e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/ecolmodel
Using simulated historical time series to prioritize fuel
treatments on landscapes across the United States: The
LANDFIRE prototype project夽
Robert E. Keane a,∗ , Matthew Rollins a,b,1 , Zhi-Liang Zhu c,2
a
USDA Forest Service, Rocky Mountain Research Station, Fire Sciences Laboratory, 5775 Hwy 10 West, Missoula,
MT 59808, United States
b USDA Forest Service, Rocky Mountain Research Station, Fire Sciences Laboratory, 5775 Hwy 10 West, Missoula,
MT 59808, United States
c US Geological Survey, EROS Data Center, Sioux Falls, SD 57198, United States
a r t i c l e
i n f o
a b s t r a c t
Article history:
Canopy and surface fuels in many fire-prone forests of the United States have increased
Received 7 February 2006
over the last 70 years as a result of modern fire exclusion policies, grazing, and other land
Received in revised form
management activities. The Healthy Forest Restoration Act and National Fire Plan establish
25 January 2007
a national commitment to reduce fire hazard and restore fire-adapted ecosystems across the
Accepted 1 February 2007
USA. The primary index used to prioritize treatment areas across the nation is Fire Regime
Published on line 8 March 2007
Condition Class (FRCC) computed as departures of current conditions from the historical
fire and landscape conditions. This paper describes a process that uses an extensive set of
Keywords:
ecological models to map FRCC from a departure statistic computed from simulated time
Integrated ecological modeling
series of historical landscape composition. This mapping process uses a data-driven, bio-
Historical range and variation
physical approach where georeferenced field data, biogeochemical simulation models, and
Landscape modeling
spatial data libraries are integrated using spatial statistical modeling to map environmen-
LANDFIRE
tal gradients that are then used to predict vegetation and fuels characteristics over space.
Fire management
These characteristics are then fed into a landscape fire and succession simulation model
Fire regimes
to simulate a time series of historical landscape compositions that are then compared to
the composition of current landscapes to compute departure, and the FRCC values. Intermediate products from this process are then used to create ancillary vegetation, fuels, and
fire regime layers that are useful in the eventual planning and implementation of fuel and
restoration treatments at local scales. The complex integration of varied ecological models
at different scales is described and problems encountered during the implementation of
this process in the LANDFIRE prototype project are addressed.
© 2007 Elsevier B.V. All rights reserved.
夽
The use of trade or firm names in this paper is for reader information and does not imply endorsement by the U.S. Department of
Agriculture of any product or service. This paper was written and prepared by U.S. Government employees on official time, and therefore
is in the public domain and not subject to copyright.
∗
Corresponding author. Tel.: +1 406 329 4846; fax: +1 406 329 4877.
E-mail addresses: rkeane@fs.fed.us (R.E. Keane), mrollins@fs.fed.us (M. Rollins), zhu@edcmail.cr.usgs.gov (Z.-L. Zhu).
1
Tel.: +1 406 329 4960; fax: +1 406 329 4877.
2
Tel.: +1 605 594 6131; fax: +1 605 594 6529.
0304-3800/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.ecolmodel.2007.02.005
486
1.
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Introduction
Wildland fire suppression activities over the last 70 years have
significantly reduced wildland fire occurrence in many historically fire-prone forests of the United States resulting in
excessive buildups of ground surface and canopy fuels which
may now tend to foster fires of abnormally high intensities
and severities (Mutch, 1994; Ferry et al., 1995; Kolb et al.,
1998; Keane et al., 2002c). The United States’ Healthy Forest
Restoration Act (http://www.fireplan.gov/) and the National
Fire Plan’s Cohesive Strategy (http://www.fs.gov/) establishes
a national commitment to reduce fire hazard and restore fire
to those ecosystems where it has been excluded for decades.
The implementation of this national fire management effort
will require extensive multi-scale spatial data for prioritizing
and planning fuel reduction and ecosystem restoration treatments across the entire nation (GAO, 2002, 2003, 2004). These
spatial data layers must provide essential fuel, fire regime,
and vegetation information critical for designing treatments
and activities at spatial scales compatible with effective land
management.
National fire management agencies have identified the
Fire Regime Condition Class (FRCC) as one of the primary
metrics used for distributing resources and prioritizing treatment areas for the implementation of the Cohesive Strategy
across the USA (Hann, 2004). FRCC is an ordinal index with
three categories that describe how far the current landscape
has departed from historical conditions (Hann, 2004) (see
http://www.frcc.gov/ for complete details). The FRCC’s primary use is to identify and prioritize those areas that are in
need of treatment so fire management can distribute funding, fire fighting resources, and personnel to those regions
with undesirable FRCC distributions to implement restoration activities and fuel reduction treatments (Laverty and
Williams, 2000) (see http://www.fireplan.gov for detail/). Ultimately, these treatments can protect homes, save lives, and
restore declining fire-adapted ecosystems. The challenge then
is to create a scientifically credible and ecologically meaningful national map of FRCC that can be used across multiple
organizational scales to distribute fire resources and prioritize
necessary fuel and restorative treatments.
This paper describes a research effort, called the LANDFIRE prototype project, that developed a process to map FRCC
using a complex integration of several ecological models. FRCC
was derived from a departure statistic computed from a comparison of current landscape conditions with simulated time
series of historical landscape composition. Maps of departure,
and many other vegetation, fire, and fuels attributes, were
created for two large test areas in the western US using a
data-driven, biophysical approach where georeferenced field
data, ecosystem simulation models, and spatial data libraries
were integrated to map environmental gradients that are then
used to predict current vegetation characteristics over space
(Franklin et al., 2000). These characteristics were then input
into a spatially explicit landscape fire and succession model
to simulate a time series of historical landscape compositions
that were then compared to the composition of current landscapes to compute departure, and eventually, FRCC values.
Intermediate products from this process were then used to
create additional fuels and fire regime layers that are use-
ful in the eventual planning and implementation of fuel and
restoration treatments at local scales.
1.1.
Background
The LANDFIRE prototype project found its origins in a Hardy
et al. (2001) project that created a set of coarse scale maps
of FRCC and other fire related variables for the contiguous
United States at 1 km2 spatial resolution (Schmidt et al., 2002)
(http://www.fs.fed.us/fire/fuelman). They used simplistic succession sequences pathways to represent possible historical
reference conditions for broad land cover types and tree density classes. Although these maps provided fire management
with a first-ever, broad picture of ecosystem condition across
the conterminous US, the layers had limited use because their
resolution and information content was too coarse and the
succession sequences were too general and subjective for use
at scales finer than a regional extent.
Fire management in the US needs a set of national data layers that are developed at a spatial resolution appropriate for
use across all government agencies and levels of management
(GAO, 2003, 2004) (selected as 30 m to match available satellite imagery). These layers need to be spatially consistent so
that historical departures are interpreted the same across the
country, and developed using the best available science so that
the process could be repeated to monitor success of fire management policies. It was also important that a set of additional
maps be created from the intermediate products of this effort
to aid fire managers in the implementation of the prioritized
fuel treatment and ecosystem restoration activities.
Since FRCC is based on the departure of the current conditions from historical conditions, we needed to find a concept
or construct to serve as the foundation of FRCC calculations.
A three category ordinal index seemed overly simplistic so we
thought that the computational foundation of FRCC needed to
accommodate additional detail so that additional resolution
could be employed to effectively prioritize across a wide range
of landscape sizes. We decided to use the notion of historical
range and variability (HRV) as the premise of all FRCC calculations instead of the qualitative assessment used in the coarse
scale effort (Landres et al., 1999; Swetnam et al., 1999; Keane
et al., 2002b). HRV is defined in this study as the quantification
of temporal and spatial fluctuations of ecological processes
and characteristics prior to western European-American settlement (Hann and Bunnell, 2001; White and Walker, 1997).
The assumption is that the quantification of HRV can serve
as an ecologically meaningful and objective reference for
computing the departure of current landscape compositions
from historical trends. To use HRV in an operational context,
it was necessary to select a historical time span that best
reflects the range of conditions to use as reference for the
management of future landscapes; an assumption that may
be overly simplistic because of documented climate change,
exotic introductions, and human land use (Hessburg et al.,
1999a, 2000; Swetnam et al., 1999). We limited the temporal
context of HRV to the time span from circa 1600 to 1900 ad
because we felt this period was best represented by the historical data, such as fire scar chronologies, that were needed to
quantify parameters in the landscape simulation model used
to generate HRV statistics.
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
We selected landscape composition as the primary evaluation criteria for computing FRCC (Hann and Bunnell, 2001).
Landscape composition is measured as the area occupied by
the combination of the following three attributes: (1) biophysical setting described by Potential Vegetation Type (PVT), (2)
vegetation composition described by cover types, and (3) vertical stand structure described by structural stage. We used
the concept of PVTs to map biophysical settings (areas of similar environmental conditions defined by biotic and abiotic
attributes) because it is the primary construct in the landscape
model used to simulate historical time series. The PVT concept approximates biophysical setting by assuming that the
“climax” vegetation that would eventually develop on an area
in the absence of disturbance can be used to identify unique
environmental conditions (Pfister et al., 1977; Daubenmire,
1962; Daubenmire, 1966). This approach has a long history in
vegetation mapping, and potential vegetation type classifications have been developed for most forests in the western U.S.
(Huschle and Hironaka, 1980; Pfister and Arno, 1980; Pfister,
1981). However, the PVT approach has had limited success
in non-forested environments because extensive disturbance
histories in rangelands has eliminated many climax indicator
species (Westoby, 1980).
We modified this potential vegetation concept to match the
scale and limitations of the LANDFIRE prototype mapping process. We assumed that the potential vegetation for forested
ecosystems could be keyed from plot-level tree data based
on tree species’ shade tolerance using the hypothesis that
the tree species with the highest shade tolerance will eventually, without disturbance, become dominant on that plot and
therefore will tend to have a high fidelity to a unique biophysical setting (Daubenmire, 1966). We made no inference that
the shade-tolerant species is the “climax” species since the
term climax has many limitations and misconceptions. One
common misconception is that the PVT classification is a vegetation classification; in this study, the PVT classification is a
biophysically based site classification that uses plant species
names as indicators of unique environmental conditions. And
because of succession modeling protocols described later, the
PVT classification categories had to match the existing cover
type classification categories.
