USDA Forest Service
 

Mapped Atmosphere-Plant-Soil System Study

 
 
Pacific Northwest Research Station Homepage
   
USFS Research & Development
   
Evaluate Our Service
Your comments and suggestions are very important to our service improvement.

Pacific Northwest Research Station
Mapped Atmosphere-Plant-Soil System Study


Corvallis Forestry Sciences Laboratory
3200 SW Jefferson Way
Corvallis, Oregon 97331

United States Forest Service.

MAPSS Home > Fire Risk Forecasts > Forecast Methods


MAPSS Team Experimental Seasonal Fire Risk Forecasts

Forecast Methods

Midterm Forecasts of Fire Risk for National Strategic Planning MAPSS Team, USDA Forest Service, Pacific Northwest Research Station, Corvallis, OR


Introduction

Wildland fire management agencies are increasingly interested in more long-term forecasts of fire business. Several short-term fire potential assessments are currently available ranging from WFAS next-day fire danger ratings to 28-day wildland fire potential assessments from the National Interagency Coordination Center (NICC). Missing from this collection are midterm fire potential forecasts for strategic planning of large prescribed fire events and prioritization of suppression resources.


The MAPSS ecosystem modeling team is uniquely prepared to develop midterm forecasts of fire risk. As long-time contributors to many different global climate change assessments, the team has much experience in developing spatially distributed, high-resolution climate data and potential future climate forecasts from a variety of state-of-the-art climate models. The team is also experienced in modeling fire behavior and effects under various long-term future climate scenarios as part of dynamic ecosystem simulations for global change studies.


Historical Weather Observations

The development of the fire risk forecasts requires not only weather forecasts, but also a record of observed weather for simulating fuel loading and the moisture status of longer time-lag fuel classes. Existing regional and national-scale estimates of fuel moisture and fire danger often suffer from underlying climate data based on the observations of too few weather stations and the use of relatively unsophisticated spatial interpolation techniques. The MAPSS team together with long-time collaborator Chris Daly of the Spatial Climate Analysis Service is using Daly's PRISM model to produce high-resolution data grids of observed fire weather. The PRISM model produces interpolations of weather station data that are sensitive to topography, which is especially important in the complex, fire-prone terrain of the mountainous West. Input station data are gathered primarily from the National Weather Service (NWS) Cooperative Observer Program (COOP) and U.S. Department of Agriculture-Natural Resources Conservation Service (USDA-NRCS) SNOTEL networks. For mapped examples of the PRISM-generated historical weather data grids see the Spatial Climate Analysis Service's Web site (www.ocs.orst.edu/prism).


Weather Forecasts

The spatial resolution of most weather forecast models is relatively coarse and does not adequately represent the topographic complexity of mountainous terrain and its influence on weather. This, in part, is why weather forecast models are often unable to simulate past observed weather with an acceptable degree of accuracy. The MAPSS team follows the procedure illustrated in figure 1 to downscale and increase the accuracy of weather model output. The long-term average for a weather model variable is calculated over a length of simulation time that is model-dependent. The long-term average for an observed variable is calculated from PRISM-generated climate data extending back to 1895.


The MAPSS team's experience with global change assessments has shown that forecasts of climate-driven processes are more credible when model simulations are conducted for more than just a single climate scenario. We are producing an ensemble of fire risk simulations based on different weather forecasts to better capture the range of potential fire risk over the midterm. We are currently processing 7-month forecasts generated by four climate models. Monthly updates of the different climate model forecasts are provided by the International Research Institute for Climate Prediction. Descriptions of the different weather models and maps of their monthly weather forecasts can be found at the Climate, Ecosystem and Fire Applications (CEFA) Web site (www.cefa.dri.edu/Assessment_Products/IRI/iri_index.htm).


The MC1 Fire Module

Fire risk forecasts are generated from the historical and forecast weather data by using MC1, a dynamic global vegetation model developed by the MAPSS team. MC1 consists of interacting modules that simulate biogeography, biogeochemistry, and fire disturbance. Descriptions of the biogeography and biogeochemistry modules can be found in Bachelet et al. 2000 and Lenihan et al. 2003 (Publications listed by author.)


The MC1 fire disturbance module simulates fire occurrence, area burned, and fire behavior and effects. Fire occurrence is simulated as discrete events over time given a constraint of no more than one event per year in each cell. The values of three different factors at each daily time-step of the fire module are compared against threshold values used to trigger a fire event. A threshold of the 12-month standardized precipitation index (SPI) is used as an indicator of moderate to severe drought to control the interannual timing of fire events. A threshold of the 1,000-hr 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, so as to control the timing of fire events at the daily time step. There is no constraint on fire occurrence owing to availability of an ignition source, such as lightning or human-caused ignition. The thresholds were calibrated to limit the occurrence of simulated fires to only the most extreme events. Large and severe fires account for a very large fraction of the annual area burned historically. Large events are also likely to be least constrained by heterogeneities in topography and fuel moisture and loading that are poorly represented by relatively coarse-scale input data grids.


