Summary A mechanistic, biogeochemical succession model, FIRE-BGC, was used to investigate the role of fire on long-term landscape dynamics in northern Rocky Mountain coniferous forests of Glacier National Park, Montana, USA. FIRE-BGC is an individual-tree model----created by merging the gap-phase process-based model FIRESUM with the mechanistic ecosystem biogeochemical model FOREST-BGC----that has mixed spatial and temporal resolution in its simulation architecture. Ecological processes that act at a land-scape level, such as fire and seed dispersal, are simulated annually from stand and topographic information. Stand-level processes, such as tree establishment, growth and mortality, organic matter accumulation and decomposition, and under-growth plant dynamics are simulated both daily and annually. Tree growth is mechanistically modeled based on the ecosys-tem process approach of FOREST-BGC where carbon is fixed daily by forest canopy photosynthesis at the stand level. Carbon allocated to the tree stem at the end of the year generates the corresponding diameter and height growth. The model also explicitly simulates fire behavior and effects on landscape characteristics. We simulated the effects of fire on ecosystem characteristics of net primary productivity, evapotranspiration, standing crop biomass, nitrogen cycling and leaf area index over 200 years for the 50,000-ha McDonald Drainage in Gla-cier National Park. Results show increases in net primary productivity and available nitrogen when fires are included in the simulation. Standing crop biomass and evapotranspiration decrease under a fire regime. Shade-intolerant species domi-nate the landscape when fires are excluded. Model tree incre-ment predictions compared well with field data.
Keywords: evapotranspiraton, landscape dynamics, net pri-mary productivity, standing crop biomass.
Introduction
Fire is the principal disturbance process in most Northern Rocky Mountain ecosystems (Habeck and Mutch 1973, Wright and Bailey 1982, Peet 1988). The interaction of fire with the biophysical environment and the vegetation has shaped landscapes and dictated species composition and struc-ture over the last 10,000 years (Heinselman 1981). The
im-pacts of fire are manifest across many temporal and spatial scales and affect most ecosystem components.
Fire characteristics are often described in two broad catego-ries: fire behavior and fire effects. Fire behavior is a quantifi-cation of the physical properties of a fire. Descriptors of fire behavior include fireline intensity (kW m−2), flame length (m) and scorch height (i.e., height of lethal foliar scorch, m) (An-derson 1969, Rothermal 1972, Albini 1976a). Fire behavior is chiefly related to topography (i.e., slope, aspect, exposure and landform), fuels (i.e., mass, energy content, moisture content and size distribution of forest floor organic matter) and weather (i.e., wind, temperature, precipitation and relative humidity). Fire effects are the direct and long-term conse-quences of a fire on ecosystem components. These effects may be correlated to fire behavior. Direct fire effects include fuel consumption, tree mortality and smoke generation. Secondary fire effects include plant succession, soil erosion and landscape pattern.
Distribution, size and amount of forest floor organic mate-rial are important ecosystem characteristics that influence fire behavior and effects (Albini 1976a). Forest floor organic ma-terials that directly sustains the spread of a fire are called fuels (Fosberg 1970, Brown and See 1981). Herbaceous and shrubby fuels can be live or dead and are composed mostly of leaves. The quantitative description of fuel characteristics for fire behavior prediction is often termed a ‘‘fuel model’’ and includes estimates of bulk density (kg m−3), loading (kg m−2), depth (m) and many other physical properties (Brown 1970, 1981, Anderson 1982, Brown and Bevins 1986).
Effects of fire on ecosystem processes and components can be both immediate and long-term (Reich et al. 1990, Crutzen and Goldammer 1993). Oxidation of organic matter by fire reduces forest floor carbon pools and disrupts carbon and nutrient cycling (Raison 1979, Gosz 1981, Maggs 1988, Dyrness et al. 1989, DeBano 1991). Vegetation mortality from fire often results in a reduction of stand leaf area, which may cause a temporary reduction in photosynthesis and transpira-tion. In turn, this can affect stand micrometeorology, water use and decomposition (Rundel 1982, Link et al. 1990, Mushinsky and Gibson 1991). However, plant communities succeeding fire usually have more rapid growth rates, which may increase net primary productivity. Landscape pattern, structure and
Simulating effects of fire on northern Rocky Mountain landscapes with
the ecological process model FIRE-BGC
ROBERT E. KEANE,
1KEVIN C. RYAN
1and STEVEN W. RUNNING
21
Intermountain Fire Sciences Laboratory, P.O. Box 8089, Missoula, MT 59807, USA
2 School of Forestry, University of Montana, Missoula, MT 59812, USA
Received April 10, 1995
composition resulting from fire directly affect species compo-sition by altering propagule dissemination and species migra-tion (Rowe 1983, Turner and Romme 1994).
Long-term responses of ecosystem processes to fire have not been studied in detail because such studies are costly and complex, and often inconclusive. Mechanistic ecosystem proc-ess models offer an alternative means to explore the role of fire in forest ecosystems (Reed 1980, Kercher and Axelrod 1984, Shugart and Seagle 1985, Bossel and Schäfer 1990, Dixon et al. 1990, Urban and Shugart 1992, Kimmins 1993). Therefore, we have investigated the influence of fire on ecosystem dy-namics in the 50,000-ha McDonald Drainage in Glacier Na-tional Park, Montana, USA, by means of a mechanistic ecosystem biogeochemical succession model called FIRE-BGC. FIRE-BGC simulates the interaction of disturbance processes such as fire with stand development processes of tree regeneration, growth and mortality, and landscape processes of weather, species migration and hydrology.
Model description and development
FIRE-BGC is a mechanistic, individual-tree succession model with stochastic properties implemented in a spatial domain. Tree growth, organic matter decomposition, litterfall and other ecological processes are simulated using detailed physical relationships. Tree establishment and mortality are modeled using probability functions with ecologically derived parame-ters. FIRE-BGC also includes a spatial simulation of fire behavior and fire effects on ecosystem components across the landscape. Insect and disease interactions are included in the model.
A detailed discussion of FIRE-BGC is presented in Keane et al. (1995). Briefly, FIRE-BGC, which is the union of the gap-replacement model FIRESUM (Keane et al. 1989, 1990a, 1990b) with the mechanistic biogeochemical simulation model FOREST-BGC (Running and Coughlan 1988, Running and Gower 1991), was designed to predict changes in species composition in response to various ecosystem processes over long time periods. Thus to create FIRE-BGC, we used the mechanistic design of FOREST-BGC as the framework and engine for ecosystem simulation. We then added important FIRESUM algorithms to this framework to simulate long-term, multispecies forest succession. The FIRESUM routines were refined to use the detailed information generated from the mechanistic FOREST-BGC routines. Finally, this modeling framework was implemented in a spatial context recognizing the spatial distribution of these processes across a simulation area (Busing 1991, Urban et al. 1991). This allowed detailed simulation of ecosystem processes that act across several spa-tial scales (Bonan and Shugart 1989).