The best sources for describing HRV of landscape composition are spatial chronosequences created from historical maps
or data. But, temporally deep chronosequences of historical
landscape conditions are scarce for most of the US, especially
prior to 1900 (White and Walker, 1997). Maps from many similar landscapes across a large geographic region (substitute
space for time) can also be used to quantify the HRV of landscape characteristics (Hessburg et al., 1999a,b). Unfortunately,
while these landscapes may appear biophysically similar, they
often have different environmental, disturbance, and biological histories (Swanson et al., 1997; Turner et al., 1994). We
decided on a simulation approach for quantifying HRV where
chronosequences of maps generated from a simulation model
are used to compute historical landscape statistics. Many spatially explicit ecosystem simulation models are available for
quantifying HRV (see Gardner et al., 1999; Mladenoff and Baker,
1999; Keane and Finney, 2003; Keane et al., 2004a), but most
are computationally intensive, difficult to parameterize and
initialize, and complex in design, thereby making them dif-
487
ficult to use in a national application (Keane et al., 2004a).
Simulation has an advantage over the previously mentioned
two sources in that it allows a spatially consistent method of
generating historical chronosequences over large regions by
efficiently integrating point and stand-level fire and vegetation
field data into a spatial context. This approach can also create
alternate time series reflecting different landscape histories,
such as climate change scenarios, so that a full complement of
range and variation statistics can be computed. The approach
can be easily replicated with new and improved spatial models
or with more accurate input parameters as they are measured
on the landscape.
2.
Methods
2.1.
Study areas
The LANDFIRE prototype project used two large areas in the
western United States – the Central Utah Highlands and
the Northern Rocky Mountains (Fig. 1) – to design, test, and
implement mapping methods. These landscapes were chosen because they represented a wide variety of vegetation
assemblages common the western Unites States and because
pre-processing of the Landsat 7 satellite imagery had been
completed as part of the Multi-Resolution Land Characterization project conducted primarily by the USGS in Sioux Falls
Fig. 1 – The two mapping zones used to prototype the
LANDFIRE mapping methods. (a) Zone 16 Central Utah
mapping zone and (b) zone 19 Northern Rockies mapping
zone.
488
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
South Dakota. We adopted the use of MRLC map zones for
spatially stratifying the United States into workable spatial
areas. We selected a small area within each mapping zone
to demonstrate the resolution of LANDFIRE products for this
paper.
The 69,907 km2 Central Utah Highlands mapping zone
begins at the northern tip of the Wasatch Mountains in Southern Idaho and extends through central Utah to the southern
border of Utah. Elevations range from 980 to 3750 m. Vegetation communities range from alpine forb communities in
the Uinta and Wasatch Mountains in the northern portion of
the map zone to desert scrub communities in the southern
deserts. Extensive areas of Piñon-Juniper/Mountain big sagebrush and both evergreen and deciduous shrub communities
are found at mid-elevations throughout the Central Utah Highlands mapping zone. Climate of the Central Utah Highlands
mapping zone is highly variable with 30-year average temperatures range from −4 ◦ C in the high Uinta Mountains to 15 ◦ C in
the southern deserts and average annual precipitation varying
from 10 cm in the southwestern deserts to nearly 2 m in the
northern mountains.
The 117,976 km2 Northern Rockies mapping zone begins
in at the Canadian border in northern Montana and extends
south into eastern Idaho. Elevations range from 760 to
3400 m. Vegetation communities range from alpine forbs in
the highest mountain ranges to prairie grasslands east of
the Continental Divide. Forested communities are common
with spruce-fir communities found at the upper subalpine
and extensive forests of lodgepole pine, western larch,
Douglas-fir and ponderosa pine at middle elevations. Thirtyyear average temperatures range from −5 ◦ C in the high
Mountains of Glacier National Park to 15 ◦ C in the valley
bottoms. Annual average precipitation varies from 14 cm
in the valley bottoms to nearly 3.5 m in the northern
mountains.
2.2.
Fig. 2 – Flow chart of the process used to create the FRCC
map and all other ancillary data layers for the LANDFIRE
prototype project. The acronyms are defined as follows:
DEM: digital elevation model, PVT: potential vegetation
type, PET: potential evapotranspiration, NPP: net primary
productivity (kg m−2 ), GIS AML: geographic information
systems macro language, PPT: precipitation (cm). WXFIRE
and LF-BGC are ecosystem models while LANDSUM is a
landscape dynamics model and HRVSTAT is a statistical
model.
General description of modeling effort
As mentioned, the computation of FRCC requires a comparison of current conditions with a time series of historical
reference conditions. In this study, current conditions were
described from maps of PVT, cover type, and structural
stage (Fig. 2). The PVT map was created using predictive
landscape modeling where biophysical gradients created
from mechanistic ecosystem simulation computer models
were employed as predictor variables. The current cover
type and structural stage maps were derived from a supervised classification of Landsat 7 ETM+ imagery and several
of the simulated biophysical layers. Historical conditions
were described from a simulated time series of landscape
composition generated from a spatially explicit landscape
dynamics model that used the PVT maps as input. FRCC
was quantified from a departure metric that was computed by statistically comparing the simulated historical time
series of landscape composition with the current conditions for each combination of PVT, cover type, and structural
stage.
Since the mapping protocols needed to be scientifically
credible and repeatable with a minimum of subjectivity, we
based all methods on a data-driven, empirical approach where
the majority of mapped and simulated entities were predicted
from complex spatial statistical modeling. To accomplish
this, we created the LANDFIRE reference database (Fig. 2)
that contains georeferenced data from thousands of plots
obtained from a variety of sources, most importantly, the
USDA Forest Service Forest Inventory and Analysis (FIA) program. These data were used for many purposes including
(1) developing training sites for imagery classification; (2)
parameterizing, validating, and testing simulation models; (3)
developing vegetation classifications; (4) creating statistical
models; (5) determining data layer attributes; (6) describing
mapped categories; (7) assessing the accuracy of maps, models, and classifications (Franklin et al., 2000). The LANDFIRE
reference database is the heart of the mapping process and
the source of nearly all LANDFIRE products.
All methods and procedures implemented for this study
are extensively detailed in a technical report edited by Rollins
and Frame (2006) (see Keane and Rollins, 2006). In this paper,
we present a general summary of these methods emphasizing the use of ecological modeling. A flow chart in Fig. 2 is
included as reference. All satellite image processing was done
at the EROS Data Center (EDC) in Sioux Falls, South Dakota,
USA while all other modeling and classification tasks were
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
accomplished at the Missoula Fire Sciences Laboratory (MFSL)
in Missoula, MT, USA.
2.3.
Describing current conditions
2.3.1.
Developing the map unit classifications
Previous attempts at mapping vegetation across large regions
(Keane et al., 1996; Hann et al., 1997; Schmidt et al., 2002)
revealed that it was difficult to fit most existing vegetation classifications into a large scale mapping effort because
existing classifications rarely match the resolution and the
predictive ability of the satellite imagery and there are often an
insufficient number of classification categories to describe the
diversity of vegetation across large regions. Most existing classifications are limited in application to unique ecosystems or
geographic areas, and many of the existing national vegetation
classifications were developed for description rather than for
mapping. The one possible exception was the National Vegetation Classification System (NVCS) (Anderson et al., 1998;
Grossman et al., 1998), but we found that (1) it was difficult to scale the NVCS categories to the detail required by
the HRV succession model, (2) it was difficult to key LANDFIRE reference database plots to the NVCS types, and (3)
it was difficult to use the system because it was not fully
developed.
We created our own vegetation composition (cover type)
classification using a top-down approach where our categories were designed to be (1) relevant to management,
(2) conducive to modeling, (3) discriminated from satellite
imagery, and (4) identifiable from plot data. Rather than
construct a system that exclusively classified all vegetation
types within the entire U.S., we decided to first identify
those vegetation types that we could successfully discriminate with the satellite imagery and identify in the reference
database. We developed classification categories for forests
and rangelands based on a blend of many national efforts
(Holdridge, 1947; Kuchler, 1975; Eyre, 1980; Running et al.,
1994; Shiflet, 1994; Bailey, 1995; Grossman et al., 1998) and
synthesized a list of vegetation map units within the context of the two prototype mapping zones. This list was
reviewed for each study area by conducting a series of informal workshops between project technical personnel and local
ecologists.
We based the vegetation composition classification on the
canopy cover of the dominant species and called the classes
cover types (Long et al., 2006b). We then created a set of
keys to identify cover type for each plot in the LANDFIRE reference database based on the dominance of plant cover by
species. The set of cover types were compared to available
field data and those types represented by less than 50 plots
were dropped. Remaining cover types were tested using mapping classifiers in an iterative process for their separability
in spectral and biophysical domains and were additionally
evaluated to ensure they fit within the landscape modeling
framework. This process continued until we finally had mutually agreed upon a final cover type list for each prototype area.
We cross-referenced the LANDFIRE vegetation classification
with all other national classifications to provide linkages to
other mapping efforts.
489
The structural stage classification proved somewhat easier
but contained far less detail than the cover type classification because of the inability of the imagery to detect complex
vertical canopy structure. Structure is important to define
pathways of successional development in the landscape modeling effort. We decided to base our structure classification on
two structural components that we could discriminate from
image analysis: canopy cover and average height. These two
attributes are mapped using a statistical modeling approach
where cover and height are regressed against spectral values and other ancillary data layers (Rollins and Frame, 2006).
We constructed a four-category forest structural classification
where there were two categories of height (short, tall) and two
categories of cover (low, high) and the thresholds that delineate between each structural category is based on the PVT
and cover type and calculated from the reference database.
Only two structural stage categories were used to describe
shrubland and grassland structure: low and high canopy
cover.
2.3.2.
