AN ABSTRACT OF THE THESIS OF

advertisement
AN ABSTRACT OF THE THESIS OF
Timothy P. Drake for the degree of Master of Science in Forest Resources presented
on September 3, 2008.
Title: Empirical Modeling of Windthrow Occurrence in Streamside Buffer Strips.
Abstract approved:
_____________________________________________________________________
Temesgen Hailemariam
Streamside buffer strips provide numerous benefits to stream ecosystems. The
buffer strips create shade, provide shelter for wildlife, act as barriers to logging debris
during and after timber harvest, and serve as a continued source of large woody debris.
Quantifying woody inputs resulting from windthrow provides managers with
estimates for long-term planning and habitat development strategies. A two step
modeling process was used to model stand and landscape level attributes to predict the
presence or absence of windthrow and quantify its occurrence.
A survey of windthrow was conducted on 22 stream reaches in the Coast
Range and western foothills of the Cascade Range of Oregon. Stream reaches were
located on Bureau of Land Management lands in Density Management Study sites.
Using logistic regression, elevation was found to be the most significant predictor of
the presence or absence of windthrow. Linear regression revealed that the mean of
ratios of height to diameter of trees in a stream reach was the most significant
predictor of the number of stems affected by windthrow.
Across all stream reaches windthrow was found to be inversely related to the
distance from the stream channel. Approximately 79% of recorded windthrow events
occurred within the first 20 meters upslope from the stream, while 21% occurred
between 21 and 40 meters upslope distance from the stream. Nearly 40% of the
windthrow events fell in a north to northeast direction indicating that the prevailing
winds played a large role in dictating the direction of fall. Local topography was also
observed to have an impact on the direction of fall. Trees fell toward the stream
channel approximately 25% of the time.
Windthrow did not appear to be a significant source of mortality. In the
majority of stream reaches, windthrow amounted to 1% or less of standing live basal
area over the 4 yr. period. The greatest observed loss was 5.2%. In terms of the
number of stems available to blow down, the greatest observed loss was 4%.
© Copyright by Timothy P. Drake
September 3, 2008
All Rights Reserved
Empirical Modeling of Windthrow Occurrence in Streamside Buffer Strips
by
Timothy P. Drake
A THESIS
submitted to
Oregon State University
in partial fulfillment of
the requirements for the
degree of
Master of Science
Presented September 3, 2008
Commencement June 2009
Master of Science thesis of Timothy P. Drake presented on September 3, 2008.
APPROVED:
_____________________________________________________________________
Major Professor, representing Forest Resources
_____________________________________________________________________
Head of the Department of Forest Engineering, Resources, and Management
_____________________________________________________________________
Dean of the Graduate School
I understand that my thesis will become part of the permanent collection of Oregon
State University libraries. My signature below authorizes release of my thesis to any
reader upon request.
_____________________________________________________________________
Timothy P. Drake, Author
ACKNOWLEDGEMENTS
The author expresses sincere appreciation to all those who help in this process
along the way. First, thanks go to Temesgen Hailemariam for his encouragement and
guidance as a mentor and professor; to Stephanie Larew for her assistance with data
collection; and to Bianca Eskelson for her assistance with statistical analysis and
guidance in the writing process. Also, a big thank you goes to Anna Conroy whose
love and support through the entire program has been nothing short of amazing. In
addition, thanks go to the graduate committee (Fred Swanson, Steve Fitzgerald, and
Kate Lajtha) and the numerous people with the BLM and Forest Service, including
Paul Anderson, who gave input and helped to organize the data collection phase of the
research.
TABLE OF CONTENTS
Page
Chapter 1 – Introduction ................................................................................................ 1
Chapter 2 – Literature Review ....................................................................................... 4
The Riparian Forest................................................................................................ 4
Structure and Function of Down Wood ................................................................. 7
Wind Damage in Forests...................................................................................... 11
Windthrow Models............................................................................................... 14
Chapter 3 – Methods and Analysis .............................................................................. 23
Site Description.................................................................................................... 23
Site Area....................................................................................................... 23
Topography .................................................................................................. 25
Geology and Soils ........................................................................................ 25
Climate ......................................................................................................... 25
Forest Type................................................................................................... 26
Data Collection..................................................................................................... 27
Timeline ....................................................................................................... 27
Site Selection................................................................................................ 27
Study Design ................................................................................................ 28
Description of Measurements....................................................................... 32
Compilation of Data ............................................................................................. 34
Data Analysis ....................................................................................................... 38
Step I: Logistic Regression .......................................................................... 38
Step II: Linear Regression............................................................................ 42
TABLE OF CONTENTS (Continued)
Page
Chapter 4 – Results and Discussion............................................................................. 44
Preliminary Analysis............................................................................................ 44
Research Question 1: windthrow and distance from stream ................................ 45
Research Question 2: windthrow and buffer width.............................................. 50
Research Question 3: windthrow and thinning treatment of neighboring stands 53
Research Question 4: windthrow and species composition ................................. 54
Step I: Logistic Regression .................................................................................. 54
Step II: Linear Regression.................................................................................... 60
One Step Approach ...................................................................................... 69
Additional Findings.............................................................................................. 70
Chapter 5 – Conclusions .............................................................................................. 74
Summary of Findings ........................................................................................... 74
Recommendations ................................................................................................ 76
Bibliography ................................................................................................................ 78
Appendices................................................................................................................... 86
Appendix A. Stream reach locations.................................................................... 86
Appendix B. Stocking density for stream reaches (Tpha) ................................... 87
Appendix C. Basal area of stream reaches (m2/ha) .............................................. 88
Appendix D. Potential explanatory variables....................................................... 89
Appendix E. Windthrow by direction of fall. ...................................................... 92
LIST OF FIGURES
Figure
Page
Figure 1. Density Management Study locations .......................................................... 24
Figure 2. Measurement plot design with a randomly assigned subplot ....................... 31
Figure 3. Measurement plot design without a randomly assigned subplot .................. 31
Figure 4. Frequency of windthrow by slope distance from stream, using 5 m swaths,
for all windthrow........................................................................................... 46
Figure 5. Frequency of windthrow by slope distance from stream, using 10 m swaths,
for softwoods................................................................................................. 47
Figure 6. Frequency of windthrow by slope distance from stream, using 10 m swaths,
for hardwoods ............................................................................................... 47
Figure 7. Dbh by distance from stream for all windthrow........................................... 49
Figure 8. Predicted probability of windthrow by elevation ......................................... 57
Figure 9. Scatter plot of down trees per hectare by height to diameter ratio ............... 63
Figure 10. Scatter plot of residual versus fitted values for height to diameter ratio
regressed on down trees per hectare............................................................ 66
Figure 11. Influence statistics for all observations of height to diameter ratio regressed
on down trees per hectare............................................................................ 67
Figure 13. Compass rose showing direction of fall for all observed windthrow ......... 71
LIST OF TABLES
Table
Page
Table 1. Stream reaches measured by thinning treatment and buffer width................ 29
Table 2. Variables used to describe stream reach attributes ........................................ 35
Table 3. Factorial treatment analysis showing presence of interaction ....................... 51
Table 4. Simple effect contrasts for thinning treatment and buffer width combinations
...................................................................................................................................... 51
Table 5. Direction of tree fall in relation to stream valley ........................................... 73
CHAPTER 1 – INTRODUCTION
Wind is a natural disturbance agent whose processes often impede or
complicate forest management plans. It is a constant source of tree mortality at the
stand and landscape level and, at times, occurs as large disturbance events. Natural
disturbances such as wind cannot be controlled; however, their impacts can be
mitigated. Much research has been conducted on windthrow in relation to edges of
harvest units (Ruth and Yoder, 1953; Gratkowski, 1956; Steinblums et al., 1984);
however, little has been done in thinned stands to provide estimates of small scale
windthrow, as opposed to large events. Knowledge of the factors influencing the
severity of wind damage can greatly assist in anticipating and mitigating tree and stand
damage.
An area of great importance in forest management and ecosystem management
is the riparian forest. Due to its inherent properties (biotic and abiotic) and
silvicultural practices, the riparian forest often experiences wind damage, specifically
windthrow. Riparian forests can be highly diverse in their characteristics; at times
varying greatly within a matter of meters. They may vary because of management
objectives, as prescribed by law (as with a buffer), or naturally by their location.
Streamside buffer strips provide numerous benefits to stream ecosystems. The
buffer strips create shade, provide shelter for wildlife, act as barriers to logging debris
during and after timber harvest, and serve as a continuous source of large woody
debris (Bisson et al., 1987; Andrus et al., 1988; Rashin and Graber, 1992; Brooks et
al., 2003). To ensure effective stream protection and to minimize excessive
2
windthrow, reliable and cost effective streamside buffer strips designs are required.
Quantifying windthrow events provides managers with estimates of large woody
debris inputs for long-term planning and habitat development strategies.
The objective of this study is to develop a two-step model to predict the
occurrence of windthrow. The first step will predict the presence or absence of
windthrow in a stream reach using logistic regression. The second step will predict
the number of stems affected by windthrow in a stream reach with linear regression.
Windthrow may be defined in many ways. For the purposes of this study it
will include only those live trees which have been completely toppled by wind. It will
not include broken tops, leaning trees, or snags which have blown over.
Quantifying woody inputs resulting from windthrow provides managers with
estimates for long-term planning and habitat development strategies. Cognizant of
these facts, modeling windthrow occurrence in streamside buffer strips, and evaluation
of the efficiency and suitability of selected streamside management options are
valuable.
Four research questions serve as a framework in order to guide this study.
Streams in the Density Management Study (DMS) on Bureau of Land Management
(BLM) lands serve as the population from which samples have been taken. The four
research questions guiding the study are:
1) Does the amount of windthrow vary with upslope distance from the
stream in headwater streams of the DMS?
3
2) Does buffer width affect the amount of windthrow in headwater
streams of the DMS?
3) Does residual thinning density of the neighboring forest affect the
amount of windthrow in headwater streams of the DMS?
4) Does species composition affect the amount of windthrow in headwater
streams of the DMS?
4
CHAPTER 2 – LITERATURE REVIEW
This study investigates the influence of several biotic and abiotic factors on the
susceptibility of riparian forests to windthrow. The importance of riparian forests as
key components of the forest ecosystem has become increasingly recognized, yet
much remains unknown about the processes by which they are shaped. Although this
study focuses on windthrow in riparian forests, it is helpful to review additional
literature addressing riparian functions as well as wind damage in many different
forest types.
The Riparian Forest
Riparian ecosystems are dynamic and diverse. As such, adequately defining a
riparian ecosystem can be a challenge. They may be found in many settings, including
forestland, rangeland, agricultural land, estuaries, and even urban landscapes. A
riparian community may be defined as the interface between an aquatic and a
terrestrial system; which is made up of unique plant and animal communities that
require the use of free or unbound water. They often have a high diversity of species,
both flora and fauna, and provide numerous ecosystem services such as stream bank
stability, maintaining water quality, erosion control, stream shading, and nutrient input
(Brooks et al., 2003). As an illustration of the diversity of plant life in these
ecosystems, Tabacchi et al. (1990) documented 900 different taxa along the 335 km
length of the Adour River in France. Another study observed, in forest conditions
ranging from clearcuts to old-growth forests, nearly twice as many species in forest
5
riparian areas as in the adjacent upland areas (Gregory et al., 1991). Cole et al. (2003)
noted an unexpected abundance of species richness and diversity in headwater streams
of the Oregon Coast Range. Species which were previously thought to be rare or
location specific were discovered in headwater riparian areas of managed forests.
While much is known about the physical and biological attributes that make up a
riparian zone, our knowledge is far from complete in predicting wind damage and,
therefore, devising management schemes to mitigate such damage.
General characteristics of a riparian zone may be outlined; however, as Ilhardt
et al. (2000) illustrated, it may be described or categorized in many different ways.
The names used by organizations and agencies: “riparian management zone”, “riparian
buffer”, “streamside management zone”, “buffer zone” stem from different
management designations and purposes, yet they are capable of being defined and
measured. Indeed, the definition of a riparian forest differs between organizations or
agencies and also between scientific disciplines. A forester may provide a different
definition than a soil scientist, or a wildlife biologist, or an ecologist.
A basic definition is provided by Hall (1989) in which a riparian ecosystem is
defined as those on or by land bordering a stream, lake, tidewater, or other body of
water. A more comprehensive definition states a riparian zone is the three
dimensional interface between aquatic and terrestrial ecosystems. It may be
characterized by vegetation type, topography, and hydrology, but none of the
components alone provide an adequate definition (Swanson et al., 1982; Gregory et
al., 1991). Naiman et al. (1993) presented the term riparian corridor and described it
6
as the area reaching from the stream channel upland to where vegetation is affected by
high water tables, flooding, and the ability of the soils to retain water. This corridor is
affected by the stream size and adjacent slope characteristics and may be quite small
in the headwater streams located in forested land.
A study conducted in the state of Maine suggests that forest riparian zones may
be defined in part by the presence of amphibians, which are dependent on the interface
between aquatic and terrestrial ecosystems. Their conclusion was that in headwater
streams, a riparian zone is relatively narrow and may be 7 to 9 m in width (Perkins and
Hunter, 2006). A similar study used plant communities to define the riparian zone
(Hagan et al., 2006). They documented the greatest species richness and abundance
within the first 5 m of headwater streams.
Riparian forests may be thought of as a continuum in a similar manner as
rivers; that is, riparian zones are not disjoint ecosystems, rather they are connected.
Upstream processes affect many of the conditions downstream. For example, much of
the mineral content of riparian soils is alluvial. The parent material of riparian soils is
likely an upstream source; whereas, soils in the uplands are weathered in place from
the underlying rock (Bilby, 1989). Another example of this connectedness is the
canopy cover provided by riparian forests. Noss (1987) stressed the importance of
riparian corridors as cover for the migration of animals as well as for retaining pieces
of undisturbed habitat for plants and animals. Perault and Lomolino (2000)
documented the value of riparian forest corridors for habitat connectivity for several
species of mammals. Hilty and Merenlender (2004) observed an increased tendency
7
of mammalian predators to utilize cover provided by riparian corridors in locations
where adjacent land provided unsuitable habitat resulting from land use change.
Stream shading is another important process of the riparian forest. Shading has
been cited as a factor related to stream temperature; although results have not been
consistent. Many studies have documented the inverse relationship between shading
and stream temperature (Rashin and Graber, 1992; Macdonald et al., 2003; Johnson,
2004). Another study found stream temperature to be un-related to the amount of
shading afforded by buffer width (Brosofske et al., 1997), although there was a
significant effect on humidity and microclimate. In some instances, such as steep
valleys and canyons or aspects facing away from the sun, the total solar exposure may
be such that the absence or presence of a riparian forest makes little difference in
stream temperature.
Structure and Function of Down Wood
Down woody debris provides habitat for numerous organisms, structural
diversity, and serves many other functional roles in riparian ecosystems of the forest.
Franklin et al. (1986) stated that large woody debris provides a major structural
component in streams and rivers and generates over half of the habitat found in small
forested streams. Woody debris is a part of all forested ecosystems. As trees die, they
transition, however slowly, from a vertical position among the forest canopy to a
horizontal position among the litter on the forest floor. One estimate of boles and
branches on the forest floor in coniferous forests of western Oregon and Washington
8
ranges between 1.5 to 4.5 megagrams (1,500 to 4,500 kg) per ha per year (Sollins,
1982). Woody debris comes in many sizes from many sources. In reference to woody
debris, it may be helpful to distinguish between size classes. Descriptions vary widely
depending on the scientific discipline, forest type, or other factors; however, Harmon
et al. (1986) proposed that coarse woody debris, also known as large woody debris,
have a minimum diameter in the range of 7.5 to 15 cm. This definition focuses on
woody debris that will persist and be of value as structural components of the forest.
One of the many important functions of woody debris in riparian forests is for
animal cover and habitat. Maser et al. (1984) noted that sloughed bark from woody
debris provides cover for vertebrates such as salamanders, shrews, shrew-moles, and
voles. Associated with rotten wood in the later stages of decay are lungless
salamanders (family-Plethodontidae) including Oregon salamanders, Oregon slender
salamanders, and clouded salamanders. A variety of organisms, both vertebrate and
invertebrate, utilize woody debris for cover and habitat; however, not all woody debris
is created equal. Organisms may have preferences for certain species, size classes, or
other characteristics of down wood. It then becomes important to further describe the
presence of woody debris A study by Swanson et al. (1984) found in a survey of
streams on Prince of Wales Island in southeast Alaska, that coarse material (>10 cm
diameter) in channels accounted for 73 to 93% of woody debris by weight, but only 20
to 53% by surface area. Knowledge of type and size of woody debris will help
determine the uses it will have.
