Woody fuel piles

Threat of fire severity and spread is better understood by using prediction models and other risk assessment tools. One of the initial attempts to implement a more generic assessment of fire risk was the National Fire Danger Rating System (Burgan 1988, Deeming and others 1972 ). This system relied on expert judgment to evaluate the risk from a set of explanatory variables, principally fuels, topography, and weather. Prediction of fire occurrence is just the first part of a complete analysis of fire risk. Fire occurrence risk must be combined with models that determine the risk of the effects of fire.

The severity and spread of wildland fire is key in understanding its threat to forest health. This section reviews many of the models created to help us better manage fire through biophysical principles, probability-based predictions, as well as the effects of vegetation changes on fire behavior. Modeling fire proves to have great promise predicting and managing fire risks. Further, goals to assess ecological risks in the context of fuel loading, which can lead to uncharacteristically severe wildlfire, has shifted fire management to being a proactive rather than reactive practice. Risk management decisions rely on a mix of science and policy; the most important role for science is providing information to be used in environmental decision making (Power and McCarty 1997). Studying fire severity and spread has led to several policy changes recently, including the Healthy Forests Initiative, and may someday shift the scales from fire as a threat to fire a tool in forest management in many ecosystems.

There are several methodologies currently in use or those under development containing features for estimating fire occurrence risk assessment. Two major categories of fire risk assessment tools are: those that predict fire under current conditions, assuming that vegetation, climate, and the interactions between them and fire remain relatively similar to their condition during recent history, and those that anticipate changes in fire risk as climate and vegetation communities change through time. Three types of models have proven useful for predicting fire under current conditions: (1) biophysical models that predict fire from vegetation type, fuel load, and climate; (2) statistical models; and (3) fire behavior models. Programs such as LANDFIRE have great promise for using biophysical properties to estimate risk. Statistical models that use historical data to predict fire probabilities if landscape-fire relationships continue to remain relatively unchanged, are gaining interest as more data becomes available. Fire behavior models are producing accurate predictions of the ways individual fires will move across the landscape. For longer time periods, fire risk needs to be evaluated by models that predict the ways vegetation communities will change over time because these changes will alter fire probabilities. Models capable of being used to track changes in vegetation are identified and the resulting effect on changes in fire frequency. Risk systems need to be designed to track changes in fire susceptibility as the climate changes, using models such as MAPSS.

Prediction of fire occurrence is just the first part of a complete analysis of risks associated with fire. Fire occurrence risk needs to be combined with models that determine the risk of the effects of fire. Models that predict mortality, fuel consumption, smoke production, and soil heating caused by prescribed fire or wildfire should be used, as well as those capable of evaluating second order effects, such as changes in site productivity, animal use, insects, and disease. Fire must be looked at in the context of other stresses, such as invasive insects and pathogens, encroaching urbanization, and loss of critical habitat. There are interactions among stresses that play a role in affecting the frequency and intensity of fire, and fire, in turn, can affect the probability of those stresses. Consequently, risk evaluation systems need to be created that can simultaneously estimate the probability of other major stresses influencing ecosystem development.

Methodologies are described here that may be useful for estimating fire occurrence risk assessment, including the probability of ignition and the spatial spread and intensity of the fire during its lifetime. Two types of risk need to be assessed: (1) fire risk occurrence (hereafter referred to as fire risk), and (2) risk to the ecosystem as a result of fire (hereafter ecosystem risk) (Finney 2005). For purposes of this review, fire risk includes the probability of ignition and the spatial spread and intensity of the fire during its lifetime. Ecosystem risk includes all of the consequences to plant and animal populations and to the soil during the recovery period once the fire has concluded. There is significant understanding of the factors that influence fire risk, with many studies analyzing the long-term consequences to ecosystems (Fairbrother and Turnley 2005). Tools to predict the likely distributional frequency of fire risk under combinations of various conditions are the focus of this review.

The two major categories of fire risk assessment tools are: (1) those that predict fire under current conditions, assuming that vegetation, climate, and the interactions among them and fire remain relatively similar to their condition during recent history; and (2) those that anticipate changes in fire risk as climate and vegetation communities change through time.

Many models are available to evaluate fire risk under current conditions, although the majority of these were designed for analysis in specific locations or under specific sets of conditions. One of the initial attempts to implement a more generic assessment of fire risk was the National Fire Danger Rating System (Burgan 1988, Deeming and others 1972 ). This system relied on expert judgment to evaluate the risk from a set of explanatory variables, principally fuels, topography, and weather. It allowed land managers to estimate fire danger for today or tomorrow for a given rating area. It characterized fire danger by evaluating the approximate probable upper limit of fire behavior in a fire danger rating area during a 24-hour period. A relative rating of the potential growth and behavior of any wildfire was based on a loose correlation between the date of fire discovery and the eventual size of a fire.

Attempts to improve upon the National Fire Danger Rating System fall into one of three categories:

1. biophysical models that predict fire from vegetation type, fuel load, and climate
2. statistical models that rely on relationships extracted from historical data
3. fire behavior models that emphasize the role of spatial distributions of wind, topography, and fuel in determining what portions of a landscape will burn

"Simple" calculation of fire risk

As an example, the first and third of these can be combined in a fire risk calculation in which fire is predicted from the biophysical influences on fire and from the spread of fire once it gets established.

Biophysical models combine local weather patterns (temperature, humidity, wind), vegetation (fuel type, moisture level), and topography (elevation, slope) to arrive at an estimate of fire risk. Biophysical models range from deterministic models in which a given set of input variables will always yield the same prediction, to systems that include simulation. Some of these models combine quantitative and qualitative criteria to arrive at a fire danger index. Biophysical models designed to estimate fire risk across large regions tend to base their predictions on fewer variables because the number of variables that have been quantified across large regions is limited.

Statistical models use regression models developed from historical data to estimate probabilities of fire occurrence under various local environmental conditions. Statistical models are valuable for understanding general historical trends. They are used to provide predictions for such factors as the expected number of fires in an area from explanatory variables such as vegetation patterns, fuel moisture conditions, meteorological variables, number of people visiting a forest, and past history of fires. We describe these types of models only briefly because there are currently few published examples of statistical fire models available. However, the increasing availability of large historical data sets on fire frequency will undoubtedly cause a great expansion of studies using this method in the near future.

Fire behavior models attempt to characterize the propagation and spread of fires under various environmental conditions. Fire behavior probabilities are dependent on ignitions occurring off-site and the fuels, topography, weather, and relative fire direction allowing each fire to reach that location. Because these models require input variables that may not be available over large regions such as entire national forests, they tend to be applied to specific watersheds in which these properties have been mapped or quantified. Because there are many possible interactions of weather with spatial landscape features, fire behavior predictions require the use of spatial fire spread simulations.