FIRE-BGC models the flow of carbon, nitrogen and water across various ecosystem components to calculate individual tree growth (Figure 1). Carbon and nitrogen are allocated to each tree’s stem, roots and leaves at year’s end. Stem carbon allocation is used to calculate diameter and height growth.
Temporal scales
FIRE-BGC has a mixed time resolution built into the simula-tion design. Primary canopy processes of intercepsimula-tion, evapo-ration, transpievapo-ration, photosynthesis and respiration are simulated at a daily timestep. Secondary canopy processes of carbon and nitrogen allocation are simulated at a yearly time-step. Landscape processes of fire, insects and disease are simulated annually as are tree mortality, regeneration and growth. Because of computational complexity, seed dispersal is simulated at decadal timesteps.
Spatial scales
Two spatial scales are explicitly implemented in FIRE-BGC. Ecosystem processes that occur at the landscape level, such as seed dispersal and fire, are modeled in a spatial domain using raster data layers. These landscape processes are simulated by external programs directly linked to FIRE-BGC. Many stand-level processes, such as tree growth and regeneration, are modeled independently of the spatial environment. Dynamic databases provide the linkage between landscape and stand-level process simulation.
Organizational scales
There are five hierarchical levels of organization within the FIRE-BGC design (Figure 2). The coarsest level is the simula-tion landscape, defined as a large expanse of land (greater than 10,000 ha) delineated by the natural boundaries that control the major properties of that ecosystem including climate, vegeta-tion and disturbance. This landscape is divided into units called sites that have similar topography, soils, weather and potential vegetation. Boundaries of each site are static and do not change in a FIRE-BGC simulation.
The third level of organization is the stand. Each site is composed of a number of stands that differ in vegetation composition and structure (Figure 2). By definition, stand boundaries cannot extend past site boundaries. Stand bounda-ries are not stationary in FIRE-BGC because the processes of succession, fire and pathogens can serve to modify stand boundaries within a site.
FIRE-BGC does not explicitly model all entities across the spatial extent of an entire simulation stand because of compu-tational limitations. Instead, the model simulates ecosystem processes in a small portion of the stand called the simulation plot (Figure 2). The simulation plot is assumed to be repre-sentative of the entire stand.
The fourth organization level is the species level. Any num-ber of species can inhabit a stand. Many modeled processes, such as canopy dynamics and tree regeneration, are performed at the species level. The finest level of organization is the tree level (Figure 2). Each tree within a simulation plot is explicitly modeled in the FIRE-BGC architecture. Many attributes of each tree, such as leaf carbon, diameter and height, are recog-nized in FIRE-BGC. However, tree locations are not spatially defined in the model. Competition between trees is primarily modeled at the stand level.
Scale interactions
FIRE-BGC links many across-scale interactions in the simula-tion of ecosystem processes. For example, weather provides process linkages that progress downward in organizational scale. A particular weather year from a weather record is selected for the entire landscape. Daily weather values for that year are gathered from site-specific climate files. These weather data are used to compute photosynthesis, respiration and other processes at the stand level for that site. Important weather events such as frosts and drought are computed at the stand level for the simulation of species dynamics. Carbon fixed through photosynthesis at the stand level is allocated to the trees based on the distribution of radiation in the forest canopy, which is computed from the site weather file and the stand’s canopy structure.
FIRE-BGC also accounts for interactions that occur upward in organizational scale. At the end of the simulation year, FIRE-BGC sums all carbon and nitrogen tree compartments for a new estimate of stand carbon and nitrogen components. The abundance of a stand’s seed crop trees by species is used in the landscape application of the seed dispersal model. Simu-lated fires burn a stand’s forest floor compartments (i.e., fuels) but use site-level weather data and landscape topography for
computation of fire spread and intensity.
FIRE-BGC program
A generalized FIRE-BGC program flowchart is shown in Fig-ure 3. FIRE-BGC reads all initial values and parameters from a collection of specified input data files. State variables are then initialized based on the input information. Simulation starts with the spatial simulation of cone abundance and seed dispersal. Daily simulation of stand-level processes is then accomplished using FOREST-BGC routines in the following procedure. First, weather data are gathered by site, and daily photosynthesis, respiration and water budgets are calculated from site weather to obtain yearly carbon gains. At the end of the simulation year, stand carbon is allocated to each tree in the stand based on species, canopy position and leaf area. Tree carbon is then portioned to the leaf, stem and root compart-ments of each tree based on dynamic equations (Running and Gower 1991). Carbon allocated to the tree’s stem is converted Figure 2. The five levels of organization implemented in FIRE-BGC.
to a diameter increment growth. Establishment of new trees is then assessed by species in the regeneration routine. Tree death and its consequences are evaluated in the mortality algorithm. Fire occurrence, behavior and effects are dynamically modeled on the landscape. Forest floor decomposition is simulated daily, the forest canopy characteristics are recomputed at year’s end, and the process is repeated again.
FIRE-BGC application
LOKI modeling platform
The LOKI modeling architecture is used to link and schedule execution of the FIRE-BGC program and the associated mod-els SEEDER (seed dispersal model), MAPMAKER (an eco-logical mapping routine), FIRESTART (a fire occurrence simulator) and FARSITE (fire behavior model) at the appropri-ate time intervals (Bevins et al. 1994, Bevins and Andrews 1994) (Figure 4). LOKI allows FIRE-BGC and other models to query, modify and create digital landscape maps during simulation. The GRASS spatial GIS package (USA CERL 1990) is used for organizing, displaying and analyzing raster files created by LOKI. Linkage of these models in the LOKI architecture to simulate long-term ecosystem dynamics for this study is called the FIRE-BGC application.
Fire occurrence model
Occurrence and points of origin of simulated fires are stochas-tically predicted yearly on the simulation landscape using the model FIRESTART (Figure 4). FIRESTART uses site-level fire frequency inputs (i.e., mean fire-free interval) to compute a probability of fire ignition (Keane et al. 1989, 1995). Fire occurrence is computed by pixel from fire frequency prob-abilities calculated from Weibull probability distributions and average fire size estimates (Van Wagner 1978, Johnson 1979, Baker 1989).
An important parameter in the fire occurrence Weibull equa-tions is fire frequency. Average fire-free intervals for this study were taken from Davis (1981), Barrett (1986), Fischer and Bradley (1987) and Barrett et al. (1991). Average fire sizes and other Weibull parameters were estimated from Johnson (1979), Baker (1989) and Clark (1990), and also from fire information compiled by Glacier National Park. Once fire origins have
been computed, FARSITE simulates fire intensity and spread rate from each fire origin (Figure 4).