9
Although woody debris provides many functions in the riparian forest, it is
perhaps most important for contributing to habitat by adding to in-stream structure
(Bisson et al., 1987). The quantity of woody debris in a stream channel is a function
of inputs and outputs (Swanson et al., 1976). The large woody debris inputs which
contribute to the habitat found in riparian forests may come from several sources.
Debris may be blown in (windthrow), undercut by banks, or deposited by mass
movements. Debris may land directly in the stream, or it may land near the stream
and slide in due to the steep sidewalls surrounding the stream. All other things being
equal, streams with steeper slopes have a greater potential to receive debris than
streams in flat terrain. It was also noted that most pieces greater than 30 ft in length
did not move after the time of windfall (Swanson et al., 1976).
Another function of woody debris is that it helps to accumulate other woody
debris. This is especially true with large woody debris. It serves as a trap for fine
litter and branches. Large woody debris also helps dissipate stream energy and allows
more complex stream channels to be developed. If debris is removed or not replaced,
a stream may become more channelized (Swanson et al., 1976). The factors affecting
debris outputs or loss are: the ability of the stream to float debris downstream,
decomposition rates, physical abrasion and deterioration, and the probability for debris
torrents.
Bisson et al. (1987) asserted that large woody debris enhances the quality of
fish habitat in all sizes of streams and that the single most important function of large
woody debris in streams is in the formation of rearing pools for salmonid habitat. It
10
can help retain sediment and gravel to form rearing pools as well as provide cover for
adult fish from terrestrial and aerial predators. Piece length is also important for
quality of habitat. Much of the debris from logging is un-merchantable and is short in
length; however, windthrow often produces substantially longer piece lengths.
Pools created by woody debris provide important spawning and rearing habitat
for salmonids. Large organic debris has been shown to exhibit a positive relationship
with the number of parr (1+ yr old salmonid) in a stream (Murphy et al., 1985). The
large organic debris is especially beneficial to the winter survival of salmonids.
Keller and Swanson (1979) asserted that windthrow is a natural process
affecting streams of all sizes, although, it is likely to be a more significant source of
woody debris in small to medium size streams than in large rivers. Lienkaemper and
Swanson (1987) observed that in steep mountain streams where old-growth Douglasfir (Pseudotsuga menziesii) was present that it was not bank undercutting that was
predominantly responsible for input of wood; rather wind was the principal agent for
debris entry. Andrus et al., (1988) noted that much of the large woody debris found in
streams of the Pacific Northwest where old-growth was present was contributed by
old-growth trees. The large piece size found in old-growth forests is of great benefit
to the riparian ecosystem in that it is able to persist for many decades, even a century,
whereas smaller conifers or hardwoods are relatively short lived in the moist
environment.
11
Wind Damage in Forests
Beginning in the early 1900’s foresters in the Pacific Northwest, realizing the
catastrophic nature of wind, began to address wind damage in their management plans
and investigate the causal factors. Early literature provided observational notes about
wind damage; citing diameter distribution, crown structure and tree spacing as
important factors (Smith, 1915). Silvicultural methods also came into question as
factors that contributed to the susceptibility of trees to windthrow. Weidman (1920)
hypothesized that severity of wind damage was largely due to residual thinning
density in certain silvicultural systems such as single selection as was implemented
with ponderosa pine (Pinus ponderosa).
The small winter storms which frequent the Coast Range of Oregon and the
Pacific Northwest as a whole provide ample opportunities to observe windthrow.
These storms often reveal patterns of wind damage; but larger storms provide an even
better opportunity to observe the effects of wind damage and hypothesize about its
causes. On December 4, 1951 a storm along the Oregon coast blew down 3.7 billion
board feet of timber, nearly $50 million in stumpage value at the time (Ruth and
Yoder, 1953). Several important observations and recommendations were made to
guide foresters designing harvest units. Among the key findings were a ranking of
overall resistance to windthrow among tree species, influence of local topography on
wind direction, effects of disease and insects, and thinning as a silvicultural tool to
develop wind firm stands.
12
On October 12, 1962 the most severe wind storm to date (the “Columbus Day
storm”) hit the Coast Range of Oregon and Washington claiming some 11.2 billion
board feet of timber (Orr, 1963). Winds higher than 170 mph were recorded on the
coast, and as high as 160 mph inland before instruments ceased to function. A
massive survey and research effort was undertaken to quantify the damage, and
understand the widespread effects of the storm. The study addressed the potential for
insect damage following storms; noting that shaded windfalls produced Douglas-fir
beetle (Dendroctonus pseudotsugae Hopkins) broods up to 6 times larger than down
trees exposed to sunlight. Thus, the greatest potential for insect damage was in areas
where windfall was several layers thick. The study also addressed species
susceptibility, and salvage potential based on deterioration rates of different tree
species.
Most recently, a wind storm along the northwest Oregon coast on December 2
and 3, 2007 blew down approximately 390 million board feet of timber in Clatsop and
Tillamook Counties. The storm brought both gale force winds with gusts up to 129
mph and heavy rainfall (Forest and Debris Recovery Team, 2007).
Other regions have experienced severe wind damage as well. In May, 1916,
the largest storm within recent years blew down approximately 5% of the standing
timber in the Adirondacks of New York. Among the findings of surveys were notes
about the importance of rooting architecture, differences in damage based on
landscape position, and the increased damage as a result of gaps in thinned stands
(Behre, 1921). A hurricane in New England in 1938 causing several billion board feet
13
of damage led to a hypothesis regarding the significance of the form point (one-third
the distance from the base of the crown to the top of the tree) in the stability of a given
tree (Curtis, 1943). All other things being equal, trees with a higher form point
(smaller crown ratio) were more susceptible to windthrow.
Wind damage is a problem that affects forests across the nation and around
the world. In the period from 1950-1980 it is estimated that 6 percent of timber
harvested in the United Kingdom was as a result of catastrophic storms, and another
20 percent was harvested in order to prevent losses from windthrow (Quine and Bell,
1998 quoting Atterson, 1980). Mitchell (1995) stated that wind damage affects
approximately 4% of the provincial annual allowable cut in British Columbia, a
similar amount as that lost to insects or wildfire.
As a result of the countless losses from wind damage, research on wind
behavior, stand dynamics, and tree response has been conducted in many regions of
the world, involving a variety of conditions and species. Early exposure to wind
resulting in increased stem diameter and root stability was documented by Jacobs
(1954) on radiata pine (Pinus radiata) in Australia. Putz et al. (1983) observed wind
damage on hardwoods in Panama and cited wood strength, or modulus of rupture
(MOR) and modulus of elasticity (MOE), as important contributing factors. Trees
with higher MOEs (less flexible), tended to uproot; whereas trees with lower MOEs
(more flexible) tended to break. Lohmander and Helles (1987) observed softwoods in
Denmark and related damage to several variables including height, diameter, stand
age, and species composition.
14
Wind damage is an issue facing forest managers across the region and around
the world. It has been a challenge and a problem as long as we have managed forests
for their resources. Understanding the ways in which it affects individual trees and
stands will help managers to reduce the risk of wind damage through careful planning.
Windthrow Models
Windthrow models have been constructed in a wide range of forest settings,
with a wide range of motivations. Much of the early work was focused on the fiscal
repercussions of wind damage in standing timber (Ruth and Yoder, 1953; Orr, 1963).
Furthermore, much of the early literature is composed of observational studies; and
identifies the underlying factors related to windthrow susceptibility rather than
predicting its occurrence (Smith, 1915; Weidman, 1920). Wind was seen as an agent
of damage and its consequences were to be avoided. Later, as ecological objective
became a component of overall forest management goals, the process of wind as a
natural disturbance agent was recognized as was its potential for contribution to forest
structure and nutrient cycling (Franklin et al., 2002). The subject literature is far
broader than that included in this thesis. One of the main areas of focus is on the
underlying processes that are explained through the modeling process rather than
through observational study.
Wind damage does not always occur in a predictable manner. It is a complex
process that varies from location to location. Quine (1995) argued that distinction of
severity should be made when classifying or describing wind damage. Extreme storm
15
events often override site differences, and do not follow the patterns or relationships
observed on a year to year basis. The small wind disturbance events characterized by
a single tree or small cluster of trees are the focus of this thesis. The underlying
processes contributing to this small scale wind damage may be divided into three main
levels: tree-level, stand-level, and landscape-level.
At the tree-level, tree form, or more specifically height (H), diameter (D), and
the ratio of height to diameter (H/D), are key factors. Many models have included
these variables as useful predictors of damage and describe the relationship they have
to windthrow susceptibility (Putz et al., 1983; McDade et al., 1990; VanSickle and
Gregory, 1990; Valinger and Fridman, 1999; Jalkanen and Matilla, 2000; Lekes and
Dandul, 2000; Canham et al., 2001; Peterson, 2004; and Scott and Mitchell, 2005).
Lohmander and Helles (1987) found improved results when D2 and H2 were included.
Based on the studies listed above, general consensus is that taller trees and
consequently trees of larger diameter have been found to be more susceptible to wind
damage than their smaller counterparts in a stand. Although large and small trees in a
stand can have the same value of H/D, this variable holds a great potential for
explanatory power as it incorporates the relationship between height and diameter.
The H/D variable has an added level of importance in that it is often a readily
available stand attribute from simple inventory data.
In general, as H/D increases, so also does the susceptibility to windthrow.
Cremer et al. (1982) observed that H/D is better at predicting stem failure than
16
uprooting; and that H/D is a better measure of susceptibility than H/D2, H3/2/D, or
H/D3.
Other ways to consider H/D are at the stand level. Lekes and Dandul (2000)
utilized average stand height and diameter to provide a stand level attribute. The
average height of the 100 largest trees (largest diameter) per hectare (or 40 largest
trees per acre) could also provide a stand level attribute. The ratios H 100 /QMD
(Quadratic Mean Diameter) or H 100 /D 100 , may provide better variables for stand level
analysis since the largest trees will often be the most exposed trees 1 . While H/D is a
useful predictor, the position in the canopy is also important. A tree in the understory,
for example, may have a higher H/D than the overstory trees; however, shelter
provided by the overstory may be more important than simply the H/D. In his
observations of storm damage, Peterson (2004) noted that damage is better predicted
by relative canopy position than by diameter, because canopy position (e.g. crown
class) often determines the exposure to wind for an individual tree.
Various measures of crown attributes have been used as explanatory variables.
Crown length (Valinger and Fridman, 1999), crown ratio (Valinger and Fridman, 1999
and Scott and Mitchell, 2005), height to crown base (Valinger and Fridman, 1999 and
Scott and Mitchell, 2005), and crown density (Scott and Mitchell, 2005) have been
employed. The various measures of crown attributes while all significant in their
respective model formulations, did not exhibit an easily generalized definitive positive
or negative relationship to the occurrence of windthrow.
1
Personal correspondence with David Hann, Professor, Department of Forest Engineering, Resources,
and Management Oregon State University.
17
The distance of a given tree from the stream channel has also been commonly
used as a tree-level explanatory variable (McDade et al., 1990; VanSickle and
Gregory, 1990; Bragg et al., 2000). Although this is not used specifically as a measure
of susceptibility to windthrow, but rather as a measure of likelihood of hitting the
stream, there is an established relationship between the quantity of windthrow and the
distance from the stream channel. The closer a tree is to the stream channel, the
higher the probability its angle of fall will place the tree into the stream channel,
though this is also dependent on the height of the trees near the stream. McDade et al.
(1990) observed that over 70% of large woody debris in streams originated from
within 20 m of the stream channel. Martin and Grotefendt (2007) observed even
higher percentages. They found that 81% of large woody debris originated within 10
m and 95% originated within 20 m of the stream.
Physiological tree-level variables were used by Putz et al. (1983) to determine
whether a tree will suffer damage in the form of snapping or uprooting. They found
that the wood strength (MOR) was an important predictor. Tree species with stronger
stems tended to uproot while trees with weaker stems tended to snap. They also
pointed to crown and root architecture as important factors related to wind resistance
and soil anchoring respectively. Other studies (Byrne and Mitchell, 2007; Elie and
Ruel, 2005) discovered that stem mass (as measured by destructive sampling) was the
best predictor of overturning resistance regardless of species or stem form.
At the stand-level, one of the most important attributes used to explain or
predict windthrow is species composition. Different tree species have differing stem
18
form, root architecture and crown architecture among other attributes. The
composition or mix of species in a forest can greatly affect the wind damage incurred
at the stand-level. At the tree level, species may be used as an indicator variable
(Canham et al., 2001); while at the stand level, it may be included as the percentage of
a particular highly susceptible species (Valinger and Fridman, 1999; Lekes and
Dandul, 2000). Regionally, individual species that are most susceptible to windthrow
can be determined and, in turn, stands with a higher percentage of susceptible species
tend to be more susceptible at the stand-level (Valinger and Fridman, 1999; Lekes and
Dandul, 2000). In the Pacific Northwest, western redcedar (Thuja plicata) and
Douglas-fir have been ranked among the most stable species, while western hemlock
(Tsuga heterophylla) has consistently been one of the least windfirm species with
intermediate species varying depending on topographic setting and geographic
location (Ruth and Yoder, 1953; McLintock, 1954; Gratkowski, 1956; Scott and
Beasley, 2001; Scott and Mitchell, 2005; Martin and Grotefendt, 2007). Lohmander
and Helles (1987) illustrated how the probability of windthrow as a function of tree
height varied among species. They found true fir and Douglas-fir to be more windfirm
than spruce. Ruth and Yoder (1953) also noted that stands of mixed species tended to
be more wind firm than single species stands, whereas Scott and Beasley (2001)
witnessed in British Columbia that windthrow decreased as the proportion of the stand
dominated by western redcedar increased.
The age of a stand has been included for use as an explanatory variable by
Lohmander and Helles (1987), Evans et al. (2007), Jalkanen and Matilla (2000), and
19
Lekes and Dandul (2000). All noted a positive relationship between stand age and
vulnerability to wind damage. This relationship is undoubtedly influenced by both
stand height and density. The relationship may remain positive over the age range of
most managed stands and then may become negative after a certain point. No
transformations of the age variable were encountered in the literature.
Stand height has been used by Lekes and Dandul (2000), and Lanquaye-Opoku
and Mitchell (2005). Similar to the tree-level attribute, taller stand heights often
indicate greater susceptibility to wind damage; however, this variable is confounded
by other factors such as topographic setting, species, and stand density.
Stand density, as measured by trees per hectare, has proven to be one of the
more powerful explanatory variables. Many have included this variable in models
(VanSickle and Gregory, 1990; Valinger and Fridman, 1999; Lekes and Dandul, 2000;
Mitchell et al., 2001), and had similar results. Where stand densities are such that
height to diameter ratios are high, wind damage is also high. Stand density serves as
an important variable because of its effect on many tree-level attributes such as the
crown architecture, stem mass, and height to diameter ratio. The use of a stand
density index, such as Reineke’s Stand Density Index (Reineke, 1933), as an
explanatory variable was not encountered in the literature, although its application
may prove quite valuable as it incorporates additional information about the
competitive status of trees and their overall health. Ruel et al. (2001) observed that in
thinned riparian buffer strips that the effect of topography was greater than the effect
of thinning. It is generally accepted that heavy thinning will temporarily reduce the
20
stability of a stand; however, this depends on the topographic setting. The result may
be more significant, such as on ridge tops, or less significant, such as in sheltered
valley bottoms.
The time since harvest can also be a valuable piece of information.
Observations show that the rate of windfall in cut block edges decreases as time since
harvest increases (Mitchell et al., 2001; Lanquaye-Opoku and Mitchell, 2005; Scott
and Mitchell, 2005). Immediately following harvest, trees are increasingly exposed to
wind and often succumb to windthrow. As time passes, the surviving trees become
more wind-firm and the rate of windthrow declines. Where riparian buffers exist,
Martin and Grotefendt (2007) concluded that, given sufficient time and similar
characteristics, wind damage will be similar among riparian stands; losses only in the
first few years will differ. They also found that windthrow may not be the dominant
cause of mortality in riparian areas over the long term. Bank erosion, mass wasting,
and other mortality agents were found to have a larger contribution to large woody
debris recruitment in streams.
