Fire-prone landscapes (i.e., fire-dependent forest ecosystems with historically frequent, < 100-yr wildfire) are globally widespread and produce valuable ecosystem services, including wood fiber, fuel, recreation, regulation of carbon emissions, and biodiversity (Noss et al. 2006, Bowman et al. 2009). Coupled human and natural systems (CHANS) in fire-prone landscapes are characterized by complex interactions between fire-dependent natural systems and nearby rural and urban human communities where high-severity wildfire is undesirable. In some parts of the world, such landscapes are characterized as the “wildland-urban interface” (WUI), a transition or contact zone between unpopulated fire-adapted natural vegetation and populated areas having elevated levels of risk for loss of homes and lives to wildfire. The spatial extent of ecological and socioeconomic dynamics of fire-prone landscapes is broader than just the WUI, however, encompassing the full extent of wildlands in which the WUI is embedded (Pyne 2008, Ager et al. 2015). Although many studies of social-ecological interactions in fire-prone ecosystems have focused on the WUI, far fewer have focused on the larger landscape and system, of which the WUI is only one part.
The application of a CHANS approach can help reveal complexities and interactions in social-ecological systems that are not visible when ecological or social systems are studied in isolation or if spatial heterogeneity and temporal lags are not taken into account (Liu et al. 2007). Understanding of fire-frequent CHANS is limited by gaps in knowledge about fundamental social and ecological processes and how they interact, as well as a lack of models that integrate human and natural systems across spatial and temporal scales. Retrospective studies of individual fire events can be quite informative about how fuels management, vegetation, topography, and weather influence fire behavior (e.g., Thompson and Spies 2009), and how people respond to specific fires (McCaffrey et al. 2012). However, it is nearly impossible to evaluate the variability and cumulative effects of fire on landscapes managed under multiple alternative management scenarios or climate regimes without simulation models.
Here, we use an agent-based landscape model to explore how forest vegetation management alternatives and policies affect fire behavior and ecosystem services in a large multiownership fire-prone landscape in the eastern Cascade Mountains of Oregon, USA. Our questions and management issues are similar to those that prevail in many landscapes of the western United States and other countries where natural resource and amenity-based economies and communities are present within a fire-prone, multiownership forest landscape, and where the goals of actors include both wildfire protection and forest restoration (Stephens and Ruth 2005, Reinhardt et al. 2008). Historical forest policies (e.g., fire suppression) have had a major effect on forest landscapes and ecosystems (Langston 1995, Stine et al. 2014). One of our goals was to develop tools that could be used to evaluate new policies and potentially avoid future undesirable or unintended outcomes of wildfire management actions.
Although many landscape models linking vegetation dynamics, management activity, climate, and fire behavior have been developed (e.g., Beukema et al. 2003, Chew et al. 2004, Keane et al. 2004, Scheller et al. 2007, O’Connor et al. 2011), relatively few models have been used to evaluate policy scenarios on multiowner landscapes over large scales (see Gustafson et al. 2007, Spies et al. 2007). For instance, most prior efforts to model fire-prone landscapes have lacked adequate representation of human decision making with respect to land management actions. Only recently have landscape models been developed to evaluate the spatial and temporal effects of land management across multiple ownerships on fire and ecosystem services over large areas (e.g., Millington et al. 2008, Scheller et al. 2011, Syphard et al. 2011, Loudermilk et al. 2013, Conlisk et al. 2015).
We used an existing agent-based modeling framework, Envision (Bolte et al. 2006), to link fire behavior, human decision making, and landscape outcomes related to wildfire, forest landscape conditions, and ecosystem services (Spies et al. 2014), and to examine broader social and ecological effects of different federal management strategies across all ownerships. Our main objective was to evaluate how alternative forest management strategies might affect wildfire, vegetation dynamics, human exposure to fire, biodiversity, and ecosystem services across a multiownership landscape. More specifically, we ask: (1) Do current or alternative accelerated restoration scenarios on federal lands, and associated linked changes on private corporate forestlands, reduce the occurrence of high-severity fire and fire hazard across the entire landscape compared with a scenario of no federal management? And (2) What are the social and ecological trade-offs, if any, associated with alternative forest management actions and wildfire? The social and ecological values and perspectives in this landscape are broad and diverse but fall into two main areas: fire protection (homes and forest values and forest restoration; Fischer and Jasny 2017) and ecosystem services, including timber production and wildlife habitat. We evaluated outcomes related to these perspectives, paying particular attention to critical social-ecological issues on federal lands, which dominate the landscape (nearly 40% of forest area). The major goals on federal lands are to protect homes in the WUI, restore fire-resilient forest structures, maintain dense forest habitat (which is sensitive to loss from fire) for the northern spotted owl (Strix occidentalis; listed under the U.S. Endangered Species Act;), and provide timber to support local communities and forest management infrastructure.
We focus on aggregate effects across all ownerships and plan to explore differences and spatial interactions among ownerships and climate change in subsequent work. Ager and colleagues describe the creation of the fire subcomponent of the model by linking existing fire models to Envision, and then use the model to explore fire feedbacks (Ager, Barros, Day et al. unpublished manuscript hereafter Ager et al. unpublished manuscript). Barros et al. (2017) examine how fuel treatments on federal lands affect fire size and behavior. Charnley et al. (2017) focus on how adaptation strategies to fire-prone landscapes differ among large landowners (e.g., federal, state, and corporate).
The 12,529-km² study area is characterized by mountainous topography and steep environmental gradients running from cool, wet subalpine mountain forests to moist and dry mixed-conifer and pine forests to semi-arid juniper woodlands that occur on more gently sloping topography at lower elevations (Fig. 1; Appendix 1 and 2). These forest types have different pre-EuroAmerican fire regimes (Agee 1993) and vary in effects of fire exclusion and logging (Merschel et al. 2014). Forest ownership patterns are also heterogeneous but dominated by federal lands. Landowners’ management objectives include wilderness experiences, producing timber, and maintaining residential homesteads (Spies et al. 2014). The WUI occupies a relatively small area (9.7%) within the larger region, but receives substantial attention from policy makers and forest managers. Historical land-use activities have left a strong imprint on the vegetation and fire regimes. Euro-American activities, which began in the mid-1880s, included grazing, logging, road building, and disruption of Native American resource use practices (Robbins 1997, Hessburg and Agee 2003). The disruption of Native American cultures would also have reduced fire ignitions, though natural ignitions from lightning are common in this region (Ager et al. unpublished manuscript). Studies of existing older pine and mixed-conifer patches indicate that the density of shade-tolerant understory trees (e.g., grand fir [Abies grandis]) has increased several fold since 1900, and the presence of large, old shade-intolerant pines has been reduced by as much as 70% as a result of partial cutting for timber (Merschel et al. 2014).
Management issues in the study region include: (1) balancing the potentially competing goals of restoring or managing forests for resilience to fire, protecting structures in the WUI, and meeting goals for producing timber and maintaining or restoring wildlife habitat (Spies et al. 2006); (2) maximizing effectiveness of limited funding for forest restoration and wildfire protection activities; and (3) engaging public land managers with stakeholder collaborative groups to advance the understanding of the need for landscape-scale restoration.
We follow the “overview, design concepts, and details” (ODD) protocol for describing our agent-based model (Grimm et al. 2010). Our model has two purposes: (1) to advance scientific understanding of the dynamics and interactions of forest management, fire, and vegetation across landscapes characterized by multiple owners; and (2) to contribute to management and collaborative restoration of fire-prone landscapes by serving as a tool for managers and stakeholders to evaluate ecological and social outcomes of different management, policy, and climate scenarios. Other components of the ODD template are described in Appendix 3.
The conceptual model of the CHANS (Spies et al. 2014) was implemented using Envision (Bolte et al. 2006). Envision is a spatially explicit agent-based modeling framework that is able to simulate forest management and wildfire disturbances and changes in vegetation and fuel that, in turn, affect fire behavior, forest succession, and ecosystem services over long time frames (Fig. 2). By integrating models of forest succession, wildfire occurrence and spread, and forest management into a spatially explicit model, we can examine how these process interact over space and time and explore possible trade-offs among ecosystem services. We started with the existing Envision basic structure for actor decisions, forest succession, and wildfire, but modified these to add more management options, successional states, and pathways that are relevant to this forest region, as well as a wildfire submodel (see Wildfire submodel) that was customized for the historical ignitions, climate, and wildfire history of this landscape. Agents are the major landowners (see Management submodel) who initiate forest management activities (e.g., silviculture, prescribed fire) that affect vegetation and fuel conditions on the landscape delineated by individual decision units (IDUs). Agents are spatially distributed (Fig. 1B) across forested landscape characterized by variability in forest composition and structure (Fig. 1A). They make decisions about forest management (e.g., silviculture and prescribed fire) based on goals (see Management submodel). Wildfires are modeled by coupling a semiempirical mechanistic fire spread algorithm with a stochastic fire event simulator representing ignitions and weather. Fires occur and spread with variable behavior based on vegetation, topography, and weather. Changes in the environment (e.g., vegetation structure and composition) create feedbacks that influence human decisions about future management activities and wildfire behavior. For example, if forests grow into new successional stages or are burned by fire, the type and probability of management changes according to empirically developed rules (Appendix 4). Some of the human actors (family forest owners and homeowners) respond to the experience of nearby wildfire by changing their likelihood of conducting certain activities such as thinning and reducing fire hazard at homesites (Olsen 2017; Kline, White, Fischer, et al. unpublished manuscript hereafter Kline et al. unpublished manuscript). The actions of federal, tribal, and private corporate landowners are based on targets of timber volume produced or area treated, and preferences for certain vegetation types and landscape positions (e.g., distance to roads). Agent actions only affect other agents indirectly through the effect of management on an IDU as it might affect the spread of wildfire across many IDUs and ownerships. We also sought to incorporate the role of social networks in influencing actor decision making (Fischer and Jasny 2017); however, network dynamics are not included in the current model version. External influences such as changing federal policy and economic conditions are modeled as scenarios.