Spatially explicit fire model
Fire is dynamically modeled on the simulation landscape using the FARSITE spatial fire model (Finney 1994). This model predicts fire intensity and rate of spread as it moves across a landscape (Rothermal 1972, Albini 1976b). FARSITE uses the spatial data layers of topography, vegetation, weather and fuels to predict fire dynamics. Topography is described using a digital elevation model (DEM) raster layer. Other input raster layers are created by MAPMAKER from FIRE-BGC output or user-defined input. Fuel biomass is computed from carbon pools comprising the forest floor compartments. Stand struc-ture values (e.g., canopy height) required by FARSITE are computed from stand attributes explicitly computed by FIRE-BGC. Because FIRE-BGC does not simulate daily fuel water content and canopy winds to the level of detail needed by FARSITE, a static set of inputs taken from weather data meas-ured at the West Glacier weather station for a typical summer having high fire activity was used in FARSITE.
FARSITE creates a raster layer of computed fire intensity (kW m−2) and flame length (m) from the spatial predictions of fire spread (m h−1). FIRE-BGC creates new stands in areas where fire intensity is greater than a user-defined threshold (i.e., burned over areas). Fire effects at the stand and tree levels, such as fuel consumption and tree mortality, are calculated from average fire intensity and flame length estimates within the burned stand (Keane et al. 1995).
Seed dispersal model
The probability that a tree species will disperse seed to each simulation stand is computed by the model SEEDER (Fig-ure 4). The spatial distribution of stand composition and struc-ture dictate species’ seed distribution (Greene and Johnson 1989, Anderson 1991). Seed production potential by species is computed from the number of cone-producing trees on each simulation stand by FIRE-BGC. Then probability of seed dispersal by wind to every landscape pixel is computed using a form of the equations described by McCaughey et al. (1985). Probabilities of seed dispersal for the bird-disseminated white-bark pine seed are calculated as described by Tomback et al. (1990). Currently, topographic and wind effects on seed dis-persal are not accounted for in the simulation.
Simulation study area
Biophysical environment
The McDonald Drainage in Glacier National Park (MD-GNP) is a long, narrow glaciated valley surrounded by rugged moun-tains (Figure 5). This 50,000-ha landscape has great diversity in vegetation, topography and climate. Western hemlock (Tsuga heterophylla (Raf.) Sarg.) and western red cedar (Thuja plicata J. Donn ex D. Don) forests dominate low elevation, lakeside settings (Habeck 1968). Western larch (Larix occiden-talis Nutt.), Douglas-fir (Pseudotsuga menziesii (Mirb.) Figure 4. Diagram of the LOKI modeling platform and the various
Franco) and lodgepole pine (Pinus contorta Dougl.) comprise the mixed forests at moderate altitudes (Habeck 1970a, Kessell 1979). Upper subalpine forests (1680 to 2130 m asl) consist primarily of subalpine fir (Abies lasiocarpa (Hook.) Nutt.), Engelmann spruce (Picea engelmannii Parry) and whitebark pine (Pinus albicaulis Engelm.) (Habeck 1970a). Krummholz conifer and forb communities are found in the alpine environ-ments (2130 to over 3000 m asl) (Habeck and Choate 1963, 1967).
Geology of MD-GNP mostly consists of Precambrian layers of limestones, mudstones and sandstones that have been folded and uplifted during a mountain-building period beginning some 75 million years ago (Dyson 1967, Alt and Hyndman 1973). These layers contain sedimentary rocks of the Middle Proterozoic, Cretaceous and Tertiary age, local lava flows of Middle Proterozoic and diorite sills of late Proterozoic age (Earhart et al. 1989). Thick glacial till deposits surround McDonald Lake, whereas limestones, argillites and quartzites are found in the upper elevations.
Climate in MD-GNP is moderate maritime with cool wet winters and short, warm dry summers (Finklin 1986). Average annual precipitation ranges from 762 mm year−1 at West Gla-cier (Figure 5) to over 2032 mm year−1 at Flattop Mountain (Soil Conservation Service 1981, Finklin 1986). Maximum July daily temperatures range from 28 °C in the valleys to 24 °C at 2000 m elevation.
Fire regimes
Two distinct fire regimes are evident on the MD-GNP land-scape. Over the last four centuries, large stand-replacement fires were most common on moist sites with return intervals of 120--350 years (Habeck 1970b, Barrett et al. 1991). Some
stands in the drier areas of MD-GNP contain evidence of less severe surface fires occurring at intervals of 25--75 years (Bar-rett 1986, Bar(Bar-rett et al. 1991). This mixed fire regime is characterized by passive crown fires that kill all trees in some areas and nonlethal underburns that kill only the smallest trees in other areas (Habeck 1970b). Underburns often tend to kill most fire-intolerant species and favor the continued dominance of species resistant to fire damage (Ryan and Reinhardt 1988).
Methods
Field data collection
Ecological characteristics of all plant communities present in MD-GNP were assessed in thirty-six 0.05-ha circular plots located in a portion of a stand representative of a plant commu-nity. Site descriptions (e.g., elevation, aspect and slope), tree structure, age and composition, forest floor biomass (i.e., duff, litter and woody material loadings), soil depth and texture, and undergrowth canopy cover were measured on each plot using ECODATA methodology (Keane et al. 1990c, Jensen et al. 1993). Leaf area index was measured with a ceptometer (Pierce and Running 1988) and an LAI-2000 (Li-Cor Inc., Lincoln, NE) (Welles and Norman 1991). A soil pit was exca-vated at each plot to determine soil depth, and a sample of soil was taken for analysis of soil texture and water holding capac-ity. Each plot was georeferenced by means of a GPS receiver. Four additional plots were permanently established just out-side MD-GNP on the Coram Experimental Forest (Figure 5) to assess the accuracy of FIRE-BGC ecophysiological predic-tions. ECODATA techniques were also used to evaluate eco-logical characteristics on these plots. In addition, nine, 1-m2
litter traps and four soil respiration sampling sites were in-stalled at each plot. Litter traps were emptied monthly, and the extracted material was sorted by size and type (i.e., needlefall, 0--1-cm diameter woody material, 1--3-cm diameter woody material, 3--7-cm diameter woody material, cones, foliage and other matter), oven-dried and weighed. Soil respiration was measured monthly from changes in dry weight of canisters containing soda lime placed within covered tubes at the soil surface (Edwards 1982, Bowden et al. 1993). Leaf area indices were measured monthly with a ceptometer and an LAI-2000.
Spatial data layer creation
Site-level data layer Sites comprising the MD-GNP simula-tion study area were delineated by vegetasimula-tion types (Pfister et al. 1977) based on satellite imagery and plant autecology information. Plant community information from Kessell’s (1979) gradient model and the field data were used to map late-successional tree species distribution across the study area. Kessell’s (1979) gradient model predicts vegetation composi-tion from several environmental gradients including elevacomposi-tion, aspect, landform position and slope. Sites were classified by reducing the gradient information into a terrain model (Table 1) and then selecting the most shade-tolerant tree species (Minore 1979) as predicted by Kessell’s (1979) model and field data analysis (Pfister et al. 1977). These topographic gradients were spatially generated from the MD-GNP elevation (DEM) raster layer using GRASS GIS software (USA CERL 1990).
