately and
subjected to rigorous review.
Forest planning can be defined as the identification of
activities and the timing of those activities to reach the goals
of forest management. Connelly (1996) has defined hierarchical analysis for forest planning as “the organization of
information for making decisions at different levels when
the quality of the decision at one level is dependent upon
decision or information at other levels. Levels may be defined temporally or spatially where the scope of the higher
level fully encompasses the scope of the lower level”
In this paper we concentrate on the application of hierarchical planning with relation to long range and shorter range
plans for forests, where the longer range plans (strategic plans)
typically deal with planning horizons of several rotation
lengths and shorter plans (tactical plans) deal with implementing strategic plans over the near term. Many strategic
plans since the late 1960’s have used linear programming to
allocate activities for forest planning. Notable developments
include MAX MILLION (Clutter 1968), Timber RAM
(Navon 1971), and FORPLAN (Kent et al. 1985, Kent et al.
1991). Other strategic plans have used some variant of binary search including Chapple (1966) and Chambers and
Pierson (1973). The land was classified according to the issues to be addressed, but usually all land with similar vegetation characteristics within a given land allocation were
grouped to reduce problem size. Spatial constraints were
either ignored or converted to output reduction factors for
harvest outputs that are passed from the strategic planning
Underlying the hierarchical approach are two important
considerations:
1.
Including spatial considerations substantially in
creases model complexity and solution time.
2.
Spatially explicit schedules require spatially explicit
data.
185
Adjacency constraints, particularly those permitting
groups of polygons to harvested together, corridors for wildlife or recreation, floating patches of mature habitat, roads,
core habitat requirements, and road access are several of the
spatial requirements often left for tactical planning due to
the inability of strata-based strategic models to consider them.
Similarly, issues such as logging feasibility, and the decisions that relate to logging feasibility, including the coordination of upslope and adjacent riparian operations, the effect of economics on setting size and volume, and cost as a
function of distance from road are generally left for tactical
planning.
Strategic planning often uses forest inventories designed
to support accuracy statements for individual strata or total
inventories at the forest level. Data requirements at the substrata or parcel level require considerably better knowledge
of the vegetation to approach accuracy standards typically
assumed for the forest level. With the advances of technology discussed at this symposium, the prospects for obtaining higher resolution, spatial data are becoming much
brighter. One might cynically question much of the economic
analysis done with non-spatial strata level data and reflect
upon whether the results are artifacts of the stratifcation process.
Model complexity and data requirements are important
issues that need careful consideration. However, the major
driver that is working to blur the line between strategic and
tactical planning is the need to evaluate cumulative spatial
effects. Strata-based plans that cannot portray the distribution of forest structure across the landscape hinder evaluation of progress toward watershed, wildlife, and social goals.
If spatially based goals and/or constraints are real, in the
sense that they affect goal attainment or plan feasibility, ignoring spatial relationships brings into question plan feasibility and efficiency.
spatial harvest scheduling algorithms. Others are experimenting with the ideas. Companies that have developed spatial
scheduling algorithms include Weyerhaeuser Company,
Willamette Industries, and Pacific Lumber. To support the
perceived requirements of the American Forest and Paper
Association (AF&PA) Sustainable Forestry Initiative process (AF&PA 2001), NCASI has supported development of
a spatial harvesting scheduling model (Van Deusen 1999)
and forest planning consultants including Forest Analytics,
Inc., James Arne and Associates, Forest Planning Systems,
and D.R. Systems, Inc. are developing software for industrial users. Both the Washington Department of Natural Resources and Oregon Department of Forestry are using spatial scheduling models to bridge strategic and tactical forest
planning goals.
An Example
We illustrate an example using the Elliott State Forest
where the hierarchies of strategic and tactical planning have
been collapsed into a single planning process. Both strategic and tactical goals are considered, over long time horizons, and with very precise spatial data as well as growth
and yield data.
The 93,000-acre Elliott State Forest is located east of Coos
Bay and Reedsport in the Coast Range. The Oregon Department of Forestry is initiating a planning process there to
update and revise the state’s plans for managing habitat for
wildlife and other resources on the Elliott State Forest.
The Elliott State Forest Habitat Conservation Plan (HCP),
approved by the U.S. Fish and Wildlife Service in 1995, is
used to help implement the Elliott State Forest Management
Plan, which was approved by the Board of Forestry in 1993.
The northern spotted owl and the marbled murrelet are the
two threatened species covered in the current HCP.
The Oregon Department of Forestry is revising the Elliott
State Forest’s HCP to update conservation strategies for
marbled murrelets (including renewal of an incidental take
permit), address protection of additional threatened and endangered species (such as coho salmon) and provide increased certainty and consistency for harvest and revenue
levels. The Department is also looking at revising the Elliott
State Forest’s management plan to possibly take advantage
of newer forest strategies that would allow greater flexibility for management of the lands.
In the current HCP the incidental take permit for owls is
60 years, but the marbled murrelet portion is only for six
years, expiring in October 2001. The permit was limited to
six years because little was known about marbled murrelets
at the time. Since then, ODF has funded significant research
on the murrelet to help revise strategies that might support a
longer-term HCP for the marbled murrelet.
ODF is developing seven variants of two harvest scheduling models to provide an analysis of a broad range of “conceptual management alternatives”. The differences between
the harvest scheduling models are due to structural differences in modeling take avoidance as compared to reaching
What is the Alternative to Hierarchical
Planning?
Planners have accepted that where linkages exist between
forest resources, higher quality solutions can be identified
by simultaneous consideration of resources than by solution
in stages. The difficulty has been how to do this. Recent
development of heuristic solution procedures (Davis et al.