Vegetation succession was modeled using state-and-transition models that have been developed and applied in the eastern Cascades of Oregon (Hemstrom et al. 2007, Burcsu et al. 2014, Halofsky et al. 2014). Each vegetation state (vegetation class) is a combination of dominant cover type (based on dominant tree or shrub species) and structure stage (based on tree size, canopy cover, number of canopy layers). Vegetation succession and other state changes occur through deterministic and probabilistic transitions that were developed for 39 potential vegetation types (PVTs) from four ecoregions using expert opinion of agency ecologists and, in some cases, calibration with the Forest Vegetation Simulator (Burcsu et al. 2014, Dixon 2015; Appendix 2). PVTs represent environmental conditions that support different late-successional or climax vegetation (e.g., ponderosa pine [Pinus ponderosa], grand fir). The states and transitions were applied in a spatially explicit manner to IDUs. Deterministic successional transitions occur when a vegetation state reaches its maximum age. Each vegetation state also has the potential for one of several probabilistic transitions at each time step to introduce alternative successional pathways (e.g., understory development). Some probabilistic transitions must also meet a minimum threshold for time since previous transition or disturbance. For disturbances from fire and forest management, we did not use the probabilistic transitions built into the existing state-and-transition models. Instead, the timing and spread of wildfire was determined stochastically by the fire submodel, whereas forest management, including prescribed fire, was scheduled and distributed based on separate disturbance rule sets. We did not implement the probabilistic transitions for insects and disease described previously (Burcsu et al. 2014) because we did not possess the information needed to model these processes spatially. Moreover, fire-insect interactions are rare in this region (Meigs et al. 2015).
We defined the landscape and vegetation structure of IDUs by the spatial extent of major vegetation classes (averaged over 4 ha), county tax-lot boundaries, and development zone boundaries identified from local land-use planning maps. IDUs < 1 ha were merged into similar adjacent IDUs, and IDUs > 8 ha were broken into a 4-ha grid pattern to allow for fine-scale representation of vegetation conditions likely to affect fire behavior. Initial vegetation conditions were modeled using an imputed spatial vegetation layer called Gradient Nearest Neighbor (GNN; Ohmann et al. 2011) that estimates the structure and composition of the live and dead woody vegetation based on 2006 Thematic Mapper satellite imagery, forest inventory plot data, and environmental data at a resolution of 30 m. Accuracy assessments for vegetation classes in GNN have been conducted for the study area using 4560 plots (http://lemma.forestry.oregonstate.edu/data/structure-maps). Overall accuracy for 11 structure classes at the pixel scale was 55%, and fuzzy accuracy (± 1 class) was 87%. These vegetation layers are not intended to be used for predicting the conditions on individual pixels, and accuracy improves with aggregation to coarser scales even at the scale of a few hectares (Matt Gregory, personal communication).
Fuel models (i.e., fuel conditions; Scott and Burgan 2005) were assigned to vegetation states based on a fuel model layer from the Deschutes National Forest (Lauren Miller, personal communication) and on Landfire-assigned fuel models (Rollins 2009; http://www.landfire.gov/viewer) for the rest of the study area. Often, multiple fuel models occurred within each IDU or vegetation state, so the majority fuel model was assigned to that state from spatial data layers. Fuel models change with changing vegetation states and disturbance. For disturbances (e.g., surface fire) that affect surface fuels but not the vegetation state or canopy fuels, fuel model variants were developed based on expert opinion (see Barros et al. 2017 for description of fuel models). A new fuel model is assigned after prescribed fire or surface wildfire, mixed-severity fire, high-severity fire, mowing and grinding, or timber harvest. The fuel model variant remains associated with the postdisturbance vegetation state for a set period of time or until the vegetation state transitions to a new state.
We incorporated wildfire into Envision using the minimum travel time fire spread algorithm (Finney 2002) and a wildfire prediction system (Ager et al. unpublished manuscript). We initialized this wildfire submodel at the beginning of each Envision run. The spatial input data included fuel model, canopy cover, canopy height, canopy base height, and canopy bulk density. The wildfire submodel then read the fire list to be simulated along with fire-specific simulation parameters (e.g., weather, ignitions) from external files. A different fire list was used for each of the 15 replicates. Fires were simulated within the submodel, and the submodel then returned to the main Envision landscape model a grid of flame lengths, which was used to update the IDU polygons that burned based on vegetation information for each individual polygon (from vegetation class lookup files). The wildfire prediction system forecasted daily fire occurrence, location, and size using empirically derived relationships between energy release component (ERC), a measure of potential fire severity, and historic fires. Thus, the wildfire submodel was stochastic, and its effects emerged from the interactions of fire parameters and vegetation conditions at IDU and landscape scales.
The historical reference period for predicting fire occurrence was 1992–2009, a period characterized by relatively frequent, large fires. The statistical relationships between ERC and fire parameters were determined from 25 remote automatic weather stations (RAWS) in the study area that incorporated a range of historical values and fire sizes. Fire weather and fuel moistures were modeled independently from fire probabilities using data from the Lava Butte RAWS for the years 1987–2011. Wind direction was randomly sampled based on day of year. Wind speed was sampled from a probability distribution of maximum gust speed generated from the same Lava Butte data but restricted to days when fires exceeded 500 ha. Fuel moistures were averaged across historical ERCs by fuel size class (i.e., 1, 10, 100, and 1000 h; woody, herbaceous).
Four actor groups were represented in our model: federal, tribal, corporate (i.e., private industrial), and family (i.e., nonindustrial private; Fig. 1). A fifth actor group representing homeowners who can treat their homesites to reduce fire hazard is also possible (Olsen et al. 2017) but was not included in the analysis described here because those decisions do not affect fire behavior in the model. The effects of homeowner decisions on fire risk will be examined in subsequent work. Oregon State lands, which consist of relatively young forests, were not managed in our scenarios, but succession and wildfire could occur there. Actors can modify forest vegetation by commercial timber harvests, tree thinning for fuels reduction, mechanical surface-fuel treatments, and prescribed fire, all of which can alter forest structure and fuel models (Appendix 4).
Management decisions for family forest owners were based on econometric models derived from surveys collected in the study area (Appendix 5). The decisions of family forest owners to implement fuels treatments are influenced by stand density, experience of wildfire near property, and presence of a structure on the property. The specific type of fuels treatment (e.g., thinning, mowing and grinding, prescribed burning) is determined from a probabilistic distribution of fuels treatments developed from rates identified from surveys of family owners within the study area.
For large landowners (federal, tribal, corporate), decisions were made based on a timber volume or treatment area target. The targets were based on interviews with public and private landowners (Charnley et al. 2017; Kline et al. unpublished manuscript). The target approach allows for implementation of actions (e.g., different types of fuels treatments or different types of harvests) that may be subject to a budget constraint or an implementation target. For example, federal managers have a fixed target for timber harvest volume, regardless of whether it comes from salvage logging after wildfire or planned timber harvests. When volume from a salvage harvest after a wildfire is available to federal managers, that salvage volume may be used to meet annual timber targets instead of planned timber harvests. The target approach has four components: (1) the “target” (either timber volume or area treated), (2) “constraints” that determine where on the landscape an action can occur based on biophysical or management criteria, (3) “preferences” that determine which IDUs have the highest likelihood for action during a given year, and (4) an “expand” function that spreads actions from an initial IDU to build logical treatment units that meet historical unit size distributions and constraints (Appendix 6). Examples of constraints include rules that disallow mechanical fuels treatment in Wilderness and designated roadless areas, areas with low canopy cover, or IDUs that had prior treatment within the last 14 years. Examples of preferences include giving more weight to treatment of IDUs in ponderosa pine vegetation types (where federal managers prioritize fuel treatments because of greater probability of fire than in other vegetation types) and near the WUI, and giving less weight to treatment of IDUs in moist mixed-conifer types or with lower basal area. IDUs with low preference scores for an action may still be selected if they meet the constraints and are adjacent to an IDU with a higher preference score. See Appendix 6 for details of targets, constraints, and preferences used in the scenarios.
We developed several evaluative metrics that describe landscape fire characteristics and ecosystem services, including merchantable wood, carbon storage, and other landscape outputs during each time step of the model (Tables 1 and 2). Some of these metrics (e.g., stand structure, fire occurrence, and management activity) act as landscape feedbacks. For example, fire can affect the pattern and amount of timber removal and fuel treatments by actors, and vice versa. Initial and simulated conditions for these metrics are linked to the vegetation states, with some provision for historical effects (e.g., in the case of dead wood and some wildlife species whose habitat is dependent on time since wildfire; Appendix 7 and 8). Evaluative metrics associated with live forest structure (e.g., live-tree basal area, wood volume, carbon, etc.) were estimated with regression models built from forest inventory data for each combination of PVT group (groups of similar PVTs) and vegetation class (Appendix 9). Most of these combinations were represented by tens to hundreds of inventory plots, although for some, the sample size was > 2000 plots. For the three PVT group–cover-type combinations with inadequate sample sizes (< 15 plots), we used the predictive equation for a comparable cover type. Finally, the density of juniper trees (Juniperus spp.) and cover of bitterbrush (Purshia tridentata) were estimated from published data for arid lands in the region (Stebleton and Bunting 2009).
Fire-related metrics included burned area (ha), fire hazard (potential; ha) and “resilient structure” (ha; Table 1). Fire occurrence was based on simulated fire spread boundaries in the model. Hazard is potential high-severity fire based on flame-length thresholds and fire submodel outputs that replicate FlamMap (Finney 2006) “static” fire behavior calculations and were performed for weather associated with 97th percentile conditions (ERC = 60, wind azimuth = 220°, wind speed = 29 km/h). Homesites, the number of homes on an IDU, are determined by Envision’s development function, which is based on a fixed ratio of population to houses, existing land-use zoning, and population projections developed by the State of Oregon. Resilient forest structure was defined for fire-frequent PVTs as the area of larger trees with open canopies (tree size ≥ 50 cm and canopy cover 10–40%; or tree size ≥ 38 cm, moderate canopy cover of 40–60%, and single canopy layer). This structure was considered resilient to fire effects and representative of the predominant historical structure of fire-frequent forests of the ponderosa pine, dry mixed, and moist mixed-conifer PVTs (Halofsky et al. 2014, Merschel et al. 2014).