Satellite imagery from a July 15, 1988, Thematic Mapper (TM) scene was classified according to land cover (e.g., forest, shrub, herb and rock categories) (White and Running 1994) to refine the site delineations. All nonforested lands above 1600 m elevation were assumed to be incapable of maintaining tree cover (Habeck 1970a). The shrub and herb lands below 1600 m were assumed to be early seral stages of potential
forest vegetation types. It was also assumed that all classified rock, bare soil, permanent snowfields and glaciers, regardless of elevation, were unable to sustain tree communities.
The final MD-GNP site layer contains six forest and three nonforest potential vegetation classes described with respect to elevation, slope and aspect (Table 1). A comparison of this classification with georeferenced MD-GNP field data showed that the site layer is approximately 79% accurate. About 35% of MD-GNP is covered by nonforest types, and only 7% of the land area is comprised of subalpine sites (north and south, Table 1).
FIRE-BGC input values for each site were derived from field data. Forty years of weather data taken from nearby West Glacier weather station were extrapolated to the nine deline-ated sites by the MTCLIM climate model (Hungerford et al. 1989). FIRE-BGC cycles the 40 years of weather data in sequence to account for temporal autocorrelation of weather years. Fuel characteristics and understory species ecophysi-ological parameters were compiled from the literature (Keane et al. 1995) and field data. Mean fire intervals were taken mostly from Barrett et al. (1991) (Table 1).
Stand-level data layers A land cover type raster map of dominant vegetation based on a plurality of canopy cover was also created from the TM scene (White and Running 1994). Field data and Kessell’s (1979) gradient model were used to assign cover type descriptions based on the unsupervised spec-tral classification. This cover type raster map was combined with the site layer to produce a raster layer of stands hierarchi-cally nested within the sites.
Tree age and size structure data as measured on field plots representing these stands were used to provide the initial quantitative descriptions needed by FIRE-BGC. Other stand-level input data, such as fuel loading and understory biomass, were either quantified from field data or obtained from the
Table 1. Environmental characteristics of delineated FIRE-BGC potential vegetation sites in the MD-GNP study area.
Site Elevation Aspect Slopes Mean fire interval Comments
(m) (%) (years)
TSHE/THPL North1 900 to 1350 North < 70% 300 All tree species,
also lakeside settings
TSHE/THPL South 900 to 1500 South 10 to 70% 250 All species,
highly productive
ABLA Lower subalpine north 1350 to 1830 North < 70% 250 PICO, PIMO, PSME
associated species
ABLA Lower subalpine south 1500 to 2000 South 10 to 50% 250 PICO, LAOC, PSME
associated species
ABLA Upper subalpine north 1830 to 2300 North 10 to 50% 300 Mostly ABLA, PIEN
and rarely LALY
ABLA Upper subalpine south 2000 to 2450 South 10 to 70% 300 Mostly PIAL, ABLA
Shrub lands > 1600 All All 400 Permanent
Herb lands > 1600 All All 500 Permanent
Rock lands 900 to 3000 All 0 to 100% 500 Bare soil,
snow, glaciers
1 TSHE = Tsuga heterophylla, THPL = Thuja plicata, ABLA = Abies lasiocarpa, PICO = Pinus contorta, PSME = Pseudotsuga menziesii, LAOC
literature (Habeck 1970a).
Simulation methods
Two simulation scenarios were modeled by FIRE-BGC to investigate the effects of fire on MD-GNP ecosystem dynam-ics. (1) No Fires (NOFIRE)----a fire exclusion alternative where all fires are suppressed within the study area. Only the process of succession affects plant community composition and struc-ture. This scenario simulates potential impacts of a fully suc-cessful fire suppression program. (2) Pre-1900 Historical Fire Occurrence (HISTFIRE)----fires are stochastically simulated on the study area at approximately the same frequency as they occurred prior to European/American settlement. Each sce-nario was simulated for 200 years using the same initial con-ditions. No insect or disease epidemics were simulated in this application.
Output
The MAPMAKER program included in the LOKI FIRE-BGC application was used to create and summarize spatial distribu-tions of predicted net primary productivity (NPP), evapotran-spiration (ET), standing crop biomass (SC), available nitrogen (AVAILN) and leaf area index (LAI) predictions for the entire MD-GNP simulation landscape. Net primary productivity (NPP) is defined as gross photosynthesis minus leaf, stem and root respiration, and growth respiration. Evapotranspiration includes both transpiration from trees and evaporation from intercepted precipitation and from the ground. Standing crop (Mg C ha−1 year−1) is defined as aboveground standing biomass and includes only tree leaf and stem components. Available nitrogen (Mg N ha−1 year−1) is the amount of nitro-gen available for tree growth. Leaf area index is computed as a projected leaf area index using species-specific projection factors. Dominance type was also computed across the land-scape and it was evaluated as the tree species comprising the majority of basal area (m2 ha−1). Rock, shrub and herb lands are collectively called the nonforested dominance type.
Model test and verification
Tree growth comparison
Few data were available to compare long-term model predic-tions with actual condipredic-tions observed in MD-GNP. However, predicted diameter increments were contrasted with measured ring widths to evaluate FIRE-BGC short-term behavior.
Approximate stand conditions 40 years ago (around 1954) were reconstructed from the tree structure and age data based on sampled diameter and growth estimates from field plots (Keane et al. 1990a, 1995). These reconstructed stand condi-tions were used as starting condicondi-tions for the 40-year FIRE-BGC simulations from 1954 to 1994. This 40-year period is the length of the weather record from the West Glacier weather station. Simulated ring width increments and diameters of 10 trees from all plots were compared to the actual growth record taken from sampled trees representing modeled trees to
esti-mate model accuracy (Keane et al. 1995).
Stand compartment comparison
Monthly ecosystem process data collected from the Coram Experimental Forest plots in 1994 were also used to test the validity of monthly and annual FIRE-BGC predictions. Litter-fall, leaf area, woody fuel accumulation and soil respiration measured on the Coram plots were compared with FIRE-BGC estimates.
Results
Fire simulation
The FIRESTART model simulated the ignition of two fires at simulation years 47 and 163. The stand-replacement fire simu-lated at Year 47 burned 4321 ha of hemlock--cedar forests at the northwestern end of MD-GNP and killed most trees. Fire intensities ranged from 40 to 140,000 kW m−2 and averaged approximately 1496 kW m−2. The mixed-intensity fire mod-eled in Year 163 burned 5187 ha in the high elevation subalpine fir forests in the northeastern portion of MD-GNP. This fire was less intense (10 to 80,000 kW m−2) and created a more complex pattern of tree mortality and fire severity than the stand-replacement fire.