2001) has shown promise for developing good, feasible solutions to large spatial problems. The tradeoff at present is
the choice of between an “optimal” solution of a non-spatial
problem with an attempt to place the non-spatial solution on
the ground during the tactical planning stage, or finding a
feasible, spatially feasible solution at the strategic level that
includes the considerations normally left to the tactical level.
Is There a Trend?
The importance of spatial requirements in harvest planning has prompted several of the major forest industry companies in the Northwest to develop their own proprietary
186
ated stochastically in potential murrelet habitat across the
landscape. The amount and location of the new reserves
depends on the distribution of age classes in the vicinity of
the potential murrelet habitat.
Restriction of thinning by distance from road – To recognize logging feasibility, thinnings were limited to areas
within 700 feet from an existing or planned road.
HCP goals. The comparative analysis of economic and habitat outputs of these alternatives will assist ODF in development of “refined management alternatives” that will form
the basis for a revised Elliott State Forest Management Plan.
The array of conceptual management alternatives include:
strategies in the current Plan/HCP, several alternatives with
additional habitat protection for salmonids and murrelets, a
“take avoidance” alternative as if no incidental take permits
were issued, an alternative for increased reserve acres, and
an alternative employing the structure-based management
approach similar to the recently approved Northwest Oregon
State Forests Management Plan.
2. Nonspatial Issues
Depending upon the scenario, nonspatial issues included:
1.
2.
3.
4.
1. Spatial Issues
Five spatial issues were considered important to realistically
portray the various alternatives for the Elliott Plan. These
are:
1.
2.
3.
4.
5.
5.
Size of regeneration openings
Logical harvest units for thinnings and regeneration harvests
Control of activities around owl/murrelet reserves
Creation of likely new reserves.
Restriction of thinning by distance from road
6.
Nondeclining flow of timber harvest
Maximization of total harvest
Maximization of present net value
Achievement and maintenance of nesting roosting and foraging (NRF) habitat goals by management basin.
Achievement and maintenance of dispersal habitat goals by management basin.
Achievement of forest structure goals forestwide .
3. Land Allocation
To address spatial and economic issues, nine data layers were
intersected to create individual parcels of lands that were
homogeneous with respect to nine variables. Depending on
the alternative, 58,000 to 85,000 separate parcels were created. The nine variables include:
Size of regeneration openings – The size of regeneration
openings is limited to 120 acres to comply with the Oregon
Forest Practices Rules. The green-up period is 5 years. This
requirement determined the 5-year period length for the harvest scheduling model.
Logical harvest units for thinnings and regeneration harvests – Logical harvest units are defined at two levels. First,
to permit efficiency of regeneration harvests, all parcels
within a logging setting must be harvested at the same time.
The average logging setting is about 40 acres and consists
of 20-30 subparcels defined by slope class, vegetation type,
distance from stream. Adjacent logging settings can be harvested during the same time period, as long as the 120 acre
limitation is not exceeded. Second, all thinnings with a vegetation type must be coordinated. Within a logging setting,
there may several different existing vegetation types. These
vegetation types may span upslope and riparian areas that
are eligible for different prescriptions. The prescriptions must
be coordinated such that all thinnings within a vegetation
type within a logging setting are done at the same time, but
the thinning intensity can vary.
Control of activities around reserves - Depending upon
the scenario, the maximum level of activity that can take
place around the 30 owl/murrelet reserves distributed over
the landscape is limited by two types of constraints: the number of logging settings that can have a regeneration harvest
adjacent to a reserve, and the number of acres in the regeneration harvest adjacent to a reserve. The limits depend upon
the management basin in which the reserve is situated.
Creation of new reserves – In some scenarios designed
to simulate take avoidance strategies, new reserves are cre-
1.
2.
3.
4.
5.
6.
7.
8.
9.
Vegetation type
Slope class
Presence of Swiss needle cast disease
Riparian/upslope
High risk and other reserve areas
Logging setting
Owl/murrelet reserves or owl circles
Management basins
Road buffers
Vegetation types- Vegetation types were derived from
delineation of stands from aerial photos. A description of
the existing stand (trees per acre, species, age, tree height)
was developed from ground plot data within each vegetation type. Sixty-nine vegetation types were defined.
Slope Class- Four slope classes were defined for the purpose of assigning regeneration activities and green tree retention levels.
Swiss Needle Cast Disease – Each parcel of land was
either assessed as having been affected by Swiss Needle Cast
disease or not. This determined eligibility for specific prescriptions and affected the choice of vegetation at regeneration time and the growth rate of Douglas-fir in existing
stands.
Riparian/Upslope – Each parcel was classified as to
whether it was a riparian polygon or upland polygon. For
several alternatives a separate spatial layer was used to de187
fine Special Stewardship (inner-riparian), Focused Stewardship (outer-riparian), and upslope polygons.
High Risk and Other Reserves – Areas of high risk for
slope instability and other areas off limits for harvesting due
to the Forest Land Management Classification (including
non-productive sites, scenic conservancy, administrative
sites) were reserved from harvest.
Logging Settings- Logging settings were defined by
Elliott Forest staff from considerations of logging feasibility. Each setting was assigned a logging system. There were
approximately 2000 logging settings. At regeneration time,
all parcels within a logging setting were harvested.
Owl/murrelet reserves or owl circles – Depending upon
the alternative each parcel was classified as to whether it
was in an owl/murrelet reserve. Thirty reserve areas were
identified. For alternatives involving take avoidance, each
parcel was classified as to whether it was in an owl circle or
overlapping group of owl circles.
Management Basins – The Elliott forest was divided into
17 management basins, that depending upon the alternative
have specific goals that may be unique to a basin, or group
of basins.