To provide a sense of how the different types of approaches are applied and which resources each might require, examples from each category are discussed in detail. Biophysical models are first discussed, with particular attention to two such systems, LANDFIRE and WALTER. Because LANDFIRE is the newest, and, perhaps, most complex approach being pursued, we discuss this approach in depth. WALTER is described as an example of an approach that makes extensive use of expert opinion. An example of the statistical modeling approach relying on large historical data sets is described with a discussion of a probability-based model. Discussed finally is a suite of fire behavior models?FARSITE, BehavePlus, FlamMap, NEXUS, and Visualized Fire Simulation (VFS).

Many models have been constructed with capabilities of projecting changes in vegetation composition over time and the way these changes alter fire risk (Keane and others 2004). Other models are capable of simulating how changes in climate will change biological processes and interactions in ecosystems. Slow changes in vegetation or climate or both can be incorporated into the fire risk calculation using these models. Recent reviews (Keane and others 2003) have concluded that simulation modeling produces the best predictive ability under changing climate and vegetation conditions compared to prediction based on either biophysical properties or statistical prediction from historical data. Over longer time periods, climate and vegetation change is very likely to occur.

The advantages of these models are discussed, building on the extensive evaluations of succession fire models (Keane and others 2004). Further, the models considered are placed in the context of a wider range of tools. Discussed are the four most widely used succession models that contain processes that link vegetation change to fire prediction: SIMPPLLE, MAGIS, VDDT, and TELSA. Many other available models perform well for specific locations for which they were designed. These four are focused on because they have the potential to be easily applied to many different areas in the Western United States.

A final category of simulators includes those that utilize broad concepts of biological processes to project the ways forests are likely to change and where various forest types are likely to be found under future climate conditions. Although this type of model has had a long history of development in forest ecology, it is only recently that several models have had fire intimately incorporated into the model structure. The MAPSS model is described as an example of this category.

Biophysical fire risk models traditionally use regional characteristics of weather patterns (temperature, humidity, wind), vegetation (fuel type, moisture level), and topography (elevation, slope) to produce a prediction index of fire risk based on historical correlations among these variables and fire. More recently, models are being developed that substitute spatial simulations for some of these correlations.

In response to a need for a national evaluation of the spatial distribution of fire risk, the USDA Forest Service developed a partnership with 4 other agencies to develop LANDFIRE ( http://www.landfire.gov/index.html). The goal of LANDFIRE is to identify areas at risk due to accumulation of hazardous fuel for the purpose of prioritizing hazardous fuel reduction projects and improving hazardous fuel treatment coordination between agencies. The program is designed to produce landscape-scale maps and data describing vegetation, fire, and fuel characteristics across the United States (at a 30-m grid resolution) (Keane and others 2004, Rollins and others 2002, Rollins and others 2004, Schmidt and others 2002). LANDFIRE is providing many of the raw materials that will be necessary to produce an estimate of fire risk. Although it is not yet in general use or publicly available, it is important to discuss it here because of the key role LANDFIRE will play in fire risk assessment over the next decade.

Fig. 1 - A diagram of the Landscape
Ecosystem Inventory System

The spatial distribution of potential vegetation, existing vegetation, canopy height, and canopy cover is mapped using gradient-based field inventories coupled with gradient modeling, remote sensing, ecosystem simulation, and statistical analyses. Biophysical gradient maps have been created containing 38 geographical information system (GIS) layers describing the direct and indirect gradients affecting the distribution of vegetation and fire regimes. The vegetation of the continental United States is divided into approximately 500 biophysical units, based on plant composition and the Ecological Systems categorization ( http://www.natureserve.org/). Each successional stage of each biophysical type is separately tracked. Fuel models are assigned to each mapping region, producing fuel maps for fire behavior models, canopy fuel projections, and fuel characterization classes. Crown bulk density and height to crown base are calculated at the plot level from tree lists.

Inputs consist of coarse-scale, 1-km² resolution, spatial data layers. These include potential natural vegetation type, current cover type, site characteristics (such as soils, climate, and topography), historical natural fire regime (fire frequency and severity), and Fire Regime Current Condition Class (layer depicting the degree of departure from historical fire regimes). There are three additional databases that are used as input:

1. National Fire Occurrence, Federal and non-Federal fire occurrences from 1986 to 1996
2. Potential Fire Characteristics? the number of days of high or extreme fire danger calculated from 8 years of historical National Fire Danger Rating System (NFDRS) data
3. Wildland Fire Risk to Flammable Structures?the potential risk of wildland fire burning flammable structures based on an integration of population density, fuel, and weather spatial data

Landscape characteristics that are used to determine fire occurrence and behavior include height to crown base, crown bulk density, fuel loadings, cover type, percent cover, height for each of the forest, shrub, and herbaceous layer, and an estimate of the departure from reference normal fire regime condition class. The interaction between these characteristics and fire probability is estimated by using a suite of computer models, WXFIRE, BIOME-BGC, LANDSUMv4, FARSITE, and HRVStat, discussed in the next section.

LANDFIRE produces three fire regime maps:

1. simulated historical fire frequency and severity
2. fire regime condition class (FRCC)
3. indices of departure from reference conditions

Fig. 2 - LANDFIRE National Fire
Regime Condition Class Map

These tools are used for determining the degree to which current landscape conditions have departed from historical reference condition vegetation, fuel, and disturbance regimes. Figure 2 shows an example map of fire regime condition classes produced by LANDFIRE. An example map of reference fire regimes is shown in Figure 3.

The data layers being produced by LANDFIRE will provide basic information from which a risk assessment can be calculated. However, there are drawbacks to this system. LANDFIRE'S classifications of fire conditions may be too coarse. The calculation of probabilities may require data on a continuous scale. Further, it is not clear whether this data is sufficient to predict the average likelihood of a fire at a location or the distribution probabilities of fires of different sizes and intensities. The probability of a worst case scenario would be difficult to estimate from LANDFIRE'S products.

Fig. 3 - LANDFIRE rapid assessment reference

Because the past condition and fire susceptibility had much to do with past climate, it is unclear whether LANDFIRE will correctly predict the relationship between vegetation, fuel loadings, and fire that will be shaped by future climates. LANDFIRE places less emphasis on the importance of the heterogeneity of types of fire and the key differences among these types, such as crown fires vs. ground fires. LANDFIRE focuses on classes that are relevant to management and current vegetation classifications. There may be advantages to having the flexibility to make changes in these classifications. Natural variability in landscape and fire characteristics, and their influence on fire (and the uncertainty with which predictions can be based on these characteristics) is not treated implicitly within the system.