Ecosystem simulation
Landscape summary predictions for the two fire scenarios of the five ecosystem characteristics of NPP, ET, SC, AVAILN and LAI for forested stands are shown in Figures 6 to 10. Net primary productivity declined from approximately 25 to < 5 Mg C ha−1 year−1 over the 200-year simulation for the NOFIRE scenario (Figure 6). Long-term trends of NPP under the HISTFIRE scenario indicated higher values toward the end of the simulation, although there was a reduction in NPP immediately after the fires (Figure 6). Evapotranspiration in-creased in the absence of fire and dein-creased under the pre-1900 fire regime (Figure 7). Standing crop simulation estimates
indicated higher aboveground biomass values with fire in the long term (Figure 8). Available nitrogen tended to increase directly after fire and remained higher than predicted under the fire exclusion scenario. The rapid increase in available N directly after fire was short lived (Figure 9). Landscape leaf area indices in the two scenarios were comparable until the fire in Year 47 (Figure 10). After Year 47, the LAI of the HISTFIRE landscape tended to remain higher than the LAI predicted for the NOFIRE landscape.
Estimates of NPP were often higher in the low-elevation, hemlock--cedar forests during cool years than during warm years, and greater in high-elevation, subalpine fir forests dur-ing warm years than durdur-ing cool years. Analysis of tree core data shows high-elevation forests experiencing high rates of NPP (> 10 Mg C ha−1 year−1) over the last decade, which is one of the warmest and driest on record (Finklin 1986).
Ecosystem process predictions for simulation Year 100 are summarized by site in Table 2 for the NOFIRE scenario. Predicted ET and SC were greatest in the low-elevation
hem-lock--cedar forests and lowest in the shrub, herb and rock lands. The most productive stands (i.e., greatest NPP) were usually found on southern aspects, except at the highest elevations. Predicted standing crop biomass was greatest in the low-eleva-tion forests (> 100 Mg C ha−1) where trees can reach 35 m in height and 150 cm in diameter. These productive forests also had the highest LAI values ranging from 4 to 9. Mid- and high-elevation forests had LAI values ranging from 1 to 6 (Table 2). Early seral stands created by fire and high elevation shrub and herb lands had low rates of ET.
Predicted available nitrogen showed a threefold increase in abundance after each fire, especially after the fire in Year 47. In the absence of fire, AVAILN seemed to parallel the 40-year climate cycle but increased over the 200-year simulation. Pre-dicted AVAILN was highest in the low-elevation hemlock--ce-dar forests and riparian forests bordering major streams. Shrubfields and forbfields also had high AVAILN, but this could be a result of high initial AVAILN input values.
FIRE-BGC predicted that shade-tolerant species comprise a
Figure 7. Evapotranspiration (Mg H2O ha−1 year−1) predictions over the 200-year FIRE-BGC simulation with and without fire for forested sites across the entire MD-GNP landscape.
Figure 8. Standing crop biomass (Mg C ha−1 year−1) predictions over the 200-year FIRE-BGC simulation with and without fire for forested sites across the entire MD-GNP landscape.
Figure 9. Available nitrogen (Mg N ha−1 year−1) predictions over the 200-year FIRE-BGC simulation with and without fire for forested sites across the entire MD-GNP landscape.
majority of the landscape (65%) after 200 years of simulation under the exclusion of fire; however, the proportion of land-scape occupied by shade-intolerant species (43% at year 200) was much greater when fires were included in the simulation. Approximately 35% of the initial landscape is dominated by shade-tolerant species and 30% is comprised by shade-intoler-ant species. Shade-intolershade-intoler-ant species like L. occidentalis died more rapidly during the NOFIRE simulation than observed in the field. Conversely, simulated growth rates of shade-tolerant species like A. lasiocarpa were higher than measured rates, and these trees seemed to live longer and become larger than expected. Consequently, especially in the highly productive low- and midelevation forests, predicted stem basal areas were higher than those measured, ranging from 40 to 110 m2 ha−1.
Discussion
Model predictions
Values of NPP seem unreasonably high at the start of a simu-lation because of inaccurate initialization of the model (Fig-ure 6). Because FIRE-BGC must initialize all model compartments from either field data or regression equations, this often results in inconsistencies between the simulation environment and initial vegetation conditions. For example, crown weight regression equations used in FIRE-BGC some-times overpredict tree leaf carbon, which will overpredict leaf area for a stand. Consequently, NPP and other process simula-tions may be overpredicted until ecosystem states correspond to the biophysical environment. Initial available nitrogen esti-mates are another source of initialization error. Because AVAILN was not measured, the initial estimate of 0.4 Mg N ha−1 might be too high for current ecosystem conditions. However, FIRE-BGC ecosystem components seem to equili-brate by simulation Years 50 to 60 and produce reasonable NPP estimates thereafter (White and Running 1994). As a result, differences in NPP after the fire of Year 163 are probably more reliable than NPP estimates after the fire of Year 47.
Evapotranspiration predictions (Figure 7) under the pre-1900 fire scenario showed a marked reduction after fires, which seemed to remain for 50 years. It is assumed in FIRE-BGC that water not transpired or evaporated contributes to
stream runoff. Most of the increased runoff due to fires occurs during spring snowmelt. The reduction in leaf interception of snow and rain after fires may also contribute to increased runoff.
Standing crop predictions (Figure 8) showed a reduction in aboveground biomass after fires due to fire mortality. Major decreases in SC in nonfire years can be attributed to large tree (> 30 cm diameter) mortality resulting from climate or compe-tition-related stress. This large decrease in standing crop is probably not representative of the stand, but due to the simula-tion plot area specificasimula-tion. It is assumed that the simulasimula-tion plot represents the entire stand, and the size of this plot is specified at the beginning of simulation. Smaller plots result in fewer trees to simulate and faster execution times; however, small simulation plots do not represent the stand if simulated trees are large (> 50 cm diameter) because the death of one tree results in major changes in ecosystem compartments of the plot.
The increase in available N under the fire exclusion scenario from Years 70 to 170 was caused by the high rates of decom-position predicted by FIRE-BGC. It seems litter, duff and woody material deposited on the forest floor from the trees was decomposed faster than observed in MD-GNP. Duff and litter depths of 2 to 7 cm are common in the low- and midelevation forests, but FIRE-BGC predicts duff and litter depths of 1 to 3 cm, indicating that adjustments to modeled decomposition rates are needed.
Model validation
Measured ring widths of 10 trees over the last 40 years are compared with FIRE-BGC predicted tree ring widths in Fig-ure 11. Ring widths for shade-intolerant species were under-predicted, whereas ring widths of shade-tolerant species were overpredicted. For example, ring width predictions for the lodgepole pine in Figure 12 varied widely from observed trends during the first years of simulation but were generally underpredicted during the last years of simulation. Predicted ring widths for the shade-tolerant subalpine fir (Figure 13) seemed to agree well with measured values.