Road Buffers – A road right of way was identified in addition to buffers of 700 feet and 1000 feet that determined
thinning feasibility and logging costs.
nealing, a Monte Carlo neighborhood search technique that
has been applied in forestry planning since the early 1990’s
(Lockwood and Moore 1993). Depending upon the alternative, a modified goal programming objective function was
used to either minimize weighted deviations from targets or
to subtract constraint penalties from net present value. Solutions started with the no-cut prescription assigned to all
polygons (spatial feasibility) and during the neighborhood
search, only moves that met all spatial constraints were admissible. To assure the model is computationally correct, a
parallel model is being independently developed as part of a
graduate thesis to provide model validation. The harvest
scheduling model is undergoing evaluation and databases,
prescriptions, goals, and constraints are being refined.
Figures 1-4 illustrate the hierarchy of management basins, logging settings, and homogeneous parcels. Figure 5
is a display of the harvest units from a non-declining flow
harvest schedule across the entire forest where activity near
reserves is being constrained according to the goals for the
individual management basin.
CONCLUSIONS
Hierarchical planning is one approach to forest planning.
It is a useful approach yet its success depends on the goals
involved, the resources available (both planning tools and
precise spatial and growth and yield data) and deadlines. It
includes one set of constraints at a strategic level and introduces additional constraints at a tactical level usually at a
different spatial scale and temporal scale. If disaggregation
of the activity levels at the tactical level is not satisfactory
(that is cannot meet the constraints) an adjustment is made
at the tactical level. If the adjustment is serious enough, a
feedback loop can be made to the strategic level.
Hierarchical planning can provide an organizationallyconvenient method for problem solving, scoping out alternatives strategically, and identifying areas and issues to concentrate on further. However, eliminating the hierarchy provides the opportunity to solve the strategic and tactical planning problems simultaneously, yet this is not without its own
challenges. For example, the use of a single hierachy requires more detailed information up-front. The trade-off
for this additional investment is that “exact solutions” produced in a hierarchical planning process, that later turn out
to be infeasible when disaggregated, can be avioded. The
downside, at least at present, is that the availability and cost
of information may be prohibitive and the resulting solutions are not solved “exactly.” The success of precision forestry may well determine the future of hierarchical planning.
4. Growth and Yield
For each vegetation strata (unique combination of vegetation type, site class, slope class, presence or absence of
Swiss Needle Cast disease) the ORGANON (Hann, 1995)
model was used to project stand structure and harvest volumes for a group of thinning prescriptions that were judged
to be reasonable candidates to contribute to landscape goals
for an alternative. For alternatives involving only one type
of riparian zone there were 576 unique vegetation strata;
552 to represent existing stands and 24 to represent regeneration stands. For each of these strata up to 63 unique thinning prescriptions were developed including the grow-only
prescription. For alternatives involving an inner and outer
riparian zone there were 1152 vegetation strata.
5. Scheduling Model
Due to the spatial goals identified for the Elliott forest,
developing forest plans using exact methods would be intractable, forcing planners to use a hierarchial approach.
However, a heuristic scheduling model was developed to
allow us to span the hierarchy, and to use a variety of specialized data to account for spatial goals.
Seven integer decision variables were defined for each
parcel; three to define the regeneration time for the three
possible final harvests that could occur within the 150 year
planning horizon; and four to define the maximum of four
thinning prescriptions that could be assigned to each parcel.
Depending on the spatial layer being used (58,000 or 85,000
parcels) there were 400,000 to 600,000 integer decision variables. The scheduling model heuristic used simulated an-
REFERENCES
American Forest & Paper Association. 2001. Sustainable
Forestry Initiative (SFI) program.
http://
www.afandpa.org/forestry/sfi_frame.html (accessed 05/
06/01).
188
Figure 1. Overview of the Elliott State Forest showing
management basins.
Figure 4. Enlargement showing individual parcels of
homogeneous vegetation; riparian and upslope.
Figure 2. Portion of the Elliott State Forest showing
individual parcels of homogeneous vegetation stratified into
riparian and upland areas. Larger aggregations are management basins.
Figure 5. Display of harvest units in one portion of the Elliott
State Forest during the first five-year planning period.
Regeneration units must be less than 120 acres and in this
alternative, the maximum number and contiguous area of
regeneration units adjacent to a reserve is limited by the
management basin (shown in purple) in which the reserve is
located.
Figure 3. Parcels aggregated into logging settings for
regeneration harvest. All parcels within a logging setting that
are eligible for a regeneration harvest (not-reserved) must be
treated during the same period.
189
Davis, L. K.N. Johnson, P. Bettinger, and T. Howard. 2001.
Forest Management. McGraw Hill. 804 p.
Barber K., R. Butler, D. Caird, and M. Kirby. 1996. Hierarchical approach for national forest planning and implementation. In Proceedings: “Hierarchical Approaches to
Forest Management in Public and Private Organizations”,
May 25-29, 1995, Toronto, Canada, Petawawa National
Forestry Institute, Information Report PI-X-124, Canadian Forest Service, p. 36-44.
Hann, D., A. Hester, and C. Olsen. 1995. ORGANON User’s
Manual Edition 5.0, Department of Forest Resources,
Oregon State University, Corvallis. 125 p.
Kent, B. , J. Kelly, and J. King. 1985. FORPLAN Version 1:
Mathematical Programmer’s Guide. Land Management
Planning Systems Section, USDA Forest Service, Fort
Collins, CO. 350 p.
Chambers, C. and R. Pierson (Eds.) 1973. Sustainable harvest analysis, 1971 and 1972. Washington Department
of Natural Resources Harvest Regulation Report 5, Olympia, WA.