Simulating biophysical gradients
Many studies have shown that augmenting satellite imagery
with a quantitative description of the biophysical environment (temperature, elevation, precipitation, for example)
improved the mapping of ecological characteristics, such as
vegetation and fuels (Franklin et al., 1986, 2000; Keane et al.,
2002a; Ohmann, 1996; Ohmann and Spies, 1998; Rollins et
al., 2004). Environmental gradients are important for mapping
vegetation and ecosystem characteristics because they define
potential species niches by describing the biophysical environment. Indirect gradients, such as elevation, were included
in the biophysical mapping effort because they can be proxies
for other unknown or immeasurable gradients. The addition
of biophysical layers to predict ecosystem attributes increased
mapping accuracy by over 20% in some cases (Rollins et al.,
2004; Cibula and Nyquist, 1987).
In the mid 1990’s, we developed a mapping process called
Landscape Ecosystem Inventory System (LEIS) that used data
from local weather stations, coupled with various input spatial data layers, to simulate and summarize a wide variety of
climate and ecosystem processes (Keane et al., 2002a; Rollins
et al., 2004). This system used a collection of computer models, including the BGC biogeochemical model (Running and
Coughlan, 1988; Running and Gower, 1991; Thornton and
White, 1996; White et al., 2000; Thornton et al., 2002), to create
a set of biophysical layers that were then used to map fuels
and vegetation. We redesigned two of the major programs in
LEIS to access the 1 km national DAYMET weather database
built by Thornton et al. (1997) (see http://www.daymet.gov/ for
details) so that simulated environmental gradient data layers
would be nationally consistent and comprehensive. DAYMET
is a computer program that generates daily spatial surfaces
of temperature, precipitation, humidity, and radiation over
the complex terrain of the conterminous US by extrapolating weather data from over 1500 weather stations across a
1 km grid using a Gaussian kernel (Running and Thornton,
1996; Thornton et al., 1997, 2000). The DAYMET program was
executed for 18 years of weather data (1980–1997) across the
contiguous US to create the DAYMET weather database that
contains daily estimates of minimum and maximum tempera-
490
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
ture, precipitation, radiation, and vapor pressure deficit across
a 1 km grid. The 18-year period has the highest number of
weather stations in mountainous terrain with SNOWTEL and
RAWS weather records becoming common.
We modified the WXGMRS program in LEIS to create the
WXFIRE program and modified the Biome-BGC program to
create the LF-BGC program (Fig. 2). The WXFIRE program
scales DAYMET weather data to finer spatial resolutions using
biophysical principles and summarizes weather into important climate descriptors, such as potential evapotranspiration,
over the 18-year record (Keane et al., 2006a; Holsinger et al.,
2006). The LF-BGC program simulates ecosystem processes
at a daily time step to compute carbon, water, and nitrogen
fluxes and outputs summaries of these fluxes for the 18-year
DAYMET record. These two programs provided the ability to
create consistent and comprehensive biophysical data layers
for LANDFIRE mapping and modeling tasks, especially for creating the PVT map.
2.3.3.
Creating the PVT map
We keyed the PVT for each forested plot in the reference
database based on the presence (>1% cover) of the most shade
tolerant tree species from canopy cover or tree density data
measured for that plot. All tree species present on a plot were
sorted by shade tolerance using the information in the literature (Fowells, 1965; Minore, 1979; Burns and Honkala, 1990).
We arranged the rangeland cover type categories along a moisture gradient from xeric to mesic communities and keyed
plots in the reference database on species presence, rather
than dominance. Although this approach for rangelands had a
number of flaws (Westoby, 1980), most importantly the inability to consistently model successional development, it proved
to be the best considering the limited resources and data
available.
Maps of PVT were created using an integrated landscape
modeling and statistical analysis approach (Franklin, 1995;
Franklin et al., 2000; Ohmann and Spies, 1998; Rollins et al.,
2004). Classification trees were used to create the statistical
models for mapping PVT from the simulated environmental gradients. The PVT classification trees were developed
using the See5 machine-learning algorithm (Quinlan, 1993,
1986, http://www.rulequest.com/) and were applied within
an ERDAS Imagine interface. See5 uses a Classification and
Regression Tree (CART) approach for constructing a tree, generating a tree with high complexity and pruning it back to a
more simple tree by merging classes (Breiman et al., 1984).
We selected CART-based classification methods because, as a
non-parametric alternative for regression, CART may be more
appropriate for broad scale mapping than parametric methods
(De’ath, 2002; Iverson and Prasad, 1998; Michaelsen et al., 1994)
and CART-based statistical models may be trained hundreds
of times faster than some other non-parametric classifiers
like neural networks and support vector machines (Huang et
al., 2001; Huang and Townshend, 2003; Moisen and Frescino,
2002), yet CART is comparable to and performs similarly with
regard to accuracy to these methods (Friedl and Brodley, 1997;
Huang et al., 2001; Franklin, 2003; McDonald and Urban, 2006;
Moisen and Frescino, 2002). Last, CART has been successfully
used for modeling and mapping vegetation at broad scales as
in the MRLC 2001 project so the EROS Data Center had exten-
sive expertise in this method (Homer et al., 2002; Huang et al.,
2001).
2.4.
Creating cover type and structural stage maps
The mapping of current vegetation conditions using the cover
type and structural stage vegetation classification as map
units was accomplished using Landsat 7 TM satellite imagery
and the set of simulated biophysical data layers. Landscape
metrics (patch shape and size) and the PVT layer were also
used to guide cover type and structural stage mapping. Many
classification algorithms have been developed for deriving
vegetation cover type from satellite imagery (Knick et al., 1997;
Homer et al., 1997; Cihlar, 2000). In this study, we created cover
type maps using a training database developed from the reference database, Landsat TM 7 satellite imagery, biophysical
gradient layers, the PVT map (for limiting the types of vegetation that may exist in any area), and classification tree
(CART) algorithms similar to those used to map PVT. Landsat TM 7 images were acquired on three different dates over
the time period between 1999 and 2001 to capture growing
season dynamics to maximize the discrimination between
cover types (Yang et al., 2001). Raw satellite digital numbers
were converted to at-satellite reflectance for the six Landsat TM 7 bands according to Markham and Barker (1985) and
the Landsat TM 7 Science Data User’s Handbook (Irish, 2000).
We used at-satellite reflectance based coefficients to calculate
Tasseled-cap brightness, greenness and wetness, which have
been found useful for vegetation characterization (Cohen et
al., 1998; Huang et al., 2002).
The structural stage map was developed from maps of
canopy closure and canopy height that were created using
satellite imagery coupled with output from physically based
models, spectral mixture models, and empirical models. EDC
used regression trees in a CART framework, applied through a
ERDAS Imagine interface, to map the relationships between
canopy closure and height and spectral information from
Landsat TM 7 imagery, biophysical gradient layers, and the
reference database (Zhu and Evans, 1994). The resultant
maps represented these two structural variables continuously
mapped across each prototype mapping zone (Huang et al.,
2001). The final structural stage layer was created by keying
the continuous canopy closure and height layers to structural
stage categories based on threshold values assigned to each
PVT based on an analysis of the species data in the LANDFIRE
reference database. This integrated height and closure information was also used to compute wildland fuel characteristics
and successional status of exiting landscapes. The cover type
and structural stage classes formed the structural framework
for the vegetation modeling described in the next section
2.5.
Describing historical conditions
2.5.1.
The simulation model
The simulation model selected to quantify the HRV of landscape composition for the LANDFIRE prototype effort is in a
class of models called Landscape Fire Succession Models that
are unique in that they spatially simulate the complex process
of fire and vegetation succession across landscapes (Keane et
al., 2004a). We researched many of these models to determine
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
491
Fig. 3 – An example of a spruce-fir (SF) succession pathway model used in LANDSUMv4 for the LANDFIRE prototype project.
Cover type names are as follows: SH: mountain shrub, WP: whitebark pine, SF: subalpine fir. Structural stage names are
defined as—LCLH: low cover low height early succession stage, HCLH: high cover low height mid-seral stage, HCHH: high
cover high height late succession stage, LCHH: low cover high height disturbance maintained late succession stage. T1–11
identify unique succession classes. Cover type categories are linked along pathways of successional development based on
shade tolerance. The terminal cover type (SF: spruce-fir HCHH) is potential vegetation type (PVT) category because it
represents the most shade tolerant tree species.
if which would be appropriate for LANDFIRE implementation and found most were inadequate for a variety of reasons
including (1) extensive input data requirements, (2) difficult
parameterization, and (3) excessive computation demands.
Once we identified the limited data sources available for model
parameterization and initialization (i.e., LANDFIRE reference
database and subsequent map products) and evaluated the
landscape models in this context, we found that the LANDSUM
model appeared to have the best fit, primarily because it stratifies landscapes by biophysical settings and it has minimal
input requirements (Keane et al., 1996, 1997a,b, 2002b).
LANDSUMv4 is the fourth version of the LANDSUM model
developed specifically for the LANDFIRE prototype project
(Keane and Lisa, 2006a) (Fig. 2). It contains a deterministic simulation of vegetation dynamics where successional
communities are linked along multiple pathways of development that converge in an end-point community (Fig. 3).
All disturbances, except fire, are stochastically modeled at
the stand-level from probabilities specified by the user. Fire
ignition is computed from input fire frequency probabilities
(fire parameters in Fig. 2) specified by PVT, cover type, and
structural stage categories. Fire is spread across the landscape based on simplistic slope and wind factors. LANDSUM
does not mechanistically simulate fire growth because of the
lack of fuel and daily weather inputs and limited computing
resources for national simulations. The effects of simulated
fires are stochastically simulated based on the fire severity
types as specified in disturbance input files (Keane and Lisa,
2006a). Finally, LANDSUM outputs the area occupied by PVT-
cover type-structural stage combinations within small areas
called landscape reporting units (described later).
2.5.2.