WXFIRE (Keane and Holsinger 2006) computes spatially explicit, climate-based biophysical variables at any landscape scale or resolution using daily weather data, topography, and soils parameters, and a diverse set of integrated environmental functions. WXFIRE computes over 50 biophysical attributes such as potential evapo-transpiration for each simulation unit. The user must estimate all input parameters for each simulation unit to create an input file to WXFIRE. WXFIRE then calculates, record by record, all biophysical attributes by accessing the DAYMET spatial weather database (Thornton and others 1997) (www.daymet.org) and using the daily weather to compute important climate and ecosystem biophysical variables. The DAYMET database consists of 18 years of daily temperature, humidity, radiation, and precipitation estimates at a 1-km spatial resolution for the contiguous United States. Output from WXFIRE can be used to digitally map those ecosystem characteristics needed by land management including fire regime, fuel model, and vegetation cover type.

BIOME-BGC (Thornton and others 2002) is used to calculate expected forest productivity in response to a given set of environmental conditions. BIOME-BGC is an ecosystem process model that simulates carbon, water, energy, and nitrogen budgets for both vegetation and soil. BIOME-BGC is inherently non-spatial and can be run on an area of any size: a single point, a watershed, a continent. BIOME-BGC computes a set of carbon and water budget metrics on a daily time step driven by daily weather data such as gross primary production, net primary production, evapo-transpiration, and runoff. The model requires three types of information as input: site physical characteristics, plant physiological characteristics, and daily weather data. A set of generic plant functional type (PFT) eco-physiological parameter files has been developed from the literature for use with BIOME-BGC (White and others 2000). The PFTs include evergreen needle leaf forest, deciduous broadleaf forest, evergreen broadleaf forest, deciduous needle leaf forest, evergreen shrub, c₃grass, and c₄grass.

The range and variation of historical landscape dynamics are estimated with LANDSUMv4 (Landscape Succession Model version 4.0) and HRVStat (Historical Range and Variation Statistics) of landscape characteristics ( Keane and Holsinger 2006, Keane and others 2002). LANDSUMv4 simulates fire and succession on fine-scale landscapes for land management applications. Species composition and stand structure are assumed to change at a predefined rate, although disturbance initiation is modeled stochastically, and disturbance effects are based on the current vegetation conditions.

The model FARSITE projects the spread of fire across landscapes, using slope relationships and wind vectors (discussed under the Fire Behavior Modeling heading). FIREHARM (http://ams.confex.com/ams/pdfpapers/66069.pdf) is used by LANDFIRE to produce probabilities of a given fire event from long-term weather. It identifies areas of highest risk based on fuel consumption, smoke production, tree mortality, soil heating, crown fire index, and proximity to urban areas. FIREHARM will produce a probability of fire for each category of weather conditions;(i.e., the number of days the fire potential is above a certain level). FIREHARM will calculate four fire behavior variables (fire line intensity, spread rate, flame length, crown fire potential), five fire danger variables (spread component, burning index, energy release component, Keetch-Byram drought index, ignition component), and five fire effects variables (smoke, fuel consumption, soil heating, tree mortality, scorch height) for every day in the DAYMET 18-year record for each spatial location. The program will simulate moisture conditions for each dead fuel component, (e.g., duff, litter, downed woody, logs) and live fuel component, (e.g., shrubs, herbs, trees) using a complex set of biophysical equations.

The Wildfire Alternatives (WALTER; http://walter.arizona.edu/index.asp) system for estimating fire risk is being developed at the University of Arizona. It is an interdisciplinary research initiative aimed at improving our understanding of the processes and consequences of interactions among wildfire, climate and society. WALTER seeks to capitalize on advances in geospatial, analytical, and web delivery technology to provide access to scientific research activities and findings and educational materials using the decision support tool, Fire-Climate-Society Strategic Fire Model (FCS-1).

Components of the Fire-Climate-Society Model

FCS-1 is an online, spatially explicit strategic wildfire planning model with an embedded multi-criteria decision process that facilitates the construction of user-designed fire risk assessment maps for different climates. The resulting maps show spatially explicit information about the geographical distribution of fire probability and values at risk for the selected study area.

FCS-1 was developed for the varying vegetation, climate, and topography as well as the unique human dimensions of wild fire found in southeastern Arizona and northern New Mexico. The model currently is made up of five fire probability and four values at risk model components. FCS-1 can be run under differing climate and corresponding fuel moisture conditions. Through an analytical hierarchy process, FCS-1 allows users to assign weights to individual model components. The on-line application can be used to help understand differing views and build consensus.

Like many expert systems, WALTER has the framework for considering an array of disparate factors in evaluating the ways they may influence fire. However, expert opinion can have uncertain predictive ability. The main value of this type of system is to identify interrelationships that may require management to control fire frequency and pattern.

A model has been developed for calculating the expected net value change (ENVC) as the product of the probability of a fire at a specific location and the resulting change in financial or ecological value (Ager and Finney, in press). The model calculates the sum of the product of the probability of the i-th fire behavior at a specific location over N fires multiplied by the benefits and losses afforded for the j-th value of M values received from the i-th fire behavior. The expected net value change (ENVC) can include financial, ecological, or other values at present day or future discounted values. Assumptions about the effect of wildfire suppression on wildfire probability and value change can also be incorporated into the (ENVC) equation. The estimation of wildfire probabilities for a specific future time period is derived from a calculation of the minimum travel time a hypothetical fire would take moving across the landscape, using algorithms from the FARSITE fire behavior model (discussed under the Fire Behavior Modeling heading). These models are run in parallel using a set of networked computers and an estimate of the spatial pattern of forest vegetation and fuels.

This model has been applied to a 16,000 ha wildland-urban interface in eastern Oregon simulating 12 fuel management scenarios and four land value schemes. Burn probabilities were estimated by simulating 200 randomly ignited wildfires, and then the average net value change for each fire and pixel on the landscape was estimated. The results indicate that fuel reduction on a relatively minor percentage of the landscape (20 percent) resulted in a 20-percent to 50-percent positive change in ENVC for most of the scenarios simulated.

Probability-based fire risk models (Brillinger and others 2006) define risk using three probabilities:

1. the probability of fire occurrence,

2. the conditional probability of a large fire assuming an ignition occurs

3. the unconditional probability of a large fire

An illustration of the general approach shared by these types of models is the probability-based fire risk model (Preisler and others 2004), which estimates fire probability by fitting a nonparametric logistic regression to data grouped in cells of 1 sq. km. with a temporal resolution of one day. The input used by this model is:

1. the probability of fire occurrence (historical data)

2. the conditional probability of a large fire given ignition (daily values of precipitation, lightning, temperature, wind speed, and humidity)

3. the unconditional probability of a large fire (10-hour lag fuel moisture).

It produces maps of predicted probabilities and estimates of the total number of expected fires in a given region and time period. This method is particularly useful for assessing the utility of explanatory variables, such as fire, weather, and danger indices for predicting fire risk. It has the advantage of basing fire predictions on rigorous use of probability statistics.