Kent B., B. Bare, R. Field and G. Bradley 1991. Natural
resource land mangement planning using large-scale linear programs: the USDA Forest Service experience with
FORPLAN. Operations Research, 39:13-27.
Chapple, D. 1966. A computer program for scheduling allowable cut using either area or volume regulation during sequential planning periods, USDA Forest Service
Research Paper PNW-33.
Navon, D. 1971. 1971. Timber RAM. A long range planning method for commercial timber lands under multiple
use management, USDA Forest Service Research Paper,
PSW-70.
Church, R., A. Murray, K. Barber, and R. Dyrland. 1995. A
spatial decision support system for doing hierarchical
planning. In Proceedings “Analysis in Support of Ecosystem Management: Analysis Workshop III”, April 1013, Fort Collins, CO, USDA Forest Service, p. 283-292
REMSOFT. 2001. Intelligent software for the environment.
www.remsoft.com (accessed 05/06/01).
Clutter, J. L. 1968. A computerized forest management planning system. Athens: School of Forest Resources, Univ.
of Georgia.
Van Deusen, P. 1999. Multiple solution harvest scheduling.
Silvica Fennica 33:207-216.
Connelly, W. 1996. A definition for hierarchical analysis
for forest planning. In Proceedings: “Hierarchical Approaches to Forest Management in Public and Private
Organizations”, May 25-29, 1995, Toronto, Canada,
Petawawa National Forestry Institute, Information Report PI-X-124, Canadian Forest Service, p. 1.
ACKNOWLEDGMENTS
The authors wish to thank Pam Overhulser, Forest Analyst, Oregon Department of Forestry, Salem and the Elliott
Forest staff for their guidance, cooperation and support in
the development of the Elliott Forest Planning model.
190
Chapter 26
Use of Airborn LASER System for Forest Inventory in Siberia
IGOR DANILIN
EVGENY MEDVEDEV
Abstract—The use of laser scanning method provides a number of principally new possibilities on remote sensing of forest
vegetation. The high productivity of laser sensing survey (up to 30 thousand original measurements per second) combining with
the spatial resolution and accuracy of tens centimeters allow making the effective algorithms of morphology analyses, ensuring
automatic extraction of many important information characteristics of forest cover.
The analysis of stand structure integrated with GPS data, digital aerial video and highly accurate (10-15 cm of actual linear
resolution) photographic images makes it possible to highly reliable interpretation of different types and layers of forest vegetation separating it by a tree species, density and the other parameters. The consequent processing of a laser profiling data using
developed in Russia ALTEX software, integral calculations, the Fourier and mean free path analysis makes possible to get such
important and precise information on vegetation as a timber stock, forest type, NDVI at a direct way or by mediate – on correlation
with tree crown diameter, density, crown vertical extent and tree stand height. The regression method provides high accuracy
assessment of a stand biomass when processing a laser profiling data. The methodology aspects of airborne laser sensing method
use for forest survey are considered in the article. The description of the morphology algorithms is shown. The results of a
practical application of airborne laser sensing by ALTM 1020 machine to forest inventory in Central Siberia are discussed in the
paper.
Systems Engineering for Biomass Feedstock
SHAHAB SOKHANSANJ
Oak Ridge National Laboratory (ORNL) provides leadership for the U.S. Department of Energy (DOE) Biopower Feedstock
Development program and Biofuels Feedstock Development Program. These programs perform research, development, and analysis
to establish that biomass supply systems can be environmentally beneficial and commercially viable. They emphasize developing
new, sustainable energy resources based on solar energy captured by living plants. The research which is supported by funds from
the DOE’s Office of Transportation Technologies and Office of Power Technologies is carried out in partnership with universities, other government agencies, and the private sector.
The Systems Engineering Task supports and conducts research on collection, storage and transportation of biomass feedstock.
The near term objective of the Task is to ensure cost competitiveness of biofuels. The overall goal is to establish safe and
sustainable supply systems in support of the entire bio-based industry. The Task initiates fundamental engineering research,
supports field trials and plays an active role in product specific initiatives. Research on physico-chemical characterization of
biomass for the design and operation of handling equipment requires careful laboratory experiments. The work in the field is
focused on testing and validation of improved or innovative concepts for storage and handling systems. The Task also supports
multidisciplinary platforms that require engineering expertise in systems optimization and scale-up technologies.
Current research is focused on efficient handling of crop residue (straw and stover) for biofuel; harvesting and treatment of
energy crops (grasses and woody crop) for bio-power. Research topics include development of moisture relations for corn stover,
data on physical characteristics of grasses, and storage stability of corn stover and switchgrass. The results will provide critical
data for optimizing machinery and systems for the entire feedstock supply chain.
191
Chapter 27
Strategic Research for Inventorying, Monitoring and Modeling
our Managed and Natural Ecosystems
JACK SJOSTROM
CLAIRE RUTLEDGE
PAUL GESSLER
PETER GORSEVSKI
AMY POCEWICZ
ASHLEY COVERT
ERDENE SAIKHAN
JEFF CRONCE
Abstract—This poster will illustrate the variety of research projects underway that cover a broad range of spatio-temporal
scales and science objectives for mapping, monitoring and modeling managed and natural ecosystems in the Northern Rocky
Mountain Region. The featured sensors and associated science objectives will include MODIS, Landsat, IKONOS and the
AVIRIS instrument. Each project is integrating advanced remote sensing and GIS tools and datasets to improve our understanding of ecosystem structure and function. Specific projects will include:
•
•
•
•
•
Change detection for the Northern Rocky Mountain Region
Analysis of MODIS imagery and derivative data products (NDVI, LAI) for understanding the temporal dynamics of
ecosystem and landscape function in the Northern Rocky Mountain Region
GIS and remote sensing for process-based modeling to derive soil moisture, slope equations, and climatic zones for
the prediction of landslides on forested landscapes
Developing new vegetation mapping methods for the Salmon-Challis National Forest
Mapping forest vegetation and biophysical characteristics using the low-altitude Airborne Visible Infrared Imaging
Spectrometer (AVIRIS) hyperspectral sensor
Forest Density Classification in Logged-Over Forest Using
Airborne Radar Data
SAFIAH YUSMAH M. Y.