Simulation design
We divided the two mapping zones (Fig. 1) into square
20,000 ha core areas (471 × 471 pixels of 30 m width) and then
surrounded these core areas by a 3 km buffer (Fig. 4) based on
an analysis of results from numerous simulations (Pratt et al.,
2006a,b). The combined core and buffer area was called the
simulation landscape. We divided the core area into 900 m by
900 m squares called landscape reporting units that defined
the spatial extent of the comparison of historical to current
landscape composition conditions units (Fig. 4). Landscape
reporting unit size was selected to correspond to the coarse
scale analysis (Hardy et al., 2001; Schmidt et al., 2002) and to
be applicable to land managers. Simulated landscape composition (area by PVT-cover type-structural stage combinations)
and total area burned were summarized for these 0.81 km2
landscape reporting units every 50 years for 5000 years of simulation for zone 16 and 10,000 years for zone 19. The time
interval and span were selected based on results of a sensitivity analyses that showed that reporting intervals less than 50
years contained significant temporal autocorrelation and that
reporting time spans of less than 5000 years did not generate
enough observations (>100) to obtain a statistically significant
sample for computing departure (Keane et al., 2002a,b,c; Steele
et al., 2006). We initialized simulation landscapes using the
most dominant cover type-structural stage combination for
each PVT on the landscape and then ran the model for 500
492
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Fig. 4 – An illustration of simulation landscapes with the core area surrounded by a buffer zone. The interior area was then
stratified into landscape reporting units of 900 m × 900 m where landscape composition was summarized and compared
across historical and current conditions.
years to remove any initial effects. Current conditions were
summarized for each landscape reporting unit using the PVT,
cover type, and structural stage maps derived from satellite
imagery discussed previously.
LANDSUMv4 succession and disturbance parameters were
quantified by a team of ecologists who used the reference
database, information from published studies, and expert
knowledge (in that order) to describe and quantify the successional pathways (see Fig. 2) for each PVT (Long et al., 2006a;
Pratt et al., 2006a,b). This team built succession pathway
models using an interactive model called VDDT: Vegetation
Dynamics Development Tool (Beukema and Kurz, 1998; Kurz
et al., 1999) that allowed the testing and refinement of input
parameter sets in a non-spatial domain. Fire frequency parameters were taken from fire history studies conducted within
the two prototype areas (Heyerdahl et al., 1995). Fire size
parameters were computed from the NIFMID database as
compiled by Schmidt et al. (2002). Successional development
parameters were also taken from other modeling efforts or
estimated by analyses of data from succession studies (Stage,
1997).
2.6.
Computing departure and FRCC indices
The idea that spatially explicit historical ecological data can be
used as reference to guide land management is somewhat new
and only recently has been put into practice (Reed et al., 1998;
Hessburg et al., 1999b; Landres et al., 1999; Keane et al., 2002b;
Swetnam et al., 1999). Although the theory appears somewhat
sound, its implementation into land management is still in its
infancy so it has not been critically evaluated. There are few
standardized methods for quantitatively comparing historical
time series with today’s landscapes and objectively evaluating their differences. Standard parametric statistics may not
be appropriate for historical time series since the sequences
may be autocorrelated in time and space. Also, landscape map
units are correlated to each other because a decrease in the
area of one map unit will result in increases in others. It is also
difficult to ascertain the thresholds of significant difference
that determine whether human intervention using fuel treatments is warranted. Because of this, we decided to develop an
alternative statistically based method for comparing the simulated historical chronosequences with the imagery derived
current conditions. We also calculated departure and FRCC
using the FRCC field methods detailed by Hann (2004), which
use a simple measure of similarity to calculate departure.
The statistical methods for computing departure from historical conditions were developed by Steele et al. (2006) using
a statistical technique that is sensitive to the autocorrelation
of chronosequential data over time, space, and mapped categories (see HRVSTAT in Fig. 2). This technique, implemented
in the HRVSTAT program, computes an index of departure
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
that varies from zero (no departure) to one (most departed)
and a measure of statistical significance computed as a probability value (p-value). HRVSTAT was designed to work with
any chronosequence, not just the simulated time series, so
that local land management offices can replace the simulated
data with actual historical observations if they exist. FRCC
was calculated by stratifying departure and probability value
statistics into three classes based on an analysis of the frequency and range of values across the mapping zone (Pratt et
al., 2006a,b; Steele et al., 2006).
2.7.
Developing ancillary data products
The biophysical, vegetation, and historical data layers
developed from the above modeling process constitute a comprehensive set of spatial data sources for describing vegetation
characteristics and for use as input to simulation models
across the entire conterminous USA. However, the utility of
these data layers is limited because there are few applications
that directly use these layers in fire management. Similarly,
while the FRCC map can be used for the prioritization and
distribution of funding and resources to land management
agencies and regions, it has little use in designing and implementing fuel and ecosystem restoration treatments. It became
obvious that additional data layers and software tools were
needed to provide local land management with applications
that were useful for planning and implementing treatments
and were consistent with other LANDFIRE products.
2.7.1.
Fuels layers
Both surface and canopy fuels were mapped in the context
of fire behavior and fire effects modeling to ensure that the
spatial data could be used in many fire hazard analysis applications. The most commonly used fire behavior simulation
models require fuel models where the loadings have been
adjusted to predict expected fire behavior under common
weather scenarios (Anderson, 1982 standard 13 fire behavior fuel models). These fire behavior fuel models are quite
broad and often don’t match the resolution needed to detect
subtle changes in fuel characteristics after fuel treatments.
As a result, a new set of 40 fire behavior fuel models using
funding from LANDFIRE (Scott and Burgan, 2005), which represents a significant improvement in detail and resolution
over 13 of Anderson (1982). These new fuel models are now
implemented in common fire behavior applications such as
BEHAVE (Andrews, 1986; Andrews and Bevins, 1999) and FARSITE (Finney, 1998).
Fire effects fuel models differ from fire behavior fuel models
in that they represent real fuel loadings by fuel category, not
abstract representations of expected fire behavior. The loading
is used to calculate important fire effects such as fuel consumption, soil heating, smoke, and tree mortality (Reinhardt
et al., 1997). At the start of this study there were no national
classifications of fuel loadings so we created our own classification called Fuel Loading Models (FLM) developed from
a clustering analysis of the variance of fuel loadings by fuel
category using an extensive dataset collected on plots from
sampling efforts across the nation (Lutes et al., 2006). Sampled fuel loading was stratified by several fuel components:
four downed woody size classes, shrub, herbaceous, duff and
493
litter (Brown and See, 1981; Brown and Bevins, 1986). FLMs
match the scale of development specified in the LANDFIRE
guidelines and were designed for use in fire effects models
such as FOFEM (Reinhardt et al., 1997; Reinhardt and Keane,
1998) and CONSUME (Ottmar et al., 1993).
The three surface fuel model classifications (standard 13,
new 43, and FLMs) were mapped for each zone using a rulebased approach where each PVT-cover type-structural stage
combination was assigned classes from each of the three fuel
model classifications (Keane et al., 2006b). Assignments were
based on an analysis of the meager fuels data in the reference
database, but there were insufficient data for most combinations and fuel model classes so we were forced to assign the
fuel models based on the expert opinion of the authors of each
classification (Scott and Burgan, 2005; Lutes et al., 2006).
Most fire behavior and effects predictive models require
canopy characteristics to simulate crown fire initiation and
propagation (Finney, 1998; Rothermel, 1991; van Wagner, 1993).
These canopy characteristics include bulk density, height,
base height, and closure. Fortunately, EDC mapped canopy
height and closure using satellite imagery, but canopy bulk
density and canopy base height had to be mapped by the MFSL
using a statistical modeling approach similar to that used for
the PVT map. We did this by calculating canopy bulk density and base height from the tree inventory information in
the reference database using the FUELCALC (FUEL CALCulation system) program. This program uses tree measurements
of species, height, and diameter to derive the vertical distribution of crown biomass from a set of canopy biomass equations
this is then used to compute the canopy base height and bulk
density (Reinhardt and Crookston, 2003). These two canopy
characteristics were computed for each plot containing trees
and then statistically modeled from variables represented in
all the biophysical layers using the CART statistical analyses. FUELCALC is still under development but will be released
before the LANDFIRE project is finished (www.landfire.gov).
3.
Results
The final LANDFIRE reference database for the two prototype
areas contained vegetation and fuels information from over
10,000 georeferenced plots with over 75% of these were FIA
plots. Plots were well distributed across the two map zones
(Fig. 5). We evaluated over 20,000 legacy plots for inclusion
into the reference database and found only about 10% met
our stringent criteria.
The important intermediate and final map layers arranged
along the procedural paths of the simulation process are
shown in Fig. 6 for zone 16 leading to the simulation of
the historical time series. Limited space precludes a detailed
description of all intermediate map products here but all layers are comprehensively described in Rollins and Frame (2006)
and posted to http://www.landfire.gov/. Table 1 presents the
results of an accuracy assessment of the three most important
layers used to simulate historical time series and calculate
departures along with accuracies, calculated using regression
analysis, for the two important canopy fuels layers that had
sufficient geo-referenced plot data (FIA plots) for a statistically
valid comparison (Vogelmann et al., 2006). Although the seven
494
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Fig. 5 – The distribution of georeferenced plots in the LANDFIRE reference database over the two map zones: (a) zone 16 and
(b) zone 19.
fuels layers in Fig. 6 were not used in the simulation of the historical time series, they will be important in the planning and
implementation of fuel treatment activities and ecosystem
restoration.
The modeling and mapping process used to compute FRCC
from simulated historical time series is illustrated in Fig. 7
with the developed intermediate products. Areas of each PVTcover type-structural stage combination was simulated within
simulation landscapes and reported within each 900 m by
900 m landscape reporting unit (Fig. 4) for each output reporting interval. The historical time series was used as reference to
compute departure from current conditions and the departure
Fig. 6 – A general illustration of the important intermediate and final map products created by the LANDFIRE process to
ultimately compute departure and FRCC from the simulated historical time series.
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
495
Fig. 7 – The synthesis of historical landscape composition time series from the simulated data for the landscape reporting
unit: The simulation landscape is stratified into 0.81 km2 square areas called landscape reporting units and the
LANDSUMv4 simulation results at 50 year intervals for 5,000 years are used to compute FRCC.