Wildlife and plant habitat suitability models were developed using a combination of expert opinion and empirically derived models (Table 2). For most species, we used habitat models developed by Morzillo et al. (2014) for eastern Oregon. That approach relied on scientific literature and expert opinion to assign habitat quality scores to selected forest species based on environment (potential vegetation type) and species, cover, size, and layering of forest structure (Appendix 8 and 10). For the northern spotted owl, we developed a simple habitat suitability model based on vegetation type, canopy cover, and tree size characteristics from central Oregon northern spotted owl occurrence data (Appendix 11). For cheatgrass (Bromus tectorum), an invasive species, we used the habitat equations from Lovtang and Riegel (2012).
Evaluating the performance of simulation models, including agent-based models, can take different forms (e.g., verification and validation) and have different objectives (Rykiel 1996). Many models only become validated over time as different users apply them and gain confidence in them. Our wildfire submodel and vegetation succession models have been tested and evaluated by Ager et al. (unpublished manuscript). For example, Ager et al. (unpublished manuscript) found that the simulated relationship of ERC and fire size and frequency captured the distribution of the historical data from 1992–2009. Barros et al. (2017) doubled and tripled the rate of management to see how that affected model behavior. They found that that at higher rates of treatment, management targets could not be met because of lack of suitable forests to treat, which makes sense ecologically given the condition of the landscape and the rules we used to identify priorities and preferences for treatment. We evaluated the integration of these models with our management submodels primarily through verification exercises. First, we verified that the integrated model was working as intended through an exhaustive series of debugging tests that examined how well volume and area targets of large landowners were being met under the current management scenario. We found that the model could meet both volume targets and area targets that were identified by managers. However, in the case of volume targets for federal owners, the area required to reach those targets was approximately 25% larger than the area that the managers indicated they harvested to meet their volume targets. This suggests that the factor we use to compute volume removed by thinning from below (20% of stand volume) was conservative. We tested the model to make sure that timber volume production from planned harvest competed with timber volume from salvage following wildfire according to the ownership-specific rules. We examined individual IDUs subject to management and fire events to make sure that transitions in vegetation and fuel model states matched those expected from model logic.
Comparison of models against independent data is considered the ultimate step in validation, but there are many reasons why this cannot be done (and is often not done) for large, complex, stochastic models (Baker and Mladenoff 1999). Hindcasting can be used, but, in our case, the historical management targets, rules, and treatment types were not the same as those being used today, and historical vegetation and fuel model layers were not available to parameterize a hindcasting model. Some people argue that validation is never possible given the variety of scenarios and extent of time periods (Rykiel 1996). For stochastic spatial models, strict spatial agreement of events or succession with real landscapes should not be expected (Urban et al. 1999, Brown et al. 2005). In these cases, it may be more realistic to compare the distributional statistics of model outputs and those from real landscapes. However, this depends on having spatially or temporally independent data (e.g., harvest unit sizes), which may not be available for a large multiownership landscape. Remote-sensing-based estimates of forest disturbance (management and fire) are available for this area (Kennedy et al. 2010); however, we are unsure how well these spectral change data can be calibrated correctly with different types of management activities (Hall 2015), especially subcanopy activities such as thinning from below, which may not result in major changes in canopy reflectance. Nevertheless, we compared the simulated size distributions of management units by owner to remote sensing for the period 2006–2012 when the model was “spun-up” from initial conditions. The size distributions of remote sensing and simulated management activity patches (thinning from below, heavy partial harvest, clear cutting) were similar for federal lands, which occupy approximately 40% of the forest lands (Appendix 12). For other owners, the sizes of the simulated patches were smaller than those recorded by remote sensing, suggesting that the rules governing management unit size for nonfederal owners need further development. The challenges of model validation with independent data mean that the findings of our study should be viewed with caution and that the focus should be on the relative differences between scenarios, which are less subject to errors in model assumptions and parameterization.
We developed three plausible management scenarios for 50-yr model runs based on discussions with managers, workshops with stakeholders, and information contained in planning and management documents (Table 3). We focused scenarios on changes in federal management that, in turn, would likely change the management actions of corporate owners. The fire simulation and landscape analysis applied to all landowners. Current management is what landowners are doing now. The accelerated federal management scenario (“accelerated management”) reflects the desire of policy makers and stakeholders for increasing the pace and scale of treatments on federal lands to reduce high-severity fire hazard (e.g., North et al. 2012). The no federal management scenario (“no management”) is plausible if funds or capacity for forest management disappear on federal and corporate lands. A no fire and no federal management scenario (“no disturbance”) was used as a theoretical reference against which to evaluate the effects of fire and forest management on ecosystem services. Because our models contained probabilistic elements related to fire, vegetative dynamics, and decision-making processes, we used Envision’s capability for Monte-Carlo simulation to make 15 50-yr runs for each scenario to develop probabilistic descriptions of landscape outcomes and performance metrics (all scenarios used the same 15 fire lists; see Barros et al. 2017). The no-disturbance scenario was a single run because most of the stochasticity derives from the fire submodel. Each 50-yr simulation required ~4 h of time on a 32-GB computer system running a quad core i7-6820HQ processor with a 500-GB solid state drive.
The mean annual proportion of the forested landscape burned with high-severity (stand replacing) fire over 50 yr (15 replications) varied from < 0.01 to 1.0%, representing an area of 100–10,000 ha (Fig. 3) across all scenarios. The median annual amount of high-severity fire for all years for fire-frequent environments (where most of the federal restoration is targeted) was similar for the three scenarios, although the no-management scenario had a slightly higher median (0.05%) than the other scenarios (Fig. 4A). The maximum annual area burned in any year, across all years and replications (N = 750), was highest for the no-management and lowest for the accelerated-restoration scenarios. Taking into account only the highest 10% of annual area of high-severity fire in a year, accelerated restoration had the lowest median amount (0.5%), and no management had the highest median amount (0.7%) of high-severity fire in a year (Fig. 4B).
The median number of homes annually exposed to high severity fire (within 1 km of a fire) differed little among the three scenarios. However, the maximum number of homes exposed to a high-severity fire in any given year was largest for the no-management scenario (10,370 homes) compared to the other two scenarios, which had maximums that were approximately 20% lower (Fig. 5A). When only the top 10% of area of stand-replacing fire in a year was considered, the median annual number of homes exposed was highest for the no-management (1226 homes) and lowest for the accelerated-restoration (1065 homes) scenarios (Fig. 5B).
At time 0 (initial conditions), 53% of the landscape had potential for high-severity fire. Under current management, that proportion declined over the 50-yr model run to 46%, whereas it increased to 62% under the no-management scenario (Fig. 6). Treatments under the accelerated-restoration scenario initially reduced the potential for high-severity fire more than current management in the first 30 years of model runs. However, the two scenarios resulted in approximately the same level of high-severity fire potential after year 30.
Under current management, harvest of merchantable wood volume was approximately 430,000 m³ annually for ~35 years before declining (Fig. 7A). The accelerated-restoration scenario produced more wood volume in the first 15 years than current management but then produced less than current management (and less than targeted) for the remainder of the simulation, although the differences between two scenarios were not large. Of course, no management (federal with lagged effects on corporate landowners) produced less wood than either of the two scenarios with federal management. The dynamics in timber volume could be explained in part by ownership (Fig. 8). The strong decline in years 15–20 resulted from declines on federal lands and somewhat later on corporate lands (Fig. 8A and C); declines around year 40 resulted from declines on tribal lands (Fig. 8B). Family forest lands had small levels of timber production with noticeable fluctuations from year to year, especially for the no-disturbance scenario, which also produced more timber for that ownership. Without wildfire (no disturbance), the tribal lands were able to sustain target levels of timber production for a few additional years (Fig. 8B) than when wildfire was present.
Trends in the amount of aboveground live and dead carbon differed among scenarios (Fig. 7B). The highest amounts and strongest increase occurred, not surprisingly, under the theoretical no-disturbance scenario. The next highest increase was for the no-management scenario, in which carbon increased on the landscape despite the occurrence of wildfire. Under current management and accelerated restoration, the amount of carbon initially declined and then increased nearly back to initial conditions by year 50.
The area of forest structure classes differed by scenario for all forest environments (Fig. 9). The proportion of early successional habitat increased slightly in all scenarios with fire, but declined in the absence of fire. The proportion of large and giant trees increased in all scenarios, whereas the proportion of pole and small- and medium-diameter forests declined. Presence of wildfire decreased all of the moderate and closed canopy forested vegetation classes. Although the area of large and giant forest conditions increased under all scenarios, the increase was slightly larger for the active management scenarios (current management and accelerated restoration) than the no-management scenario.
The amount of high-resilience, older forest structure in fire-frequent environments increased over time under current management and accelerated restoration, whereas the proportion of low-resilience, older forest structure was unchanged under those scenarios (Fig. 10). Under the no-management and no-disturbance scenarios, the proportion of resilient older forests declined by nearly one-half as a result of succession, which increased canopy cover and layering.
Trends in the proportion of wildlife and plant habitat varied among species and scenarios (Fig. 11). Under current management, the proportion of the landscape providing nesting habitat for northern spotted owl and northern goshawk (Accipiter gentilis), and habitat for black-backed woodpecker (Picoides arcticus), western bluebird (Sialia mexicana), mule deer (Odocoileus hemionus), and cheatgrass increased slightly or were unchanged. The proportion of habitat for Pacific marten (Martes caurina) and white-headed woodpecker (Leuconotopicus albolarvatus) declined moderately. The general direction of trends in habitat for accelerated restoration were similar to current management for most species. Under no management, habitat increased for northern spotted owl, northern goshawk, and Pacific marten, whereas the proportion of habitat declined for the woodpeckers and remained unchanged for western bluebird and cheatgrass.