Abstract—Virgin forest that has been logged is usually known as logged-over forest. In Malaysia, some of these logged–
over forests are approaching the second felling cycle. However, much information about these logged-over forests is still not
collected and inadequate. This Study is carried out to investigate the stocking of the logged-over forest in Gunung Arong
Forest Reserve, in the District of Mersing, Johor, Malaysia. The objective of the Study is to classify the logged-over forest
into several density classes using airborne AIRSAR radar data. The image is processed using unsupervised classification to
produce the density classes. Then, ground-thruthing is carried out to validate the processed image. The image is classified
once again using the supervised classification. The result is a forest map showing areas of low, medium and high-density
trees. This map will provide information about the stocking of the area for the next felling cycle.
192
List of Contributors
A. Lynn Abbott
The Bradley Department of Electrical
& Computer Engineering
Virginia Tech
Blacksburg, VA 24061
USA
Marvin E. Bauer
University of Minnesota
Department of Forest Resources
1530 Cleveland Avenue North
St. Paul, MN 55108
USA
Neil Clark
USDA Forest Service
Souther Research Station
1650 Ramble Road
Blacksburg, VA 24060
USA
Abdullah E. Akay
Oregon State University
Department of Forest Engineering
213 Peavy Hall
Corvallis, OR 97331
USA
Gero Becker
University of Freiburg
Institute of Forest Utilization and
Work Science
Werderring 6
79085 Freiburg,
Baden-Wurttemberg,
GERMANY
Woodam Chung
Department of Forest Engineering
213 Peavy Hall
Corvallis, OR 97331
USA
Han-Erik Anderson
University of Washington
College of Forest Resources
PO Box 352100
Seattle, WA 98195
USA
Philip Araman
USDA Forest Service
1650 Ramble Road
Blacksburg, VA 24060
USA
Kazuhiro Aruga
University of Tokyo
Graduate School of Agricultural and
Life Sciences
Laboratory of Forest Engineering
1-1-1 Yayoi Bunkyo-ku
Tokyo, 113-8657
JAPAN
Jamie Barbour
USDA Forest Service
PNW Research Station
1221 S.W. Yamhill, Suite 200
Portland, OR 97205
USA
Bruce Bare
University of Washington
College of Forest Resources
PO Box 352100
Seattle, WA 98195
USA
Pete Bettinger
Oregon State University
Department of Forest Resources
280 Peavy Hall
Corvallis, OR 97331
USA
Kevin Boston
Carter, Holt, Harvey Fibre Solutions
3 Omega Street,
Albany, NEW ZEALAND
David Briggs
University of Washington
College of Forest Resources
PO Box 352100
Seattle, WA 98195
USA
Thomas Burk
University of Minnesota
Department of Forest Resources
1530 Cleveland Avenue North
St. Paul, MN 55108
USA
Ward Carson
University of Washington
College of Forest Resources
PO Box 352100
Seattle, WA 98195
USA
194
Elizabeth Dodson Coulter
Oregon State University
Department of Forest Engineering
215 Peavy Hall
Corvallis, OR 97331
USA
Igor Daniline
Laboratory of Forest Inventory and
Forest Planning
V.N. Sukachev Institute of Forest
Russian Academy of Sciences
Siberian Branch
Akademgorodok
Krasnoyarsk, 660036,
RUSSIA
Peter Farnum
Weyerhaeuser Company
P.O. Box 9777
Federal Way, WA 98063-9777
USA
Seca Gandaseca
Bogor Agricultural University
Bogor 16001
INDONSEIA
Walter Garms
WoodVision
In Vision Technologies, Inc.
7151 Gateway Blvd.
Newark, CA 94560
USA
Joel Gillet
Applanix Corporation
85 Leek Crescent
Richmond Hill, Ontario
CANADA , L4B 3B3
François A. Gougeon
Natural Resources Canada
Canadian Forest Service
Pacific Forestry Centre
506 West Burnside Rd.
Victoria, British Columbia
CANADA, V8Z 1M5
Ton. E Grift
Auburn University
214 Tom Corley Building
Biosystems Engineering Department
Auburn , AL 36849-5417
USA
Dave Harrison
Canadian Forest Products Ltd.
2900-1055 Dunsmuir St.