496
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Table 1 – Accuracy assessment of the three critical map data layer products (categorical maps) developed by the
LANDFIRE prototype project that were used to simulate historical time series to compare with current conditions to
determine departure and eventually FRCC
Map layer
Categorical maps
PVT
Created from
Biophysical layers
Cover type
PVT, satellite imagery
Structural stage
Canopy cover and canopy height from
imagery
Continuous maps
Canopy bulk density
Biophysical layers, cover type, imagery
Canopy base height
Biophysical layers, cover type, imagery
Used for
Mapping current vegetation; landscape
simulation modeling, fire regime
calculation
Describing current conditions and
computing departure and FRCC
Describing current conditions and
computing departure and FRCC
Modeling fire behavior, growth, and
spread, predicting smoke, calculating
carbon stores
Modeling crown fire behavior
Map accuracya
Z16
Z19
61.2
58.4
63.0
62.6
79.3
79.0
76.0
66.0
63.0
38.0
Accuracies for two canopy fuels layers (continuous maps) are provided for reference. A full accuracy assessment is presented in Rollins and
Frame (2006) and detailed in Vogelmann et al. (2006).
a
Map accuracy estimates represent agreement between mapped values on the final LANDFIRE maps and measured values for the same pixel
using geo-referenced field data. For categorical maps, agreement is measured using standard cross validation analysis and contingency tables
so the number represents a percent agreement of mapped value to plot value. For continuous maps, the agreement is measured as a coefficient
of determination (R2 ) calculated by regressing predicted mapped values to observed values calculated for the plot.
(number between zero and one) was divided into categories to
create the FRCC map using HRVSTAT output and the FRCC map
using procedures specified in Hann (2004).
An important end product of the LANDSUMv4 simulations
is the creation of a set of four fire historical regime maps compiled from the simulations (shown for zone 19 in Fig. 8) (Keane
et al., 2003, 2004b). The fire frequency map represented by
the mean fire return interval (Fig. 8d) was computed from the
number of fires experienced by each pixel, while the three fire
severity maps represent the simulated probability of a stand
replacement fire, a mixed severity fire, and a non-lethal surface fire, respectively (Fig. 8a–c). These maps provide a useful
reference for locating those areas having frequent fire so that
fuel treatments can be designed (using severity maps), located,
and scheduled.
4.
Discussion
The methods presented here demonstrate the value of ecological modeling to land management. A complex integration
of three modeling approaches (biogeochemical, statistical,
and landscape dynamics modeling) were used to successfully
derived a useful time series of historical landscape composition that was then used to compute departure of current
landscape conditions from the historical landscape conditions
and ultimately to estimate FRCCs across a large region at a spatial resolution useful to land management. FRCCs can then be
used to prioritize regions, National Forests, and landscapes for
fuel treatment and ecological restoration (GAO, 2003). Results
from this prototyping effort have been refined, revised, and
implemented in comprehensive national mapping effort to
create maps critical for fire management for the entire United
States at 30 m resolution (http://www.landfire.gov/). This inte-
grated ecological modeling approach is perhaps the only
feasible means to provide consistent, comprehensive, and
temporally deep historical record across large regions to evaluate whether today’s landscapes are within historical bounds.
Moreover, this approach can be used at many scales and resolutions as the landscape reporting unit is adjusted to reflect
the scale of the application, and it can use landscape attributes
other than vegetation composition to compute departure such
as fuel model composition, landscape pattern, and canopy
bulk density. Last, the approach can be repeated so that landscapes can be monitored for changes in FRCC, departure, or
fuel conditions to evaluate the efficacy of the National Fire
Plan.
Limited geo-referenced field data across all mapped categories proved to be the most critical problem throughout the
project. The FIA data represents an important source of forest plot data but there were few FIA plots in rangelands and
non-forest types. Most non-FIA legacy georeferenced field data
compiled from government and university sources were sampled with such a wide variety of methods and resolutions
that they were useless for statistical analysis and vegetation
classification. Some cover types important to management
did not have sufficient plots to warrant their mapping while
others were over-represented in the reference database. And,
some landscape succession model parameters and pathways
needed to be estimated from expert opinion because of the
lack of pertinent fire and succession data. The lack of current, geo-referenced field data also precluded an extensive
assessment of the accuracy of many mapped layers, especially
the fuels layers (Table 1) (Vogelmann et al., 2006). We needed
most of the plots in the LANDFIRE reference database to build
the empirical CART models so very few plots could be set
aside to estimate map accuracy. While somewhat low (60–80%
for the three critical data layers), these accuracies could eas-
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
497
Fig. 8 – The fire regime maps produced by the LANDSUMv4 simulation model for mapping zone 16 Northern Rockies. Maps
show the percent of total fires that are (a) non-lethal surface fires, (b) mixed severity fires, and (c) stand-replacement fires.
The mean fire return interval (MFRI) is shown in the last map (d).
ily be improved with additional field data (Vogelmann et al.,
2006). The success of the LANDFIRE national project, and other
future national mapping efforts, will ultimately depend on the
availability, quality, and quantity of georeferenced field data,
especially surface and canopy fuels data which were lacking
in this project.
Despite our best efforts, some elements of subjectivity
crept into the set of methods described here. The cover type
layer probably contains the highest bias because the vegetation map unit classification was derived from an integration
of expert opinion and data analysis. It was difficult to create a comprehensive list of cover types that matched the
sensitivity of the satellite imagery and simulation models
without some subjectivity because of the wide variety of cover
type classifications and vegetation communities possible for
each individual mapping zone and the inherent variability of vegetation across the mapping landscape. A possible
solution would be to use an existing cover type classification for the entire US so that bias is minimized and the
list of cover types remains constant across mapping zones;
the LANDFIRE national effort (http://www.landfire.gov/) has
decided to use the NCVS (Anderson et al., 1998) as the basis
for developing map units for creating layers of vegetation
composition.
Development of the fuels layers presented a special problem because mapped fuel model classifications are somewhat
new to management so fuel models were never assessed
on many geo-referenced plots in the LANDFIRE reference
database (Keane et al., 2006b). This meant that we could
not map fuel models with the statistical approach that was
used to create the other maps (i.e., CART), so we were forced
to use a more subjective, rule-based strategy where fuel
model categories were assigned to combinations of PVT-cover
type-structural stage categories. Most mapped fuel model
classifications also lacked a dichotomous key to consistently
identify fuel model from commonly collected field data. Future
efforts should attempt to derive a key that objectively assigns
fuel models based on the data collected in the field or present
in the reference database.
The landscape reporting unit size (0.81 km2 ) was probably
too small to adequately describe fire and vegetation dynamics.
Subsequent analyses found that reporting unit sizes ranging
from 15 to 30 km2 appear to capture sufficient variation in both
burned area and vegetation composition to produce ecologi-
498
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
cally meaningful historical time series (Karau and Keane, in
press). We also found that the landscape reporting unit size
should probably be variable and sized to fit the fire and disturbance dynamics of each mapping region or landscape. For
example, flat areas with frequent fire may need larger reporting units to effectively describe the high variation of fire size
and vegetation diversity.
There were some major limitations in the calculation of
the departure statistic and ultimately the quantification of
FRCC. The ranges used to create the FRCC index from classes
of the departure statistic were somewhat arbitrary and often
misleading. Some landscapes have such great variability in
historical landscape composition that departure statistics
rarely exceeded a value of 50 of 100 and the HRVSTAT probability values rarely exceeded 0.2. This resulted in maps where
most areas were in one or two FRCC classes and the ranges
needed to be adjusted to get adequate discrimination across
the mapping zone. The most extreme departures occurred on
landscape reporting units that contained a high proportion of
exotic plant cover types because exotic types were absent in
the historical time series. As a result, the high departures and
high FRCCs did not reflect the lack of fire from the landscape
and the need for fuel treatments, but rather the replacement
of native species with exotics. Some ecosystems experienced
changes in species composition resulting in high departures
but there were few changes in fuel characteristics indicating
no change in fire hazard. For these reasons, FRCCs are most
useful as an ecological measure of departure from historical
vegetation conditions rather than an index of fire hazard or
risk.
While the FRCC index has problems, such as a limited
three category index and spatial scale limitations, the departure statistic, on which FRCC is based, seems to have value
as a comprehensive index to determine the ecological status
of landscapes. Moreover, the idea of using simulated historical range and variation to drive assessments of ecosystem
and landscape condition appears to have great value to land
management (White and Walker, 1997). Wimberly et al. (2000)
uses a similar simulation approach to quantify the historical
variation in landscape composition to quantify optimal old
growth forest levels. The HRV of landscape composition and
structure were simulated for an Oregon Coast Range landscape using the LADS model (Nonaka and Spies, 2005). The
LANDIS model was used to simulate time series of fuel loads
(Shang et al., 2004) and fires (Yang et al., 2004) under various management scenarios. Hummel and Cunningham (2006)
used the FVS model to estimate variation in forest structure for a landscape in the Cascade Range in Washington,
USA, while Boychuk and Perera (1997) used the FLAP-X model
to simulate variability in age classes for Canadian boreal
forests. Many spatially explicit ecosystem simulation models are available for quantifying HRV in patch dynamics (see
reviews by Gardner et al., 1999; Keane et al., 2004b; Mladenoff
and Baker, 1999).
Acknowledgements
We thank Lisa Holsinger, Bruce Jeske, Eva Karau, Alisa Keyser,
Kris Lee, Don Long, Jim Menakis, Melanie Miller, Maureen
Mislevits, Russ Parsons, Sarah Pratt, Matt Reeves, Kevin Ryan
and Dennis Simmerman, USDA Forest Service Rocky Mountain
Research Station; Wendel Hann and Rich Lasko, USDA Forest
Service Fire and Aviation Management; Tracey Frescino, USDA
Forest Service RMRS FIA; Carl Skinner, USDA Forest Service
Pacific Southwest Research Station; Don Bedunah, University
of Montana; and John Caratti, Christine Frame, Larry Gangi,
Jennifer Key, Jack Losensky, Scott Mincemoyer, Karen Short,
and Amy Vitateau, Systems for Environmental Management.
references
Anderson, H.E., 1982. Aids to determining fuel models for
estimating fire behavior. General Technical Report INT-122,
USDA Forest Service, 22 pp.
Anderson, M., Bourgeron, P., Bryer, M.T., Crawford, R., Engelking,
L., Faber-Langendoen, D., Gallyoun, M., Goodin, K., Grossman,
D.H., Landaal, S., Metzler, K., Patterson, K.D., Pyne, M., Reid,
M., Sneddon, L., Weakley, A.S., 1998. International
Classification of Ecological Communities: Terrestrial
Vegetation of the United States. The National Vegetation
Classification System: List of Types. The Nature Conservancy,
vol. II. Arlington, VA, USA.