Fire behavior models predict the propagation of fire by assuming that the landscape is subdivided into cells, and each cell has a probability of burning that depends on conditions in the cell and in surrounding cells, (e.g., Beer 1991). Many of these models use a deterministic version of the elliptical growth model (Green and others 1983, Richards 2000) to simulate spread of forest fires. The input data for these models include a base vegetation map, which can be generated with a vegetation simulator, such as Forest Vegetation Simulator (FVS) (Dixon 2002) and a fuel load simulator, such as Fuels and Fire Extension ( http://www.essa.com/downloads/prognosis/ffe.pdf). Fire behavior predictors were initially built for local scale projections, where initial ignitions and spatial distribution of landscape properties could be precisely specified.

Among the more widely applied fire behavior models is FARSITE, which predicts fire spread and intensity across a specific landscape as a continuous fire over multiple time steps at a user-specified resolution, commonly 30 m. The focus of FARSITE is to simulate fire growth and the changes that occur over time for a specific fire. It uses spatial information on topography and fuels along with weather and wind files to simulate surface fire, crown fire, spotting, postfrontal combustion, and fire acceleration. FARSITE requires spatial landscape information from a GIS to run, including slope, aspect, surface fuel, canopy cover, crown base height, crown bulk density, and stand height. The outputs include a prediction of fire perimeter length and location, spread rate, intensity, flame length, and heat per unit area. Although the focus is on single fire, a regional prediction could be made by combining output from all fires each year. Fire spread predictions are based on a snapshot of current vegetation structure. In order to simulate fire behavior under future conditions, the changes that will occur in the vegetation structure must be known.

BehavePlus ( http://fire.org/) predicts fire behavior based on fuels, weather, topography, and wildfire situations. BehavePlus uses a minimum amount of site-specific input data to predict fire behavior for a point in time and space; (i.e., spatially explicit data layers are not used). The present version of BehavePlus simulates only surface fire spread, but later versions may include crown fire simulation capability.

FlamMap ( http://fire.org/) is a fire behavior mapping and analysis program, based in a geographic information system (GIS) that computes potential fire behavior characteristics (spread rate, flame length, fire line intensity, etc.) over an entire FARSITE landscape for constant weather and fuel moisture conditions. FlamMap uses spatial information on topography and fuels to calculate fire behavior characteristics at one point in time, assuming that all cells in the landscape function independently of one another. A map is produced for an area of any modeled value, such as: fuel moisture, fire line intensity, scorch height, or fuel consumption. Comparisons can be made between locations, or the effect of fuel treatment can be examined.

NEXUS ( http://fire.org/) is an Excel spreadsheet linking surface and crown fire prediction models to compute indices of relative crown fire potential. NEXUS is useful for evaluating alternative treatments for reducing crown fire risk and assessing the potential for crown fire activity. NEXUS includes several visual tools useful in understanding how surface and crown fire models interact.

VFS ( http://fire.org/) is a graphical user interface- (GUI-) based computer program to simulate and animate fire on heterogeneous landscapes. VFS captures fire spread behaviors based on fuel configuration, wind regime and topographical effects using percolation algorithms such as static percolation, depth first search recursive algorithm, and dynamic percolation with fire front. Users can compare the simulation capability of each method, (e.g., burned pattern maps). Furthermore, output from VFS can be linked to GIS and used to cross-validate other fire simulation models. To evaluate the sensitivity of the input parameters of each method, users can specify a range for each parameter in VFS and test the influence of changes in parameters on model predictions. VFS can be used as a parameterization tool for the forest landscape models that incorporate fire spread simulation.

Fire risk over long time periods cannot be adequately evaluated without projecting the ways that the structure and composition of forest vegetation and fuels will change over time. Fire is very sensitive to vegetation structure and composition (Clark 1993, Swetnam 1997, Swetnam and Baisan 1996). Fire will, in turn, affect the rate and direction of vegetation change (Lenihan and others 1998).

Keane and colleagues classified 44 linked fire succession models, based on their complexity, principle mechanism, and stochasticity in their simulation design (Keane and others 2004). Most of these models fall into the category of fire behavior models because they attempt to simulate, at a fine scale, the dynamics that cause fire to spread across a landscape. These authors provide a handy table for deciding which models to use based on: (1) whether the purpose is management or research, (2) the importance of predicting fire patterns, (3) whether information is needed on specific species or just on stand characteristics, and (4) what spatial scale is of interest (Table: Model selection key).

Complex models such as Fire-BGC (Keane and others 1989) and LANDIS (Mladenoff 2004, Mladenoff and others 1996) simulate vegetation change as a complex function of either the development of nutrient cycling conditions or the driving force of individual life history characteristics. The most complex of these are the gap models such as ZELIG-SP (Miller and Urban 1999) that predict successional development by simulating the dynamics of each individual tree on representative plots in the forest. Simpler models, such as SIMPPLLE (Chew and others 2004), represent vegetation changes as a predictable succession of stages following a resetting disturbance. A few of these models that are in common use were selected for further discussion in the following sections.

SIMPPLLE (SIMulating Vegetative Patterns and Processes at Landscape scaLEs) is a stochastic non-spatial simulation model for projecting vegetative change over time in the presence of natural processes, either with or without management treatments. It models interaction of various natural processes on a landscape. Because it is stochastic, multiple simulations are run to generate a record of the frequency of natural disturbances for each polygon in the landscape. These frequencies represent an estimate of the risk of these natural processes occurring over a given period of time and are used to develop a risk index for each section of the landscape.

MAGIS (Multiple-resource Analysis and Geographic Information System ( http://www.fs.fed.us/rm/econ/magis/) is a spatial decision support system for using the risk index to find optimal management practices spatially and temporally for a landscape. MAGIS uses optimization toselect the spatial arrangement and timing of treatments that fit user-determined objectives and constraints. A variety of resource effects, management targets, and economic costs or benefits can be used to specify the objective and constraints for scheduling both vegetation treatments and road activities.

With these two programs, the user can evaluate a variety of management alternatives. However, their projections are dependent on the estimate of current fuel distributions across the landscape. When projecting into the future, they will be increasingly in error as these fuel loads change.

Vegetation Disturbance Dynamics Tool (VDDT) ( http://www.essa.com/downloads/vddt/) was developed to support the Interior Columbia River Basin Assessment. This nonspatial tool uses a state and transition matrix approach to predict changes in vegetative composition and structure using disturbance probabilities and successional pathways, including infrequent large-scale disturbances such as stand replacement fires.

VDDT models typically apply to potential vegetation types. For each of these types, succession classes are defined according to the cover type and structural stage. In the absence of disturbance, vegetation community assemblage changes from one succession class to the next. Both natural and man-caused disturbances that affect vegetation can be examined. In VDDT, disturbances are defined for each succession class according to type, (e.g., wildland fire, harvest, etc.), succession class destination, probability of occurrence, and the relative ages for which each probability applies. For each year of the simulation, VDDT determines whether each landscape unit is subjected to a disturbance.