The previous factors at the tree- and stand-level may be influenced through
careful planning and stand management activities. Lekes and Dandul (2000) have
divided factors into temporary and permanent factors, or those factors which are
subject to human control (those mentioned above) and those beyond human control
such as soil characteristics, terrain, and climatic factors. They found that soils with
lower nutrient and water content tended to be more stable and, consequently trees
were more resistant to uprooting and breakage. It is reasonable to assume that the
21
slower growth rates associated with poor site conditions result in shorter stand heights
as well as allow trees to slowly become adapted to the conditions. Mitchell et al.
(2001) observed that damage was more frequent on moist sites than dry sites and more
frequent in mineral soil than organic soil. Others, including Lanquaye-Opoku and
Mitchell (2005) and Scott and Mitchell (2005), have observed similar results.
Lohmander and Helles (1987) observed a negative relationship between windthrow
and soil drainage; that is to say, the probability of windthrow increased as soil
drainage decreased. Sinton et al. (2000) observed shallow soils as a contributing
factor to windthrow events.
Terrain also plays an important role in the susceptibility of trees to wind
damage. While terrain can influence physiological attributes of trees, it can, perhaps,
also influence wind patterns. Measures of exposure to wind include elevation or
aspect, but a more comprehensive metric is the TOPographic EXposure (TOPEX)
index. This index measures the sum of the angles to the skyline in the 8 major
directions of the compass. This measure has been in use for several decades; however,
it has increased in popularity with the advent of geographic information system (GIS)
technology. Those sites which are more exposed tend to have higher rates of wind
damage (Valinger and Fridman, 1999; Sinton et al., 2000; Lanquaye-Opoku and
Mitchell, 2005; Mitchell et al., 2001; Scott and Mitchell, 2005; Evans et al., 2007).
Scott and Mitchell (2005) alluded to refining TOPEX by limiting the index to the
direction of prevailing winds. This transformation has potential to provide a more
meaningful index of exposure to damaging winds.
22
The TOPEX index is a useful metric on the landscape level, but does not
perform well at a fine scale, as with individual stands, as it inadequately accounts for
the complexities of local terrain (Ruel et al., 1997; Mitchell et al., 2001). While there
is a positive relationship between exposure and wind damage in general, Mitchell et al.
(2001) observed that moderately exposed sites experienced more damage than severe
or sheltered sites. Sheltered sites are seldom exposed to strong winds necessary to
cause wind damage, thus relatively little damage would occur. On severely exposed
sites however, trees are well adapted to harsh conditions. They are often stunted by a
combination of climatic conditions resulting in a very low form point, and
consequently a very windfirm tree.
Related to exposure is the mean annual wind speed. Lanquaye-Opoku and
Mitchell (2005) and Scott and Mitchell (2005) both found good results by including
local climate data. Given that wind is the causal force in windthrow, it seems wise to
include it as an explanatory variable where possible. Other landscape metrics included
latitude and longitude (Valinger and Fridman, 1999), elevation and slope (LanquayeOpoku and Mitchell, 2005 and Scott and Mitchell, 2005), and aspect (Scott and
Mitchell, 2005). Less common explanatory variables include indicators for the harvest
method and subsequent method of regeneration as well as the sum of daily high
temperatures above 5°C (Jalkanen and Matilla, 2000). Lohmander and Helles (1987)
created a variable to characterize the protection of a stand by multiplying the height of
the neighboring stand by the stand density of the neighboring stand by the distance to
the stand edge in the direction of windthrow.
23
CHAPTER 3 – METHODS AND ANALYSIS
Site Description
Site Area
Twenty-two stream reaches located in the Coast Range and western foothills of
the Cascade Range of Oregon were examined to determine the prevalence of
windthrow. The stream reaches were located at DMS locations (Figure 1) on BLM
lands in the Salem, Eugene, and Roseburg districts. One stream reach was located at
both Bottom Line (BL) and Callahan Creek (CC). Two stream reaches were located at
Delph Creek (DC); and three were located at Green Peak (GP). Seven stream reaches
were located at Keel Mt. (KM); three at O.M. Hubbard (OMH); and five at Tenhigh
(TH). Exact locations of stream reaches may be found in Appendix A. The northern
most stream reach was located near the city of Estacada at approximately N
45°15’56’’, W 122°09’33” while the southern most stream reach was near the city of
Sutherlin, at approximately N 43°17’30’’, W 123°35’00”.
24
Figure 1. Density Management Study locations.
25
Topography
Stream reaches in the study area are headwater streams in seven major
drainages: Umpqua River (BL, OMH), Siletz River (CC), Alsea River (GP),
Willamette River (GP), South Santiam River (KM), Clackamas River (DC), and
Siuslaw River (TH). The terrain is generally rugged and characterized by steep
valleys with stream gradients between 10% and 40%. Stream reaches in the study area
range from 300 m to 800 m in elevation.
Geology and Soils
The geology of the study area has been influenced primarily by volcanic
activity and Missoula floods. The Tyee formation, sandstone and siltstone, make up
much of the visible geology in the central and southern portions of the study area in
the Coast Range. To the north, undifferentiated flows and clastic rock, basalt and
andesite, and tuffs are more prevalent. There are 24 different soil series in three soil
orders (Inceptisol, Andisol, and Ultisol) within the study area. For further detail the
interested reader is referred to Cissel et al. (2006) for series names and locations.
Climate
The climate of western Oregon may be characterized by wet winters and dry
summers. Precipitation within the study area ranges from an average of 40 to 60
inches per year in the lower elevation sites near the Willamette Valley up to an
average of 80-100 inches per year at the higher elevation sites in the Coast Range and
26
western foothills of the Cascade Range. Precipitation at the lower elevation valley
sites comes mostly in the form of rain; whereas the higher elevation sites (at the south
end of the study area and in the Cascade foothills) experience much of the
precipitation as snow (National Weather Service, 2006).
Prevailing storm winds along the coast of Oregon come from the south to
southwest. Ruth and Yoder (1953) documented prevailing winds coming from an
average of S 30° W. Inland valleys may experience deviations from the prevailing
direction due to local topography; however, the majority of wind, especially during
storms comes from the south to southwest. In Northwest Oregon, winds also come
from the east as they are funneled through the Columbia Gorge or from the north as
they move down the Willamette Valley (National Weather Service, 2006).
Forest Type
Coniferous dominated forests cover much of western Oregon. Much of the
land at the study sites has had a history of management and, as a result, there is little
old-growth present. Scattered old-growth trees may be found at all sites; however, the
study sites may best be described as mature forests. The majority of stands at the
study sites were naturally regenerated following clearcut harvests, and range from 40
to 60 years in age. Stream reaches, or streamside forestsy, vary slightly in species
composition, although they are all dominated by conifers.
The majority of the DMS sites are similar in forest type and species
composition, and fall under one of four plant association groups (Cissell et al., 2006):
27
1) Tsuga heterophylla/Achlys triphylla-dry, 2) Tsuga heterophylla/Mahonia nervosaOxalis oregano, 3) Tsuga heterophylla/Oxalis oregano, or 4) Tsuga
heterophylla/Vaccinium alaskense-Oxalis oregano. The southern most DMS site
(O.M. Hubbard), differs from the rest. It is closer to the mixed coniferous forests of
southern Oregon and northern California. This one site may be described by the
following three plant association groups: 1) Abies grandis/Toxicodendron
diversilobum, 2) Abies grandis/Mahonia nervosa-Gaultheria shallon, or 3)
Pseudotsuga menziesii/Holodiscus discolor-Whipplea modesta.
Data Collection
Timeline
Data were collected for twenty-two stream reaches during the summers of
2006 and 2007. In 2006, nine BLM DMS stream reaches were located and surveyed.
In 2007, the original nine stream reaches were re-visited to observe changes and take
additional measurements, and an additional thirteen stream reaches were surveyed.
Site Selection
The selection method for the stream reaches involved two stages. First, there
was a subjective selection of the stream, and second, a random location of the stream
reach. In the first stage, the streams were subjectively selected from all streams in all
DMS locations. Entire streams or large segments of streams were selected to
28
represent a variety of buffer widths, residual thinning treatments, slopes, and
geographic location. In the second stage, following the selection of a stream or stream
segment, the length of the stream was determined using a geographic information
system (GIS).
From the total length, 200 m was subtracted and a random number was
generated between 0 and the resultant number. For example, suppose stream ‘A’
measures 461 m in length. Given that 461-200 = 261, a random number was
generated between 0 and 261. The random number then served as the distance from
the beginning of the stream reach to the start of the stream reach. The stream reach
then extended for 200 m (topography and treatment designation permitting) from that
point. It should be noted that streams in the DMS had been surveyed prior to this
study, and the beginning of each stream was marked.
Because the random component in this study is the stream reach location and
not the stream itself, the scope of inference is limited to streams of a certain treatment
designation, not necessarily all streams in the DMS. Any model based inference is
limited to this population; however, the intended biological area of representation
encompasses similar forest types and management regimes throughout western
Oregon.
Study Design
Stream reaches were selected to represent a range of treatments present at
DMS locations. Treatments may be considered to be more representative of federal
29
and state forest land than private forest land. Thinning treatments where stream
reaches were measured were: control (~500 tpha), high density (~300 tpha), and
moderate density (~200 tpha) thinning treatments. Buffer widths where stream
reaches were measured include: control, two site potential tree heights, one site
potential tree height, variable width (minimum 15.2 m), and streamside retention (6.1
m). Buffer widths were implemented with a hard edge; that is to say, the boundaries
of the buffers were clearly defined, not tapered or feathered. Table 1 provides a
breakdown of the stream reaches in the combinations of thinning treatments and buffer
widths.
Table 1. Stream reaches measured by thinning treatment and buffer width.
Treatment Buffer Width
Control
Control
Streamside
Variable
Moderate
One Tree
Two Tree
Streamside
Variable
High
One Tree
Two Tree
GP42
GP 7-17
KM21
GP 7-18
BL13
KM17
DC 9-10
OM36
TH 7-30
KM18
TH46
CC 9-11
TH75
DC 8-28
TH 8-1
KM19
KM 7-2
KM 7-5
OMH 8-20
OMH 7-23
TH 7-10
KM 8-15
The term “stream reach” refers to a given length of stream and the associated
riparian forest. The length of the stream reach varied depending on topographic
restrictions or DMS treatment designations Attributes of the original nine stream
reaches were provided from prior surveys; USGS 7.5 minute quadrangle maps and
GIS were used to characterize the thirteen additional stream reaches. Stream reaches
30
in the original survey were required to be at least 100 m in length. The location of the
starting point for each stream reach was assigned using a randomly generated number
representing distance downstream from the beginning of seasonal flow. The random
number was between 0 and the stream length minus 100 for the original nine stream
reaches. The point at which seasonal flow began was determined during
implementation of the DMS, and was marked by painting the stream reach number on
a tree.
Once the starting point was located, a transect was installed which was
oriented along the same azimuth as the stream. The plot design for the original nine
stream reaches called for an initial transect length of 100 m along the stream and a
width of 36 m on either side of the stream, for a total plot size of 100 m by 72 m (with
the exception of Green Peak 42 which measured 100 m by 100 m). As discussed
previously, a riparian forest may be defined in many different ways; however, for the
purposes of this study, it was defined by the plot boundary of 36 m (horizontal
distance) to either side of the stream.
Measurement plots for stream reaches are illustrated in Figures 2 and 3. Two
designs were used in this study. Measurement plots installed in 2006 were defined by
a 200 m by 72 m rectangle intersected length-wise by the stream. The rectangle was
then divided in half, width-wise, at the 100 m mark, and then length-wise, at the
stream to form four equally sized quarters. For streams where the plot was less than
200 m in length, the plot was divided width-wise at the 100 m mark and length-wise at
the stream to form two pairs of equally sized quarters in the same manner as
31
previously described. A smaller 72 m by 72 m subplot was installed within the first
100 m of the plot. The 72 m by 72 m subplot was randomly assigned to either the first
72 m of the 100 m division, or the last 72 m of the 100 m division. Measurement plots
installed in 2007 are of the same design as those in 2006 with one exception; no 72 m
by 72 m subplot is present.
Figure 2. Measurement plot design with a randomly assigned subplot. For a stream
reach of 200m.
Figure 3. Measurement plot design without a randomly assigned subplot. For a stream
reach of 200m.
32
Description of Measurements
Measurements were taken on standing trees, windthrown trees, and site
characteristics. Within the larger plot boundary (200 m by 72 m) a tally of all
windthrown trees since 2003 was recorded along with several attributes. For plots
installed in 2006, measurements of standing trees larger than 10 cm were taken in the
72 m by 72 m subplot. For plots installed in 2007, measurements of standing trees
larger than 10cm were taken in 2 of the 4 quarters. In the initial 100 m of the plot, one
36 m by 100 m half was randomly selected and measured. Then, the quarter diagonal
from the one initially selected was also measured.
The diameter at breast height of all standing trees greater than 10 cm within the
specified measurement plot was recorded. Diameter measurements were taken to the
nearest 0.1 cm. Species was then recorded, along with crown class. Crown class was
determined by the height of the tree relative to the surrounding canopy, crown ratio,
and overall appearance. Trees were assigned to one of four categories: dominant, codominant, intermediate, or suppressed. Snags were assigned a decay class (1 through
5) similar to the decay classes described in Maser et al. (1984). Decay class 1 was
assigned to sound and intact recently dead trees, and decay class 5 was assigned to
very rotten, unstable or broken snags.
On those plots installed in 2006, two randomly located 10 m radius plots
served as subsamples for height. The height and crown ratio of all trees in the 10 m
radius plots were measured with a laser rangefinder to the nearest 0.1 m. The two
largest trees on the plot were cored with an increment borer to determine breast height
33
age. On those plots installed in 2007, heights and crown ratios, measured with a
clinometer to the nearest 1.0 m, were sampled from the range of diameter classes on
approximately 10 percent of the measured trees. The laser rangefinder used in 2006
was un-available in 2007, thus a clinometer was used. Age was recorded for at least
two of the largest trees on the plot.
The tally of windthrown trees was restricted to those trees which had fallen
since 2003 in order to capture the small scale windthrow events. DMS sites were
implemented and treated between 1997 and 2000. It was not desired to capture
windthrow resulting from initial exposure after thinning. For each windthrown tree
the number of years since blowdown, diameter at breast height, direction of fall,
height, diameter at break (if broken upon impact), slope distance from stream, percent
slope up and downhill from root wad, side of stream or quarter of plot (in reference to
cardinal directions), and presence of root rot were all noted. The presence and nature
of broken tops were also noted as were general landscape characteristics, stand
conditions, and a visual assessment of inherent site stability. A subjective rating of
stability class was also assigned based on understory species presence and abundance,
presence/absence of jackstrawed trees, slumps or slides, soil characteristics,
presence/absence of rock outcroppings, and observed rooting depth.
A set of criteria was developed for this study to determine time since
blowdown. Key elements included appearance of vegetation beneath the bole of the
tree; presence/absence and size of plant life on the root ball and on the upturned soil;
sharpness and brittleness of broken branches and roots; presence/absence of needles
34
and fine branches; presence/absence and orientation of fruiting fungal bodies (i.e.,
perpendicular or parallel to bole); and condition of the bark. Though the criteria used
were consistent throughout the whole study, it is a subjective method which includes
some amount of human error.
In addition to the above listed measurements, nine hemispherical photographs
of the canopy cover were taken at each stream reach in order to determine the percent
open sky beneath the canopy (SKY). Site characteristics such as slope of the stream,
orientation of the stream, and general condition of the stand were also recorded.
Compilation of Data
The review of literature has illustrated many potential attributes to include as
explanatory variables in a model. The four research questions proposed at the
beginning of the thesis were intended to guide the direction of the study and outline
meaningful attributes on which measurements could be taken. The three stand-level
variables addressed in the research questions (buffer width, thinning treatment, and
species composition) provide a framework, but are not a comprehensive list of
explanatory variables considered in the analysis. Data were recorded by hand in the
field and entered into Microsoft ® Office Excel 2003 Copyright © 1985-2003
Microsoft Corporation for further processing. Table 2 provides a list of stand level
and site attributes, which have been calculated from the collected data, to serve as
explanatory variables. Many of the potential explanatory variables have been included
35
as a result of the literature review; however, some variables such as SKY and
transformations of the traditional H/D ratio have not yet been utilized by others.