Year-to-year trends in ring width predictions generally agreed with observed trends (Figures 11 and 12), although the magnitude of variation was larger for FIRE-BGC predictions. Table 2. Predictions of ecosystem process variables for the NOFIRE simulation scenario at Year 100.
Site NPP ET SC AVAILN LAI
(Mg C ha−1 year−1) (Mg H2O ha−1 year−1) (Mg C ha−1 year−1) (Mg N ha−1 year−1)
TSHE/THPL North1 5.48 7.66 133.7 1.83 6.69
TSHE/THPL South 6.81 6.45 145.2 1.00 7.01
ABLA Lower subalpine north 5.38 3.46 141.7 1.45 6.07
ABLA Lower subalpine south 6.04 4.56 70.5 1.12 4.97
ABLA Upper subalpine north 1.02 2.58 45.0 0.96 5.00
ABLA Upper subalpine south 1.82 3.33 36.7 0.94 2.14
Shrub lands > 0.50 > 1.00 > 1.0 0.10 1.00
Herb lands > 0.50 > 1.00 > 1.0 0.10 0.50
Rock lands > 0.1 > 1.00 > 1.0 0.05 0.20
This discrepancy could be related to the treatment of tree carbon storage in the model. Currently, FIRE-BGC sets aside only 10% of the previous year’s net carbon gain as storage for next year’s growth. Running and Hunt (1993) used storage percentages of 25% in BIOME-BGC.
Model limitations
The FIRE-BGC initialization process often creates incompat-ible conditions within the model. The large amount and speci-ficity of input data needed to initialize model compartments preclude the use of field data to quantify all starting conditions. Regression equations provide efficient and economical esti-mates of initial tree components, but often do not account for biophysical influences on tree and stand characteristics. As a result, ecosystem compartments are in conflict with simulated ecosystem processes. There are two possible solutions to this difficulty. Field measurements can be made of all ecosystem components in the simulation area. However, this solution is almost intractable because of the cost and time required to complete such a task for an entire landscape. A more reason-able solution may be to run the model using initial values computed from allometric equations until ecosystem states have stabilized and then use the stabilized simulated ecosys-tem component output values as input on subsequent runs.
FIRE-BGC does not simulate undergrowth ecosystem proc-esses at the detail provided for tree species. Consequently, processes such as NPP and ET are often underpredicted for early seral forest stands and perennial shrub and herb lands because of the lack of trees in the stand. Undergrowth contri-butions to the forest floor are also overpredicted because there is no feedback to regulate growth based on environmental conditions.
FIRE-BGC does not adequately relate life cycle charac-teristics of trees with ecophysiological dynamics of the stand. The link between tree regeneration, mortality and growth with stand carbon, nitrogen and water cycling needs a more com-prehensive and mechanistic treatment. There needs to be a closer relationship to NPP and tree growth cycles, and tree regeneration and mortality must be connected to NPP and other stand-level processes.
Carbon allocation from the stand-level to individual trees needs improvement. Contributions of stand photosynthate pro-duction by trees of different species and sizes is a complex computation in the model. FIRE-BGC calculates a species-specific allocation factor from daily ecosystem process calcu-lations assuming stand leaf area was wholly comprised of the species in question. The species with the greatest photosyn-thetic production would receive the highest allocation factor. Then another allocation factor is computed to represent a tree’s position in the forest canopy. This factor is based on available light and the species’ shade tolerance. The last factor is based on the proportion of stand leaf area comprised of the tree in question. Currently, these allocation factors cause more carbon allocation to shade-tolerant trees and understory trees. Modifi-cation of this procedure should take into account the efficiency and rate of photosynthate production relative to the amount of leaf area by species and tree size.
Figure 11. Measured versus observed tree ring growth increments for 10 trees over 40 years (R2 = 0.89, slope = 0.94, intercept = 0.015).
Figure 12. Predicted versus observed tree ring growth increments for a 13-cm diameter, shade-intolerant lodgepole pine (Pinus contorta) tree.
The weather data we used for simulating the MD-GNP ecosystem processes are probably not appropriate for long-term successional modeling. First, a 40-year weather record is too short for multicentury simulations. Second, the raw weather data used in the simulation came from one weather station just outside the drainage boundaries. Extrapolation of these data to the complex mountain terrain in the MD-GNP may not fully represent the true range of weather conditions that controlled vegetation development. Many local weather events such as cold air drainage, frost pockets and wind-fun-neling are not predicted by the MTCLIM weather model (Hungerford et al. 1989). Long-term weather records from many weather stations are needed for long-term successional modeling in FIRE-BGC.
Conclusions
The mechanistic succession model FIRE-BGC is still in the developmental stage and many model process algorithms and parameters need improvement. However, general trends pre-dicted by the model seem reasonable. FIRE-BGC simulation of the MD-GNP landscape under fire exclusion predicts de-creases in net primary productivity and available nitrogen, and increases in standing crop and evapotranspiration. The same landscape under a pre-1900 fire regime simulation scenario shows increases or stability in net primary productivity and available N, and decreases in standing crop and evapotranspi-ration. Shade-tolerant tree species eventually dominate the landscape in the absence of fire.
Acknowledgments
This research was performed, in part, in cooperation with the National Biological Service and Glacier National Park’s Global Change Re-search Program under Interagency Agreements 1430-1-9007 and 1430-3-9005. We thank Joseph White, Peter Thornton, Polly Thornton and Mike White of the University of Montana, Dan Fagre and Carl Key of National Biological Service, Glacier Park Field Station, Stephen Arno, James Menakis, Todd Carlson, Kirsten Schmidt, Sephanie Katz, Donald Long, Janis Garner and Tom Cook of USDA Forest Service, Intermountain Research Station, and Mark Finney and Collin Bevins of Systems for Environmental Management, Missoula, Montana.
References
Albini, F.A. 1976a. Estimating wildfire behavior and effects. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Gen. Tech. Rep. INT-30, 92 p.
Albini, F.A. 1976b. Computer-based models of wildland fire behavior: a user’s manual. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, 68 p.
Alt, D.D. and D.W. Hyndman. 1973. Rocks, ice and water: the geology of Waterton-Glacier Park. Mountain Press Publishing Co., Mis-soula, MT, 104 p.
Andersen, M. 1991. Mechanistic models for the seed shadows of wind-dispersed plants. Am. Nat. 137:476--497.
Anderson, H.E. 1969. Heat transfer and fire spread. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Res. Pap. INT-69, 20 p.
Anderson, H.E. 1982. Aids to determining fuel models for estimating fire behavior. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Gen. Tech. Rep. INT-122, 22 p.
Baker, W.L. 1989. Effect of scale and spatial heterogeneity on fire-in-terval distributions. Can. J. For. Res. 19:700--706.
Barrett, S.W. 1986. Fire history of Glacier National Park: middle fork Flathead River drainage. Intermount. Fire Sci. Lab., Missoula, MT, 32 p.