Once a fire event is triggered, the MC1 fire module determines the area burned as a nonspatial fraction of the cell (i.e., the location of the area burned within the cell is not specified). Fire area is not an explicit function of fire behavior, but rather a dynamic function of the current vegetation type, the current drought condition, and the number of years since fire. Every vegetation type simulated by the MC1 biogeography module is associated with a minimum and maximum fire return interval. A return interval between the minimum and maximum values for the current vegetation type is selected as a function of the current drought condition as indicated by the standardized precipitation index (SPI). The selected return interval gets closer to the minimum value as the drought condition gets more extreme. The reciprocal of the selected fire-return interval is used as the estimate the current annual fraction of the cell burned, which is multiplied by the years since fire to the determine the fraction of the cell burned by a given fire event. For example, if the current vegetation type simulated for a cell is Mediterranean Shrubland, then the set minimum and maximum return intervals return are 20 and 43 years, respectively. If the current value of SPI is -1.5, the current return interval is 32 years (i.e., approximately halfway between the minimum and maximum values for the vegetation type), and the current annual fraction of the cell burned is 0.031 (i.e., the reciprocal of 32). If 20 years have elapsed since fire occurrence in the cell, then the simulated fraction of the cell burned is 0.63 (i.e., 0.031 multiplied by 20).


The module calculates potential fire behavior for a given event based on the current weather and estimates of the mass, vertical structure, and moisture content of several live and dead fuel size classes. The aboveground live and dead biomass simulated by the biogeochemistry module is proportionally allocated to the live and dead fuel classes as a function of the current vegetation type. Allometric functions keyed to the different life forms and vegetation types are used to simulate the depth of surface fuels and the vertical structure of the overstory in terms of crown height, length, and shape. The moisture content of each dead fuel size class is a function of antecedent weather conditions averaged over a period of days dependent on size class. The moisture content of each live fuel class is a function of the soil moisture content to a specific depth in the profile. The simulated fuel characteristics together with current weather and several fire behavior functions are used to estimate key aspects of fire behavior (i.e., fireline intensity, height of lethal crown scorch, and the transition from surface to passive crown fire) used for simulating fire effects.


The direct effects of fire simulated by the MC1 fire module are the mortality and consumption of aboveground carbon and nitrogen stocks. The mortality and consumption of overstory biomass are simulated as a function of fire behavior and the vertical structure of the canopy. The consumption of surface fuels is simulated by using functions of moisture content that are fuel-class specific. Gaseous and particulate fire emissions from flaming and smoldering fuel consumption are simulated by using emission factors that are a function of the current vegetation type. Fire-induced fluxes of carbon and nitrogen, simulated per unit area, are multiplied by the burned fraction of the cell to prorate fire effects to the entire cell.


Fire effects extend beyond the direct impact on carbon and nutrient pools to more indirect and complex effects on life-form competition. Fire tends to tip the competitive balance toward grasses in the model because much, or all, of the grass biomass consumed is reestablished the year following a fire event. Tree or shrub biomass is more gradually replaced. The more rapidly growing grasses 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.


Simulating Historical Fire Area

Prior to 1950, the total annual area burned simulated by MC1 is comparable to the observed long-term historical record for the coterminous United States (available as average annual totals by decade) as shown in figure 2. Increased support for fire suppression and the rapid mechanization of fire control after World War II contributed to the abrupt decline in the observed annual area burned after 1950.


There is a moderately strong and highly significant correlation (r = .56, p < 0.001) between the simulated and observed trends over the period of available annual observations (1960-2003) as shown in figure 3. Observed annual values are, on average, 12.5 percent of the simulated values during the last four decades of the post-1950 suppression era. To simulate the suppression effect, each annual cell fraction burned was multiplied by 0.125 after the year 1950 in a separate model run.


The MC1 fire simulations also show considerable accuracy in the spatial domain. The maps of the simulated fire occurrence for the years 2000-2003 are shown are figure 4. The accuracy of the maps is hard to assess with much rigor given the paucity of compiled, georeferenced observations documenting the distribution of recent fires in the United States. However, narratives describing past fire seasons and annual statistics compiled by geographic region (NICC) describe interannual, broad-scale shifts in the distribution of wildland fire similar to those simulated by the fire module. For example, the focus of observed and simulated fire activity in 2000 was in the southern tier of states early in the fire season, as is typical of a La Niña year. A weakening of La Niña later in the season brought hot and dry conditions to the northern Rocky Mountains where large fires were both observed and simulated. A different pattern was both observed and simulated in 2001, when the focus of fire activity was in the Pacific Northwest, especially in Washington, Oregon, and Nevada. The 2002 fire season was particularly severe in the four corners region of the Southwest, with very large fires in Colorado and Arizona, a pattern well simulated by the model. The 2003 fire season (both observed and simulated) was notable for being much milder than that of the 3 preceding years.


Fire Risk Forecast Maps

Each month four fire risk forecasts are generated, from the updated 7-month weather forecasts, by different weather models. Maps showing the distribution of the MC1-simulated fire occurrence and area burned for each weather forecast (e.g., figure 5) are posted on the MAPSS team's Web site. A consensus forecast map is also generated showing regions where MC1 predicts fire under one or more of the weather scenarios, and the relative size of the predicted fires. Warmer colors on a forecast consensus map indicate regions where agreement among the four individual fire risk forecasts is highest (e.g., figure 6).



Jim Lenihan
Fire and Ecosystem Modeler
USDA Forest Service
3200 SW Jefferson Way
Corvallis, OR 97331

Phone: (541) 750-7432
FAX: (541) 750-7329

jlenihan@fs.fed.us

 

Ronald P. Neilson
Bioclimatologist
USDA Forest Service
3200 SW Jefferson Way
Corvallis, OR 97331

Phone: (541) 750-7303
FAX: (541) 750-7329

rneilson@fs.fed.us

 

 

US Forest Service - Pacific Northwest Research Station, Mapped Atmosphere-Plant-Soil System Study
Last Modified: Monday, 16 December 2013 at 14:18:49 CST


USDA logo which links to the department's national site. Forest Service logo which links to the agency's national site.