Comparing scenarios, no management resulted in more habitat than active management scenarios by the end of 50 years for northern spotted owl, Pacific marten, northern goshawk (nesting habitat), and pileated woodpecker (Dryocopus pileatus), species associated with dense, multistoried forests with trees > 25 cm in diameter (Fig. 11). In contrast, no management resulted in less habitat than the other scenarios for white-headed woodpecker, western bluebird, and mule deer. Compared with current management, accelerated restoration resulted in more habitat for white-headed woodpecker and western bluebird, species associated with open forests, and early successional vegetation and meadows, respectively. A comparison of the no-disturbance scenario and the scenarios with wildfire indicates that relatively large amounts of owl habitat and pileated woodpecker habitat were lost to wildfire, whereas scenarios with wildfire created some habitat for western bluebird, mule deer, and cheatgrass relative to the no-disturbance scenario.
Treatment of forest fuels to reduce the potential for high-severity fire, a common goal of many landowners, may or may not be compatible with goals for forest restoration, wildlife habitat, or carbon storage (Reinhardt et al. 2008, Ager et al. 2010b). Our study examines how alternative landscape management scenarios could influence future fire area and severity, high-severity fire potential, wood, carbon, and wildlife habitat. By integrating forest succession, wildfire, and forest management in a spatially explicit model, we were able to reveal interactions and trade-offs that the separate submodels could not reveal. Our projections indicated that current management of federal lands by thinning and surface fuel treatments can reduce both the amount of high-severity fire, overall fire hazard (potential), and exposure of dwellings to fire, although effects were small across all fire years compared to the no-management scenario. However, when years with extreme area of fire are considered, the effects of current management and accelerated restoration were larger compared to no management on federal lands. Years with extremely high fire area would have more fire-fuel treatment encounters than in years with less fire, making higher fuel treatment scenarios more effective during those years (Rhodes and Baker 2008, Syphard et al. 2011, Barros et al. 2017). Considering only extreme fire years, the variation in amount of high-severity fire among the scenarios was quite large, and distributions of high-severity fire overlapped among the scenarios.
The relatively modest effects of current management and accelerated restoration may be a result of the small area treated. The current management scenario, which is based on volume targets, converts to approximately 6500 ha of thinning and surface fuel treatments per year on federal lands and another 1000 ha on tribal and corporate lands, which are the other two major landowners. Those rates translate to treatment rates of approximately 12% per decade for the fire-frequent forest environments and approximately 9% per decade for all forest lands, including higher elevation forest types (e.g., mountain hemlock) that have inherently longer fire return intervals and higher fire severities. Assuming the effectiveness of fuel treatments is approximately 10–20 yr, this means that at any one time, 12–24% of the fire-frequent landscape has reduced fuels from treatments. The actual proportion of low fuel conditions would be higher than this because of the presence of recent postwildfire areas and unburnable sites such as rocky soils and roads. Other studies have shown that optimized fuel treatments that cover at least 20% of the landscape will significantly reduce fire spread rates (Finney et al. 2011) and reduce risk to large, fire-resistant trees (Ager et al. 2010b). Given that the spatial distribution of treatments was not optimized in our simulations, it seems likely that alternative spatial designs and treating additional area would have a greater effect on the amount of high-severity fire and the amount of fire-resilient vegetation. See Barros et al. (2017) and Ager et al. (unpublished manuscript) for further analysis of these topics.
Federal forest treatments to reduce the potential for high-severity fire are prioritized for the WUI (as defined by federal managers) over the less human-settled parts of the landscape (Appendix 7). The federally defined WUI extends many kilometers away from homes and does not appear to place a strong spatial constraint on federal management, leading to dispersion of fuel treatments. Defining the WUI in a more restricted way (e.g., Silvis WUI; Radeloff et al. 2005) would concentrate treatments nearer to homes. Our results suggest that current management activities (based on federal WUI) reduce exposure of homes to high-severity fire (within 1 km) by almost 15% for the years with the most fire. However, exposure of homes to flames and fire brands is only one part of the risk framework for loss of homes to wildfire and may not be as important as susceptibility of the home to ignition (e.g., Calkin et al. 2014). Home ignition is a function of the live and dead fuel on and surrounding the home site as well as home building materials, especially flammability of the roof (Cohen 2000). Consequently, a more complete risk analysis should include homes’ susceptibility to ignition. We did not report the results of actions by homeowners to adopt defensible space behaviors that reduce risk at the homesite scale. Olsen et al. (2017) found that almost 80% of the homes in the study area adopted some component of defensible space practices, and that these actions were more likely in locations characterized by heighted fire hazard, suggesting that a high level of risk mitigation occurs across the population of homes. Our results suggest that land management actions in this area can reduce the potential for losses.
Much of the federal restoration activity in the study area includes federal timber sales that contribute to reaching the federal harvest volume target. Our simulations suggest that efforts to increase the area of restoration (the accelerated restoration scenario) could be limited by availability of stands with suitable ecological and topographical conditions. For example, the accelerated restoration scenario, which scheduled a doubling of the volume target for the first 25 years before returning to the current management volume target level, was not able to sustain the doubling beyond 18 yr and was not able to reach the current management volume target after that given lack of treatable area. This finding suggests that efforts to increase harvesting on federal lands might not be successful in the long run without expanding the vegetation types and land allocations that can be treated, with or without harvesting. For example, we limited thinning treatments to stands dominated by trees > 25 cm diameter at breast height (dbh), canopies with > 60% cover, and multistory structure. Sensitivity analyses (not shown) indicate that accelerated restoration might be sustainable for ≥ 50 yr if treatments were allowed in less dense stands or other areas of the landscape where we assumed that treatments would not occur (see Barros et al. 2017).
Although timber production from federal lands has direct economic benefits in the jobs and business activity it generates for local communities, there are also important indirect connections between current timber production and the future capacity for businesses to do ecological restoration (Kelly and Bliss 2009). Timber from federal lands can help to maintain the forestry infrastructure and mill capacity necessary to conduct merchantable harvest-based restoration treatments (Charnley 2014). The trends we observed, however, suggest that as the densest and highest volume forests are thinned and then treated with prescribed fire, the levels of timber production will decline as management activities focus more on using prescribed fire or mechanical methods to reduce surface fuels. Barros et al. (2017) indicate that the area suitable for prescribed fire will dramatically increase under active management strategies because current area targets for prescribed fire are relatively small and fixed compared to those for thinning. Increases in prescribed fire targets may be needed to keep up with the increasing areas that have had the first installment of restoration treatments. As future timber volume may be less able to offset the costs of restoration treatments, other restoration funding mechanisms may be needed. Economics may not be the only barrier to increasing the area of prescribed fire; air quality standards and public concerns about smoke and fire escape also can occur (Charnley et al. 2015) and will require special efforts on the part of managers to build trust for this important restoration action (McCaffrey 2006). We have a smoke production model and plan to present the results in a future paper.
An eventual shortfall in harvest volume production (and area treated) was also projected for tribal lands at approximately 35 yr, when the model could not find enough stands that met the treatment criteria. This appeared to result from future lack of older stands (> 70 yr) for clearcutting, which is practiced in some forest types on tribal lands. The Envision model does not optimize harvest schedules. If the current management plan for tribal lands were put into a harvest scheduling model built to select optimally which stands to harvest while also meeting the annual timber target between now and 2050, a schedule might be found to meet the timber target in all years. Without wildfire, the volume production for tribal lands was steady and could be met for a few more years than with wildfire. The effects of wildfire on timber production have been observed (Armstrong 2004), but typically, stochastic natural disturbance agents that remove existing merchantable timber volume or shift age class distributions are not factored into forest management planning.
Carbon storage was sensitive to management scenarios. More carbon was stored in aboveground live and dead biomass under no federal management and no fire (no management and no disturbance) scenarios. Landscape-level treatments to reduce fire effects and increase restoration through thinning and surface fuel treatments reduced stored forest carbon relative to no management and no disturbance, but the relative differences were small (~5–10%) depending on year and scenario; by the end of 50 yr, the amount of stored carbon in the active management scenarios was slightly less than at the start of the simulation. Others have also found that active management in fire-prone forest landscapes can reduce carbon storage by ~5–25% (Ager et al. 2010a, Loudermilk et al. 2013) relative to no management depending on timing and intensity of management and assumptions about future wildfire. These results occur because treatments always reduce carbon stored in forests, and although they typically reduce carbon losses to high-severity fire at stand levels, at landscape scales many treatments do not experience fires and do not realize carbon losses associated with such fires. However, the effect of fuel treatments in reducing landscape fire spread and intensity can confer carbon sequestration benefits outside of treated areas. However, our results and those of others (Ager et al. 2010a) indicate that the magnitude of benefits outside of treated areas does not make up for carbon removed by harvesting and prescribed fire. Accounting for the carbon stored in wood products manufactured from timber harvested in the course of fuels treatments would increase the total carbon stored under the treatment scenarios (Bergman et al. 2014), potentially altering the carbon accounting results.
Our projections suggest that the area of larger sized trees will increase and the area of small and mid-size trees will decrease across all ownerships (though not necessarily for all individual ownerships; Charnley et al. 2017) under current management constraints and despite wildfires. This happens despite a significant area of large trees being killed by high-severity wildfire and an increase in the area of early successional forests because of wildfire and clearcutting on tribal lands. Similar trends in medium and larger diameter forest were observed in another modeling study for a landscape in this region (Halofsky et al. 2014). The net increase in mature and older forests is likely a result of ingrowth from large areas of cutover lands, timber plantations, and areas of insect outbreaks and wildfire during the 20th and early 21st centuries in ponderosa pine and dry mixed-conifer forests on public and private lands that are now maturing (Hessburg and Agee 2003). This highlights the importance of landscape legacies and age class distributions in controlling future landscape development (Wallin et al. 1994). These past land-use effects are especially important in low-productivity forest environments, where recovery from stand replacement disturbances can be slow.