Table 2. Variables used to describe stream reach attributes.
Variable
Description
BUFF
Indicator -- Buffer width (Streamside, Variable, One tree, Two tree, Control)
TRT
Indicator -- Residual thinning density (Moderate, High, Control)
WH
Proportion western hemlock in stream reach, for live trees
TOPEX
TOPEX of 8 directions limited to 500m (includes negative angles)
TOPEXSW
TOPEX limited to 500m to the W, SW, S, and SE (includes negative angles)
TOPEXPOS
TOPEX of 8 directions limited to 500m (negative angles are counted as zero)
TOPEXPOSSW
TOPEX limited to 500m to the W, SW, S, and SE (negative angles are counted as zero)
H
Mean height (m) of trees in stream reach, for live trees
LH
Lorey's Height (m) (height weighted by basal area)
H100
Mean height (m) of 100 largest trees per hectare in stream reach, for live trees
D
Mean diameter(cm) of trees in stream reach, for live trees
QMD
Quadratic mean diameter (cm)
D100
Mean diameter (cm) of 100 largest trees in stream reach, for live trees
HD
Mean of ratios of height to diameter of trees in stream reach, for live trees
MHMD
Ratio of mean height to mean diameter of trees in stream reach, for live trees
LHQMD
Ratio of Lorey's Height to Qmd, for live trees
H100D100
Ratio of meanH100 to meanD100 of trees in stream reach, for live trees
TPHA
Trees per hectare in stream reach, for live trees
BAHA
Basal area per hectare in stream reach, for live trees
SKY
Percent open sky from hemispherical photographs
SI
Site Index (m)
STAB
Visual assessment of inherent site stability
ELEV
Elevation of center of stream reach (hundred m)
ORIENT
Aspect of stream in stream reach (°)
GRAD
Stream gradient (%)
UTMN
Universal Transverse Mercator Northing coordinate
UTME
Universal Transverse Mercator Easting coordinate
DOWNHA
Down trees per hectare in stream reach
PRES
Indicator -- Presence of windthrow
36
The DMS establishment report and study plan by Cissel et al. (2006) provided
some of the summary variables including: buffer width (BUFF), thinning treatment
(TRT), and site index (SI). Heights were predicted using height-diameter equations
from Hanus et al. (1999b); except for O.M. Hubbard which is located in Southwest
Oregon, for which equations from Hanus et al. (1999a) were used. Predicted heights
were calibrated to better fit the data using the measured heights and weighted least
squares regression in SAS software, Version 9.1 of the SAS System for UNIX.
Copyright © 2003 SAS Institute Inc. SAS and all other SAS Institute Inc. product or
service names are registered trademarks or trademarks of SAS Institute Inc., Cary,
NC, USA.
Microsoft Excel was used to compile many of the summary variables
including: proportion of western hemlock (WH), mean height (H), Lorey’s height
(LH), mean height of the 100 largest trees by diameter per hectare (H100), mean
diameter (D), quadratic mean diameter (QMD), mean diameter of the 100 largest trees
per hectare (D100), mean of ratios of height to diameter (HD), ratio of means of height
to diameter (MHMD), ratio of LH to QMD (LHQMD), ratio of H100 to D100
(H100D100), SKY, stream gradient (GRAD), and down trees per hectare
(DOWNHA). R Version 2.6.1, Copyright © 2007 The R Foundation for Statistical
Computing (http://www.r-project.org/) software was used to calculate the variables:
trees per hectare (TPHA) and basal area per hectare (BAHA).
Hemispherical photographs were processed using the Gap Light Analyzer
(GLA) version 2.0 (Frazer, 1999). Nine photos were taken at each location and
37
averaged to provide a single value of SKY. The variables: elevation (ELEV),
Universal Transverse Mercator Northing (UTMN) and Universal Transverse Mercator
Easting (UTME) were taken directly from 7.5 minute quadrangles from the United
States Geological Survey (USGS). The variable for stability (STAB), as previously
described, was a subjective rating assigned to each stream reach. The variable for
stream orientation (ORIENT) was recorded in the field using a hand compass.
The TOPEX scores (and variations thereof) were computed using ESRI’s
ArcGIS. View shed analyses were performed on digital elevation models (DEMs) of
7.5 minute USGS quadrangles. Quine and White (1998) implemented several limiting
distances on the projected angle to skyline and found 500 m to be the most appropriate
distance. Using trigonometric relationships, the angle from the point of origin to the
skyline along each line was determined. All eight values were summed to produce a
total index value.
A second method was used whereby the angles in the direction of prevailing
winds were summed. In western Oregon, the prevailing winds come from south and
southwesterly directions (National Weather Service, 2006). This variable was denoted
as TOPEXSW. By focusing on the directions facing the prevailing storm winds (W,
SW, S, and SE) a better relationship between windthrow occurrence and topographic
exposure may be determined. TOPEXPOS is a variation on TOPEX whereby negative
angles are treated as 0. TOPEXPOSSW treats negative angles as 0 and includes only
the four south to southwesterly directions.
38
The topographic information was obtained using 30 meter DEMs. This
resolution is appropriate for the given application. The large plot size (0.9 to 1.45 ha)
and length of TOPEX lines (500 m) lend themselves well to the coarse scale of the 30
m DEM. Although 30 m DEMs are thought to be sufficient, finer resolution DEMs
may provide greater precision.
Data Analysis
A two-step approach was chosen to analyze the windthrow data at a stand
level. The first step of the analysis models the binary response of windthrow absence
or presence through logistic regression and estimates a probability of the occurrence of
windthrow. The second step of the analysis uses multiple linear regression to model
the frequency or count of windthrow within the stream reaches where windthrow is
present. This type of two-step approach has been used by Hamilton and Brickell
(1983) to model cull volume in standing trees and by Woollons (1998) to model stand
mortality.
Step I: Logistic Regression
The science of forest management includes managing risk. Managers strive to
minimize the risk of mortality during regeneration, the risk of insect damage, the risk
of fire, or the risk of windthrow. In order to manage risk, the probability or likelihood
that the event will occur must be known or estimated. Logistic regression was chosen
because of its model form with the response as a predicted probability constrained
39
between 0 and 1. In addition, the model form may be transformed using the logit
transformation which results in the logarithm of the odds of an event occurring.
Logistic regression gained much of its acceptance and popularity from the field
of epidemiology (Hosmer and Lemeshow, 2000), and has since been used in many
fields of natural resource research. It characterizes binary or dichotomous data in such
a manner that the response, the logit, may be directly applied to situations in a useful
and interpretable way. It was chosen to represent the probability of windthrow for this
very reason. An objective of this study is to provide an interpretable and applicable
result in the form of a response variable. The importance of the underlying biotic and
abiotic mechanisms should not be underestimated; however, the focus rests in
prediction rather than inference regarding parameters.
The most widely encountered method of modeling windthrow in the literature
was logistic regression. Logistic regression uses the method of maximum likelihood
to estimate parameters. Logistic regression takes the model form:
Y=
e g ( x)
1
=
g ( x)
1+ e
1 + e −g ( x)
where:
g(x) = β 0 + β 1 x 1 + β 2 x 2 + …+ β p x p
and:
β 0 , … , β p = parameter estimates, and
x 1 , … , x p = predictor variables.
This non-linear model form may be transformed into:
40
⎛ Y ⎞
g ( x) = ln⎜
⎟ = β 0 + β1 x1 + β 2 x 2 + ... + β p x p
⎝1− Y ⎠
This transformation results in the logit, g(x), which is linear in its parameters.
Logistic regression requires certain assumptions to be met in order to be valid.
Assumptions of least squares regression may be more familiar to the reader than
assumptions of logistic regression. Some of the assumptions of logistic regression are
similar to least squares regression; however, logistic regression does not assume a
linear relationship between the independent variables and the dependent variable. It
does not require variables to be normally distributed. It does not assume
homogeneous variance. It does, however, require that observations be independent of
each other (non-correlated independent variables) and that the independent variables
be linearly related to the logit of the dependent variable (Hosmer and Lemeshow, 2000
p.# 6-7).
Logistic regression has proven to be a useful tool to estimate the probability of
windthrow. It has been utilized by Valinger and Fridman (1999), Jalkanen and Matilla
(2000), Canham et al. (2001), Mitchell et al. (2001), Peterson (2004), LanquayeOpoku and Mitchell (2005), and Scott and Mitchell (2005) to predict the probability of
windthrow on an individual tree level. Though the probability is calculated on a treelevel basis, it is expanded to a stand-level basis to provide the risk of a given stand to
windthrow. Although the response is at the tree-level, the explanatory variables for
these models may include tree-level, stand-level, and landscape-level attributes.
41
Lohmander and Helles (1987) utilized both tree-level and stand-level variables
to create a logistic model to predict the probability of windthrow at a stand level.
Sinton et al. (2000) used logistic regression to model the odds of windthrow as
functions of landscape position and features. Logistic regression has also proven
useful in other areas of forestry such as modeling cull volume in standing trees
(Hamilton and Brickell, 1983), modeling tree mortality (Temesgen and Mitchell,
2005), and estimating wildfire risk (Preisler et al. 2004). Its ability to provide
interpretable data with direct application has made this a popular model choice.
Logistic regression was performed using SAS v. 9.1 with the logistic
procedure. Given the data set of 22 stream reaches and a list of 27 potential
explanatory variables, the list was refined to include only the most relevant variables.
Pearson correlation coefficients were used to help identify variables which were
highly correlated with other explanatory variables, so as to minimize multicollinearity.
With a suitable list of potential explanatory variables the forward selection and
stepwise selection methods were utilized to aid in the formulation of a parsimonious
model. Greenland (1989) cautioned that when using stepwise regression, which is the
most common form of variable selection; that biologically important variables may not
be included. This warning to the analyst emphasizes the importance of scrutinizing
the results and making sure that the biologically relevant variables are included, and
conversely that irrelevant or noise variables are not included.
42
Step II: Linear Regression
Linear regression uses the method of ordinary least squares to estimate
parameters. The model is of the form:
Yi = β 0 + β 1 x1 + β 2 x 2 ... + β p x p
where:
Y i = volume of windthrow, or other metric of wind damage; β 0 , … , β p = parameter
estimates, and x 1 , … , x p = predictor variables.
It requires that several assumptions be met; depending on the intent of the user.
In order for the least squares estimators of the model parameters to be unbiased it is
assumed that: (1) the model is linear in its parameters and the error terms are additive;
(2) the number of sample observations (n) is greater than the number of parameters to
be estimated (k + 1 if the model has an intercept); (3) all independent variables are
non-stochastic variables measured without error; (4) no perfect correlation (often
called perfect multicollinearity) exists between any linear combination of the
independent variables; and (5) the model is correctly specified, including the
appropriate explanatory variables.
There are additional assumptions necessary for the least squares estimators of
the model parameters to be Best Linear Unbiased Estimators (BLUE’s). It is assumed
that (6) the variance, or mean square error (MSE) about the model is homogeneous,
and (7) the random errors are uncorrelated. This property is also called nonautoregression or lack of serial correlation. Further assumptions necessary for the
application of exact tests, the developments of exact confidence intervals, and the least
43
squares estimators of the model parameters to be Uniformly Minimum Variance
Unbiased Estimators (UMVUE’s) are: (8) the random errors are normally distributed,
and (9) the form and number of independent variables in the model are known before
parameter estimation.
Least squares regression has been utilized by Elling and Verry (1978), and
Steinblums et al., (1984) to formulate multiple linear regression windthrow models.
Elling and Verry (1978) fitted a model to estimate the total volume of wind caused
mortality in black spruce strip-cuts. Steinblums et al. (1984) fitted a model to estimate
the volume of windthrow in riparian buffer strips within or bordering clearcuts.
Linear regression was performed using SAS v. 9.1. Selected graphical
displays were created in R v. 2.6.1. Pearson correlation coefficients were utilized to
narrow down the list of explanatory variables to only those most relevant to the
response, the number of down trees per hectare (DOWNHA). Variables which were
not correlated with the response were eliminated; as well as variables which were
highly correlated with other explanatory variables, so as to minimize multicollinearity.
With a suitable list of potential explanatory variables the forward selection and
stepwise selection methods were again utilized to aid in the formulation of a
parsimonious model.
44
CHAPTER 4 – RESULTS AND DISCUSSION
Preliminary Analysis
To explore the characteristics of each of the stream reaches, and to
conceptualize and understand the processes occurring in each respective location,
tables of general stand attributes were compiled in Appendices B and C. Values for
the list of potential explanatory variables were then compiled (Appendix D) and
exploratory analysis was conducted to gain an understanding of basic relationships
among the attributes. Stream reaches covered a range of densities from 138 tpha to
856 tpha. All stream reaches were dominated by conifers, although hardwoods
comprised up to 25% of live trees.
As previously mentioned, the timeframe for which measurements were taken
was limited to windthrow having occurred between the years of 2003-2007. During
this time it should be noted that no major storm events occurred. Storm events in the
study region may be considered to be representative of that which may occur on a year
to year basis. Results of the study, and any conclusions, would likely change had
there been a large storm event.
Of the 22 stream reaches measured, 19 had at least one windthrow event
present. There was no windthrow present at 3 of the stream reaches. Out of the 19
where windthrow did occur, the number of windthrown trees ranged from 1 to 22. A
total of 145 windthrown trees were sampled across all stream reaches.
45
Research Question 1: windthrow and distance from stream
Based on the data collected for the windthrown trees, the frequency of
windthrow decreased as distance from the stream channel increased. Approximately
40% of the surveyed windthrow occurred within the first 5 m of slope distance from
the stream (Figure 4). An additional 17% occurred between 6-10 m from the stream
for a total of 57% of windthrow occurring within the first 10 m. Another 22% of
windthrow occurred between 11-20 m from the stream for a total of 79% of
windthrow events within 20 m slope distance from the stream channel. These results
are similar to others such as McDade et al. (1990) who observed greater than 70% of
large woody debris in streams originated within 20 m of the stream channel; and
Martin and Grotefendt (2007) who observed that 81% of large woody debris
originated within 10 m of the stream, and 95% originated within 20 m of the stream.
These results undoubtedly are dependent on the height of trees bordering the stream.
Wood may be input into streams from greater distances where tall trees border the
stream than where short trees border the stream.
46
100
Frequency (%)
n = 145 trees
80
60
40
20
0
1--5
6--10
11--15
16--20
21--25
26--30
31--35
36--40
40+
Slope Distance from Stream (m)
Figure 4. Frequency of windthrow by slope distance from stream, using 5 m swaths,
for all windthrow.
Softwoods and hardwoods exhibited a similar trend of decreased frequency of
windthrow with increasing distance from stream (Figure 5 and Figure 6). Although
the number of windthrow events is greater among softwoods, this does not imply that
softwoods are more susceptible to windthrow; rather, it is a reflection of the number of
softwoods to hardwoods present in the stream reaches. The ratio of softwoods to
hardwoods ranged from approximately 3:1 to 198:1, with two stream reaches
completely absent of hardwoods.
47
100
Frequency (%)
n = 127 trees
80
60
40
20
0
1--10
11--20
21--30
31--40+
Slope Distance from Stream (m)
Figure 5. Frequency of windthrow by slope distance from stream, using 10 m swaths,
for softwoods.
100
Frequency (%)
n = 18 trees
80
60
40
20
0
1--10
11--20
21--30
31--40+
Slope Distance from Stream (m)
Figure 6. Frequency of windthrow by slope distance from stream, using 10 m swaths,
for hardwoods.
48
Several factors contribute to the decreased occurrence of windthrow as
distance from the stream increased. Soil moisture has been cited as an influence on
the occurrence of windthrow (Mitchell et al., 2001; Steinblums et al., 1984; LanquayeOpoku and Mitchell, 2005; and Scott and Mitchell, 2005). Soil drainage has also been
identified as a contributing factor by Lohmander and Helles (1987). Sites with
shallow depth to bedrock, and rocky sites have also been cited as likely contributors to
windthrow (Sinton et al., 2000). Soil properties undoubtedly affect the stability of a
given tree to withstand the forces of wind. Shallow rooting was observed in many
sites with saturated poorly drained soils, and offered little in the way of anchoring
strength for the roots of a tree. Such soils may often be found close to a stream.
Although the crown architecture of a tree plays a large role in the amount of bending
force applied to the bole of a tree, it is the roots that ultimately must withstand such
forces.