Post Office Box 49420
Bentall Postal Station
Vancouver, BC
CANADA, V7X 1B5
Loren Kellogg
Oregon State University
Department of Forest Engineering
215 Peavy Hall
Corvallis, OR 97331
USA
Bob McGaughey
USDA Forest Service
PNW Research Station
Box 352100
Seattle, WA 98195-2100
USA
Thomas Kerns
University of Minnesota
Department of Forest Resources
1530 Cleveland Avenue North
St. Paul, MN 55108
USA
Evgeny Medvedev
Opten Limited
Semenovskii preulok, 15
Moscow, 105023
Russia
Hiroshi Kobayashi
University of Tokyo
Graduate School of Agricultural and
Life Sciences
Laboratory of Forest Engineering
1-1-1 Yayoi Bunkyo-ku
Tokyo, 113-8657
JAPAN
Donald G. Leckie
Pacific Forestry Centre
506 West Burnside Rd.
Victoria, British Columbia
CANADA, V8Z 1M5
Hisashi Hasegawa
Kyoto University
Graduate School of Information
Kyoto 606-8502
JAPAN
Ted Leininger
USDA Forest Service
P.O. Box 227
Stoneville, MS 38776-0227
USA
Brian Holly
GeoTech Systems
PO Box 3459
Tallahassee, FL 32315
USA
Erik Lithopoulos
Applanix Corporation
85 Leek Crescent
Richmond Hill, Ontario
CANADA , L4B 3B3
Sean Hoyt
University of Washington
Department of Electrical Engineering
Box 352500
Seattle, WA 98195-2500
USA
Juang Rata Matangaran
Bogor Agricultural University
Faculty of Forestry IPB
Kampus IPB Darmaga
PO Box 168 Bogor
Bogor 16001
INDONESIA
Masahiro Iwaoka
Tokyo University of Agriculture and
Technology
Faculty of Agriculture
3-5-8 Saiwai-cho
Fuchu-shi
Tokyo 183-8509
JAPAN
Timothy McDonald
USDA Forest Service
Southern Research Station
520 Devall Drive
Auburn, AL 36849-5418
USA
195
Mark Milligan
LandMark Systems
P.O. Box 3459
Tallahassee, FL 32315
USA
Koji Nakamura
University of Tokyo
Graduate School of Agricultural and Life
Sciences
Laboratory of Forest Engineering
1-1-1 Yayoi Bunkyo-ku
Tokyo, 113-8657
JAPAN
John Nelson
Faculty of Forestry
University of British Columbia
2424 Main Mall
Vancouver, BC
CANADA, V6T 1Z4
Toshio Nitami
University of Tokyo
Graduate School of Agricultural and
Life Sciences
Laboratory of Forest Engineering
1-1-1 Yayoi Bunkyo-ku
Tokyo, 113-8657
JAPAN
Luis G. Occeña
Dept. of Industrial and Manufacturing
Systems Engineering
E3437 Engineering Bldg. East,
University of Missouri-Columbia,
Columbia, MO 62211
USA
Francis Pierce
Washington State University
Center for Precision Agricultural Systems
24106 North Bunn Road
Prosser, WA 99350-8694
USA
Timothy J. Rayner
WoodVision
InVision Technologies, Inc.
7151 Gateway Blvd
Newark, CA 94560
USA
Michael Renslow
Spencer B. Gross, Inc.
13545 NW
Science Park Drive
Portland, OR 97229
USA
Steve Reutebuch
USDA Forest Service
PNW Research Station
Box 352100
Seattle, WA 98195-2100
USA
Robert B. Rummer
USDA Forest Service
Forest Operations Research Unit
520 Devall Drive
Auburn , AL 36849-5418
USA
Hideo Sakai
The University of Tokyo
Graduate School of Agricultural and
Life Sciences
Tokyo University Forest in Chichibu
Chichibu-shi, Saitama 368-0034
JAPAN
Rin Sakurai
University of Tokyo
Graduate School of Agricultural and
Life Sciences
Laboratory of Forest Engineering
1-1-1 Yayoi Bunkyo-ku
Tokyo, 113-8657
JAPAN
Gerard Schreuder
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
John Sessions
Oregon State University
Department of Forest Engineering
213 Peavy Hall
Corvallis, OR 97331
USA
Bruno M. Scherzinger
Applanix Corporation
85 Leek Crescent
Richmond Hill, Ontario
CANADA , L4B 3B3
Daniel L. Schmoldt
USDA Forest Service
Southern Research Station
University of Wisconsin
460 Henry Mall
Madison, WI 53706
USA
Jack Sjostrom
University of Idaho
510 Clearwater Loop,
Suite 2
Post Falls, ID 83854
USA
Shahab Sokhansanj
Oak Ridge National Laboratory
Environmental Sciences Division
Oak Ridge, TN 37831-6422
USA
Johan Stendahl
Swedish University of Agricultural
Sciences
Department of Forest Soils
PO Box 7001
SE-75007 Uppsala
SWEDEN
Bryce Stokes
USDA Forest Service
VMPR, RPC - 4
PO Box 96090
Washington, DC 20090-6090
USA
Doug St. John
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
T. Sweda
Forest Resources Planning, Ehime
University, 3-5-7 Tarumi,
Matsuyama, 790-8566
JAPAN
Frank H. Tainter
Clemson University
Department of Forest Resources
Clemson, SC 29634
USA
196
Steven Taylor
Auburn University
Biosystems Engineering Department
214 Tom Corley Building
Auburn , AL 36849--5417
USA
Jorge Valenzuela
Auburn University
Industrial Engineering Department
Auburn , AL 36849
USA
Matthew W. Veal
Auburn University
Biosystems Engineering Department
214 Tom Corley Building
Auburn , AL 36849-5417
USA
Dale Weyermann
USDA Forest Service
PNW Northwest Research Station
333 SW First Avenue
Portland, OR 97208-3890
USA
Denise Wilson
University of Washington
Department of Electrical Engineering
Box 352500
Seattle, WA 98195-2500
USA
Michael Wing
Oregon State University
Department of Forest Engineering
215 Peavy Hall
Corvallis, OR 97331
USA
Tetsuhiko Yoshimura
Kyoto University
Graduate School of Information
Kyoto 606-8501
JAPAN
Safiah Yusmah M. Y.