Our metric for resilient older forest structure (i.e., low canopy cover forests with larger trees in fire-frequent PVTs) increased under the two active management scenarios. This metric reflects the fact that the need for forest restoration varies by environment as a function of historical fire regime and logging history (Merschel et al. 2014). In this landscape, the environments with the highest fire frequencies (and therefore potentially the greatest departures from historical regimes under fire suppression policy) are ponderosa pine, dry mixed-conifer, and moist mixed-conifer. The first two types are generally known to have fire return intervals of < 20–25 yr (Agee 1993). At present, these types have much higher densities and many fewer large fire-tolerant species such as ponderosa pine as a result of fire exclusion and logging, which both selectively removed large pine or created clearcuts that are now occupied by younger stands of pine. The history of the moist mixed-conifer type is less well understood, but recent studies in the area (Hagmann et al. 2014, Merschel et al. 2014) indicate that these environments had similar low density, large-tree structure and species composition to the dry mixed-conifer, implying a similar relatively high frequency of fire that kept grand fir and other shade-tolerant tree densities low. Baker (2012) suggests that forests in these environments were historically denser than indicated by these other studies, but his estimated historical densities are still considerably lower than current densities (Merschel et al. 2014).
Despite the increase in area of forests with highly resilient forest structure, such forests only accounted for approximately 19% of the landscape by the end of 50 yr (and less in the no-management scenarios). It is unclear what the proportion of this type of structure would have been under the historical disturbance regime. However, it probably was quite high, e.g., > 75%, given the recent historical work of Hagmann et al. (2014) and Merschel et al. (2014) and that of Baker (2015), who found that > 76% of the forests of all levels of canopy cover contained trees > 53 cm dbh in central Oregon. An earlier simulation study by Kennedy and Wimberly (2009) estimated that under historical regimes, 35% of all forest types on the Deschutes River, Oregon was covered by old forest structure, of which approximately 25% was in closed canopy conditions. The relatively low estimate of old-forest structure by Kennedy and Wimberly (2009) may result from their use of a relatively high diameter threshold (53 cm rather than 50 cm) and their assumption that historical fires had more stand replacement (high severity) patches in them than recent historical analysis indicate. If we assume that > 75% of the frequent-fire environments had large and giant trees (> 50 cm) with any canopy cover, then our analysis suggests that 50 yr and current practices will not be enough to get back to that level; we estimated that large and giant trees, regardless of canopy cover and layering (both resilient and nonresilient vegetation), increased from 28 to 32%. Without any human or natural disturbance on federal lands (no disturbance), this class is projected to compose approximately 35% of the landscape at 50 yr. The results indicate that given the management history, current age and size distributions (~48% of the area has tree size < 37 cm), relatively low site productivity, and diversity of management goals, it will take much longer than 50 yr to recover a larger portion of the original forest structure.
The management scenarios affected wildlife habitat in different ways. Habitat of species associated with relatively dense, multilayered, older forests (e.g., northern spotted owl, Pacific marten, and pileated woodpecker) was reduced by treatments to decrease stand density. For example, we found that nesting habitat for the northern spotted owl was less under the active management scenarios compared with no management. However, wildfire removed many times more owl habitat than was lost through management.
Of the species associated with dense, multilayered forests, the northern spotted owl is of particular importance because it is listed under the U.S. Endangered Species Act and is often used as a surrogate for threatened old-growth forest ecosystems. We emphasize the northern spotted owl here because it is both an ecological and social issue that drives federal forest management and reflects changing social values related to timber vs. biodiversity values of forests (Lee 2009). The primacy of conserving northern spotted owl habitat in federal forest management goals means that other social and ecological values (e.g., thinning to support local timber-based economies or managing for more open, fire-resilient, older forests) are constrained by concerns over providing habitat for the owl.
Only a few other modeling studies have explored questions related to owl habitat, management, and fire. Ager et al. (2007) found that fuel treatments would reduce expected loss of northern spotted owl habitat when the treatment area reached at least 20% of the landscape. The reduction in expected loss of owl habitat in that study went from approximately 2.4 to 1.3% between 0% treated and 20% of landscape treated. Ager et al.’s (2007) analysis did not allow treatment in areas that were defined as owl habitat and did not assume that succession or stand development would occur (static vegetation). The relatively lower amount of owl habitat in our study under the active management scenarios compared with no management may occur because of thinning in younger forests that reduced the potential for development of owl habitat in the future. Thus, thinning, although it was not allowed in existing habitat, reduced recruitment of future owl nesting habitat in addition to what was lost to wildfire. Wildfire was the dominant cause of habitat loss, although habitat increased under all scenarios, suggesting that it may be possible to both increase restoration and increase habitat for northern spotted owl in this landscape. Roloff et al. (2005) modeled active management and no management in fire-prone landscapes in southwestern Oregon and found that active management in owl foraging areas reduced owl habitat compared with no management (only losses to wildfire). However, in a second analysis, Roloff et al. (2012) found that active management “was more favorable to spotted owl conservation...than no management.” They used FlamMap to estimate crown fire potential and assumed that if 50% of the owl territory had crown fire potential, then all of the territory would be lost to a fire. This assumption may overestimate the loss of habitat to fire. Odion et al. (2014) also found that thinning treatments in fire-prone landscapes reduced owl habitat more than did wildfire.
These studies, along with ours, suggest that the question of how to sustain northern spotted owl habitat in fire-prone landscapes is complex and needs further evaluation (Lehmkuhl et al. 2015). While stand-replacing fire has been observed to reduce owl occupancy (Clark et al. 2013), patchy fires may not have detrimental effects on northern spotted owl as long as patterns are heterogeneous and adequate amounts of nesting and roosting habitat remain. It is clear that some fuel treatment designs intended to reduce loss of owl habitat to high-severity fire will result in reduced owl habitat compared to a no-treatment option. However, several key questions remain unanswered, including: How do the rates and patterns of fuel treatment affect high-severity fire in landscapes with different initial conditions of forest structure and age? How do the amounts and landscape patterns of fuel treatments inside and outside existing and potential owl habitat affect dynamics of owl habitat, owl prey, and owl populations? And how do different landscape management strategies affect owl habitat outcomes under different future fire scenarios?
The results from other species also demonstrate the variability in effects of treatment and wildfire on species with different habitat needs. Habitat for the white-headed woodpecker declined under all scenarios, though the decline was lowest for the accelerated restoration scenario. The habitat for this species is in relatively open (< 40% canopy cover) ponderosa pine forests; however, our thinning prescription typically did not produce this low level of canopy cover because it focused on stands with high canopy cover (> 60%) and reduced it to medium canopy cover (40 to 60%). Changing the thinning prescription to promote open conditions would probably produce more habitat for this species.
Like all modeling studies, the results of our study are a product of the assumptions and limitations of the models used. Our fire submodel has been evaluated, and simulations show reasonable correspondence to historical frequencies and spatial patterns (Barros et al. 2017; Ager et al. unpublished manuscript). However, an area of uncertainty is the relative amounts of different fire severities, which needs further evaluation. We did not evaluate a climate change scenario. Halofsky et al. (2014) and Case, Kerns, Kim, et al. (unpublished manuscript) found that projected changes in area of forested PVTs in central Oregon under climate change scenarios do not differ much from current conditions until at least 40–60 years from present, so some of our 50-yr findings might not be that different for climate change scenarios. However, increasing frequencies of fire may come much sooner than 50 yr under climate change and could alter our findings and conclusions. For example, our finding that treatments are more effective in high-fire years (see also Barros et al. 2017) suggests that with increasing fire under climate change, management actions could be more effective, although the opposite could be the case for severe weather events that may cause fuel treatments to be less effective (Fernandes and Botelho 2003). In addition, using different scenarios in terms of stand treatments, landscape allocation, and rate, and using different habitat models could produce different results. We also did not report on how treatments on one ownership affect fire behavior and habitat on other ownerships (Ager et al. 2014), which is an important question for policies such as the National Cohesive Wildland Fire Management Strategy (Wildland Fire Leadership Council 2014) that seek to coordinate fire and fuels management across multiownership landscapes. We plan to report the cross-ownership effects of fire and fuels management in a future paper focused on all-lands management and landscape-level planning.
The structure of this agent-based model enabled us to explore some of the interactions between social and ecological components of this fire-prone system, but we have not fully exploited its capabilities. For example, we learned that efforts to reduce the occurrence of high-severity fire, which affects both social and ecological values, show the greatest effectiveness in years with the most fire. This indicates that for restoration actions to be most effective at landscape scales, the number of encounters between managed areas and fires must increase, either through more fire, more management, or both. We also learned that other management goals (e.g., wilderness provision, roadless areas designation, habitat conservation for northern spotted owl) could constrain the area available for thinning and restoration and may limit the potential to reach restoration goals for the entire landscape that has been affected by past management practices and fire exclusion. Although agents in this model do not “learn” in the sense of changing their management goals or rules as a result of feedbacks from the environment, their management actions do respond to how the landscape changes from the cumulative effects of their management, vegetation succession, and wildfire. For example, federal managers may not be able to reach their treatment and timber production goals if wildfire occurs in stands that are suitable for treatment, or if the cumulative effects of thinning and wildfire reduce the area of forest that is suitable for timber harvest. The model also indicates that trade-offs will occur among fire, timber, carbon storage, and wildlife habitat. More thinning will reduce fire occurrence and increase timber production and the areas for white-headed woodpeckers and other species favoring more open forests, but will also reduce habitat for northern spotted owl and carbon sequestration. How these potential trade-offs can be dealt with in management and policy is a question that federal managers are wrestling with, but they lack analytical frameworks and tools that can help reveal the spatial and temporal patterns and scales of the interactions. We plan to use the model and its results with stakeholder groups in collaborative landscape projects in the study area to determine if and how stakeholders can learn from our model and if this process changes the nature of the discussions and debates around forest landscape restoration.
We have also not yet fully exploited the capabilities of this model to examine social-ecological interactions. For example, the model could be used to examine more thoroughly how management and fire on one ownership affect ecological and social outcomes on adjacent ownerships. Such cross-boundary effects are a major concern among some private landowners, who view federal forests that abut or surround their lands as the major source of fire risk (Charnley et al. 2017). The model also could be used to focus on homeowners and family forest owners to examine how changing forest structure and fire occurrence influences their behaviors. In the current analysis, the dynamics of these agents remained unresolved given the broad focus on the entire landscape dominated by large landowners.