Tool for Exploratory Landscape Scenario Analyses (TELSA) ( http://www.essa.com/downloads/telsa/index.htm) is a spatially explicit extension to VDDT that simulates forest succession, natural disturbances, and forest management activities. It is designed to simulate up to 250,000-ha landscape units. TELSA can be used to simulate multiple scenarios, each characterized by different assumptions about management actions and natural disturbances. Because wildfires and other natural disturbance events that affect vegetation dynamics are inherently unpredictable, the model can use multiple stochastic simulations of each scenario to provide estimates of the mean, range, and variability of the selected performance indicators. Unlike many other strategic planning models of landscape dynamics, TELSA takes into account natural disturbances so users can explore how their proposed management strategies will interact with vegetation succession and disturbances to alter landscape composition and structure. This model has been used to define the transition times between various succession classes (combinations of species composition and structural stage), the probabilities and impacts of disturbance by insects, fire or other agents, and the impacts of landscape management actions on structure and composition.

Biophysical process models can be used to estimate the effect of vegetation change on fire risk. The representation of basic processes in biophysical process models allows them to project the consequences of known relationships under future conditions. These models are capable of examining the effects of long- term changes in conditions such as climate change. An example of these model types, MAPSS ( http://www.fs.fed.us/pnw/corvallis/mdr/mapss/) predicts vegetation distributions from either by the availability of water in relation to transpirational demands or the availability of energy for growth (Neilson and Wullstein 1983, Neilson and others 1989, Stephenson 1990, Woodward and Williams 1987).

MAPPS and its recent derivative, MAPSS-CENTURY 1 (MC1), simulate life-form mixtures and vegetation types, fire disturbance, and ecosystem fluxes of carbon, nitrogen, and water (Lenihan and others 2003). MC1 is routinely implemented on spatial data grids of varying resolution, (i.e., grid cell sizes ranging from 900 m² to 2500 km²) where the model is run separately for each grid cell; (i.e., there is no exchange of information across cells) (Bachelet and others 2000, Bachelet and others 2001, Daly and others 2000). MAPSS has been implemented at a 10 km resolution over the continental United States and at a 0.5 degree resolution globally (Neilson 1993, Neilson 1995, Neilson and Marks 1994). It has also been implemented at the watershed scale (MAPSS-W, 200 m resolution) through integration with a distributed catchment hydrology model (Daly and others 1994, Wigmosta and others 1994).

Using climate data at a monthly time step, the model calculates the leaf area index (LAI) of three generic life form groups?trees, shrubs, and grasses?in competition for both light and water given a site water balance consistent with observed runoff (Neilson 1995). Water in the surface soil layer is apportioned to the two life forms in relation to their relative LAIs and stomatal conductance.

MAPSS simulation of fire probability

MAPSS is used to develop midterm forecasts of fire risk by using spatially distributed, high resolution climate data and potential future climate forecasts from climate models. An example of these predictions is given in. The PRISM model (www.ocs.orst.edu/prism) is used to produce high-resolution data grids of observed fire weather extending back to 1895 and interpolations of weather station data that are sensitive to topography. Fire risk forecasts, including fire occurrence, area burned, and fire behavior are generated from the historical and forecast weather data (Bachelet and others 2000, Lenihan and others 2003) (www.fs.fed.us/pnw/corvallis/mdr/mapss/pubs.html). A fire event is triggered in any given cell on any day if one of three thresholds is exceeded. A threshold of the 12-month standardized precipitation index (SPI) is used as an indicator of moderate to severe drought to control the inter-annual timing of fire events. A threshold of the 1000-hour fuel moisture content of dead fuels is used as an indicator of extreme fire potential to control the seasonal timing of fire events. A threshold of fine fuel flammability is used as an indicator of the sustainability of fire starts to control the timing of fire events at the daily time step. There is no constraint on fire occurrence due to the availability of an ignition source, such as lightning or human-caused ignition. Once a fire event is triggered, the MC1 fire module determines the fraction of each cell burned, which is dependent on the current vegetation type, the current drought condition, and the number of years since fire.

Potential fire behavior is also influenced by estimates of the mass, vertical structure, and moisture content of several live and dead fuel size-classes. The consumption of aboveground biomass, carbon, and nitrogen stocks are simulated as a function of the moisture content of each live and dead fuel size-class and the vertical structure of the canopy. The more rapidly growing grasses are assumed to gain an advantage over woody life forms in the competition for water and nutrients, promoting even greater grass production which, in turn, produces a more flammable fuel bed and more frequent fire. MAPSS is capable of producing predictions of future vegetation change and fire risk over a large geographical area.

Much effort has gone into creating a capability of predicting fires throughout the region, both in their likely location and frequency. We described two major categories of fire risk assessment tools: those that predict fire under current conditions, assuming that vegetation, climate, and the interactions between them and fire remain relatively similar to their condition during recent history; and those that anticipate changes in fire risk as climate and vegetation communities change through time. Three types of models have proven useful for predicting fire under current conditions:

1. biophysical models that predict fire from vegetation type, fuel load, and climate

2. statistical models that use historical data to predict fire probabilities if landscape-fire relationships continue to remain relatively unchanged

3. fire behavior models that produce predictions of the ways individual fires will move across the landscape

Programs such as LANDFIRE have great promise for using biophysical properties to estimate risk. The LANDFIRE program is creating base data sets of fuel loadings, biophysical variables, and vegetation composition. Risk assessments conducted with LANDFIRE tools will depend heavily on well-tested models such as FARSITE. Since the intention is to implement the LANDFIRE project nationally at a fine-scale resolution, its data sets could provide the framework around which to build other risk evaluations. However, the data sets produced by LANDFIRE could be made even more valuable if they contained information on the probabilities of fires of different sizes, intensities, and heterogeneity of fire types at any given location.

For longer time periods, fire risk must be evaluated by models that predict the ways vegetation communities will change over time because these changes will alter fire probabilities. Risk systems must be designed to track changes in fire susceptibility as climate changes, using models such as MAPSS. LANDFIRE is not currently designed to track these changes, so it is unclear whether it will correctly predict the relationship between vegetation, fuel loadings, and fire that will be shaped by future climates. MAPSS would have a much higher likelihood of being able to track these changes in relationships.

Prediction of fire occurrence is just the first part of a complete analysis of fire risk. Fire occurrence risk must be combined with models that determine the risk of the effects of fire. For example, one such effects model is FOFEM (a First Order Fire Effects Model) ( http://fire.org/), which predicts mortality, fuel consumption, smoke production, and soil heating caused by prescribed fire or wildfire. It uses an algorithm key that selects different functions for different geographic areas and cover types. It can be used for setting acceptable upper and lower fuel moistures for conducting prescribed burns, determining the number of acres that may be burned on a given day without exceeding particulate emission limits, assessing effects of wildfire, developing timber salvage guidelines following wildfire, and comparing expected outcomes of alternative actions. There are second order effects, such as changes in site productivity, animal use, insects, and disease that need to be evaluated by other models.