Natural Forest Division
Forest Research Institute Malaysia
(FRIM)
Kepong, 52109 Kuala Lumpur
MALAYSIA
LIST OF ATTENDEES
Sencer Alkan
Forintek Canada Corporation
2665 East Mall
Vancouver, BC V6T 1W5
CANADA
Craig Campbell
Boise Cascade
8057 W. Scardale Ct.
Boise, ID 83704
USA
Burton Dial
The Forest Technology Group
125 Crosscreek Drive
Summerville, SC 29485
USA
Phil Allen
Air Logistics
P.O. box 302072
North Harbor, Auckland 1330
NEW ZEALAND
Phil Cook
Weyerhaeuser
1701 First Street
Cosmopolis, WA 98537
USA
Stephen Aulerich
Forest Engineering, Inc.
620 SW 4th Street
Corvallis, OR 97333
USA
Luis Gustavo Costa
International Paper
Rodovia SP 340 Km 171
Mogi Guacu
Sao Paulo 13840-970
Brazil
Jo Diamond
University of Washington
Department of Electrical Engineering
Box 352500
Seattle, WA 98195-2500
USA
Brian Boswell
Forest Engineering Research
Institute of Canada
2601 East Mall
Vancouver, BC V6T 1Z4
CANADA
Wade Boyd
Longview Fibre Company
Box 667
Longview, WA 98632
USA
Dennis Brandt
Weyerhaeuser Company
31002 Chinook Pass HWY
Enumclaw, WA 98022
USA
Anne Briggs
P.O. Box 663
Issaquah, WA 98027
USA
Angus Brodie
Washington State Department
of Natural Resources
1111 Washington Street SE
P.O. Box 47016
Olympia, WA 98504-7019
USA
Betty Cowan
32733 - 30th Ave. SW
Federal Way, WA 98023
USA
Andrew Cowan
32733 - 30th Ave. SW
Federal Way, WA 98023
USA
David Crooker
Plum Creek Timber Company
999 - 3rd Ave
Suite 2300
Seattle, WA 98104
USA
Steve Curry
Washington State Department of
Natural Resources
1111 Washington Street SE
P.O. Box 47016
Olympia, WA 98504-7016
USA
Florentiu Damian
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
197
Steve Dowman
Rayonier New Zeland
Level 5
49 symonds Street
Auckland
NEW ZEALAND
Kelley Duffield
University of Washington
College of Forest Resouces
Box 352100
Seattle, WA 98195-2100
USA
Stephen E. Fairweather
Mason, Bruce & Girard, Inc.
707 S.W. Washington Street
Suite 1300
Portland, OR 97205
USA
Jeff Foster
US Dept. of the Army
Forestry Public Works
AFZH-PWE
PO Box 339500 ms17
Fort Lewis, WA 8433-9500
USA
Jim Fridley
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Fred Friesz
Confederated Salish & Kootenai
Tribes Forestry
104 Main Street SE
Ronan, MT 59864-2803
USA
Chih-Lin Huang
Weyerhaeuser Company
WTC 2B2
Box 9777
Federal Way, WA 98063-9777
USA
Dave Furtwangler
Cascade Timber Consulting
P.O. Box 446
Sweet Home, OR 97386
USA
Rick Hulce
Confederated Salish & Kootenai
Tribes Forestry
104 Main Street SE
Ronan, MT 59864-2803
USA
David Galt
Seattle Public Utilities
19901 Cedar Falls Rd. SE
North Bend, WA 98045
USA
David Gilluly
Weyerhauser
WWC ZEZ
33405 8th Ave S
Federal Way, WA 98003
USA
Robert W. Grenley
IDmicro, Inc.
1019 Pacific Ave
13th Floor
Tacoma, WA 98402
USA
Richard Grotefendt
University of Washington
P.O. Box 1794
North Bend, WA 98045
USA
Patrick Higgins
Canadian Consulate General
412 Plaza 600
Sixth and Stewart
Seattle, WA 98101
USA
Olav Hoibo
Agricultural University of Norway
Department of Forest Sciences
Box 5044
AAS 1432
NORWAY
Robert Hutcheson
Mason, Bruce & Girard, Inc.
1615 Continental Street
#100
Redding, CA 96001
USA
Finn Krogstad
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Bruce Larson
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 9 8195-2100
USA
Edwin Lewis
Bureau of Indian Affairs
P.O. Box 632
Toppenish, WA 98948-0632
USA
Bruce Lippke
University of Washington
Rural Technology Institute
Box 352100
Seattle, WA 98195-2100
USA
Zhenkui Ma Ma
Weyerhaeuser Company
WWC-2E2
P.O. box 9777
Federal Way, WA 98063-9777
USA
198
Andy Malmquist
Forest Technology Group
601 Clarkline Road
Paducah, KY 42003
USA
James McCarter
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Robert McCormack
CSIRO
P.O. Box E4008
Kingston
AUSTRALIA
Rex McCullough
Weyerhaeuser Company
WTC 1A5
PO Box 9777
Federal Way, WA 98063-9777
USA
Gerald Middel
Duck Creek Associates, Inc.
301 SW 4th Street
Suite 270
Corvallis, OR 97333
USA
Esko Mikkonen
University of Helsinki
P.O. Box 27
University of Helsinki
Helsinki, F1 - 00014
Finland
Duncan Munroe
Seattle Public Utilities
19901 Cedar Falls Rd. SE
Northbend, WA 98045
USA
Chris Nelson
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Walt Obermeyer
Washington State Department of
Natural Resources
1111 Washington Street SE
P.O. Box 47016
Olympia, WA 98504-7019
USA
Kelly O’Brian
University of Washington
Box 352100
Seattle, WA 98195-2100
USA
Hirotaka Okabayashi
University of Washington
2238 70th SE
Mercer Island, WA 98040
USA
Chad Oliver
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Elaine Oneil
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Megan O’Shea
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Pam Overhulser
Oregon Department of Forestry
Salem, OR 97310
USA
Robert Parker
Oregon State University
2610 Grove Street
Baker City, OR 97814
USA
Charles Petersen
USDA Forest Service
PNW Research Station
1221 SW Yamhill, Suite 200
Portland, OR 97205
USA
Mike Phelps
Haglof Inc.