Andrews, P.L., 1986. Behave: fire behavior prediction and fuel
modeling system, Burn subsystem. General Technical Report
INT-194, USDA Forest Service.
Andrews, P.L., Bevins, C.D., 1999. BEHAVE Fire Modeling System:
redesign and expansion. Fire Manage. Notes 59, 16–19.
Bailey, R.G., 1995. Description of the Ecoregions of the United
States. Miscellaneous Publication Number 1391. USDA Forest
Service, Washington, DC.
Beukema, S.J., Kurz, W.A., 1998. Vegetation Dynamics
Development Tool—Users Guide Version 3.0. ESSA
Technologies, #300-1765 West 8th Avenue. Vancouver, BC,
Canada.
Boychuk, D., Perera, A.H., 1997. Modeling temporal variability of
boreal landscape age-classes under different fire disturbance
regimes and spatial scales. Can. J. For. Res. 27, 1083–1094.
Breiman, L., Frieman, J.H., Olshen, R.A., Stone, C.J., 1984.
Classification and regression trees. Belmont, CA, Wadsworth,
International Group, p. 432.
Brown, J.K., Bevins, C.D., 1986. Surface Fuel Loadings and
Predicted Fire Behavior for Vegetation Types in the Northern
Rocky Mountains. Research Note INT-358. U.S. Department of
Agriculture, Forest Service, Intermountain Forest and Range
Experiment Station, Ogden, UT, USA.
Brown, J.K., See, T.E., 1981. Downed dead woody fuel and biomass
in the northern Rocky Mountains. Gen. Tech. Rep. INT?117,
U.S. Department of Agriculture, Forest Service, Intermountain
Forest and Range Experiment Station, Ogden, UT, USA.
Burns, R.M., Honkala, B.H., 1990. Silvics of North America:
Conifers. Agriculture Handbook 654, vol. 1. USDA Forest
Service, Washington, DC.
Cibula, W.G., Nyquist, M.O., 1987. Use of topographic and
climatological models in a geographical data base to improve
Landsat MSS classification for Olympic National Park.
Photogram. Eng. Remote Sens. 53, 67–75.
Cihlar, J., 2000. Land cover mapping of large areas from satellites:
status and research priorities. Int. J. Remote Sens. 21 (6/7),
1093–1114.
Cohen, W.B., Fiorella, M., Gray, J., Helmer, E., Andersen, K., 1998.
An efficient and accurate method for mapping forest clearcuts
in the Pacific Northwest using Landsat imagery. Photogramm.
Eng. Remote Sens. 64, 293–300.
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Daubenmire, R., 1962. Forest vegetation of northern Idaho and
adjacent Washington, and its bearing on concepts of
vegetation classification. Ecol. Monogr. 22, 301–329.
Daubenmire, R., 1966. Vegetation: identification of typal
communities. Science 151, 291–298.
De’ath, G., 2002. Multivariate regression trees: a new technique
for modeling species–environment relationships. Ecology 83,
1105–1117.
Eyre, F.H.E., 1980. Forest Cover Types of the United States and
Canada. Society of American Foresters, Washington, DC, USA.
Ferry, G.W., Clark, R.G., Montgomery, R.E., Mutch, R.W.,
Leenhouts, W.P., Zimmerman, G.T., 1995. Altered Fire Regimes
within Fire-adapted Ecosystems. U.S Department of the
Interior—National Biological Service, Washington, DC.
Finney, M.A., 1998. FARSITE: Fire Area Simulator—model
development and evaluation. Research Paper RMRS-RP-4,
United States Department of Agriculture, Forest Service Rocky
Mountain Research Station, Ft. Collins, CO, USA.
Fowells, H.A., 1965. Silvics of the Forest Trees of the United
States. Agricultural Handbook Number 271. USDA Forest
Service, Washington Office.
Franklin, J., 1995. Predictive vegetation mapping: geographic
modelling of biospatial patterns in relation to environmental
gradients. Progr. Phys. Geogr. 19, 474–499.
Franklin, J., 2003. Clustering versus regression trees for
determining Ecological Land Units in the Southern California
Mountains and foothills. Forest Sci. 49, 354–368.
Franklin, J., Logan, T.L., Woodcock, C.E., Strahler, A.H., 1986.
Coniferous forest classification and inventory using Landsat
and digital terrain data. In: IEEE Transactions on Geoscience
and Remote Sensing, vol. GE-24, pp. l39–l49.
Franklin, J., Woodcock, C.E., Warbington, R., 2000. Digital
vegetation maps of forest lands in California: Integrating
satellite imagery, GIS modeling, and field data in support of
resource management. In: Photogrammetric Engineering and
Remote Sensing, vol. 66, no. 10, pp. 1209–1217.
Friedl, M.A., Brodley, C.E., 1997. Decision tree classification of
land cover from remotely sensed data. Remote Sens. Environ.
61 (3), 399–409.
GAO, 2002. Severe wildland fires: leadership and accountability
needed to reduce risks to communities and resources. Report
to Congressional Requesters GAO-02-259, United States
General Accounting Office, Washington, DC.
GAO, 2003. Additional actions required to better identify and
prioritize lands needing fuels reduction. Report to
Congressional Requesters GAO-03-805, United States General
Accounting Office, Washington, DC.
GAO, 2004. Forest Service and BLM need Better Information and a
Systematic Approach for Assessing Risks of Environmental
Effects. GAO-04-705. United States General Accounting Office,
Washington, DC.
Gardner, R.H., Romme, W.H., Turner, M.G., 1999. Predicting forest
fire effects at landscape scales. In: Mladenoff, D.J., Baker, W.L.
(Eds.), Spatial Modeling of Forest Landscape Change:
Approaches and Applications. Cambridge University Press,
Cambridge, United Kingdom, pp. 163–185.
Grossman, D.H., Faber-Langendoen, D., Weakley, A.S., Anderson,
M., Bourgeron, P., Crawford, R., Goodin, K., Landaal, S.,
Metzler, K., Patterson, K., Pyne, M., Reid, M., Sneddon, L., 1998.
International classification of ecological communities:
terrestrial vegetation of the United States. In: The National
Vegetation Classification System: development, status, and
applications, vol. I. The Nature Conservancy, Arlington, VA,
USA.
Hann, W.J., 2004. Mapping fire regime condition class: a method
for watershed and project scale analysis. In: Engstrom, R.T.,
Galley, K.E.M., De Groot, W.J. (Eds.), Proceedings of the 22nd
Tall Timbers Fire Ecology Conference, Fire in Temprate, Boreal,
499
and Montane Ecosystems. Tall Timbers Research Station, pp.
22–44.
Hann, W.J., Bunnell, D.L., 2001. Fire and land management
planning and implementation across multiple scales. Int. J.
Wildland Fire 10, 389–403.
Hann, W.J., Jones, J.L., Karl, M.G., Hessburg, P.F., Keane, R.E., Long,
D.G., Menakis, J.P., McNicoll, C.H., Leonard, S.G., Gravenmier,
R.A., Smith, B.G., 1997. An Assessment of Ecosystem
Components in the Interior Columbia Basin and Portions of
the Klamath and Great Basins, vol. II. Landscape Dynamics of
the Basin. USDA Forest Service Pacific Northwest Research
Station General Technical Report PNW-GTR-405,
172 pp.
Hardy, C.C., Schmidt, K.M., Menakis, J.P., Sampson, N.R., 2001.
Spatial data for national fire planning and fuel management.
Int. J. Wildland Fire 10, 353–372.
Hessburg, P.F., Smith, B.G., Krieter, S.G., et al., 1999a. Historical
and current forest and range landscapes in the Interior
Columbia River Basin and portions of the Klamath and Great
Basins. Part II. Linking vegetation patterns and landscape
vulnerability to potential insect and pathogen disturbances.
General Technical Report PNW-GTR-458, USDA Forest Service.
Hessburg, P.F., Smith, B.G., Salter, R.B., 1999b. A method for
detecting ecologically significant change in forest spatial
patterns. Ecol. Appl. 9, 1252–1272.
Hessburg, P.F., Smith, B.G., Salter, R.B., Ottmar, R.D., Alvarado, E.,
2000. Recent changes (1930’s–1990’s) in spatial patterns of
interior northwest forests, USA. Forest Ecol. Manage. 136,
53–83.
Heyerdahl, E.K., Berry, D., Agee, J.K., 1995. Fire History Database of
the Western United States. Interagency Agreement
DW12934530, Report EPA/600/R-96/081 NHEERL-COR-851, US
Environmental Protection Agency, USDA Forest Service, and
University of Washington, Seattle.
Holdridge, L.R., 1947. Determination of world plant formations
from simple climatic data. Science 105, 367–368.
Holsinger, Lisa, Keane, R.E., Parsons, R., Karau, E., 2006.
Development of biophysical gradient layers for the LANDFIRE
prototype project. In: Rollins, M.G., Frame, C. (Eds.), The,
LANDFIRE, Prototype Project: Nationally Consistent, Locally
Relevant Geospatial Data for Wildland Fire, Management,
USDA, Forest Service Rocky Mountain Research Station
General Technical Report, RMRS-GTR-175.
Homer, C.G., Ramsey, R.D., et al., 1997. Landscape cover-type
modeling using a multi-scene thematic mapper mosaic.
Photogramm. Eng. Remote Sens. 63 (1), 59–67.
Homer, C.G., Huang, C., et al., 2002. Development of a circa 2000
landcover database for the United States. In: Proceedings 2002
ASPRS-ACSM Annual Conference and FIG XXII Congress, April
2002. Washington, DC, The American Society for
Photogrammetry and Remote Sensing, pp. 111–123.
Huang, C., Townshend, J.R.G., 2003. A stepwise regression tree for
nonlinear approximation: applications to estimating subpixel
land cover. Int. J. Remote Sens. 24 (1), 75–90.
Huang, C., Yang, L., Wylie, B., Homer, C.A., 2001. A strategy for
estimating tree canopy density using Landsat 7ETM+ and high
resolution images over large areas. Proceedings of the Third
International Conference on Geospatial Information in
Agriculture and Forestry, Denver, Colorado.