Further complicating the issue is the interaction between soil properties and
species. Differences in the stability of species have been well documented (Ruth and
Yoder, 1953; McLintock, 1954; Gratkowski, 1956; Scott and Beasley, 2001; Scott and
Mitchell, 2005; Martin and Grotefendt, 2007), and have shown, for instance, western
hemlock to be one of the least windfirm species. In the stream reaches measured in
this study, western hemlock was often found nearer to the stream than Douglas-fir.
The observed difference in stability may be due in part to inherent characteristics and
properties of a given species. It may also be due in part to the site characteristics and
soil properties where a given species is found.
49
The implications of the relationship between windthrow occurrence and
distance from stream may help forest managers understand the potential for the supply
of down woody debris in streams and in riparian forests. For instance, VanSickle and
Gregory (1990) stated that trees entering the stream from a closer distance tended to
be larger and thus deliver a greater volume/quantity of wood to a stream. Growing
conditions near streams are often conducive to producing larger trees compared to the
uplands. If large piece size is desired in the management plans for a stream, it follows
that leaving larger trees closer to the stream would be more beneficial than leaving
large trees far away from the stream. The data from this study, however, does not
reveal a strong relationship between diameter and distance from the stream (Figure 7).
80
70
Dbh (cm)
60
50
40
30
20
10
0
0
10
20
30
Distance from stream (m)
Figure 7. Dbh by distance from stream for all windthrow.
40
50
50
The long-term supply of wood is also an issue that must be considered. One
recommendation comes from a study in the Coast Range of Oregon where Andrus et
al. (1988) suggest that trees retained in riparian areas should be at least 50 years old in
order to be large enough to ensure delivery of large material to the stream. The
desired size of wood for a particular management goal may help managers determine
the necessary age distribution of trees left in buffer strips.
Research Question 2: windthrow and buffer width
In order to determine the effect of buffer width on the frequency of windthrow,
the data were analyzed as a factorial treatment design to test for an interaction between
buffer width and thinning treatment. There was convincing evidence to suggest the
presence of an interaction between buffer width and thinning treatment (p-value =
0.0040) (Table 3). Due to the presence of the interaction only simple effects of a
factor may be estimated. Simple effects of a factor are contrasts between levels of one
factor at a single level of another factor.
To clarify the relationship between buffer width and thinning treatment, the
buffers in this study have not been cut. The thinning treatment includes the forest
surrounding the buffer in which the trees have been thinned to a specific density
according to the DMS treatment prescriptions.
Contrasts were arranged in order to determine if the number of windthrow in
any of the buffer width and thinning treatment combinations was significantly
different from the number of windthrow in the control treatment. The contrasts
51
revealed that there was not enough evidence to suggest a difference, with the
exception of one-tree buffer width combined with the high thinning density (Table 4).
The difference in means, however, is based upon only one observation for the one-tree
buffer high thinning density combination and therefore should not be considered when
comparing means. All other contrast combinations were also examined and none were
found to have significantly different numbers of windthrow from each other; again
with the exception of the one-tree buffer in the high thinning treatment which was
different from all other combinations.
Table 3. Factorial treatment analysis showing presence of interaction.
Source
Thinning treatment
Buffer width
Thinning treatment X Buffer width
DF
2
3
1
SS
111.76
52.72
132.03
MS
55.88
17.57
132.03
F-value p-value
4.88
0.0233
1.54
0.2463
11.54
0.004
Table 4. Simple effect contrasts for thinning treatment and buffer width combinations.
Contrast
Control - Moderate Streamside
Control - Moderate Variable
Control - Moderate One-tree
Control - Moderate Two-tree
Control - High Variable
Control - High One-tree
Estimate Std. Err.
0.85
2.58
-1.75
2.39
0.20
2.93
1.30
2.39
-2.45
2.39
-16.75
3.78
t-value
0.33
-0.73
0.07
0.54
-1.02
-4.43
p-value
0.7467
0.4757
0.9465
0.5948
0.3220
0.0005
It is possible that there is indeed no difference in frequency of windthrow
between buffer width and thinning treatment combinations; however, the lack of
difference may be explained by a number of possibilities. First and most plausible are
52
the small sample size and unequal replication. Second is the impact of the fixed width
plot design. Plots at each stream reach were uniform in width; though the width of the
buffer was not always uniform. For example, at stream reaches where a streamside
buffer (20 ft) was implemented, the measurement plot extended beyond the boundaries
of the buffer. Thus, the data is confounded by the effect of the thinning treatment of
neighboring stands, and does not reflect the effect of the buffer width alone. At stream
reaches where a two site tree potential tree height was used as the buffer width, the
measurement plot did not encompass the whole width of the buffer. Although the
width of the measurement plot was not based on site potential tree height, it did
adequately reflect the observed height of trees. In addition, trees originating greater
than 36 m horizontal distance from the stream would likely deliver only small amounts
of woody debris and foliage.
The lack of difference in windthrow between buffer widths may also be
explained by the forest cover. One might expect a larger buffer to afford more shelter
from the wind; however, in continuous forest cover, the edge effect so clearly
observable along clearcuts is not present. Although this may be a contributing factor,
it is likely not the entire story since similar results were observed by Steinblums et al.
(1984) and Ruel et al. (2001) who both found no significant relationship between
mortality and buffer widths along clearcut edges. The underlying mechanism or
explanation for this occurrence may not be apparent; however, it is apparent that if no
buffer strip is present, no wood can be delivered to the stream.
53
Research Question 3: windthrow and thinning treatment of neighboring stands
Similar to the discussion above regarding the effect of buffer width, the
presence of an interaction between buffer width and thinning treatment limits
inference about thinning treatments to simple effects. The same results apply to
thinning treatment as applied to buffer width. There is no significant difference
between the control and any of the combinations (Table 4).
Similar to the issues surrounding the buffer width, the lack of evidence may be
due to the small sample size, the sample design, or the plot design. The plot locations
are not representative of the thinning treatments of the neighboring forest; rather they
are focused on the riparian buffers.
The question addressed here has received little attention in the literature. The
issue of windthrow in the thinned forest is often addressed rather than windthrow
adjacent to the thinned forest. Weidman (1920), Alexander and Buell (1955), and
McLintock (1954) have all linked windthrow to thinning practices; however, the link
was windthrow within the thinned forest. Regarding thinning in buffer strips
specifically, Ruel et al. (2001) observed no significant difference in mortality of buffer
strips adjacent to clearcuts due to thinning intensity.
It was hypothesized in this study that the neighboring forest would exhibit a
sheltering effect from the oncoming wind. This, however, was not observed. The
sheltering effect would be of particular utility should the objective be to prevent
windthrow. The effectiveness of the neighboring forest’s ability to shelter the buffer
54
strip likely depends a great deal on the residual thinning density, age, species present,
and height of the stand.
Research Question 4: windthrow and species composition
As illustrated in the review of literature by Ruth and Yoder (1953), McLintock
(1954), Gratkowski (1956), Scott and Mitchell (2005), and others, tree species differ
in their susceptibility to windthrow. In the Pacific Northwest, western hemlock has
consistently been ranked as one of the least resistant species to windthrow. For this
reason, it was used as a metric for species composition. Based on the data however,
there was not enough evidence to suggest a difference in the amount of windthrow by
percent western hemlock present in the stream reaches (p-value = 0.2533). Percent
hardwood present was also a poor explanatory variable for the data. There was not
enough evidence to suggest a difference in the amount of windthrow by percent
hardwood present in the stream reaches (p-value = 0.7826).
Step I: Logistic Regression
In determining the presence or absence of windthrow in a stream reach logistic
regression was utilized. First the list of possible explanatory variables was narrowed
down by removing variables which were highly correlated with other explanatory
variables. Variables which were better captured by other means were eliminated as
well. For instance mean diameter, quadratic mean diameter, and diameter of the 100
largest trees per hectare all explain the same basic attribute and were therefore all
55
highly correlated with each other. During the exploratory analysis only one of the
diameter variables was included at a time. The same procedure was used with the
three height variables. Trees per hectare and basal area were also highly correlated,
therefore only one of the two was included in the preliminary model fitting procedure
at a time.
As a result of the stepwise method of variable selection, ELEV, D100, and
D100H100 were found to be the most significant explanatory variables. Using the
maximum likelihood ratio statistic, elevation was found to be the most significant
variable (p-value = 0.0028). The intercept term however, was not significant at the
alpha = 0.05 level and was therefore dropped from the model. The parameters were
estimated using the method of maximum likelihood in SAS v.9.1 with the LOGISTIC
procedure; resulting in the final model form: logit ( pˆ ) = 0.3515(ELEV) where ELEV is
the elevation of the midpoint of a stream reach in hundred meters. The probability of
observing windthrow (y) is then calculated as: y =
e g ( x)
where g(x) is the
1 + e g ( x)
logit ( pˆ ) . A 95 % confidence interval for the estimated slope parameter is 0.1549 to
0.6363.
Brown et al. (2002) demonstrated that the Wald confidence interval often
provided poor coverage, and recommend the likelihood ratio test interval as one of the
superior alternatives. The odds ratio of windthrow for the above logistic regression
equation was estimated to increase by 1.421 for every 100 m gain in elevation. A 95%
confidence interval for this estimated odds ratio is 1.129 to 1.178. As an example to
56
illustrate this odds ratio, a stream buffer strip within the scope of the study area at 600
m elevation is estimated to be 1.4 times as likely to suffer windthrow over a similar 4
year period as a stand at 500 m elevation. Also, a stream buffer strip within the scope
of the study area at 600 m elevation has a probability of 0.8918 of suffering
windthrow over a similar 4 year period. Mean predicted probabilities for the range of
elevation values in the study are illustrated in Figure 8.
Regarding the assumptions of logistic regression, observations must be
independent of each other. This assumption is believed to be upheld. As discussed in
the Methods and Analysis section, streams were selected independently of each other.
The second assumption is that the explanatory variables are linearly related to the logit
of the response variable. It is believed that the logit transformation of the ith
individual’s event probability is indeed capable of being expressed as a linear function
of the explanatory variables.
Although other variables such as the ratio of height to diameter, or TOPEX
were believed to be more powerful explanatory variables, elevation affects or is
associated with many of the factors influencing windthrow. Elevation has been linked
to gradients in soil properties and species composition (Laughlin and Abella, 2007). It
also affects the amount of exposure, and often slope, of a landscape. While individual
sites at high elevation may be sheltered, the higher the elevation, the less protection is
afforded by surrounding terrain. Elevation cannot be controllable by management
practices, but it can be accounted for when conducting silvicultural treatments; and its
relevance as a predictor of windthrow should not be underestimated.
57
Predicted probability of windthrow
1
0.9
0.8
0.7
2
3
4
5
6
7
8
Elevation (hundred m)
Figure 8. Predicted probability of windthrow by elevation.
The fit of the model to the data may be evaluated by means of several
statistics. The Hosmer and Lemeshow goodness-of-fit statistic is one such measure of
how well the model fits the data. The test sorts the observations by their estimated
probabilities in to g groups. It then calculates a Chi-square statistic from the g x 2
table of observed and expected frequencies. Under the null hypothesis that the model
provides a good fit of the data, the Hosmer and Lemeshow goodness-of-fit test
resulted in a p-value of 0.9219; suggesting that the null hypothesis should not be
rejected and that the model is indeed a good fit.
58
Although the Hosmer and Lemeshow goodness-of-fit statistic indicates that the
model fits the data well, there are other diagnostic measures that must be considered.
Given 19 event occurrences and 3 non-occurrences in the data set, the 3 nonoccurrences are likely to be influential in the outcome of the model. A drop in
deviance test for individual observations reveals that two of the three non-occurrence
observations (DC828 and GP718) have a large influence compared to the other data
points. This suggests that the model is highly dependent on two data points, and,
although the model may be a good fit to the observed data, it is very fragile and
changes significantly upon the removal or addition of any observations.
The percent concordant and discordant are other measures of the fit of the
model. In these measures, all possible pairs of observations where one is an event
occurrence and the other is a non-occurrence are arranged. If the event occurrence has
a higher predicted probability than the non-occurrence then the pair is said to be
concordant. If the event occurrence has a lower predicted probability than the nonoccurrence then the pair is said to be discordant. Based on the data set, 75.4% of the
pairs were found to be concordant, 19.3% of the pairs were found to be discordant,
and 5.3% of the pairs were neither concordant nor discordant. With 19 event
occurrences and 3 non-occurrences, there were 57 pairs in all.
Examination of fit statistics is an important part of the model building process
in order to determine the strength of the model based on the observed data. As
Chatfield (1995) points out, the training set, or data set used to build the model, will
59
consistently overstate the fit of the model. In order to extend the model to future
predictions, model validation becomes important.
One method of model validation is known as controlled division or data
splitting. In this method, the data is split into two sets, one for model building and one
for validation (Stone, 1974). Data may be split 50/50, 60/40, 70/30, or any other
number of ways depending on the quantity of data. This method is preferred, but is
restrictive in that large data sets must be collected in order to have enough data for
both the model building and the validation.
Another method is known as cross-validation. There are variations of this
method; however, the main idea is to build the model upon n-1 observations and then
test the model on the omitted observation. This process is repeated n times until each
data point has been omitted once. The predictions are then averaged over all n
omissions. Still another method is known as the bootstrap. This involves re-sampling
with replacement from the original data set and building the model from the
bootstrapped sample. The model is then validated on the original data set, or those
subjects not included in the bootstrap sample. This method relies heavily on the
quality of the underlying data set to produce meaningful results.
Validating the predictive accuracy of a logistic regression model presents some
difficulty. Predictions using logistic regression assume the form of a probability. In
the observed data set observations are binary; the event either did or did not occur.
One consideration is selecting an appropriate cutoff for predicted probabilities for
which an event would or would not occur. For instance, does a predicted probability
60
of 0.35 suggest that windthrow will occur or not? A second consideration is the level
of certainty regarding the quality of the underlying data set. For these reasons, it is
recommended that validation of the windthrow data be conducted on an independent
data set.
Although the model provides useful insight into the processes of windthrow,
the logistic model developed in the first step of the modeling process should be taken
with some caution. Logistic regression has great potential in windthrow modeling and
should not be underestimated; however, this particular model has been developed with
a relatively small sample size, and is highly influenced by a few data points. It is
recommended that further data be collected as a validation set in order to strengthen
the conclusions regarding the model.
Step II: Linear Regression
The second step of the analysis used linear regression to relate the explanatory
variables to the response (number of windthrown trees per hectare) only in those sites
where windthrow was observed. Similar to the process used with logistic regression,
the list of explanatory variables was refined in the exploratory analysis. Correlation
coefficients were utilized not only between explanatory variables but also between
explanatory variables and the response, due to the linear relationship of the model
form.
The stepwise method of variable selection was again used to determine the
most relevant explanatory variables. At the alpha = 0.05 level, the variables TOPEX,
61
TOPEXSW, HD, and UTMN were found to be significant predictors of the frequency
of windthrow. The relationships between each of the variables and the response were
then examined. The variables TOPEX and TOPEXSW were eliminated from
consideration because their relationship to the response was incorrect. The
relationship reflected in the observed data was weak, but suggested that as the
exposure of a stream reach increased, the number of windthrown tree decreased.
Because these variables did not make sense on a practical level, they were removed
from consideration in the model.
The variable HD was found to be the most significant predictor (p-value =
0.0199). No additional variables had an F-value to enter that was significant in the
second step of the selection process. At the 0.10 significance level however, ELEV +
HD, and UTME + HD were found to be significant. The relationships between the
combinations of explanatory variables and the response were examined; however, HD
was chosen as the most useful predictor because of its significance level, and its
biological meaning.
Model parameters were estimated in SAS v. 9.1 using the method of ordinary
least squares. The resulting model for the mean number of down trees per hectare (y)
was estimated to be: y = -24.96 + 0.36(HD), where HD is the mean of ratios of height
(m) to diameter (m) for trees in a stream reach. The standard errors for the estimate of
the intercept and slope parameters are 11.9 and 0.14 respectively. The root mean
square error (RMSE) is 4.06. Approximately 28% of the variation in windthrow over
the last 4 years was explained by the variable HD (R2 = 0.2797).