P.O. Box 2548
Madison, MS 39130
USA
John St. Pierre
Colville Confederated Tribes
P.O. Box 72
Nespelem, WA 99155
USA
Lester Power
Weyerhaeuser Company
P.O. Box 977
WWC 2E2
Federal Way, WA 98063
USA
Pierre Turcotte
Forest Engineering Research
Institute of Canada
580 Boul. St. - Jean
Pointe-Claire, Quebec H9R 3J9
CANADA
Joannes Ressmann
University of Freiburg
Institut fur Forstbenutzung und
Forstiche Arbeits
Werderring 6 Freiburg
Baden-Wurttemberg 79085
GERMANY
Eric Turnblom
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Don Robinson
Northwest Timber Technology
704 - 228th Ave. NE
Suite 102
Redmond, WA 98024
USA
G.H. Weaver
Louisiana Tech University
School of Forestry
P.O. Box 10138
Ruston, LA 71272
USA
Luke Rogers
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
William Wheeler
Geotiva, Inc
P.O. Box 1792
Poulsbo, WA 98370
USA
Patrick Rusher
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195-2100
USA
Andrew Wilson
Rayonier
3033 Ingram
PO Box 200
Hoquiam, WA 98550
USA
Alastair Sarre
International Tropical Timber
Organization
USA
Chrissy Scannell
University of Washington
College of Forest Resources
Box 352100
Seattle, WA 98195
USA
George Sereno
IDmicro, Inc.
1019 Pacific Ave.
13th Floor
Tacoma, WA 98402
USA
199
Safiah Yusmah
Forest Research Institute of
Malaysia Natural Forests Division
Kepong 52109
Kuala Lumpur
MALAYSIA
First International Precision Forestry Symposium
Agenda
Monday, June 18, 2001
Bruce Bare, Dean, University of Washington College of Forest Resources
Introductory Remarks
Peter Farnum, Weyerhaeuser Company
Keynote Address:
Weyerhaeuser Industry Perspective on the Potentials of Precision Forestry
Francis Pierce, Director, WSU Precision Agriculture Cooperative
Keynote Address:
Developments and Benefits in Precision Agriculture
Plenary Session A: Remote Sensing of Forest Land & Vegetation
Anderson, Reutebuch and Schreuder
Use of Automated Individual Tree Crown Recognition and Measurement Algorithms in Forest Inventories
Sakai
Forest Investigation Using High-Resolution Photographs
Carson, Weyermann and Akay
An Inventory of Juniper Through the Automated Interpretation of Aerial Digital Imagery
Stendahl
Spatial Variability Within Managed Forest Stands
Gougeon and Leckie
Individual Tree Crown Image Analysis - A Step Toward Precision Forestry
Burk, Kerns and Bauer
Taking GIS/Remote Sensing into the Field
Milligan
RTI - Real Time Inventory: A new Approach to an Old Problem
Plenary Session B: Sensing, Measuring & Tagging Trees
Wilson, Hoyt and St. John
Passive Sensing for Precision Forestry Applications
Occena, Schmoldt and Abbott
Cooperative Use of Advanced Scanning Technology for Low-Volume Wood Processors
Clark
Applications of an Automated Stem Measurer for Precision Forestry
Leininger, Schmoldt and Tainter
Using Ultrasound to Detect Defects in Trees: Current Knowledge and Future Needs
Rayner, Garms and Scheinman
The Development of an Internet-based Log Marketing Tool
200
Tuesday, June 19, 2001
Gero Becker, University of Freiburg
Keynote Address: Precision Forestry in Central
Europe: New Perspectives for a Classical Management Concept
Plenary Session C: Machinery, Monitoring, Road Layout
Gandaseca
Evaluating the Performance of GPS Surveying Under Different Forest Conditions in Japan
Coulter, Chung, Akay and Sessions
Optimization of Forest Road Layout Using a High Resolution Digital Terrain Model Generated from
LIDAR Data
Gillet
Tightly coupled inertial/GPS system for Precision Forestry Surveys Under canopy: test results
Holley
Real Time Harvester: The Future of Logging
Aruga, Matangaran, Sakurai, Nitami, Sakai and Kobayashi
Vehicle Management System for Forest Environmental Conservation
Wing and Kellogg
Using a Laser Range Finder to Assist Harvest Planning
Taylor, McDonald, Veal and Grift
Using GPS to Evaluate Productivity and Performance of Forest Machine Systems
Plenary Session D: Decision Support Systems
Nelson and Harrison
Global Demands Drive Advances in Data Management and Hierarchical Decision Support Systems in
Northern British Columbia
Salam and Sasaki
The Role of Precision Forestry: Case Study for Timber Harvesting Planning in Peninsular Malaysia
Boston
Precision Log Making for Plantation Operations
Araman, Baumgras, Schmoldt and Clark
A Look to Future Precision Tree Length Stem Analysis and Processing
McGaughey
Data Requirements for Precision Visualization of Forest Operations
McDonald and Rummer
Information Needs for Increasing Log Transportation Efficiency
Sessions and Bettinger
Heirarchical Planning: Pathway to the Future?
Wednesday, June 20, 2001
Vans depart from UW College of Forest Resources for Capitol State Forest field tour
201