Few other landscape models have the range of capabilities that Envision does to represent wildfire, succession, multiowner forest management, and homeowner fire mitigation across large landscapes. Many landscape fire models exist, but few have the potential to represent human decision making in great detail. The closest example may be Landis (Scheller et al. 2007), which has many of the same capabilities as Envision, although with somewhat different submodel processes, especially for succession and forest management. Envision has multiple ways to represent agent-based behavior, include policies that guide agents based on value-to-action rules and potential for learning, and target-based approaches to decisions. We used both approaches in this effort, e.g., policies for family forest owners and homeowners and a target (subject to constraints and preferences) approach similar to that used in Landis.
The agent-based model we have developed fits within the broad set of agent-based model applications. Central features of agent-based models include large numbers of “active objects” that interact with their environment and with each other (Borshchev and Filippov 2004). Although the objects are typically people or households, they may be animals, business units, vehicles, or even spatial units (Box 2002). In our model, people occur in nested spatial structures from large ownership blocks to smaller individual landowner parcels. They interact with their environment through vegetation manipulations based on global goals and rules. They interact with each other largely through how management on their lands might affect fire spread and occurrence on other ownerships. In reality, they also interact with each other through social networks (Fischer and Jasny 2017) and indirectly through timber production as it might affect timber mill capacity and timber markets; these other relationships are not currently in the model.
The agents in our landscape are relatively few and the system we studied is relatively slow and noninteracting compared to many systems modeled with agent-based models (e.g., cities or agricultural landscapes). For example, our landscape has a low number of human agents, least in terms of the dominant area of the landscape (e.g., 80% is controlled by six agents). Also, large landowners have historically shifted behaviors slowly because of institutional factors (Steen-Adams, Charnley, and Adams unpublished manuscript). Another distinctive feature of this landscape is the slowness of change and infrequency of wildfire and management events. Vegetation can stay in some states for decades without change, and frequency of wildfire and management have low probabilities at an IDU level (e.g., 0.05 annual probability for some kinds of management to 0.003 for occurrence of wildfire). The real complexity and heterogeneity of this system lies in the interactions of hundreds of thousands of spatial units of vegetation, fire, managed actions of a handful of human agents, and time lags that occur over decades to centuries. The scale, elements, and interactions of our model seem appropriate to this system given that there are great expectations and concerns about how individual vegetation treatments scale up to affect behavior of fire, restoration, and outputs of ecosystems services across multiownership landscapes.
Using a spatial agent-based model, we were able to examine interactions between human and natural systems across spatial and temporal scales in a fire-prone landscape. We gained several insights that would have been very difficult to achieve without this type of model. Overall, our study reveals that alternative approaches to vegetation management can affect fire area, fire outcomes, exposure of homes to fire, wood production, carbon storage, vegetation conditions, and wildlife habitat. More specifically, we found that current practices can reduce fire severity compared to no management, but the magnitude of fire effects is very small for average fire years. Management to reduce fuel loads appears to have greatest effects in extreme fire years, i.e., years with large areas burned by fire. We found that current timber targets and restoration programs may be at the limit of what is sustainable under the stand structure and land allocation constraints we assumed. Reducing the area of high-severity fire and creating more resilient forest structures through current and accelerated restoration programs will result in trade-offs for carbon and wildlife habitat for some species, including the northern spotted owl, a species of critical concern in the region. Finally, our study demonstrates the importance of legacies of past disturbances, expressed in current forest structure and age, on the pattern and dynamics of future forest characteristics. Ultimately, managers and the public will need to decide what the most socially, economically, and ecologically viable strategies for landscape-scale management are in this and other fire-prone landscapes. We hope the model and our initial application can contribute to that social process and provide a way for stakeholders to understand better the landscape-scale and longer term implications of forest management decisions.
This research was funded by the National Science Foundation, Coupled Human and Natural Systems Program (NSF Grant CHH-1013296), the USDA Forest Service PNW Research Station, and the Interagency Joint Fire Sciences Program (Grants 09-1-08-31 and 14-1-01-22). We thank Stu Brittain of Alturas Solutions for his development work on the Envision wildfire submodel and the development of the FlamMap DLL code libraries. We also acknowledge advice from Ashely Steel of the Pacific Northwest Research Station, and cooperation from managers on the Confederate Tribes of Warm Springs Branch of Natural Resources and the Deschutes National Forest.
Agee, J. K. 1993. Fire ecology of Pacific Northwest forests. Island Press, Washington, D.C., USA.
Ager, A. A., M. A. Day, M. A. Finney, K. Vance-Borland, and N. M. Vaillant. 2014. Analyzing the transmission of wildfire exposure on a fire-prone landscape in Oregon, USA. Forest Ecology and Management 334:377-390. http://dx.doi.org/10.1016/j.foreco.2014.09.017
Ager, A. A., M. A. Finney, B. K. Kerns, and H. Maffei. 2007. Modeling wildfire risk to northern spotted owl (Strix occidentalis caurina) habitat in central Oregon, USA. Forest Ecology and Management 246(1):45-56. http://dx.doi.org/10.1016/j.foreco.2007.03.070
Ager, A. A., M. A. Finney, A. McMahan, and J. Cathcart. 2010a. Measuring the effect of fuel treatments on forest carbon using landscape risk analysis. Natural Hazards and Earth System Science 10(12):2515-2526. http://dx.doi.org/10.5194/nhess-10-2515-2010
Ager, A. A., J. D. Kline, and A. P. Fischer. 2015. Coupling the biophysical and social dimensions of wildfire risk to improve wildfire mitigation planning. Risk Analysis 35(8):1393-1406. http://dx.doi.org/10.1111/risa.12373
Ager, A. A., N. M. Vaillant, and M. A. Finney. 2010b. A comparison of landscape fuel treatment strategies to mitigate wildland fire risk in the urban interface and preserve old forest structure. Forest Ecology and Management 259(8):1556-1570. http://dx.doi.org/10.1016/j.foreco.2010.01.032
Armstrong, G. W. 2004. Sustainability of timber supply considering the risk of wildfire. Forest Science 50(5):626-639.
Baker, W. L. 2012. Implications of spatially extensive historical data from surveys for restoring dry forests of Oregon’s eastern Cascades. Ecosphere 3(3):1-39. http://dx.doi.org/10.1890/es11-00320.1
Baker, W. L. 2015. Historical northern spotted owl habitat and old-growth dry forests maintained by mixed-severity wildfires. Landscape Ecology 30(4):655-666. http://dx.doi.org/10.1007/s10980-014-0144-6
Baker, W. L., and D. J. Mladenoff. 1999. Progress and future directions in spatial modeling of forest landscapes. Pages 333-349 in D. J. Mladenoff and W. L. Baker, editors. Spatial modeling of forest landscape change: approaches and applications. Cambridge University Press, Cambridge, UK.
Barros, A. M. G., A. A. Ager, M. A. Day, H. K. Preisler, T. A. Spies, E. White, R. Pabst, K. A. Olsen, E. Platt, J. D. Bailey, and J. P. Bolte. 2017. Spatiotemporal dynamics of simulated wildfire, forest management, and forest succession in central Oregon, USA. Ecology and Society 22(1):24. https://doi.org/10.5751/ES-08917-220124
Bergman, R., M. Puettmann, A. Taylor, and K. E. Skog. 2014. The carbon impacts of wood products. Forest Products Journal 64(7-8):220-231. http://dx.doi.org/10.13073/fpj-d-14-00047
Beukema, S. J., W. A. Kurz, W. Klenner, J. Merzenich, and M. Arbaugh. 2003. Applying TELSA to assess alternative management scenarios. Pages 145-154 in G. J. Arthaud and T. M. Barrett, editors. Systems analysis in forest resources. Springer, Dordrecht, The Netherlands. http://dx.doi.org/10.1007/978-94-017-0307-9_16
Bolte, J. P., D. W. Hulse, S. V. Gregory, and C. Smith. 2006. Modeling biocomplexity – actors, landscapes and alternative futures. Environmental Modelling and Software 22(5):570-579. http://dx.doi.org/10.1016/j.envsoft.2005.12.033
Borshchev, A., and A. Filippov. 2004. From system dynamics and discrete event to practical agent based modeling: reasons, techniques, tools. Page 45 in M. Kennedy, G. W. Winch, R. S. Langer, J. I. Rowe, and J. M. Yanni, editors. System Dynamics Society: proceedings of the 22nd international conference. System Dynamics Society, Albany, New York, USA. [online] URL: http://www.systemdynamics.org/conferences/2004/SDS_2004/PROCEED.pdf
Bowman, D. M. J. S., J. K. Balch, P. Artaxo, W. J. Bond, J. M. Carlson, M. A. Cochrane, C. M. D’Antonio, R. S. DeFries, J. C. Doyle, S. P. Harrison, F. H. Johnston, J. E. Keeley, M. A. Krawchuk, C. A. Kull, J. B. Marston, M. A. Moritz, I. C. Prentice, C. I. Roos, A. C. Scott, T. W. Swetnam, G. R. van der Werf, and S. J. Pyne. 2009. Fire in the Earth system. Science 324(5926):481-484. http://dx.doi.org/10.1126/science.1163886
Box, P. 2002. Spatial units as agents: making the landscape an equal player in agent-based simulations. Pages 59-82 in H. R. Gimblett, editor. Integrating geographic information systems and agent-based modeling techniques for simulating social and ecological processes. Oxford University Press, Oxford, UK.