Barrett, S.W., S.F. Arno and C.H. Key. 1991. Fire regimes of western larch--lodgepole pine forests in Glacier National Park, Montana. Can. J. For. Res. 21:1711--1720.
Bevins, C.D., P.L. Andrews and R.E. Keane. 1994. Forest succession modeling using the Loki software architecture. Lesn. For. 41:158--163.
Bevins, C.D. and P.L. Andrews. 1994. The Loki software architecture for fire and ecosystem modeling: a tinker toy approach. 12th Conf. on Fire and For. Meteorol., Soc. Am. For., Washington, D.C., pp 252--260.
Bonan, G.B. and H.H. Shugart. 1989. Environmental factors and ecological processes in boreal forests. Annu. Rev. Ecol. Syst. 20:1--28.
Bossel, H. and H. Schäfer. 1990. Ecophysiological dynamic simula-tion model of tree growth, carbon, and nitrogen dynamics. In Forest Simulation Systems. Eds. L.C. Wensel and G.S. Biging. Univ. of Calif., Div. of Agric. Nat. Resources, Berkeley, CA, Bull. 1927, 420 p.
Bowden, R.D., K.J. Nadelhoffer, R.D. Boone, J.M. Melillo and J.B. Garrison. 1993. Contributions of aboveground litter, below-ground litter and root respiration to total soil respiration in a temperate mixed hardwood forest. Can. J. For. Res. 23:1402--1407.
Brown, J.K. 1970. Ratios of surface area to volume for common fire fuels. For. Sci. 16:101--105.
Brown, J.K. 1981. Bulk densities of nonuniform surface fuels and their application to fire modeling. For. Sci. 27:667--683.
Brown, J.K. and C.D. Bevins. 1986. Surface fuel loadings and pre-dicted fire behavior for vegetation types in the northern Rocky Mountains. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Res. Note INT-358, 9 p.
Brown, J.K. and T.E. See. 1981. Downed dead woody fuel and biomass in the northern Rocky Mountains. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Gen. Tech. Rep. INT-117, 48 p.
Busing, R.T. 1991. A spatial model of forest dynamics. Veg. Sci. 92:167--179.
Clark, J.S. 1990. Fire and climate change during the last 750 years in northwestern Minnesota. Ecol. Monogr. 60:135--159.
Crutzen, P.J. and J.G. Goldammer. 1993. Fire in the environment: the ecological, atmospheric and climatic importance of vegetation fires. John Wiley and Sons, New York, 456 p.
Davis, K.M. 1981. Fire history of a western larch/Douglas-fir forest type in northwestern Montana. Proc. Fire History Workshop, Tuc-son, AZ, pp 69--74.
DeBano, L.F. 1991. The effect of fire on soil properties. Proc. Manage. and Prod. of Western-Montane For. Soils, Boise, ID, pp 151--157. Dixon, R.K., R.S. Meldahl, G.A. Ruark and W.G. Warren. 1990.
Process modeling of forest growth responses to environmental stress. Timber Press, Portland, OR, 320 p.
Dyrness, C.T., K. Van Cleve and J.D. Levison. 1989. The effect of wildfire on soil chemistry in four forest types in interior Alaska. Can. J. For. Res. 19:1389--1396.
Earhart, R.L., O.B. Raup, J.W. Whipple, A.L. Isom and G.A. Davis. 1989. Geologic maps, cross section, and photographs of the central part of Glacier National Park, Montana. USGS Misc. Invest. Ser. Map I-1508-B.
Edwards, N.T. 1982. The use of soda-lime for measuring respiration rates in terrestrial systems. Pedobiologia 23:321--330.
Finklin, A.I. 1986. A climatic handbook for Glacier National Park with data for Waterton Lakes National Park. USDA Forest Service, Gen. Tech. Rep. INT-204, 124 p.
Finney, M.A. 1994. Modeling the spread and behavior of prescribed natural fires. 12th Conf. on Fire and For. Meteorol., Soc. of Am. For., Bethesda, MD, pp 138--144.
Fischer, W.C. and A.F. Bradley. 1987. Fire ecology of western Mon-tana forest habitat types. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Gen. Tech. Rep. INT-223, 95 p. Fosberg, M.A. 1970. Drying rates of heartwood below fiber saturation.
For. Sci. 16:57--63.
Gosz, J.R. 1981. Nitrogen cycling in coniferous ecosystems. Ecol. Bull. 33:405--426.
Greene, D.F. and E.A. Johnson. 1989. A model of wind dispersal of winged or plumed seeds. Ecology 70:339--347.
Habeck, JR. 1968. Forest succession in the Glacier Park cedar--hem-lock forests. Ecology 49:872--880.
Habeck, J.R. 1970a. The vegetation of Glacier National Park, Mon-tana. USDI National Park Service, West Glacier, MT, 132 p. Habeck, J.R. 1970b. Fire ecology investigations in Glacier National
Park. Historical considerations and current observations. USDI National Park Service, West Glacier, MT, 80 p.
Habeck, J.R. and C.M. Choate. 1963. An analysis of krummholtz communities at Logan Pass, Glacier National Park. Northwest Sci. 37:165--166.
Habeck, J.R. and C.M. Choate. 1967. Alpine communities at Logan Pass, Glacier National Park. Proc. Montana Acad. Sci. 27:36--54. Habeck, J.R. and R.W. Mutch. 1973. Fire-dependent forests in the
Northern Rocky Mountains. Quat. Res. 3:408--424.
Heinselman, M.L. 1981. Fire intensity and frequency as factors in the distribution and structure of northern ecosystems. In Fire Regimes and Ecosystem Properties. Eds. H.A. Mooney, T.M. Bonnicksen, N.L. Christensen, J.E. Lotan and W.A. Reiners. USDA Forest Service, Gen. Tech. Rep. WO-26, pp 7--58.
Hungerford, R.D., R.R. Nemani, S.W. Running and J.C. Coughlan. 1989. MTCLIM: A mountain microclimate simulation model. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Res. Pap. INT-414, 52 p.
Jensen, M.E., W. Hann, R.E. Keane, J. Caratti, and P.S. Bourgeron. 1993. ECODATA----A multiresource database and analysis system for ecosystem description and evaluation. In Eastside Forest Eco-system Health Assessment, Vol. II. EcoEco-system Management: Prin-ciples and Applications. Eds. M.E. Jensen and P.S. Bourgeron. USDA Forest Service, Nat. For. Syst. Inform. Rep., pp 249--265. Johnson, E.A. 1979. Fire reoccurrence in the subartic and its
implica-tions for vegetation composition. Can. J. Bot. 57:1374--1379. Keane, R.E., S.F. Arno and J.K. Brown. 1989. FIRESUM----An
eco-logical process model for fire succession in Western conifer forests. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Gen. Tech. Rep. INT-266, 76 p.