62
The linear regression equation suggests that the mean number of down trees
per hectare increases by 0.36 for every one unit increase in HD. Although the variable
HD is commonly used in forestry, it is not as common to imagine a one unit change in
HD as it would be to imagine a one unit change in height or diameter. Consider the
following example. A stream reach with an average height of 32 m and an average
diameter of 40 cm is equivalent to the mean of ratios necessary to achieve an HD of
80. If the average height were increased to 36 m while the average diameter remained
constant at 40 cm, this would be equivalent to the mean of ratios for an HD of 90.
This ten unit increase in HD would result in an increase of an average of 3.6 down
trees per hectare in a stream reach. The relationship presented by the model suggests
that stream reaches with larger ratios of height to diameter will incur more windthrow.
The degrees to which the assumptions of linear regression are met affect the
strength of the model. The first five assumptions are necessary for the least squares
estimators of the model parameters to be unbiased. Some of these may be formally
tested, while others are subjectively addressed by the modeler. The first assumption, as
discussed in the Methods and Analysis section, states that the model must be linear in
its parameters. Results of previous models reviewed in the literature illustrate the
appropriateness of using linear regression to model the relationship of HD to the
occurrence of windthrow. Other model forms may be appropriate as well; however, it
was decided a priori that linear regression was to be used in the second step of the
modeling process. Figure 9 shows the relationship of HD to the occurrence of
windthrow.
63
Secondly, the number of sample observations must be greater than the number
of parameters to be estimated. Although the number of observations was less than the
number of potential parameters from which to choose, the final model form estimates
10
0
5
Down trees per hectare
15
20
only two parameters: the intercept and one slope parameter.
75
80
85
90
95
100
HD
Figure 9. Scatter plot of down trees per hectare by height to diameter ratio.
The third assumption is that the explanatory variables are measured without
error. Within reason this assumption has been met. Care has been taken during data
64
collection to be as accurate as possible with the conditions and equipment. It is
believed that no systematic or subjective errors have been introduced into the
measurements.
The fourth assumption requires that no perfect multicollinearity exists between
independent variables. In order for perfect correlation between independent variables
to exist, there must be multiple independent variables. Because the selected model is a
simple linear regression, multicollinearity is not a concern.
The fifth assumption is that the model form is correctly specified. Although
many variables have been shown to be related to windthrow, it is not feasible to
include all variables in a ‘kitchen sink’ model. The chosen model may not be a true
representation of windthrow, but it is believed to be the most parsimonious model
formulation given the data and choice of explanatory variables.
The next assumption requires a homoscedastic (homogenous variance) model.
Examination of the plot of residuals was inconclusive, thus the Glejser Test (Glejser,
1969) for homoscedasticity was used. In the test, the absolute values of the residuals
are regressed on the independent variable(s) to determine if they are significantly
different from zero. If only the intercept is significantly different from zero, the model
has pure homoscedasticity. If only the slope is significantly different from zero, the
model has heteroscedasticity. If both the intercept and slope are significantly different
from zero, then the model is said to have mixed heteroscedasticity. Based on the
Glejser Test, there is suggestive, but inconclusive evidence regarding the
homoscedasticity of the model. The p-values for the intercept and slope of the model
65
in the Glejser Test were 0.0544, and 0.0228 respectively. As previously mentioned, a
log transformation and a square root transformation were unsuccessful at correcting
the problem of non-constant variance. The least squares estimates of parameters will
still be unbiased; however, the standard errors for the parameters may not be accurate.
Thus, any tests of hypothesis or confidence intervals may be misleading.
The seventh assumption is that the random errors are uncorrelated. This
property is also called non-autoregression or lack of serial correlation. This is a
mostly a problem with time series data (Ramsey and Schafer, 2002 pp. 63-65). This
assumption does not need to be addressed at this time; however, if repeat
measurements were to be taken in the future, this would need to be addressed with the
Durbin Watson Test (Durbin and Watson, 1950, 1951, 1971).
Two further assumptions remain for the application of exact tests, the
developments of exact confidence intervals, and the least squares estimators of the
model parameters to be Uniformly Minimum Variance Unbiased Estimators
(UMVUE’s). The first of these assumptions is that the errors are normally distributed.
The second is that the form and number of independent variables in the model are
known before parameter estimation.
In order to test for normally distributed errors, a plot of residual versus fitted
values (Figure 10) was examined. The plot of residuals versus fitted values shows
roughly normal distribution, although the model appears to overestimate the quantity
of windthrow at large values of HD. A log transformation of the response was
66
performed in order to stabilize the variance; however, the resulting model was no
-5
0
Residuals
5
longer significant at the alpha = 0.05 level.
2
4
6
8
10
Fitted Values
Figure 10. Scatter plot of residual versus fitted values for height to diameter ratio
regressed on down trees per hectare.
0.0 0.5 1.0 1.5 2.0
C ooks.D istance
67
5
10
15
1
0
-1
studres
2
observation number
5
10
15
0.25
0.15
0.05
leverage
0.35
observation number
5
10
15
observation number
Figure 11. Influence statistics for all observations of height to diameter ratio regressed
on down trees per hectare.
68
Influence statistics were examined to determine the effect of each observation
on the normality of the residuals and the performance of the model. As illustrated in
Figure 11, observation number 14 (OMH820) has a relatively high leverage, which
indicates that it greatly influences the slope of the regression equation in its region of
the data points. Because the formula for Cook’s distance includes the leverage value,
the same point also has a large Cook’s distance. An observation with a Cook’s
distance larger than 1 is considered to have significant influence on the regression
equation (Ramsey and Schafer, 2002 p.320).
With observation 14 removed, the variable HD is no longer significant in the
regression equation (p-value = 0.5180). In order to remove a data point, it must be
believed to have originated from a different population, or be a recording error. In this
case it was neither; therefore, the point should not be removed. The influence exerted
on the regression equation by a single observation suggests that the model is very
fragile. It is recommended in order to strengthen the model, that more data is
collected.
The final assumption for linear regression is that the form and number of
independent variables in the model are known before parameter estimation. This was
not the case in this modeling procedure. The purpose was not to test certain
hypotheses regarding a pre-determined model; rather, it was exploratory modeling
process. Given the information gained through this process, hypotheses may be made
regarding the effect of height to diameter ratio and tested with supplemental data.
69
One Step Approach
Given the small sample size and small number of non-occurrences (stream
reaches where no windthrow occurred) a single stage approach, in which linear
regression is used to predict the number of windthrown trees based on data from all
stream reaches, was compared with the two step approach. Again the variable HD
was found to be the most significant predictor of the number of windthrown trees. A
similar result was found wherein the variance about the model was non-constant. The
two-step approach is a valuable method to model windthrow; however, it may be best
suited to larger data sets, where a greater number of non-occurrences are available.
A log transformation of the model with no intercept was chosen in order to
correct the problem of non-constant variance. The resulting model for the log of the
mean number of down trees per hectare (y) was estimated to be: log(y) = 0.0175(HD),
where HD is the mean of ratios of height (m) to diameter (m) for trees in a stream
reach. The standard error for the estimate of the slope parameter is 0.0020. The
RMSE is 0.7358. Approximately 79% of the variation in windthrow over the last 4
years was explained by the variable HD (R2 = 0.7913).
Due to the logarithm transformation of the response variable, the interpretation
of the regression equation differs slightly from that of the linear regression equation in
the two step approach. The linear regression equation above suggests that the median
number of down trees per hectare increases by a factor of 1.02 for each 1 unit change
in HD. In other words, a 10 unit change in HD is associated with a multiplicative
increase in the median number of down trees per hectare of 1.22 times.
70
Although the relationship between neighboring thinning treatment and
windthrow was not able to be detected in this study, the relationship between HD and
windthrow may provide some useful insight. The linear regression models suggest
that the number of windthrow events increases as the ratio of HD increases. This
suggests that tree density, which influences the ratio HD (Wang et al., 1998), may be
useful in creating a more windfirm stand over time. As Jacobs (1954) demonstrated,
increased exposure to wind can help strengthen to withstand the forces of wind.
Additional Findings
Beyond the a priori research questions and modeling process, there are certain
trends or observations that became apparent in the analysis process. One such trend is
the direction of fall of windthrown trees. The greatest proportion of trees,
approximately 23%, fell in a northeasterly direction; and approximately 15% of the
trees fell in a northerly direction. All told, nearly 40% of the trees fell in a north to
northeasterly direction, indicating the influence exerted by the prevailing winds from
the south to southwest. Figure 13 depicts the direction of fall for all 145 trees
observed.
71
Figure 13. Compass rose showing direction of fall for all observed windthrow.
Although a high proportion of trees fell to the north and northeast, the number
of trees which fell to the south, southeast, or east is noteworthy. A summary table of
72
windthrow by direction for each stream reach may be seen in Appendix E. Prevailing
winds played a dominant role in determining the direction of fall for windthrow;
however, the influence of local topography also appears to be quite significant.
Alexander and Buell (1955) asserted that local topography can alter the direction of
prevailing winds by as much as 90 degrees. Wind may be funneled through narrow
valleys, changing direction along with the contours. Many of the stream buffer strips
were located deep “U” or “V” shaped valleys quite capable of altering wind direction.
Strong winds may predispose trees to fall in a certain direction; however the
effect of topography or of sudden gusts may alter that direction. It was observed that
25% of the trees fell down hill toward the stream ± 15 degrees.
Wind storms are often accompanied by rain in western Oregon which may
cause further complications. Saturated soil can cause trees to be more susceptible to
windthrow. Sudden wind gusts followed by periods of calm may be enough to loosen
the roots in the soil; and as the tree recoils from the initial gust it may fall downhill;
especially on steep slopes where stems are often jackstrawed or have a slight downhill
lean.
Further investigation of the direction of tree fall indicated that it was not equal
for all directions. A null hypothesis may be formed whereby it is assumed an equal
number of trees fell up-valley, down-valley, toward the stream and away from the
stream. A test of the null hypothesis indicated that there was convincing evidence that
the direction of tree fall was not equal among the four categories (Chi-squared test pvalue <0.0001).
73
Table 5. Direction of tree fall in relation to stream valley.
Direction
Toward stream
Down-valley
Up-valley
Away from stream
# of trees
72
36
19
18
%
50
25
13
12
Table 5 indicates that a greater number of trees fell toward the stream than any
other direction. This suggests that there may be a high probability that a falling tree
will have some portion of the bole land in the stream channel. This of course is
dependant upon the distance from the stream the tree originated.
Another complicating factor affecting the occurrence of windthrow is root
disease. Several root pathogens including Phytophthora, Armillaria, and Phellinus
affect many of the tree species commonly found in western Oregon (Hamm and
Hansen, 1982; Entry et al., 1990; Lim et al., 2005). Root rot diseases were identified
on approximately 4% of the windthrown trees. Root rot has been documented to
increase the risk of windthrow by Ruth and Yoder (1953), McLintock (1954),
Alexander (1964), and others. Alexander (1964) observed root rot and butt rot in
association with approximately one-third of windthrow in spruce-fir forests in
Colorado.
Additionally, approximately 12% of the trees were thought to have been
knocked down or severely weakened by other trees as they fell. The force exerted by
a falling tree is more than enough to knock down other trees in its path as it falls.
Windthrow is seldom an isolated event; even when it is not in large proportions.
74
CHAPTER 5 – CONCLUSIONS
Summary of Findings
Windthrow is influenced by many factors at the tree, stand, and landscape
level. The interaction among factors can further complicate the issue. Managing for
windthrow requires not only knowledge of how the factors relate to windthrow, but
also how they relate to each other. It is the intent of this study to advance the
understanding of the factors affecting the occurrence of windthrow and further the
knowledge required to effectively manage for this occurrence.
The definition of windthrow provided in the introduction includes only those
live trees which have been completely toppled by wind. It does not include broken
tops, leaning trees, or snags which have blown over. This definition of windthrow
addresses the result, but may not adequately address the cause. Trees may be
weakened by root rot or previous storms for instance and therefore are more
susceptible to windthrow. In other words, wind is the eventual cause of all windthrow,
but may not be the fundamental cause.
Based on the observations of this study windthrow did not appear to be
inhibited by either buffer strips, or continuous forest cover; though it was not a
significant source of small scale mortality. In the majority of stream reaches,
windthrow amounted to 1% or less of standing live basal area over the 4 yr. period.
The greatest recorded loss was 5.2%. Often times preventing windthrow is not the
primary management objective; rather, minimizing or regulating its occurrence is
75
desirable and part of the trade offs of leaving trees to meet a variety of management
objectives. At times it may even be encouraged, as in the cases of habitat
development or for structural diversity.
Many of the positive and negative contributions of windthrow and the resulting
down woody debris have been explored. From the perspective of habitat management,
windthrow can be a vital source of down woody debris providing structure for
terrestrial and aquatic organisms. From the perspective of natural resource production,
windthrow can mean a loss of valuable wood fiber and consequently a loss of revenue.
Regardless of how it is viewed and the benefits and tradeoffs associated with it,
windthrow must be incorporated in management plans as a stochastic element.
In order to obtain the desired type and size of woody debris, the processes
affecting its input must be known. Under an active management setting, the resource
manager must be aware of the factors which can be managed to achieve a desired
result. As discussed in the review of literature some of these factors are: species
composition, stocking or density, diameter distribution, height to diameter ratio,
among others. Knowledge of how these manageable factors relate to the occurrence
of windthrow is crucial to effectively manage riparian forests. Based on the findings
of this study, in headwater streams of the DMS:
1) Windthrow decreased with increasing distance from the stream channel.
2) The effect of buffer width on the occurrence of windthrow was not detectable.
3) The effect of thinning density in neighboring stands on the occurrence of
windthrow was not detectable.
76
4) The effect of species composition on the occurrence of windthrow was not
detectable.
5) Elevation was the most significant predictor of occurrence of windthrow.
6) Height to diameter ratio was the most significant predictor of the number of
windthrown trees.
7) The direction of fallen trees was dictated primarily by prevailing storm winds, but
was also influenced by local topography.
Recommendations
Despite advances in forest sciences, or perhaps because of the advances, there
remains much to discover. As societal needs and environmental conditions change, so
too to the objectives of research. To build upon this and other research related to
windthrow the following suggestions for future projects are presented.
1) Observations in this study may be supplemented to strengthen the models, and
begin development of a tree level windthrow risk model. Although stands are
often the unit of management, silvicultural operations begin with a single tree.
2) Many studies have isolated important factors affecting windthrow and developed
models to predict its occurrence. These models may be compared and tested for
portability to other regions and forest types.
3) Models for large storm events, buffer strips, and clearcut edges have been
developed, but little has been done in the way of modeling small scale windthrow
in continuous forest.
77
4) Much is known about the use of woody debris and its importance in forest
structure, but little is known about the effect of the method of input; or rather the
time scale of input. Often times buffer strips and clearcut edges provide
immediate input of down woody debris as a result of increased exposure. The
differences in periodic inputs every harvest rotation versus slow continuous inputs
may be studied.
5) Windthrow is often not an isolated event. It often occurs in groups or clusters.
There is potential to study the effectiveness and efficiency of using adaptive
cluster sampling for detecting and quantifying windthrow.
6) Methods of surveying for windthrow by ground, as opposed to remote sensing, are
invaluable as they afford a clear picture to the researcher not only of the
windthrow but also the subtle nuances of the surrounding conditions. There is
however, a great potential to utilize remote sensing applications such as Light
Detection and Ranging (LiDAR). The spatial information associated with LiDAR
may be especially useful in monitoring change over time, because exact locations
of trees are known.
78
BIBLIOGRAPHY
Alexander, R.R. 1964. Minimizing windfall around clear cuttings in spruce-fir forests.
For. Sci. 10(2):130-142.
Alexander, R.R. and J.H. Buell. 1955. Determining the direction of destructive winds
in a Rocky Mountain timber stand. J. of For. 53(1):19-23.
Andrus, C.W., B.A. Long, and H.A. Froehlich. 1988. Woody debris and its
contribution to pool formation in a coastal stream 50 years after logging. Can.
J. Fish. Aqu. Sci. 45:2080-2086.
Atterson, J. 1980. Gambling with gales. Paper presented to British Association for the
Advancement of Science, Salford, UK.
Behre, C.E. 1921. A study of windfall in the Adirondacks. J. of For. 19(6)632-637.
Bilby, R.E. 1989. Interactions between aquatic and terrestrial systems. In K.J.
Raedeke, editor. Streamside management: riparian wildlife and forestry
interactions. Institute of Forest Resources. Contribution number 59. University
of Washington, Seattle, WA. Pp 13-43.