Huang, C., Wylie, B., Homer, C., Zylstra, G., 2002. Derivation of a
tasselled cap transformation based on Landsat 7 at-satellite
reflectance. International J. Remote Sens. 23 (8), 1741–
1748.
Hummel, S., Cunningham, P., 2006. Estimating variation in a
landscape simulation of forest structure. Forest Ecol. Manage.
228, 135–144.
Huschle, G., Hironaka, M., 1980. Classification and ordination of
seral plant communities. J. Range Manage. 33, 179–182.
500
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Irish, R.R., 2000. Landsat 7 science data user’s handbook:
http://ltpwww.gsfc.nasa.gov/IAS/handbook/handbook toc.
html, National Aeronautics and Space Administration.
Iverson, L.R., Prasad, A.M., 1998. Predicting abundance of 80 tree
species following climate change in the eastern United States.
Ecol. Monogr. 68, 465–485.
Karau, E., Keane, R.E., in press. Determining the spatial extent of
a landscape using simulation modeling. Landscape Ecology.
Keane, R.E., Finney, M.A., 2003. The simulation of landscape fire,
climate, and ecosystem dynamics. In: Veblen, T.T., Baker, W.L.,
Montenegro, G., Swetnam, T.W. (Eds.), Fire and Global Change
in Temperate Ecosystems of the Western Americas.
Springer-Verlag, New York, NY, USA.
Keane, R.E., Lisa, H., 2006a. Computing biophysical environment
for gradient modeling and ecosystem mapping using the
WXFIRE program: model documentation and application.
USDA Forest Service Rocky Mountain Research Station
General Technical Report RMRS-GTR-168DVD, 1 DVD, 61 pp.
Keane, R.E., Rollins, M.G., 2006. The scientific foundation of the
LANDFIRE project. In: Rollins, M.G., Frame, C. (Eds.), The,
LANDFIRE, Prototype Project: Nationally Consistent, Locally
Relevant Geospatial data for Wildland Fire Management.
USDA, Forest Service Rocky Mountain Research Station
General Technical Report, RMRS-GTR-175, pp. 45–
68.
Keane, R.E, Long, D.G., Menakis, J.P., Hann, W.J., Bevins, C.D.,
Intermountain Research Station (Ogden Utah), 1996.
Simulating coarse-scale vegetation dynamics using the
Columbia River Basin succession model: CRBSUM. U.S.
Department of Agriculture Forest Service Intermountain
Research Station, Ogden, UT, 94 pp.
Keane, R.E., McNicoll, C.H., Schmidt, K.M., Garner, J.L., 1997a.
Spatially explicit ecological inventories for ecosystem
management planning using gradient modeling and remote
sensing. In: Proceedings of the Sixth Forest Service Remote
Sensing Applications Conference—Remote Sensing; People in
Partnership with Technology. American Society of
Photogrammetry and Remote Sensing, Bethesda, MD, USA,
pp. 135–146.
Keane, R.E., Long, D., Basford, D., Levesque, B.A., 1997b.
Simulating vegetation dynamics across multiple scales to
assess alternative management strategies. In: Conference
Proceedings, GIS 97, 11th Annual Symposium on Geographic
Information Systems, Integrating Spatial Information
Technologies for Tomorrow. GIS World, INC, Vancouver,
British Columbia, Canada, pp. 310–315.
Keane, R.E., McNicoll, C., Rollins, M.G., 2002a. Integrating
ecosystem sampling, gradient modeling, remote sensing, and
ecosystem simulation to create spatially explicit landscape
inventories. USDA Forest Service, Rocky Mountain Research
Station, General Technical Report RMRS-GTR-92, Fort Collins,
CO, 99 pp.
Keane, R.E., Parsons, R., Hessburg, P., 2002b. Estimating historical
range and variation of landscape patch dynamics: limitations
of the simulation approach. Ecol. Modell. 151, 29–49.
Keane, R.E., Veblen, T., Ryan, K.C., Logan, J., Allen, C., Hawkes, B.,
2002c. The cascading effects of fire exclusion in the Rocky
Mountains. In: Jill Baron (Ed.), Rocky Mountain Futures: An
Ecological Perspective. Island Press, Washington, DC, USA, pp.
133–153.
Keane, R.E., Cary, G.J., Parsons, R., 2003. Using simulation to map
fire regimes: an evaluation of approaches, strategies, and
limitations. Int. J. Wildland Fire 12, 309–322.
Keane, R.E., Cary, G.J., Davies, I.D., Flannigan, M.D., Gardner, R.H.,
Lavorel, S., Lennihan, J.M., Li, C., Rupp, T.S., 2004a. A
classification of landscape fire succession models: spatially
explicit models of fire and vegetation dynamic. Ecol. Modell.
256, 2–27.
Keane, R.E., Parsons, R., Rollins, M.G., 2004b. Predicting fire
regimes across multiple scales. In: Perera, A., Buse, L. (Eds.),
Emulating Natural Disturbances: Concepts and Techniques.
Cambridge University Press, Cambridge, UK.
Keane, R.E., Holsinger, L., Sarah, P., 2006. Simulating historical
landscape dynamics using the landscape fire succession
model LANDSUM version 4.0. General Technical Report.
RMRS-GTR-171CD. Fort Collins, CO: U.S. Department of
Agriculture, Forest Service, Rocky Mountain Research Station,
73 pp.
Keane, R.E., Frescino, T.L., Reeves, M.C., Long, J., 2006b. Mapping
wildland fuels across large regions for the LANDFIRE
prototype project. In: Rollins, M.G., Frame, C. (Eds.), The
LANDFIRE Prototype Project: Nationally Consistent and
Locally Relevant Geospatial Data for Wildland Fire
Management. USDA Forest Service Rocky Mountain Research
Station General Technical Report RMRS-GTR-175, pp. 367–
396.
Knick, S.T., Rotenberry, J.T., Zarriello, T.J., 1997. Supervised
classification of Landsat Thematic Mapper imagery in a
semi-arid rangeland by nonparametric discriminant analysis.
Photogram. Eng. Remote Sens. 63 (1), 79–86.
Kolb, P.F., Adams, D.L., McDonald, G.I., 1998. Impacts of fire
exclusion on forest dynamics and processes in central Idaho.
In: Tall Timbers Fire Ecol. Conf., vol. 20, pp. 911–923.
Kuchler, A.W., 1975. Potential Natural Vegetation of the
Conterminous United States: Manual and Map, 2 ed.
American Geological Society, New York, NY.
Kurz, W.A., Beukema, S.J., Merzenich, J., Arbaugh, M., Schilling, S.,
1999. Long-range modeling of stochastic disturbances and
management treatments using VDDT and TELSA. In: Society
of American Foresters 1999 National Convention. Society of
American Foresters, Portland, OR, USA, pp. 349–355.
Landres, P.B., Morgan, P., Swanson, F.J., 1999. Overview and use of
natural variability concepts in managing ecological systems.
Ecol. Appl. 9, 1179–1188.
Laverty, L., Williams, J., 2000. Protecting people and sustaining
resources in fire-adapted ecosystems—a cohesive strategy.
Forest Service response to GAO Report GAO/RCED 99-65. USDA
Forest Service, Washington, DC.
Long, D.G., Losensky, J., Bedunah, D., 2006a. Vegetation
succession modeling for the LANDFIRE Prototype Project. In:
Rollins, M.G., Frame, C. (Eds.), The LANDFIRE Prototype
Project: Nationally Consistent and Locally Relevant Geospatial
Data For Wildland Fire Management. USDA Forest Service
Rocky Mountain Research Station General Technical Report
RMRS-GTR-175, pp. 277–314.
Long, J., Melanie, M., James, M., Keane, R.E., 2006b. Developing the
LANDFIRE vegetation and biophysical settings map unit
classifications for the LANDFIRE prototype project. In: Rollins,
M.G., Frame, C. (Eds.), The LANDFIRE Prototype Project:
Nationally Consistent and Locally Relevant Geospatial Data
for Wildland Fire Management. USDA Forest Service Rocky
Mountain Research Station General Technical Report
RMRS-GTR-175, pp. 123–180.
Lutes, D.C., Keane, R.E., Caratti, J.F., 2006. Fuel loading models: a
national classification of wildland fuelbeds for fire effects
modeling. Can. J. Forest Res., in preparation.
Markham, B.L., Barker, J.L., 1985. Spectral characterization of the
Landsat Thematic Mapper sensors. Int. J. Remote Sens. 6 (5),
697–716.
McDonald, R.I., Urban, D.L., 2006. Spatially varying rules of
landscape change: lessons from a case study. Landscape
Urban Plan. 74, 7–20.
Michaelsen, J., Schimel, D.S., Friedl, M.A., Davis, F.W., Dubayah,
R.C., 1994. Regression tree analysis of satellite and terrain
data to guide vegetation sampling and surveys. J. Veg. Sci. 5,
673–686.
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Minore, D., 1979. Comparative autecological characteristics of
northwestern tree species: a literature review. General
Technical Report PNW-87. U.S. Department of Agriculture,
Forest Service, Pacific Northwest Forest and Range
Experiment Station, Portland, OR, USA.
Mladenoff, D.J., Baker, W.L., 1999. Spatial Modeling of Forest
Landscape Change. Cambridge University Press, Cambridge,
United Kingdom.
Moisen, G.G., Frescino, T.S., 2002. Comparing five modelling
techniques for predicting forest characteristics. Ecol. Modell.
157, 209–225.
Mutch, R.W., 1994. Fighting fire with fire: a return to ecosystem
health. J. Forest. 92, 31–33.
Nonaka, E., Spies, T.A., 2005. Historical range of variability in
landscape structure: a simulation study in Oregon, USA. Ecol.
Appl. 15, 1727–1746.
Ohmann, J.L., 1996. Linking plot data, models, and maps in
regional ecological analysis. In: Society of American Foresters
Convention. Society of American Foresters, Portland, Maine,
pp. 99–103.
Ohmann, J.L., Spies, T.A., 1998. Regional gradient analysis and
spatial pattern of woody plant communities of Oregon forests.
Ecol. Monogr. 68, 151–182.
Ottmar, R.D., Burns, M.F., Hall, J.N., Hanson, A.D., 1993. CONSUME
Users Guide. USDA Forest Service Pacific Northwest Research
Station General Technical Report PNW-GTR-304, Portland, OR,
USA.