Fire must be looked at in the context of other stresses, such as invasive insects and pathogens, encroaching urbanization, and loss of critical habitat. There are interactions among stresses that play a large role in affecting the frequency and intensity of fire, and fire, in turn, can affect the probability of those stresses. Fire probability can increase in stands that have experienced large amounts of tree mortality caused by native pest infestations. Because these precursor stresses have received less attention than fire, the uncertainty for predicting their probability is much higher. The effects of fire are dramatic, but its role in shaping the future of forest systems may be equaled or exceeded by other stresses. Consequently, it is vitally important that risk evaluation systems be created that can simultaneously estimate the probability of all the major stresses influencing forest and grassland development.

In the development of fire modeling systems, data sets containing fine-scale grids of key data on the landscape have been established, with most containing detailed information on variables such as fuel loads, vegetation, and climate trends. These variables are likely to be useful in evaluations of many other kinds of risks. LANDFIRE'S classifications of fire conditions may be too coarse and inflexible to be useable to assess risks across a number of stresses at a given location. Vegetation mapping systems being developed (Kerns and Ohmann 2004, Ohmann and Gregory 2002, Wimberly and Ohmann 2004) may provide a much more flexible solution to this problem. Although there is not yet a movement toward a standard set of spatial data where models require identical inputs, such standardization could certainly aide risk comparisons. However, in order to compare fire risks to those from other threats, a flexible classification of forest types may be needed that is not just centered on fire risk characteristics.

The great proliferation of fire modeling systems in different portions of the regions suggests that each has specific strengths in simulating fires in the area for which the model was originally designed. However, maintaining so many different types of models in the fire modeling toolbox will inevitably prove unwieldy and confusing to potential users. Consequently, an effort toward consolidation (or choice of the best models) is likely to occur. The ultimate question is whether any of the tools discussed in this review could provide an initial framework on which evaluations of risks to other threats could be added. The best models would be those that facilitate comparison between fire risk and its associated ecosystem risk and the risk from other threats. It was not possible in this review to evaluate which fire models offer a better route for considering multiple interacting stresses. This question must remain a primary one to consider in the choice and use of models.

Risk is a combined statement of the probability that something of value will be damaged and some measure of the damage?s adverse effect. Wildfires burning in the uncharacteristic fuel conditions now typical throughout the Western United States can damage ecosystems and adversely affect environmental conditions. Wildfire behavior can be modified by prefire fuel treatments, thereby reducing risks to firefighters, structures, and ecosystems, but such projects pose their own environmental risks. To support fuels treatment decisions, environmental analysis of alternatives is generally required, including taking no action. How can managers determine whether risks of actively treating fuels are greater than risks posed by no action? The risk-reduction benefits of fuel treatment are often overlooked in decision processes for comparing wildfire effects with and without fuel treatment. To fill the void, a comparative ecological risk assessment conceptual model is presented. Both prefire fuels treatment and postfire events produce sediment that can adversely affect water quality and aquatic organisms. Similarly, both prescribed fire and wildfire can adversely affect air quality. The model?s trade-off diagram tests a risk management hypothesis: The benefits of restoring natural (historical) fire regimes and native vegetation in a particular location, plus the benefits of reducing the severity of wildfire effects, balance favorably against any adverse effect, either short- or long-term, from fuels treatment. Managers may believe this hypothesis, but policies require environmental analysis to support it. A trade-off diagram illustrates the conceptual model and graphically replies to the question: Which produces more sediment, wildfire burning under untreated conditions, wildfire burning after fuels are reduced, or the treatments designed to reduce wildfire risks? Similarly: Which situation would produce more fine particulate matter (PM_2.5) air pollution? Trade-off diagrams of such situations may contribute to sustainable resource management decisions by improving communications between risk assessors, public agency managers, and interested nongovernmental parties.

Compared to not taking any action, fuel treatments may or may not reduce adverse ecological and environmental effects that accompany wildfires. Risk management is central to many human enterprises and is the foundation for all fire management activities (USDA-FS/USDI and others 2001). The U.S. Department of Agriculture, Forest Service (USDA-FS), and the U.S. Department of the Interior, Bureau of Land Management (BLM), have been advised to adopt a systematic risk-based approach to target fuel reduction projects across landscapes and to make fully informed decisions about fuels treatment project alternatives and their effects (GAO 2004, USDA-OIG 2006).

Risk is usually defined as having two components: a measure of adverse effects, and the probability of the adverse effects? occurrence. Many people have difficulty comprehending risk, and its quantification has challenged and confused lay persons and professionals (Haimes 2004). A simplifying approach for fire-adapted ecosystems is to use the fire return interval as the time horizon for analyzing environmental effects. By definition, this ensures that a wildfire will occur during the analytical period, and different conditions that affect fire behavior can be compared more readily by eliminating the need for knowing the probability distribution of fire occurrence. The risk management question then becomes: What is the desired condition for a particular ecosystem and location when the inevitable lightning bolt strikes?

Wildfire risk management is a synthesis of scientific and nonscientific concerns. Information from ecological, social, managerial, and policy sciences is integrated into a decision process framework that incorporates social values and concerns. These include democratic process and institutions for governing collective decisions in our society. Risk management depends on effective communication of information between risk assessors in regulatory agencies, risk managers in land management agencies, and interested members of the public who may be directly or indirectly affected by wildfire.

Risk is related to each of the three components of the general sustainable forestry model?risk of losing ecosystem components or damaging environmental values, economic investment risk, and social risk in communities facing forest-based change (O?Laughlin 2004, 2006). This synthesis takes a problem-oriented approach (see Clark 2002) to wildfire risk management and deals primarily with the ecological and environmental aspects of wildland fire management embedded in decision making and social contexts and reflected in various institutions. Whether fuels treatment should take place on public lands is a collective decision.

Managers may lack the tools and information to demonstrate the beneficial effect that fuel treatment projects could have on environmental quality. To fill the void, a conceptual model of decision trade-offs has been developed (O?Laughlin 2005a, O'Laughlin 2005d). Herein, it is applied to the multiobjective fire/fish risk management problem in order to compare postfire sedimentation with and without fuel treatment. This same approach can be used to assess other environmental effects including fine particulate matter (PM_2.5) air pollution from either wildfire or prescribed fire smoke. The model is based on the framework of the Environmental Protection Agency?s Guidelines for Ecological Risk Assessment (EPA 1998). In a comparative risk assessment framework, the integration of sediment, smoke particles, and other environmental risks with land management objectives is accomplished not by science, but via social process, primarily the public involvement processes required by environmental and land use planning laws. Risk assessors and risk managers interact with stakeholders to determine which risks are most important. Risks to firefighters, structures, scenery and aesthetics, vegetation species and age classes, wildlife habitat, and air and water quality should be considered. Using cause-and-effect modeling results developed by risk assessors, risk managers can demonstrate for each risk whether prefire fuels reduction could potentially reduce postwildfire adverse effects.