Bisson, P.A., R.E. Bilby, M.D. Bryant, C.A. Dolloff, G.B. Grette, R.A. House, M.L.
Murphy, K.V. Koski, and J.R. Sedell. 1987. Large woody debris in forested
stream in the Pacific Northwest: past, present, and future. In proceedings of the
symposium Streamside management: Forestry and Fishery Interactions. Feb.
12-14, 1986, Seattle, WA. Edited by E.O. Salo and T.W. Cundy. University of
Washington, Seattle, WA. pp143-190.
Bragg, D.C., J.L. Kershner, and D.W. Roberts. 2000. Modeling large woody debris
recruitment for small streams of the Central Rocky Mountains. U.S.D.A. F.S.
Rocky Mountain Research Station. Gen. Tech. Rep. RMRS-GTR-55. 36 pp.
Brooks, K.N., P.F. Ffolliott, H.M. Gregersen, and L.F. DeBano. 2003. Hydrology and
the management of watersheds 3rd edition. Iowa State Press. Ames, IA. 574 pp.
Brosofske, K.D., J. Chen, R.J. Naiman, and J.F. Franklin. 1997. Harvesting effects on
microclimatic gradients from small streams to uplands in western Washington.
Ecol. App. 7(4):1188-1200.
Brown, L.D., T.T. Cai, and A. DasGupta. 2002. Confidence intervals for a binomial
proportion and asymptotic expansions. The Annals of Stat. 30(1):160-201.
79
Byrne, K.E. and S.J. Mitchell. 2007. Overturning resistance of western redcedar and
western hemlock in mixed-species stands in coastal British Columbia. Can. J.
of For. Res. 37:931-939.
Canham, C.D., M.J. Papaik, and E.F. Latty. 2001. Interspecific variation in
susceptibility to windthrow as a function of tree size and storm severity for
northern temperate tree species. Can. J. of For. Res. 31(1):1-10.
Chatfield, C. 1995. Model uncertainty, data mining, and statistical inference. J. Royal
Stat. Soc. Series A. 158:419-466.
Cissel, J.H., P.D. Anderson, D. Olson, K. Puettmann, S. Berryman, S. Chan, and C.
Thompson. 2006. BLM Density management and riparian buffer study:
Establishment report and study plan: U.S. Geological Survey Scientific
Investigations Report 2006-5087. 144 pp.
Cremer, C.W., C.J. Borough, F.H. Kinnell, and P.R. Carter. 1982. Effects of stocking
and thinning on wind damage in plantations. N. Z. J. of For. Sci. 12(2):244268.
Cole, M.B., K.R. Russell, and T.J. Mabee. 2003. Relation of headwater
macroinvertebrate communities to in-stream and adjacent stand characteristics
in managed second-growth forests of the Oregon Coast Range mountains. Can.
J. of For. Res. 33:1433-1443.
Curtis, J.D. 1943. Some observations on wind damage. J. of For. 41(12):877-882.
Durbin, J. and G.S. Watson. 1950. Testing for serial correlation in least squares
regression. I. Biometrika. 37(3/4):409-428.
Durbin, J. and G.S. Watson. 1951. Testing for serial correlation in least squares
regression. II. Biometrika. 38(1/2):159-178.
Durbin, J. and G.S. Watson. 1971. Testing for serial correlation in least squares
regression. III. Biometrika. 58(1):1-19.
Elie, J-C. and J-C. Ruel. 2005. Windthrow hazard modeling in boreal forests of black
spruce and jack pine. Can. J. of For. Res. 35: 2655-2663.
Elling, A.E. and E.S. Verry. 1978. Predicting wind-caused mortality in strip-cut stands
of peatland black spruce. The For. Chron. 249-252.
Entry, J.A., S.K. Hagle, and K. Cromack Jr. 1990. The affect of Armillaria attack on
the nutrient status of inland Douglas-fir. Eur. J. For. Path. 20(5):269-274
80
Evans, A.M., A.E. Camp, M.L. Tyrrell, and C.C. Riely. 2007. Biotic and abiotic
influences on wind disturbance in forests of NW Pennsylvania, USA. For.
Ecol. and Manage. 245(1-3): 44-53.
Forest and Debris Recovery Team. 2007. Forest and Debris Recovery Final Report
Winter Storm – December 2007. O.D.F. Aug 18, 2008.
http://egov.oregon.gov/ODF/TimberBlowdown.shtml.
Franklin, J.F., T. Spies, D. Perry, M. Harmon, and A. McKee. 1986. Modifying
Douglas-fir management regimes for nontimber objectives. In proceedings of
the symposium Douglas-fir: stand management for the future. June 18-20,
1985. Seattle, WA. University of Washington, Seattle, WA. pp 373-379.
Franklin, J.F., T. Spies, R. Van Pelt, A.B. Carrey, D.A. Thornburgh, D.R. Berg, D.B.
Lindenmayer, M.E. Harmon, W.S. Keeton, D.C. Shaw, K. Bible, and J.Chen.
2002. Disturbances and structural development of natural forest ecosystems
with silvicultural implications, using Douglas-fir forests as an example. For.
Ecol. and Manage. 155:399-423.
Frazer, G.W., C.D. Canham, and K.P. Lertzman. 1999. Gap Light Analyzer (GLA),
Version 2.0: Imaging software to extract canopy structure and gap light
transmission indices from true-colour fisheye photographs, users manual and
program documentation. Copyright © 1999: Simon Fraser University,
Burnaby, British Columbia, and the Institute of Ecosystem Studies, Millbrook,
New York.
Glejser, H. 1969. A new test for heteroscedasticity. J. Amer. Stat. Assoc. 64:316-323.
Gratkowski, H.J. 1956. Windthrow around staggered settings in old-growth Douglasfir. For. Sci. 2:60-74.
Greenland, S. 1989. Modeling and variable selection in epidemiologic analysis. Am. J.
of Pub. Health. 79(3):340-349.
Gregory, S.V., F.J. Swanson, W.A. McKee, and K.W. Cummins. 1991. An ecosystem
perspective of riparian zones. BioScience. 41(8):540-551.
Hagan, J.M., S. Pealer, and A.A. Whitman. 2006. Do small headwater streams have a
riparian zone defined by plant communities? Can. J. of For. Res. 36:21312140.
Hall, F.C. 1989. Characterization of riparian systems. In K.J. Raedeke, editor,
Streamside management: riparian wildlife and forestry interactions. Institute of
Forest Resources. Contribution number 59. University of Washington, Seattle,
WA. Pp 7-12.
81
Hamilton, D.A. Jr. and J.E. Brickell. 1983. Modeling methods for a two-state system
with continuous responses. Can. J. of For. Res. 13:1117-1121.
Hamm, P.B. and E.M. Hansen. 1982. Pathogenicity of Phytophthora species to Pacific
Northwest Conifers. Eur. J. For. Path. 12(3):167-174.
Hanus, M.L., D.W. Hann, and D.D. Marshall. 1999a. Predicting height for undamaged
and damaged trees in southwest Oregon. Research Contribution 27, Forest
Research Lab. Oregon State University, Corvallis, OR. 22 p.
Hanus, M.L., D.D. Marshall, and D.W. Hann. 1999b. Height-diameter equations for
six species in the coastal regions of the Pacific Northwest. Research
Contribution 25, Forest Research Lab. Oregon State University, Corvallis, OR.
11 p.
Harmon, M.E., J.F. Franklin, F.J. Swanson, P. Sollins, S.V. Gregory, J.D. Lattin, N.H.
Anderson, S.P. Cline, N.G. Aumen, J.R. Sedell, G.W. Lienkaemper, K.
Cromack, JR., and K.W. Cummins.1986. Ecology of coarse woody debris in
temperate ecosystems. Pages 133-302 in A. MacFadyen and E.D. Ford, editors.
Advances in Ecological Research. Vol. 15. Academic Press, New York, NY,
USA.
Hilty, J.A. and A.M. Merenlender. 2004. Use of riparian corridors and vineyards by
mammalian predators in northern California. Conserv. Bio. 18(1):126-135.
Hosmer, D.W. and Lemeshow, S. 2000. Applied logistic regression. 2nd edition. New
York, NY, USA. John Wiley & Sons, Inc. 375 pp.
Ilhardt, B.L., E.S. Verry, and B.J. Palik. 2000. Defining the riparian areas. In E.S.
Verry et al., editors, Riparian management of forests of the continental eastern
United States. Lewis Publishers. New York, NY. Pp 23-42.
Jacobs, M.R. 1954. The effect of wind sway on form and development of Pinus
radiata D.Don. Aus. J. of Bot. 2:35-51.
Jalkanen, A. and U. Matilla. 2000. Logistic regression models for wind and snow
damage in northern Finland based on the National Forest Inventory data. For.
Ecol. and Man. 135:315-330.
Johnson, S.L. 2004. Factors influencing stream temperatures in small streams:
substrate effects and a shading experiment. Can. J. Fish. Aquat. Sci. 61:913923.
Keller, E. A. and F.J. Swanson. 1979. Effects of large organic material on channel
form and fluvial processes. Earth Surface Processes. 4:361:380.
82
Lanquaye-Opoku, N., and S.J. Mitchell. 2005. Portability of stand-level empirical
windthrow risk models. For. Ecol. and Man. 216:134-148.
Laughlin, D.C. and S.R. Abella. 2007. Abiotic and biotic factors explain independent
gradients of plant community composition in ponderosa pine forests. Ecol.
Model. 205(1/2):231-241.
Lekes, V. and I. Dandul. 2000. Using airflow modeling and spatial analysis for
defining wind damage risk classification (WINDARC). For. Ecol. and Man.
135:331-344.
Lienkaemper, G.W. and F.J. Swanson. 1987. Dynamics of large woody debris in
streams in old-growth Douglas-fir forests. Can. J. of For. Res. 17(2):150-156.
Lim, Y.W., C.A.Yeung, R. Sturrock, I. Leal, and C. Breuil. 2005. Differentiating the
two closely related species, Phellinus werii and P. sulphurascens. For. Path.
35:305-314.
Lohmander, P. and F. Helles. 1987. Windthrow probability as a function of stand
characteristics and shelter. Scand. J. of For. Res. 2:227-238.
Macdonald, J.S., E.A. MacIsaac, and H.E. Herunter. 2003. The effect of variableretention riparian buffer zones on water temperatures in small headwater
streams in sub-boreal forest ecosystems of British Columbia. Can. J. of For.
Res. 33:1371-1382.
Martin, D.J. and R.A. Grotefendt. 2007. Stand mortality in buffer strips and the supply
of woody debris to streams in Southeast Alaska. Can. J. For. Res. 37(1):36-49.
Maser, C., J.M. Trappe, S.P. Cline, K. Cromack J.R., H. Blaschke, J.R. Sedell, and
F.J. Swanson. 1984. The seen and unseen world of the fallen tree. U.S.D.A.
Forest Service. P.N.W. Forest and Range Experiment Station. Gen. Tech. Rep.
PNW-164. 56 pp.
McDade, M.H., F.J. Swanson, W.A. McKee, J.F. Franklin, and J. VanSickle. 1990.
source distances for coarse woody debris entering small streams in western
Oregon and Washington. Canadian Journal of Forest Research 20:326-330.
McLintock, T.F. 1954. Factors affecting wind damage in selectively cut stands of
spruce and fir in Maine and New Hampshire. U.S.D.A. Forest Service. N.E.
Forest Experiment Station. Research Paper 70. 17 pp.
Mitchell, S.J. 1995. Windthrow in British Columbia. In Wind and Trees. Coutts, M.P.
And J. Grace. New York, NY, Syndicate Press of the University of Cambridge:
448- 459.
83
Mitchell, S.J., T. Hailemariam, and Y. Kulis. 2001. Empirical modeling of cutblock
edge windthrow risk on Vancouver Island, Canada, using stand level
information. For. Ecol. and Man. 154:117-130.
Murphy, M.L., K.V. Koski, J.Heifetz, S.W. Johnson, D. Kirchhofer, and J.F.
Thedinga. 1985. Role of large organic debris as winter habitat for juvenile
salmonids in Alaska. Proceedings, Western Association of Fish and Wildlife
Agencies 1984:251-262.
Naiman, R.J., H. Decamps, and M. Pollock. 1993. The role of riparian corridors in
maintaining regional biodiversity. Ecol. App. 3(2):209-212.
National Weather Service Forecast Office, Portland, OR. 2006. June 14, 2008.
http://www.weather.gov/climate/local_data.php?wfo=pqr.
Noss, R.F. 1987. Corridors in real landscapes: a reply to Simberloff and Cox. Conserv.
Bio. 1:159-164.
Orr, P.W. 1963. Windthrow timber survey in the Pacific Northwest, 1962. U.S.D.A.
Forest Service. P.N.W. Timber Management Insect and Disease Control. 19
pp.
Perault, D.R. and M.V. Lomolino. 2000. Corridors and mammal community structure
across a fragmented, old-growth forest landscape. Ecol. Mon. 70(3):401-422.
Perkins, D.W. and M.L. Hunter, Jr. 2006. Use of amphibians to define riparian zones
of headwater streams. Can. J. of For. Res. 36:2124-2130.
Peterson, C.J. 2004. Within-stand variation in windthrow in southern boreal forests of
Minnesota: Is it predictable? Can. J. of For. Res. 34:365-375.
Preisler, H.K., D.R. Brillinger, R.E. Burgan, and J.W. Benoit. 2004. Probability based
models for estimation of wildfire risk. Int. J. of Wild. Fire. 13:133-142.
Putz, F.E, P.D Coley, K. Lu, A. Montalvo, and A. Aiello. 1983. Uprooting and
snapping of trees: structural determinants and ecological consequences. Can. J.
For. Res. 13:1011-1019.
Quine, C.P. 1995. Assessing the risk of wind damage to forests. In: Wind and Trees.
Coutts, M.P. and Grace, J. New York, NY, Syndicate Press of the University
of Cambridge:379-403.
Quine, C.P. and P.D. Bell. 1998. Monitoring of windthrow occurrence and progression
in spruce forests in Britain. Forestry. 71(2):87-97.
84
Quine, C.P. and I.M.S. White. 1998. The potential of distance-limited topex in the
prediction of site windiness. Forestry. 71:325-332.
Ramsey, F.L. and D.W. Schafer. 2002. The Statistical Sleuth 2nd ed. Duxbury, Pacific
Grove, CA. p.436.
Rashin, E. and C. Graber. 1992. Effectiveness of Washington’s Forest Practice
Riparian Management Zone Regulations for protection of stream temperature.
Center for Streamside Studies, University of Washington, Seattle, WA. TFWWQ6- 92-001. 118 pp.
Reineke, L.H. 1933. Perfecting a stand-density index for even-aged stands. J. Agric.
Res. 46:627-638.
Ruel, J-C., D. Pin, and K. Cooper. 2001. Windthrow in riparian buffer strips: effect of
wind exposure, thinning and strip width. For. Ecol. and Man. 143:105-113.
Ruel, J-C., D. Pin, L. Spacek, K. Cooper, and R. Benoit. 1997. The estimation of wind
exposure for windthrow hazard rating: comparison between Strongblow, MC2,
Topex, and a wind tunnel study. Forestry. 70(3):253-266.
Ruth, R.H. and R.A. Yoder. 1953. Reducing wind damage in the forests of the Oregon
Coast Range. U.S.D.A. Forest Service. P.N.W. Forest and Range Experiment
Station. Res. Pap. 7. 30 pp.
Scott, R. and B. Beasley. 2001. Variable retention windthrow monitoring report. Long
Beach Model Forest Society Report. Ucluelet, BC. 42 pp.
Scott, R.E. and S.J. Mitchell. 2005. Empirical modeling of windthrow risk in partially
harvested stands using tree, neighborhood, and stand attributes. For. Ecol. and
Man. 218:193-209.
Sinton, D.A., J.A. Jones, J.L. Ohmann, and F.J. Swanson. 2000. Windthrow
disturbance, forest composition, and structure in the bull run basin, Oregon.
Ecology. 81(9):2539-2556.
Smith, K. and R.H. Weitknecht. 1915. Windfall damage in selection cuttings in
Oregon. Proceedings of the Society of American Foresters. 10:263-265.
Sollins, P. 1982. Input and decay of coarse woody debris in coniferous stands in
western Oregon and Washington. Can. J. of For. Res. 12:18-28.