Brown, D. G., S. Page, R. Riolo, M. Zellner, and W. Rand. 2005. Path dependence and the validation of agent-based spatial models of land use. International Journal of Geographical Information Science 19(2):153-174. http://dx.doi.org/10.1080/13658810410001713399
Burcsu, T. K., J. S. Halofsky, S. A. Bisrat, T. A. Christopher, M. K. Creutzburg, E. B. Henderson, M. A. Hemstrom, F. J. Triepke, and M. Whitman. 2014. Dynamic vegetation modeling of forest, woodland, shrubland, and grassland vegetation communities in the Pacific Northwest and Southwest regions of the United States. Pages 15-69 in J. E. Halosfky, M. K. Creutzburg, and M. A. Hemstrom, editors. Integrating social, economic, and ecological values across large landscapes. General Technical Report PNW-GTR-896. USDA Forest Service Pacific Northwest Research Station, Portland, Oregon, USA. [online] URL: http://www.fs.fed.us/pnw/pubs/pnw_gtr896.pdf
Calkin, D. E., J. D. Cohen, M. A. Finney, and M. P. Thompson. 2014. How risk management can prevent future wildfire disasters in the wildland-urban interface. Proceedings of the National Academy of Sciences 111(2):746-751. http://dx.doi.org/10.1073/pnas.1315088111
Charnley, S. 2014. Strategies for job creation through National Forest management. Pages 599-628 in J. W. Long, L. Quinn-Davidson, and C. N. Skinner, editors. Science synthesis to support socioecological resilience in the Sierra Nevada and southern Cascade Range. General Technical Report PSW-GTR-247. USDA Forest Service Pacific Southwest Research Station, Albany, California, USA. [online] URL: http://www.fs.fed.us/psw/publications/reports/psw_sciencesynthesis/psw-gtr-247_POSTPRINT_7-03-14.pdf
Charnley, S., M. R. Poe, A. A. Ager, T. A. Spies, E. K. Platt, and K. A. Olsen. 2015. A burning problem: social dynamics of disaster risk reduction through wildfire mitigation. Human Organization 74(4):329-340. http://dx.doi.org/10.17730/0018-7259-74.4.329
Charnley, S., T. A. Spies, A. M.G. Barros, E. M. White, and K. A. Olsen. 2017. Diversity in forest management to reduce wildfire losses: implications for resilience. Ecology and Society 22(1):22. https://doi.org/10.5751/ES-08753-220122
Chew, J. D., C. Stalling, and K. Moeller. 2004. Integrating knowledge for simulating vegetation change at landscape scales. Western Journal of Applied Forestry 19(2):102-108.
Clark, D. A., R. G. Anthony, and L. S. Andrews. 2013. Relationship between wildfire, salvage logging, and occupancy of nesting territories by northern spotted owls. Journal of Wildlife Management 77(4):672-688. http://dx.doi.org/10.1002/jwmg.523
Cohen, J. D. 2000. Preventing disaster: home ignitability in the wildland-urban interface. Journal of Forestry 98(3):15-21.
Conlisk, E., A. D. Syphard, J. Franklin, and H. M. Regan. 2015. Predicting the impact of fire on a vulnerable multi-species community using a dynamic vegetation model. Ecological Modelling 301:27-39. http://dx.doi.org/10.1016/j.ecolmodel.2015.02.004
Dixon, G. E. 2015. Essential FVS: a user’s guide to the Forest Vegetation Simulator. Revised edition. USDA Forest Service Forest Management Service Center, Fort Collins, Colorado, USA. [online] URL: http://www.fs.fed.us/fmsc/ftp/fvs/docs/gtr/EssentialFVS.pdf
Fernandes, P. M., and H. S. Botelho. 2003. A review of prescribed burning effectiveness in fire hazard reduction. International Journal of Wildland Fire 12(2):117-128. http://dx.doi.org/10.1071/wf02042
Finney, M. A. 2002. Fire growth using minimum travel time methods. Canadian Journal of Forest Research 32(8):1420-1424. http://dx.doi.org/10.1139/x02-068
Finney, M. A. 2006. An overview of FlamMap fire modelling capabilities. Pages 213-220 in P. L. Andrews and B. W. Butler, editors. Fuels management—how to measure success: conference proceedings. Proceedings RMRS-P-41. USDA Forest Service Rocky Mountain Research Station, Fort Collins, Colorado, USA. [online] URL: http://www.fs.fed.us/rm/pubs/rmrs_p041.pdf
Finney, M. A., C. W. McHugh, I. C. Grenfell, K. L. Riley, and K. C. Short. 2011. A simulation of probabilistic wildfire risk components for the continental United States. Stochastic Environmental Research and Risk Assessment 25(7):973-1000. http://dx.doi.org/10.1007/s00477-011-0462-z
Fischer, A. P., and L. Jasny. 2017. Capacity to adapt to environmental change: evidence from a network of organizations concerned with increasing wildfire risk. Ecology and Society 22(1):23. https://doi.org/10.5751/ES-08867-220123
Grimm, V., U. Berger, D. L. DeAngelis, J. G. Polhill, J. Giske, and S. F. Railsback. 2010. The ODD protocol: a review and first update. Ecological Modelling 221(23):2760-2768. http://dx.doi.org/10.1016/j.ecolmodel.2010.08.019
Gustafson, E. J., D. E. Lytle, R. Swaty, and C. Loehle. 2007. Simulating the cumulative effects of multiple forest management strategies on landscape measures of forest sustainability. Landscape Ecology 22(1):141-156. http://dx.doi.org/10.1007/s10980-006-9017-y
Hagmann, R. K., J. F. Franklin, and K. N. Johnson. 2014. Historical conditions in mixed-conifer forests on the eastern slopes of the northern Oregon Cascade Range, USA. Forest Ecology and Management 330:158-170. http://dx.doi.org/10.1016/j.foreco.2014.06.044
Hall, M. P. 2015. Remote detection and predicted locations of NIPF fuel treatments in central Oregon. Thesis. Oregon State University, Corvallis, Oregon, USA. [online] URL: http://hdl.handle.net/1957/55299
Halofsky, J. S., J. E. Halofsky, T. Burcsu, and M. A. Hemstrom. 2014. Dry forest resilience varies under simulated climate-management scenarios in a central Oregon, USA landscape. Ecological Applications 24(8):1908-1925. http://dx.doi.org/10.1890/13-1653.1
Hemstrom, M. A., J. Merzenich, A. Reger, and B. Wales. 2007. Integrated analysis of landscape management scenarios using state and transition models in the upper Grande Ronde River subbasin, Oregon, USA. Landscape and Urban Planning 80(3):198-211. http://dx.doi.org/10.1016/j.landurbplan.2006.10.004
Hessburg, P. F., and J. K. Agee. 2003. An environmental narrative of Inland Northwest United States forests, 1800–2000. Forest Ecology and Management 178(1-2):23-59. http://dx.doi.org/10.1016/S0378-1127(03)00052-5
Keane, R. E., G. J. Cary, I. D. Davies, M. D. Flannigan, R. H. Gardner, S. Lavorel, J. M. Lenihan, C. Li, and T. S. Rupp. 2004. A classification of landscape fire succession models: spatial simulations of fire and vegetation dynamics. Ecological Modelling 179(1):3-27. http://dx.doi.org/10.1016/j.ecolmodel.2004.03.015
Kelly, E. C., and J. C. Bliss. 2009. Healthy forests, healthy communities: an emerging paradigm for natural resource-dependent communities? Society and Natural Resources 22(6):519-537. http://dx.doi.org/10.1080/08941920802074363
Kennedy, R. S. H., and M. C. Wimberly. 2009. Historical fire and vegetation dynamics in dry forests of the interior Pacific Northwest, USA, and relationships to Northern Spotted Owl (Strix occidentalis caurina) habitat conservation. Forest Ecology and Management 258(5):554-566. http://dx.doi.org/10.1016/j.foreco.2009.04.019
Kennedy, R. E., Z. Yang, and W. B. Cohen. 2010. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr — temporal segmentation algorithms. Remote Sensing of Environment 114(12):2897-2910. http://dx.doi.org/10.1016/j.rse.2010.07.008
Langston, N. 1995. Forest dreams, forest nightmares: the paradox of old growth in the inland West. University of Washington Press, Seattle, Washington, USA.
Lee, R. G. 2009. Sacred trees. Pages 95-103 in T. A. Spies and S. L. Duncan, editors. Old growth in a new world: a Pacific Northwest icon reexamined. Island Press, Washington, D.C., USA.