This section is concerned with all aspects of risk analysis, including risk communication as well as risk assessment and management. First, appropriate terminology is presented. Then the wildfire situation in the Western United States is defined as a fuels management problem and placed within its decision process and social contexts. Complications from spatial and temporal dimensions, as well as risk governance, arise from policy requirements and social perceptions of risk that may require institutional changes to produce sustainable improvements in forest ecosystem conditions. This synthesis rationalizes the choice of parameters selected to adapt the EPA?s ecological risk assessment framework for the purpose of supporting fuels treatment decisions by comparing wildfire risk management alternatives using a with-or-without framing consistent with National Environmental Policy Act (NEPA 1969) requirements. A conceptual model diagram is the core of ecological risk assessment and the principle product of this synthesis, where it is used to compare wildfire effects on sediment with and without fuel reduction.

Risk terms can be a barrier to effective communications if not properly defined. Risk is a combined statement of the probability that something of value will be damaged and some measure of the damage's adverse effect. Risk simply gives meaning to the things, forces, or circumstances that pose danger to people or what they value (NRC 1996). Risk can mean both the probability of loss and the hazard or threat that might cause that loss (Harwood 2000). A hazard or threat is something that poses danger or can cause an adverse effect. Stressor, an EPA term, seems to be synonymous with hazard or threat and is any physical, chemical, or biological entity that can induce an adverse response in an ecological risk assessment endpoint. The endpoint is an explicit expression of the environmental value that is to be protected, operationally defined by an ecological entity and its attributes (EPA 1998).

Ecological risk assessment is a process that evaluates the likelihood that adverse ecological effects may occur or are occurring as a result of exposure to one or more stressors (EPA 1998). Ecological risk assessments are developed within a risk management context to evaluate human-induced changes that are considered undesirable and are used to support many types of management actions, including the management of watersheds or other ecosystems affected by multiple nonchemical and chemical stressors (EPA 1998).

Adverse effects and their potential damages and consequences are real components of risk (Haimes 2004). The meaning of risk has always been inherently controversial and political because it depends on value-based, (i.e., non-scientific) judgments about adverse effects (Slovic 1999). Changes are considered undesirable because they alter valued structural or functional characteristics of ecosystems or their components. An evaluation of adversity may consider the type, intensity, and scale of the effect as well as the potential for recovery from the risk-inducing event (EPA 1998). The probability component of risk is an imagined, mathematical human construct (Haimes 2004). Although most people understand probability in its simplest form?the likelihood of outcomes from tossing a coin or rolling dice?conditional and joint probabilities can be perplexing.

Risk analysis is usually considered to be the process of assessing, characterizing, communicating, and managing risk, (e.g., Haimes 2004, NRC 1996). Effective risk analysis is integrated into decision-making processes, not treated as a gratuitous add-on task (Haimes 2004). Risk assessment asks: What can go wrong, and what are the consequences? Human and organizational failures are sources of risk and may be caused by environmental and institutional elements. Risk management asks what can be done, and what are the impacts on future options? Risks cannot be managed until they have been assessed, and some form of model is necessary for that (Haimes 2004).

Figure 1 -

Wildfire acres burned in the 11 western states

The accumulation of fuels over time can lead to uncharacteristic wildfires and associated problems. Alteration of historic fire regimes often causes serious changes in forest ecosystem processes, resulting in unusually intense, large fires. Current forest conditions and wildfire regimes pose risks to many environmental and socioeconomic values and threaten human communities (USDA-FS 2004). Wildfires near human communities can have devastating effects, as can postfire floods (May, this volume). Wildfire poses immediate threats of damage or loss to nearby structures, and endangerment of public safety is an even more important concern during a wildfire incident and for many years thereafter. Wildfires also pose secondary threats to ecological and environmental resources (Summerfelt 2003). (See Figure 1).

The acreage affected by wildfire nationwide has steadily increased over the past four and a half decades, with a trend towards uncharacteristically severe and uncontrollable fire behavior (NIFC 2006b). Trends of increasing fire size and severity have emerged over the past 20 years (USDA-FS 2004) as wildfires in the Western States have increased, (see Figure 2). The trend is influenced by changes in climate, extreme droughts, and, in some forests, overabundant fuels (Westerling and others 2006).

Figure 2 -

Immediate and secondary wildfire threats

The combined effects of increased fuel accumulations, lengthened fire seasons, and intensified burning conditions are expected to contribute to larger and more extensive wildfires in the near future (Covington and others 1994), with increases of 74 to 118 percent in wildfire burn areas expected over the next century (Running 2006). These expectations underscore the urgency of fuels management to reduce wildfire hazards to human communities as well as actions to mitigate wildfire impacts in forests that have undergone substantial alterations due to past land uses (Covington and others 1994, Westerling and others 2006).

An understanding of the decision processes and social context for wildland fire management is necessary if risk analysis is to be integrated into land and resource management decision making. Decision processes are defined by public policies and laws, which are a function of the social environment within which wildland fire management occurs. Although various agencies and organizations may perceive wildfire risks differently, management decision processes are affected by the same evolving institutional framework of laws and policies. By engaging collaboratively with stakeholders in formal decision process forums, managers can integrate different perspectives regarding environmental risks and thereby arrive at socially acceptable decisions regarding wildland fire management and fuels reduction.

The ultimate utility of decision analysis is not necessarily articulating the best policy option, but avoiding extreme events (Haimes 2004), such as large-scale, uncharacteristically severe wildfires. Risk analysis traditionally has been used for other purposes, but it can address forest management issues in a transparent way and disclose risk trade-offs that are often not accounted for in other decision analysis techniques (Hollenstein 2001). Land and resource management decisions always involve risk, including the decision not to take action (Thomas and Dombeck 1996).

The Federal Wildfire Policy (USDA-FS/USDI and others 2001) recognizes that sound risk management is a foundation for all fire management activities. Together with the National Fire Plan (NFP), this policy provides the institutional framework for Federal agencies, States, Native American tribes, local governments, and communities to manage wildfire risk, improve land conditions, and reduce impacts to communities while ensuring sufficient firefighting capacity for the future. The NFP has established a long-term hazardous fuels reduction program in which treatments are designed to reduce wildfire risks to people, communities, and natural resources while restoring forest and rangeland ecosystems to closely match their historical structure, function, diversity, and dynamics (USDA-FS/USDI 2006a). The NFP is implemented through a collaborative framework agreed upon by Federal and State agencies, tribes, and other parties (WGA 2006). The collaborative framework is a way to address fire and fuel management problems within their social context.