Pfister, R.D., 1981. Status of forest successional studies in the
northern Rocky Mountains. Successional research and
environmental pollutant-monitoring associated with
biosphere reserves. In: Proceedings of the Second US-USSR
Symposium on Biosphere Reserves. US National Committee
for Men and the Biosphere, Washington, DC, USA, Everglades
National Park, FL, USA, pp. 80–90.
Pfister, R.D., Arno, S.F., 1980. Classifying forest habitat types
based on potential climax vegetation. Forest Sci. 26, 52–
70.
Pfister, R.D., Kovalchik, B.L., Arno, S.F., Presby, R.C., 1977. Forest
habitat types of Montana. U.S. Department of Agriculture,
Forest Service, Intermountain Forest and Range Experiment
Station, General Technical Report INT-34, Ogden, UT, USA.
Pratt, S., Holsinger, L., Keane, R.E., 2006. Using simulation
modeling to assess historical reference conditions for
vegetation and fire regimes for the LANDFIRE prototype
project.
Pratt, S., Holsinger, L., Keane, R.E., 2006. Using simulation
modeling to assess historical reference conditions for
vegetation and fire regimes for the LANDFIRE prototype
project. In: Rollins, M.G., Frame, C. (Eds.), The LANDFIRE
Prototype Project: Nationally Consistent and Locally Relevant
Geospatial data for Wildland Fire Management. USDA Forest
Service Rocky Mountain Research Station General Technical
Report RMRS-GTR-175, pp. 277–314.
Quinlan, J.R., 1986. Induction of decision trees. Mach. Learn. 1,
81–206.
Quinlan, J.R., 1993. Combining instance-based and mode-based
learning. In: Proceedings of the 10th International Conference
on Machine Learning, Amherst, Massachusetts, Morgan
Kaufmann, San Mateo, California, pp. 236–243.
Reed, W.J., Larsen, C.P.S., Johnson, E.A., MacDonald, G.M., 1998.
Estimation of temporal variations in historical fire frequency
from time-since-fire map data. Forest Sci. 44, 465–475.
Reinhardt, E., Crookston, N.L., 2003. The fire and fuels extension
to the Forest Vegetation Simulator. General Technical Report
RMRS-GTR-116. Rocky Mountain Research Station, Fort
Collins, CO.
Reinhardt, E., Keane, R.E., 1998. FOFEM—a first order fire effects
model. Fire Manage. 58, 25–28.
501
Reinhardt, E., Keane, R.E., Brown, J.K., 1997. First Order Fire Effects
Model: FOFEM 4.0 User’s Guide. USDA Forest Service Rocky
Mountain Research Station General Technical Report
INT-GTR-344, Fort Collins, CO, USA.
Rollins, M.G., Christine, F. (Technical editors), 2006. The
LANDFIRE prototype project: nationally consistent and locally
relevant geospatial data for wildland fire management. USDA
Forest Service Rocky Mountain Research Station General
Technical Report RMRS-GTR-175, pp. 834.
Rollins, M.G., Keane, R.E., Parsons, R., 2004. Mapping ecological
attributes using gradient analysis and remote sensing. Ecol.
Appl. 14, 75–95.
Rothermel, R.C., 1991. Predicting behavior and size of crown fires
in the Northern Rocky Mountains. United States Department
of Agriculture, Forest Service Intermountain Forest and Range
Experiment Station Research Paper INT-438, Ogden, UT, USA,
23 pp.
Running, S.W., Coughlan, J.C., 1988. A general model of forest
ecosystem processes for regional applications. I. Hydrologic
balance, canopy gas exchange and primary production
processes. Ecol. Modell. 42, 125–154.
Running, S.W., Gower, S.T., 1991. FOREST-BGC, a general model of
forest ecosystem processes for regional applications. II.
Dynamic carbon allocation and nitrogen budgets. Tree
Physiol. 9, 147–160.
Running, S.W., Thornton, P.E., 1996. Generating daily surfaces of
temperature and precipitation over complex topography. In:
Goodchild, M.F., Parks, B.O., Steyaert, L.T. (Eds.), GIS and
Environmental Modeling. GIS World Books, Fort Collins, CO,
USA, pp. 93–98.
Running, S.W., Loveland, T.R., Pierce, L.L., 1994. A vegetation
classification logic based on remote sensing for use in global
biogeochemical models. Ambio 23, 77–81.
Schmidt, K.M., Menakis, J.P., Hardy, C.C., Hann, W.J., Bunnell, D.L.,
2002. Development of coarse-scale spatial data for wildland
fire and fuel management. USDA Forest Service Rocky
Mountain Research Station General Technical Report
RMRS-GTR-87. Fort Collins, CO, USA.
Scott, J., Burgan, R.E., 2005. A new set of standard fire behavior
fuel models for use with Rothermel’s surface fire spread
model. General Technical Report RMRS-GTR-153. USDA Forest
Service Rocky Mountain Research Station, Fort Collins, CO, 76
pp.
Shang, B.Z., He, H.S., Crow, T.R., Shifley, S.R., 2004. Fuel load
reductions and fire risk in central hardwood forests of the
United States: a spatial simulation study. Ecol. Modell. 180,
89–102.
Shiflet, T.N.E., 1994. Rangeland Cover Types of the United States.
Society of Range Management, Denver, CO, USA, p. 151.
Stage, A.R., 1997. Using FVS to provide structural class attributes
to a forest succession model CRBSUM. In: Tech, R., Moeur, M.,
Adams, J. (Eds.), Proceedings of the Forest Vegetation
Simulator, Conference, February 3–11, 1997, Fort Collins, CO,
USA, USDA, pp. 139–147. Forest Service Intermountain
Research Station General Technical Report, INT-GTR-373, 222
pp.
Steele, B.M., Reddy, S.K., Keane, R.E., 2006. A methodology for
assessing departure of current plant communities from
historical conditions over large landscapes. Ecol. Modell. 199,
53–63.
Swanson, F.J., Franklin, J.F., Sedell, J.R., 1997. Landscape patterns,
disturbance, and management in the Pacific Northwest, USA.
In: Zonnneveld, I.S., Forman, R.T.T. (Eds.), Changing
Landscapes: An Ecological Perspective. Springer-Verlag, New
York, NY, USA, pp. 191–213.
Swetnam, T.W., Allen, C.D., Betancourt, J.L., 1999. Applied
historical ecology: Using the past to manage for the future.
Ecol. Appl. 9, 1189–1206.
502
e c o l o g i c a l m o d e l l i n g 2 0 4 ( 2 0 0 7 ) 485–502
Thornton, P.E., White, J.D., 1996. Biogeochemical characterization
of the Columbia River Basin using the BGC Model: Model
Description, Ecophysiological Inputs and Landscape
Description. USDA Forest Service, Intermountain Fire Sciences
Laboratory, Missoula, MT, USA.
Thornton, P.E., Running, S.W., White, M.A., 1997. Generating
surfaces of daily meteorological variables over large regions of
complex terrain. J. Hydrol. 190, 214–251.
Thornton, P.E., Hasenauer, H., M. White, A., 2000. Simultaneous
estimation of daily solar radiation and humidity from
observed temperature and precipitation: an application over
complex terrain in Austria. Agricult. Forest Meteorol. 104,
225–271.
Thornton, P.E., Law, B.E., Gholz, H.L., Clark, K.L., Falge, E.,
Ellsworth, D.S., Goldstein, A.H., Monson, R.K., Hollinger, D.Y.,
Falk, M., Chen, J., Sparks, J.P., 2002. Modeling and measuring
the effects of disturbance history and climate on carbon and
water budgets in evergreen needleleaf forests. Agricult. Forest
Meteorol. 113, 185–222.
Turner, M.G., Gardner, R.H., O’Neill, R.V., Pearson, S.M., 1994.
Multiscale organization of landscape heterogeneity. In:
Jensen, M.E., Bourgeron, P.S. (Eds.), Ecosystem Management:
Principles and Applications. U.S. Department of Agriculture,
Forest Service, Pacific Northwest Research Station, Portland,
Oregon, pp. 73–79.
van Wagner, C.E., 1993. Prediction of crown fire behavior in two
stands of jack pine. Can. J. Forest Res. 23, 442–449.
Vogelmann, J., Zhiliang, Z., Jay, K., Brian, T., Donald, O., 2006.
Perspectives on LANDFIRE prototype project accuracy
assessment. In: Rollins, M.G., Frame, C. (Eds.), The LANDFIRE
Prototype Project: Nationally Consistent and Locally Relevant
Geospatial Data for Wildland Fire Management. USDA Forest
Service Rocky Mountain Research Station General Technical
Report RMRS-GTR-175, pp. 397–420.
Westoby, M., 1980. Elements of a theory of vegetation dynamics
in arid rangelands. Isreal J. Botany 28, 169–194.
White, P.S., Walker, J.L., 1997. Approximating nature’s variation:
selecting and using reference information in restoration
ecology. Restor. Ecol. 5 (4), 338–345.
White, M.A., Thornton, P., Running, S.W., Nemani, R., 2000.
Parameterization and sensitivity analysis of the Biome-BGC
terrestrial ecosystem model: net primary production controls.
Earth Interact. 4, 1–85.
Wimberly, M.C., Spies, T.A., Long, C.J., Whitlock, C., 2000.
Simulating historical variability in the amount of old forest in
the Oregon Coast Range. Conserv. Biol. 14, 167–180.
Yang, L., Collin, H., Kent, H., Chengquan, H., Bruce, W., Bradley, R.,
2001. A Landsat 7 Scene Selection Strategy for a National Land
Cover Database. Proceedings of the IEEE 2001 International
Geoscience and Remote Sensing Symposium, Sydney,
Australia, CD-ROM, 1 Disc.
Yang, J., Hong, S., He, E.J., Gustafson, 2004. A hierarchical fire
frequency model to simulate temporal patterns of fire regimes
in LANDIS. Ecol. Modell. 180, 119–133.
Zhu, Z., Evans, D.L., 1994. U.S. forest type groups and predicted
percent forest cover from AVHRR data. Photogram. Eng.
Remote Sens. 60 (5), 525–531.