Steinblums, I.J., H.A. Froehlich, and J.K. Lyons. 1984. Designing stable buffer strips
for stream protection. J. of For. 82(1):49-52.
85
Stone, M. 1974. Cross-validatory choice and assessment of statistical predictions. J. of
the Royal Stat. Soc. Series B. 36(2):111-147.
Swanson, F.J., G.W. Leinkaemper, and J.R. Sedell. 1976. History, physical effects,
and management implications of large organic debris in western Oregon
streams. U.S.D.A. Forest Service. P.N.W. Forest and Range Experiment
Station. Gen. Tech. Rep. PNW-56. 15 pp.
Swanson, F.J., M.D. Bryant, G.W. Leinkaemper, and J.R. Sedell. 1984. Organic debris
in small streams, Prince of Wales Island, Southeast Alaska. U.S.D.A. Forest
Service. P.N.W. Forest and Range Experiment Station. Gen. Tech. Rep. PNW166. 12 pp.
Swanson, F.J., S.V. Gregory, J.R. Sedell, and A.G. Campbell. 1982. Land-water
interactions: the riparian zone. In R.L. Edmonds, editor, Analysis of coniferous
forest ecosystems in the western United States. Hutchinson Ross Publishing
Company. Stroudsburg, PA, USA. Pp. 267-291.
Tabacchi, E., A.M. Planty-Tabacchi, and O. Décamps. 1990. Continuity and
discontinuity of the riparian vegetation along a fluvial corridor. Land. Ecol.
5(1):9-20.
Temesgen, H. and S.J. Mitchell. 2005. An individual-tree mortality model for complex
stands of southeastern British Columbia. W. J. of App. For. 20(2):101-109.
Valinger, E., and J. Fridman. 1999. Models to assess the risk of snow and wind
damage in pine, spruce, and birch forests in Sweden. Env. Manage. 24(2):209217.
VanSickle, J. and S.V. Gregory. 1990. Modeling inputs of large woody debris to
streams from falling trees. Can. J. of For. Res. 20(10):1593-1601.
Wang, Y., S.J. Titus, and V.M. LeMay. 1998. Relationships between tree slenderness
coefficients and tree or stand characteristics for major species in boreal
mixedwood forests. Can. J. of For. Res. 28:1171-1183.
Weidman, R.H. 1920. The windfall problem in the Klamath region, Oregon. J. of For.
18:837-843.
Woollons, R.C. 1998. Even-aged stand mortality estimation through a two-step
regression process. For. Ecol. and Man. 105:189-195.
APPENDICES
Appendix A. Stream reach locations.
Stream Reach BLM District
County
Legal Description
Latitude
Longitude
UTMN
UTME
BL13
Eugene
Douglas
T21S, R5W, S1
N43º46'20.0"
W123º14'11.0"
4846036
480785
CC911
Salem
Polk
T8S, R7W, S31
N44°50'05.0"
W123°35'26.0"
4965022
453847
DC828
Salem
Clackamas
T3S, R5E, S35
N45°15'56.0"
W122°09'33.0"
5012069
565435
DC910
Salem
Clackamas
T3S, R5E, S35
N45°15'56.0"
W122°09'33.0"
5012398
566577
GP42
Salem
Benton
T14S, R6W, S7
N44º22'00.0"
W123º27'30.0"
4912718
463490
GP717
Salem
Benton
T14S, R6W, S7
N44º22'00.0"
W123º27'30.0"
4912482
463494
GP718
Salem
Benton
T14S, R6W, S7
N44º22'00.0"
W123º27'30.0"
4912034
463879
KM17
Salem
Linn
T12S, R1E, S13
N44º31'41.0"
W122º37'55.0"
4929904
528833
KM18
Salem
Linn
T12S, R1E, S13
N44º31'41.0"
W122º37'55.0"
4929597
529419
KM19
Salem
Linn
T12S, R1E, S13
N44º31'41.0"
W122º37'55.0"
4929739
529285
KM21
Salem
Linn
T12S, R1E, S13
N44º31'41.0"
W122º37'55.0"
4930053
528800
KM72
Salem
Linn
T12S, R1E, S13
N44º31'41.0"
W122º37'55.0"
4929650
529159
KM75
Salem
Linn
T12S, R1E, S13
N44º31'41.0"
W122º37'55.0"
4930491
528696
KM815
Salem
Linn
T12S, R1E, S13
N44º31'41.0"
W122º37'55.0"
4930548
528859
OMH36
Roseburg
Douglas
T25S, R7W, S19 / T26S, R8W, S24
N43º17'30.0"
W123º35'00.0"
4793039
452618
OMH723
Roseburg
Douglas
T25S, R7W, S19 / T26S, R8W, S24
N43º17'30.0"
W123º35'00.0"
4793364
452502
OMH820
Roseburg
Douglas
T25S, R7W, S19 / T26S, R8W, S24
N43º17'30.0"
W123º35'00.0"
4793546
452539
TH46
Eugene
Lane/Benton
T15S, R7W, S10,15
N44º16'50.0"
W123º31'06.0"
4902573
458569
TH75
Eugene
Lane/Benton
T15S, R7W, S10,15
N44º16'50.0"
W123º31'06.0"
4902764
458935
TH710
Eugene
Lane/Benton
T15S, R7W, S10,15
N44º16'50.0"
W123º31'06.0"
4902916
458514
TH730
TH81
Eugene
Eugene
Lane/Benton
Lane/Benton
T15S, R7W, S10,15
T15S, R7W, S10,15
N44º16'50.0"
N44º16'50.0"
W123º31'06.0"
W123º31'06.0"
4903570
4902836
457644
457896
86
Appendix B. Stocking density for stream reaches (Tpha)
Live Conifer
Live Hardwood
Live (total)
Dead Conifer
Dead Hardwood
Dead (total)
Windthrow Conifer
Windthrow Hardwood
Windthrow (total)
Live Conifer
Live Hardwood
Live (total)
Dead Conifer
Dead Hardwood
Dead (total)
Windthrow Conifer
Windthrow Hardwood
Windthrow (total)
BL13
CC911
DC828
DC910
GP42
GP717
GP718
KM17
KM18
KM19
KM21
286
48
243
3
335
41
275
0
348
6
138
1
342
21
405
39
856
46
552
16
349
42
334
246
376
275
354
139
363
444
903
568
392
77
6
29
0
28
11
11
0
160
13
33
0
88
1
29
4
106
25
112
0
37
21
83
29
39
11
173
33
89
33
131
112
58
0
0
2.8
0
0
0
4.2
0
8.3
0
1.4
0
0
0
1.4
0.7
4.2
0
4.5
0
2.3
4.6
0
2.8
0
4.2
8.3
1.4
0
2.1
4.2
4.5
6.9
KM72
KM75
KM815
OMH36
TH46
TH75
TH710
TH730
TH81
676
142
379
3
360
35
385
39
365
32
509
30
602
0
565
201
297
49
593
3
353
18
818
382
394
423
397
539
602
766
346
596
371
50
25
38
3
42
6
54
6
29
4
100
6
64
0
52
14
24
14
100
1
26
4
75
40
47
60
33
106
64
66
38
101
31
0.7
0
4.2
0
1.4
0.7
5.6
0
2.1
2.1
19.5
0.9
4.2
0
3.5
0
9.7
4.2
6.9
0
5.6
0
0.7
4.2
2.1
5.6
4.2
20.4
4.2
3.5
13.9
6.9
5.6
OMH723 OMH820
87
Appendix C. Basal area of stream reaches (m2/ha)
Live Conifer
Live Hardwood
Live (total)
Dead Conifer
Dead Hardwood
Dead (total)
Windthrow Conifer
Windthrow Hardwood
Windthrow (total)
Live Conifer
Live Hardwood
Live (total)
Dead Conifer
Dead Hardwood
Dead (total)
Windthrow Conifer
Windthrow Hardwood
Windthrow (total)
BL13
CC911
DC828
DC910
GP42
GP717
GP718
KM17
KM18
KM19
KM21
39.9
3.0
53.9
0.1
53.8
4.4
49.0
0.0
58.9
0.4
30.9
0.1
75.3
1.9
67.6
2.8
84.4
3.2
51.1
0.9
47.7
3.4
42.8
53.9
58.2
49.0
59.3
31.0
77.3
70.5
87.6
52.0
51.1
2.3
0.1
3.0
0
1.9
0.7
1.3
0
8.3
0.7
1.7
0
4.7
0.1
1.3
0.1
10.1
0.9
2.1
0
4.2
0.6
2.4
3.0
2.6
1.3
8.9
1.7
4.8
1.4
11.0
2.1
4.8
0
0
0.7
0
0
0
1.4
0
1.8
0
0.3
0
0
0
0.6
0
0.5
0
0.2
0
0.7
0.5
0
0.7
0
1.4
1.8
0.3
0
0.6
0.5
0.2
1.2
KM72
KM75
KM815
OMH36
OMH723
OMH820
TH46
TH75
TH710
TH730
TH81
54.7
3.8
57.1
0.1
46.5
3.2
35.8
2.7
41.1
2.5
50.7
1.1
85.5
0
67.0
7.2
47.1
3.1
51.6
0.1
45.6
0.9
58.6
57.3
49.6
38.6
43.6
51.7
85.5
74.2
50.2
51.6
46.5
1.0
0.6
3.3
0.2
2.2
0.3
1.9
0.3
1.1
0.1
2.3
0.2
9.2
0
2.9
0.5
1.0
0.8
7.3
0
2.3
0
1.6
3.5
2.5
2.2
1.2
2.5
9.2
3.4
1.8
7.3
2.3
0.1
0
0.2
0
0.2
0.2
1.4
0.0
0.3
0.5
2.4
0.0
1.1
0
1.3
0
1.9
0.7
0.8
0
0.5
0
0.1
0.2
0.4
1.4
0.8
2.4
1.1
1.3
2.6
0.8
0.5
88
Appendix D. Potential explanatory variables
Stream Reach
BL13
CC911
DC828
DC910
GP42
GP717
GP718
KM17
KM18
KM19
KM21
KM7-2
KM75
KM815
OMH36
OMH723
OMH820
TH46
TH75
TH710
TH730
TH81
BUFF
TRT
WH
Two tree
Streamside
Control
Streamside
Control
Streamside
One tree
Control
Two tree
Two tree
Variable
Two tree
Variable
Variable
Variable
Variable
One tree
Control
Variable
Variable
One tree
Variable
Moderate
Moderate
Control
Moderate
Control
Moderate
Moderate
Control
Moderate
Moderate
Moderate
Moderate
High
High
Moderate
High
High
Control
Moderate
High
Moderate
Moderate
0
0.272
0.322
0.447
0.144
0.014
0
0.622
0.618
0.325
0.464
0.523
0.788
0.614
0.036
0.015
0
0.292
0.038
0.064
0.238
0.326
TOPEX TOPEXSW TOPEXPOS
42
99
27
0
79
34
93
24
45
28
23
3
11
6
45
56
68
49
85
150
126
27
40
29
13
20
61
51
17
11
23
14
8
16
23
24
69
59
56
19
-6
72
71
28
54
99
28
22
87
60
106
36
51
34
29
17
34
32
72
69
74
70
102
150
135
65
TOPEXPOSSW
H
LH
H100
40
29
14
20
61
51
30
18
24
15
12
16
26
27
69
59
56
30
11
74
71
36
32.6
35.6
30.3
33.0
34.2
39.4
37.9
32.8
26.5
25.0
29.3
23.2
31.6
28.8
29.3
31.2
31.1
34.1
27.5
33.7
28.1
31.2
37.9
44.1
37.3
35.9
40.7
42.0
42.4
35.6
32.5
33.3
34.5
30.3
36.0
34.9
33.3
34.8
35.0
36.8
35.2
38.0
32.0
36
41.9
46.2
40.9
38.5
44.4
42.6
46.4
39.1
38.7
38.4
38.0
35.8
39.2
38.6
33.3
38.6
38.5
41.9
41.1
42.0
36.3
39.9
89
Appendix D. continued…
Stream Reach
BL13
CC911
DC828
DC910
GP42
GP717
GP718
KM17
KM18
KM19
KM21
KM7-2
KM75
KM815
OMH36
OMH723
OMH820
TH46
TH75
TH710
TH730
TH81
D
QMD
D100
HD
MHMD
LHQMD
H100D100
TPHA
BAHA
38.3
48.1
40.4
45.9
42.9
51.6
49.3
43.3
32.4
30.6
38.0
27.2
41.1
36.8
32.0
35.4
32.8
41.2
32.0
40.9
31.5
37.6
40.4
52.9
44.4
47.6
46.2
53.3
52.1
45.0
35.2
34.2
40.7
30.2
43.7
40.0
34.1
37.4
35.0
42.5
35.1
43.0
33.2
39.9
55.1
69.3
63.6
58.0
63.0
58.3
70.5
58.6
57.2
54.7
56.0
50.3
59.8
57.7
46.6
52.3
50.9
56.8
54.4
56.3
47.9
53.0
89
79
79
73
82
78
81
78
84
84
79
89
82
82
95
93
101
84
87
85
92
85
85
74
75
72
80
76
77
76
82
82
77
85
77
78
92
88
95
83
86
82
89
83
94
83
84
75
88
79
81
89
92
97
85
100
82
87
98
93
100
87
100
88
96
90
76
67
64
66
70
73
66
67
68
70
68
71
66
67
71
74
76
74
76
75
76
75
334
246
376
275
354
139
363
444
903
568
392
818
382
394
423
397
539
602
766
346
596
371
42.8
53.9
58.2
49.0
59.3
31.0
77.3
70.5
87.6
52.0
51.1
58.6
57.3
49.6
38.6
43.6
51.7
85.5
74.2
50.2
51.6
46.5
90
Appendix D. continued…
Stream Reach
LAI
SI
STAB
ELEV
ORIENT
GRAD
UTMN
UTME
BL13
CC911
DC828
DC910
GP42
GP717
GP718
KM17
KM18
KM19
KM21
KM7-2
KM75
KM815
OMH36
OMH723
OMH820
TH46
TH75
TH710
TH730
TH81
14.8
11.0
11.2
17.3
10.5
16.0
14.5
10.0
13.8
10.4
12.1
10.6
13.3
12.4
13.9
12.3
11.4
13.5
13.9
11.0
11.0
13.4
138
130
122
122
130
128
123
129
135
137
120
130
134
130
112
142
120
125
123
125
106
128
2
2
1
1
2
3
2
1
2
1
2
2
2
2
2
1
2
1
3
3
2
2
295
485
590
645
590
640
520
700
745
730
670
735
665
660
510
480
450
525
520
520
755
725
322
206
250
5
82
77
136
273
254
278
280
14
338
328
69
14
2
102
176
125
104
79
20
15
7
13
25
28
25
10
20
24
10
15
18
18
30
15
17
25
40
23
25
30
4846036
4965022
5012069
5012398
4912718
4912482
4912034
4929904
4929597
4929739
4930053
4929650
4930491
4930548
4793039
4793364
4793546
4902573
4902764
4902916
4903570
4902836
480785
453847
565435
566577
463490
463494
463879
528833
529419
529285
528800
529159
528696
528859
452618
452502
452539
458569
458935
458514
457644
457896
DOWNHA PRES
0
2.8
0
4.2
8.3
1.4
0
2.1
4.2
4.5
6.9
0.7
4.2
2.1
5.6
4.2
20.4
4.2
3.5
13.9
6.9
5.6
0
1
0
1
1
1
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
91
Appendix E. Windthrow by direction of fall.
Stream
Reach
BL13
CC911
DC828
DC910
GP42
GP717
GP718
KM17
KM18
KM19
KM21
KM72
KM75
KM815
OM36
OMH723
OMH820
TH46
TH75
TH710
TH730
TH81
Total
NE
E
SE
S
SW
W
NW
N
(338°-22°) (23°-67°) (68°-112°) (113°-157°) (158°-202°) (203°-247°) (248°-292°) (293°-337°) Total
2
2
1
1
2
2
2
2
1
5
1
21
3
6
1
2
1
1
2
1
1
1
1
5
1
1
1
1
1
1
1
1
1
2
2
1
3
1
10
1
5
4
8
3
5
3
3
2
1
33
22
18
1
1
2
1
1
6
1
1
1
1
1
1
1
1
1
1
1
2
11
8
4
1
6
24
8
0
4
0
6
15
2
0
3
6
4
10
1
6
3
8
6
22
6
5
20
10
8
145
92
Download