Lehmkuhl, J., W. Gaines, D. W. Peterson, J. Bailey, and A. Youngblood, editors. 2015. Silviculture and monitoring guidelines for integrating restoration of dry mixed-conifer forest and spotted owl habitat management in the eastern Cascade Range. General Technical Report PNW-GTR-915. USDA Forest Service Pacific Northwest Research Station, Portland, Oregon, USA. http://dx.doi.org/10.2737/pnw-gtr-915
Liu, J., T. Dietz, S. R. Carpenter, M. Alberti, C. Folke, E. Moran, A. N. Pell, P. Deadman, T. Kratz, J. Lubchenco, E. Ostrom, Z. Ouyang, W. Provencher, C. L. Redman, S. H. Schneider, and W. W. Taylor. 2007. Complexity of coupled human and natural systems. Science 317(5844):1513-1516. http://dx.doi.org/10.1126/science.1144004
Loudermilk, E. L., R. M. Scheller, P. J. Weisberg, J. Yang, T. E. Dilts, S. L. Karam, and C. Skinner. 2013. Carbon dynamics in the future forest: the importance of long-term successional legacy and climate–fire interactions. Global Change Biology 19(11):3502-3515. http://dx.doi.org/10.1111/gcb.12310
Lovtang, S. C. P., and G. M. Riegel. 2012. Predicting the occurrence of downy brome (Bromus tectorum) in central Oregon. Invasive Plant Science and Management 5(1):83-91. http://dx.doi.org/10.1614/ipsm-d-11-00029.1
McCaffrey, S. M. 2006. Prescribed fire: What influences public approval? Pages 192-198 in M. B. Dickinson, editor. Fire in eastern oak forests: delivering science to land managers. General Technical Report NRS-P-1. USDA Forest Service Northern Research Station, Newtown Square, Pennsylvania, USA. [online] URL: http://www.fs.fed.us/nrs/pubs/gtr/nrs_gtrP1.pdf
McCaffrey, S., E. Toman, M. Stidham, and B. Shindler. 2012. Social science research related to wildfire management: an overview of recent findings and future research needs. International Journal of Wildland Fire 22(1):15-24. http://dx.doi.org/10.1071/wf11115
Meigs, G. W., J. L. Campbell, H. S. J. Zald, J. D. Bailey, D. C. Shaw, and R. E. Kennedy. 2015. Does wildfire likelihood increase following insect outbreaks in conifer forests? Ecosphere 6(7):118. http://dx.doi.org/10.1890/ES15-00037.1
Merschel, A. G., T. A. Spies, and E. K. Heyerdahl. 2014. Mixed-conifer forests of central Oregon: effects of logging and fire exclusion vary with environment. Ecological Applications 24(7):1670-1688. http://dx.doi.org/10.1890/13-1585.1
Millington, J., R. Romero-Calcerrada, J. Wainwright, and G. Perry. 2008. An agent-based model of Mediterranean agriculture land-use/cover change for examining wildfire risk. Journal of Artificial Societies and Social Simulation 11(4):4. [online] URL: http://jasss.soc.surrey.ac.uk/11/4/4.html
Morzillo, A. T., P. Comeleo, B. Csuti, and S. Lee. 2014. Application of state-and-transition models to evaluate wildlife habitat. Pages 129-146 in J. E. Halosfky, M. K. Creutzburg, and M. A. Hemstrom, editors. Integrating social, economic, and ecological values across large landscapes. PNW-GTR-896. USDA Forest Service Pacific Northwest Research Station, Portland, Oregon, USA. [online] URL: http://www.fs.fed.us/pnw/pubs/pnw_gtr896.pdf
North, M., B. M. Collins, and S. Stephens. 2012. Using fire to increase the scale, benefits, and future maintenance of fuels treatments. Journal of Forestry 110(7):392-401. http://dx.doi.org/10.5849/jof.12-021
Noss, R. F., J. F. Franklin, W. L. Baker, T. Schoennagel, and P. B. Moyle. 2006. Managing fire-prone forests in the western United States. Frontiers in Ecology and the Environment 4(9):481-487. http://dx.doi.org/10.1890/1540-9295(2006)4[481:mffitw]2.0.co;2
O’Connor, C. D., G. M. Garfin, D. A. Falk, and T. W. Swetnam. 2011. Human pyrogeography: a new synergy of fire, climate and people is reshaping ecosystems across the globe. Geography Compass 5(6):329-350. http://dx.doi.org/10.1111/j.1749-8198.2011.00428.x
Odion, D. C., C. T. Hanson, D. A. DellaSala, W. L. Baker, and M. L. Bond. 2014. Effects of fire and commercial thinning on future habitat of the northern spotted owl. Open Ecology Journal (7):37-51. http://dx.doi.org/10.2174/1874213001407010037
Ohmann, J. L., M. J. Gregory, E. B. Henderson, and H. M. Roberts. 2011. Mapping gradients of community composition with nearest-neighbour imputation: extending plot data for landscape analysis. Journal of Vegetation Science 22(4):660-676. http://dx.doi.org/10.1111/j.1654-1103.2010.01244.x
Olsen, C. S., J. D. Kline, A. A. Ager, K. A. Olsen, and K. C. Short. 2017. Examining the influence of biophysical conditions on wildland–urban interface homeowners’ wildfire risk mitigation activities in fire-prone landscapes. Ecology and Society 22(1):21. https://doi.org/10.5751/ES-09054-220121
Pyne, S. J. 2008. Spark and sprawl: a world tour. Forest History Today Fall 2008:4-11. [online] URL: http://www.foresthistory.org/Publications/FHT/FHTFall2008/Pyne.pdf
Radeloff, V. C., R. B. Hammer, S. I. Stewart, J. S. Fried, S. S. Holcomb, and J. F. McKeefry. 2005. The wildland-urban interface in the United States. Ecological Applications 15(3):799-805. http://dx.doi.org/10.1890/04-1413
Reinhardt, E. D., R. E. Keane, D. E. Calkin, and J. D. Cohen. 2008. Objectives and considerations for wildland fuel treatment in forested ecosystems of the interior western United States. Forest Ecology and Management 256(12):1997-2006. http://dx.doi.org/10.1016/j.foreco.2008.09.016
Rhodes, J. J., and W. L. Baker. 2008. Fire probability, fuel treatment effectiveness and ecological tradeoffs in western U.S. public forests. Open Forest Science Journal 1(1):1-7. http://dx.doi.org/10.2174/1874398600801010001
Robbins, W. G. 1997. Landscapes of promise: the Oregon story, 1800–1940. University of Washington Press, Seattle, Washington, USA.
Rollins, M. G. 2009. LANDFIRE: a nationally consistent vegetation, wildland fire, and fuel assessment. International Journal of Wildland Fire 18(3):235-249. http://dx.doi.org/10.1071/wf08088
Roloff, G. J., S. P. Mealey, and J. D. Bailey. 2012. Comparative hazard assessment for protected species in a fire-prone landscape. Forest Ecology and Management 277:1-10. http://dx.doi.org/10.1016/j.foreco.2012.04.015
Roloff, G. J., S. P. Mealey, C. Clay, J. Barry, C. Yanish, and L. Neuenschwander. 2005. A process for modeling short- and long-term risk in the southern Oregon Cascades. Forest Ecology and Management 211(1-2):166-190. http://dx.doi.org/10.1016/j.foreco.2005.02.006
Rykiel, E. J. Jr. 1996. Testing ecological models: the meaning of validation. Ecological Modelling 90(3):229-244. http://dx.doi.org/10.1016/0304-3800(95)00152-2
Scheller, R. M., J. B. Domingo, B. R. Sturtevant, J. S. Williams, A. Rudy, E. J. Gustafson, and D. J. Mladenoff. 2007. Design, development, and application of LANDIS-II, a spatial landscape simulation model with flexible temporal and spatial resolution. Ecological Modelling 201(3-4):409-419. http://dx.doi.org/10.1016/j.ecolmodel.2006.10.009
Scheller, R. M., W. D. Spencer, H. Rustigian-Romsos, A. D. Syphard, B. C. Ward, and J. R. Strittholt. 2011. Using stochastic simulation to evaluate competing risks of wildfires and fuels management on an isolated forest carnivore. Landscape Ecology 26(10):1491-1504. http://dx.doi.org/10.1007/s10980-011-9663-6
Scott, J. H., and R. E. Burgan. 2005. Standard fire behavior fuel models: a comprehensive set for use with Rothermel’s surface fire spread model. General Technical Report RMRS-GTR-153. USDA Forest Service Rocky Mountain Research Station, Fort Collins, Colorado, USA. [online] URL: http://www.fs.fed.us/rm/pubs/rmrs_gtr153.pdf
Spies, T. A., M. A. Hemstrom, A. Youngblood, and S. Hummel. 2006. Conserving old-growth forest diversity in disturbance-prone landscapes. Conservation Biology 20(2):351-362. http://dx.doi.org/10.1111/j.1523-1739.2006.00389.x
Spies, T. A., K. N. Johnson, K. M. Burnett, J. L. Ohmann, B. C. McComb, G. H. Reeves, P. Bettinger, J. D. Kline, and B. Garber-Yonts. 2007. Cumulative ecological and socioeconomic effects of forest policies in coastal Oregon. Ecological Applications 17(1):5-17. http://dx.doi.org/10.1890/1051-0761(2007)017[0005:ceaseo]2.0.co;2
Spies, T. A., E. M. White, J. D. Kline, A. P. Fischer, A. Ager, J. Bailey, J. Bolte, J. Koch, E. Platt, C. S. Olsen, D. Jacobs, B. Shindler, M. M. Steen-Adams, and R. Hammer. 2014. Examining fire-prone forest landscapes as coupled human and natural systems. Ecology and Society 19(3):9. http://dx.doi.org/10.5751/ES-06584-190309
Stebleton, A., and S. Bunting. 2009. Guide for quantifying fuels in the sagebrush steppe and juniper woodlands of the Great Basin. Technical Note 430. BLM/ID/PT-09/002+2824. Bureau of Land Management, Denver, Colorado, USA. [online] URL: http://www.sagestep.org/pdfs/SageSTEPFuelsGuide.pdf
Stephens, S. L., and L. W. Ruth. 2005. Federal forest-fire policy in the United States. Ecological Applications 15(2):532-542. http://dx.doi.org/10.1890/04-0545
Stine, P., P. Hessburg, T. Spies, M. Kramer, C. J. Fettig, A. Hansen, J. Lehmkuhl, K. O’Hara, K. Polivka, P. Singleton, S. Charnley, A. Merschel, and R. White. 2014. The ecology and management of moist mixed-conifer forests in eastern Oregon and Washington: a synthesis of the relevant biophysical science and implications for future land management. General Technical Report PNW-GTR-897. USDA Forest Service Pacific Northwest Research Station, Portland, Oregon, USA. http://dx.doi.org/10.2737/pnw-gtr-897
Syphard, A. D., R. M. Scheller, B. C. Ward, W. D. Spencer, and J. R. Strittholt. 2011. Simulating landscape-scale effects of fuels treatments in the Sierra Nevada, California, USA. International Journal of Wildland Fire 20(3):364-383. http://dx.doi.org/10.1071/wf09125
Thompson, J. R., and T. A. Spies. 2009. Vegetation and weather explain variation in crown damage within a large mixed-severity wildfire. Forest Ecology and Management 258(7):1684-1694. http://dx.doi.org/10.1016/j.foreco.2009.07.031
Urban, D. L., M. F. Acevedo, and S. L. Garman. 1999. Scaling fine-scale processes to large-scale patterns using models derived from models: meta-models. Pages 70-98 in D. J. Mladenoff and W. L. Baker, editors. Spatial modeling of forest landscape change: approaches and applications. Cambridge University Press, Cambridge, UK.
Wallin, D. O., F. J. Swanson, and B. Marks. 1994. Landscape pattern response to changes in pattern generation rules: land-use legacies in forestry. Ecological Applications 4(3):569-580. http://dx.doi.org/10.2307/1941958
Wildland Fire Leadership Council. 2014. A national cohesive wildland fire management strategy. Wildland Fire Leadership Council, USDA and U.S. Department of the Interior, Washington, D.C., USA. [online] URL: http://www.forestsandrangelands.gov/strategy/documents/reports/1_CohesiveStrategy03172011.pdf