National Environmental Policy Act (NEPA) process

The National Environmental Policy Act (NEPA 1969) is the cornerstone of our environmental laws. It requires that Federal agencies analyze the short- and long-term adverse environmental consequences of a range of proposed management alternatives, including no action. The result of the NEPA process is some type of an environmental impact assessment document (see figure at right). According to an EPA scientist (Fairbrother and Turnley 2005), the NEPA process has various shortcomings that have hampered decision making and significantly reduced public acceptance of fuels treatment. Risk assessment integrated into the NEPA process could result in more meaningful environmental impact analyses, thereby providing a more technically sound and robust means for assessing and comparing potential adverse outcomes of proposed management alternatives (Fairbrother and Turnley 2005).

The healthy forests policy?comprised of the President?s Healthy Forests Initiative and the Healthy Forests Restoration Act (HFRA 2003)?has led to modification or streamlining of decision processes (see O?Laughlin 2005b). The HFRA requires that before a court can issue an injunction, there must be a weighing of the environmental effects of doing fuels treatment against not doing fuels treatment. A legacy of HFRA therefore may be the stimulation, if not institutionalization, of comparative ecological risk assessment. Courts will look to land management agencies for such analysis. At this writing, it is unclear what the analysis might look like. One option is adapting the EPA (1998) Guidelines (see EPA's Risk Assessment Framework) to a wildfire risk management problem (see Fire/Fish Risk Management Application).

Risk is a social construction, combining science and judgment with psychological, social, cultural, and political factors (Slovic 1997). What risks are taken into account, how they are framed, and what constitutes a solution to a risk problem are all matters that go beyond scientific inquiry (De Marchi 2003). Environmental risk involves value judgments that reflect much more than just the probability and consequences of the occurrence of an event (Kunreuther and Slovic 1996). Because risk assessment brings specific values into consideration, it can help reveal which values are at greater risk (Molak 1997). Interested parties can contribute knowledge that risk assessors would otherwise overlook (NRC 1996). Serious attention to citizen participation and process issues may eventually lead to more satisfying and successful ways to manage risks (Slovic 1997). If done systematically and transparently, risk assessment can help build trust (Slovic 1993) primarily by adding transparency to forest decision-making processes (Hollenstein 2001).

Social interaction is the phase of the risk analysis process during which integration of different risk assessment endpoints can be addressed. To enable management actions to improve the fire and fuel problem on Federal lands, managers should focus on the things people care about. These include:

1. the condition of forests relative to forest values associated with management objectives, (i.e., forest health)

2. forest values at risk of damage from nature?s forces as well as human actions

3. the environmental and socioeconomic effects of wildland fire management policy that protects some values while putting others at greater risk (O?Laughlin 2006)

For example, trying to protect fish habitat in fire-adapted ecosystems by not allowing a riparian vegetation management project simply because it will produce a small amount of sediment may be counterproductive in the long term. Fire is inevitable, and the magnitude of adverse effects is more important than when the effects occur (O?Laughlin and others 1998).

When interacting with stakeholders, risk assessors and managers should use qualitative and quantitative approaches as appropriate, while disclosing assumptions and potential for errors. Risk assessment can help risk managers compare the environmental effects of management alternatives. Clear objectives are needed, consistent with a long-term vision of what the land should look like. Effective communication of short- and long-term risks and risk-reduction benefits can help build trust with stakeholders (O?Laughlin 2005c). The thought process that goes into evaluating a particular hazard is more important than the application of some sophisticated mathematical technique or formula (Molak 1997). More emphasis on the risk management aspects of risk analysis would mean greater stakeholder involvement and de-emphasis on quantitative characterization of risk and uncertainty (Power and McCarty 1998). Fuels management projects on several California national forests emphasize the importance of stakeholder collaboration and modeling tools to help facilitate communications (see Strategic Fuel Treatment).

The goal for improving the wildland fire and fuels problem is to reduce the long-term risk wildfire poses to human and ecological communities. Managers need to weigh the short-term risks posed by active management against the long-term risks posed by continued inaction, and to communicate these risks in a meaningful way to the public (Bosworth 2003). Managers cannot change weather or topography, but fuels can be modified to change the burning and value-loss characteristics at specific locations as well as across large landscapes. This not only reduces the negative impacts on those forests but the wildfire itself may also provide benefits (Finney 2005). Benefits include environmental risks prevented by management actions (Davies 1996). For example, before enjoining ?an agency action under an authorized hazardous fuel reduction project, the court reviewing the project shall balance the impact to the ecosystem likely affected by the project of (1) the short- and long-term effects of undertaking the agency action; against (2) the short- and long-term effects of not undertaking the agency action? (HFRA 2003, title 1, section 106).

Fire hazard in a given area is partly a function of the combustible materials located on site. Thinning and prescribed burning are the primary fuel management activities and repeatedly have shown to reduce fire intensities and increase survival of some forest types (Finney 2005). Considering anticipated changes in Interior West forests, including climate, Covington and others (1994) concluded that the undesirable consequences of inaction far exceed those of action. Without active management of fuels, many forests will continue to be subject to uncharacteristically severe fires, and the costs of firefighting will continue to increase (Stephens and Ruth 2005). Active management can mitigate wildfire risks to watersheds in some situations, but, in others, forest management may not be effective (Bisson and others 2003, Schoennagel and others 2004, Westerling and others 2006).

Benefits from prefire management are most likely to come from prioritizing treatment areas (Dunham and others 2003). Priorities can be based on ecological value, evolutionary significance, and the risk of loss (Bisson and others 2003). The scale of the problem, however, is enormous. High priority treatment areas cover 397 million acres of forests and grasslands across all ownerships, public and private, an area three times the size of France. Some 73 million acres of forests in the low- and mixed-severity regimes are far denser than they ought to be, increasing their vulnerability to stand-replacing fires. These have been identified as high-priority treatment areas (USDA-FS 2006a).

Successful projects for reducing fire hazard depend on taking many factors into account and developing protocols for deciding which stands should be thinned and how much, with each situation evaluated on its own merit and operations planned carefully to ensure that the cure is not worse than the disease (NRC 2000). In risk management, avoiding actions in which the cure is worse than the disease means avoiding extreme events, i.e., the worst and the most disastrous situations (Haimes 2004).

Problem fires are today?s parlance for extreme fire events (NIFC 2006b). Of all ignitions, 2-3 percent escape initial attack and become the problem fires that damage resources, threaten communities, and cost millions of dollars in suppression efforts. Whereas not all wildland fires grow to such proportions, problem fires are those events that are large, destructive, dangerous, and costly to manage. Problem fires are the symptoms of a larger forest health issue, where ecological realities conflict with social expectations and economic limitations (NIFC 2006b). Spatial fire behavior models used in collaborative settings (see Strategic Fuel Treatment) offer some promise in dealing with social issues at various spatial and temporal scales. Especially on National Forest System lands (see National Forest System Issues), institutional improvements may be necessary to help put such technologies in place.