Keane, R.E., S.F. Arno and J.K. Brown. 1990a. Simulating cumulative fire effects in ponderosa pine/Douglas-fir forests. Ecology 71:189--203.
Keane, R.E., S.F. Arno, J.K. Brown and D.F. Tomback. 1990b. Mod-eling stand dynamics in whitebark pine (Pinus albicaulis) forests. Ecol. Model. 51:73--95.
Keane, R.E., M.E. Jensen and W.J. Hann. 1990c. ECODATA and ECOPAC----analytical tools for integrated resource management. The Compiler 8:24--37.
Keane, R.E., P. Morgan and S.W. Running. 1995. FIRE-BGC----A mechanistic ecological process model for simulating fire succes-sion on Northern Rocky Mountain coniferous forest landscapes. USDA Forest Service, Res. Pap. INT-RP-434. In press.
Kercher, J.R. and M.C. Axelrod. 1984. A process model of fire ecology and succession in a mixed-conifer forest. Ecology 65:1725--1742. Kessell, S.R. 1979. Gradient modeling: resource and fire management.
Springer Verlag, New York, 432 p.
Kimmins, J.P. 1993. Scientific foundations for the simulation of eco-system function and management in FORCYTE-11. Forestry Can-ada, Northwest Region, Northern Forestry Center, Edmonton, Alberta, Inf. Rep. NOR-X-328, 88 p.
Link, S.O., G.W. Gee, M.E. Thiede and P.A. Beedlow. 1990. Response of a shrub-steppe ecosystem to fire: soil water and vegetational change. Arid Soil Res. Rehabil. 4:163--172.
Maggs, J. 1988. Organic matter and nutrients in the forest floor of a
Pinus elliottii plantation and some effects of prescribed burning and superphosphate addition. For. Ecol. Manage. 23:105--119. McCaughey, W.W., W.C. Schmidt and R.C. Shearer. 1985. Seed
dis-persal characteristics of conifers of the Inland Mountain West. In
Conifer Seed in Inland Mountain. West. Ed. R.C. Shearer. USDA Forest Service, Gen. Tech. Rep INT-203, Missoula, MT, pp 50--61. Minore, D. 1979. Comparative autecological characteristics of north-western tree species: a literature review. USDA Forest Service, Pacific Northwest For. and Range Exp. Stn., Portland, OR, Gen. Tech. Rep. PNW-87, 28 p.
Mushinsky, H.R. and D.J. Gibson. 1991. The influence of fire peri-odicity on habitat structure. In Habitat Structure. Ed. S.S. Bell. Chapman and Hall, New York, pp 237--259.
Peet, R.K. 1988. Forests of the Rocky Mountains. In North American Terrestrial Vegetation. Eds. M.G. Barbour and W.D. Billings. Cam-bridge University Press, New York, pp 63--96.
Pfister, R.D., B.L. Kovalchik, S.F. Arno and R.C. Presby. 1977. Forest habitat types of Montana. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Gen. Tech. Rep. INT-34, 174 p. Pierce, L.L. and S.W. Running. 1988. Rapid estimation of coniferous
forest leaf area index using a portable integrating radiometer. Ecol-ogy 69:1762--1767.
Raison, R.J. 1979. Modification of the soil environment by vegetation fires, with particular reference to nitrogen transformations: a re-view. Plant Soil 51:73--108.
Reed, K.L. 1980. An ecological approach to modeling growth of forest trees. For. Sci. 26:33--50.
Reich, P.B., M.D. Abrams, D.S. Ellsworth, E.L. Kruger and T.J. Tabone. 1990. Fire affects ecophysiology and community dynamics of central Wisconsin oak forest regeneration. Ecology 71:2179--2190.
Rothermal, R.C. 1972. A mathematical model for predicting fire spread in wildland fuels. USDA Forest Service, Intermount. For. Range Exp. Stn., Ogden, UT, Res. Pap. INT-115, 40 p.
Rowe, J.S. 1983. Concepts of fire effects on plant individuals and species. In The Role of Fire in Northern Circumpolar Ecosystems. Eds. R.W. Wein and D.A. MacLean. John Wiley and Sons, New York, pp 135--151.
Running, S.W. and J.C. Coughlan. 1988. A general model of forest ecosystem processes for regional applications. I. Hydrologic bal-ance, canopy gas exchange and primary production processes. Ecol. Model. 42:125--154.
Running, S.W. and S.T. Gower. 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. and E.R. Hunt. 1993. Generalization of a forest
ecosys-tem process model for other biomes, BIOME-BGC, and an appli-cation for global scale models. In Scaling Physiological Processes: Leaf to Globe. Eds. J.R. Ehleringer and C. Field. Academic Press Inc., London, pp 141--157.
Ryan, K.C. and E.C. Reinhardt. 1988. Predicting postfire mortality of seven western conifers. Can. J. For. Res. 18:1291--1297.
Soil Conservation Service. 1981. Average annual precipitation, Mon-tana, based on 1941--1970 base period. USDA Soil Conserv. Serv-ice, Portland, OR, 16 p.
Shugart H.H. and S.W. Seagle. 1985. Modeling forest landscapes and the role of disturbance in ecosystems and communities. In The Ecology of Natural Disturbance and Patch Dynamics. Academic Press, San Diego, CA, pp 353--368.
Tomback, D.F., L.A. Hoffmann and S.K. Sund. 1990. Coevolution of whitebark pine and nutcrackers: implications of forest regeneration.
In Whitebark Pine Ecosystems: Ecology and Management of a High Mountain Resource. USDA Forest Service, Bozeman, MT, Gen. Tech. Rep. INT-270, pp 118--130.
Turner, M.G. and W.H. Romme. 1994. Landscape dynamics in crown fire ecosystems. Landscape Ecol. 9:59--77.
Urban, D.L., G.B. Bonan, T.M. Smith and H.H. Shugart. 1991. Spatial applications of GAP models. For. Ecol. Manage. 42:95--110. Urban, D.L. and H.H. Shugart. 1992. Individual-based models of
forest succession. In Plant Succession Theory and Prediction. Eds. D.C. Glenn-Lewin, R.K. Peet and T.T. Veblen. Chapman and Hall, London, pp 249--292.
USA CERL. 1990. GRASS 4.0 reference manual. U.S. Army Corps of Engineers, Construction Engineering Res. Lab., Champaign, IL, 208 p.
Van Wagner, C.E. 1978. Age-class distribution and the forest fire cycle. Can. J. For. Res. 8:220--227.
Welles, J.M. and J.M. Norman. 1991. Instrument for indirect measure-ment of canopy architecture. Agron. J. 83:818--825.
White, J.D. and S.W. Running. 1994. Testing scale dependent assump-tions in regional ecosystem simulaassump-tions. J. Veg. Sci. 5:686--702. Wright, H.A. and A.W. Bailey. 1982. Fire ecology: United States and
