Assessing Fire Risk Using a GIS-Based Approach
By Ruth Carapella

Since 1986 nearly a half million acres of land on the Boise National Forest (BNF) of Idaho have burned. These fires have scorched almost 20 percent of the land managed by the Forest Service and consumed about 50 percent of the Boise's Ponderosa pine forests. Many of these wildfires burned with uncharacteristic intensity, killing trees, critically damaging soils, and drastically altering habitat for terrestrial and aquatic plants and animals. Over the last decade, fire suppression and emergency watershed rehabilitation costs have exceeded one hundred million dollars.
      Managers on the BNF needed a way to assess the risk of fire hazard on national forest lands. They turned to fire experts and to the BNF's Geographic Information System (GIS) to develop a model that would help assess fire hazard risk.

Changes in Fire Intensity
Travelers and settlers in southwestern Idaho in the late 1800s and early 1900s described open, almost park-like hillsides dominated by Ponderosa pine trees. In these pine forests, grass and low herbaceous plants covered the ground between widely spaced trees. Forest densities varied from just a few trees per acre up to 40 to 50 trees per acre.
      Frequent, low-intensity fires maintained these landscapes by killing small trees and consuming forest debris without harming mid- to large-sized Ponderosa pine. Lightning strikes started most fires; some were set by Native Americans.
      Over time, Idaho became more developed and the forests changed. Fire suppression and livestock grazing helped to decrease fire frequency. Early timber harvest removed the fire-adapted and economically valuable Ponderosa pine. With fewer fires, the trees in Ponderosa pine forests became crowded. Today, dense and moderately dense forests with several hundred trees per acre of Ponderosa pine, Douglas fir, and Grand fir typify many areas where open pine forests once grew.
      Fires in these dense areas burn uncharacteristically hot. Since 1986, wildfires on the Boise National Forest have become larger and more severe than those of the past. In 1986, a large fire burned on the Emmett District. In 1989, 68,000 acres burned on the Boise's Lowman District. In 1992, on the southern end of the BNF, the Foothills Fire burned 250,000 acres.
      Managers knew they needed a way to identify areas most at risk for catastrophic fire. Cathy Barbouletos, who was temporarily serving as forest supervisor, directed an interdisciplinary (ID) team to identify the forest ecosystems and resources most at risk to severe wildfires. The team focused on 1.1 million acres of pine forest types located within the perimeter of the BNF where frequent, low-intensity fires were once the norm and where the forest conditions differed most from historic conditions. The team was led by Cydney Weiland, land management planner. The team worked with Dr. Leon Neuenschwander, fire ecologist at the University of Idaho, to develop the risk assessment.
      Halfway through the process of developing a GIS-based hazard/risk assessment, the Boise River Fire began. Weiland said, "Things got put on hold from July through December while everyone worked on fire suppression and rehabilitation." Weiland added that the 180,000-acre Boise River Fire underscored the value of the project and served as a cross check to the model.

Existing Data
To keep costs down and to speed the process, the team built the five submodels that comprise the hazard/risk assessment model from existing data layers in the BNF GIS. This introduced the risk of inaccuracy where data changed between input and use in 1994. But it offered an unexpected benefit: because these data layers were in place before the Boise River Fire, the team could use this fire as a partial validation of the model.
      The BNF GIS contains a vegetation cover type layer derived from a June 20, 1992 Landsat Thematic Mapper (TM) image. Two of the hazard/risk assessment submodels rely on this cover type layer. Additionally, the submodels use digital elevation model (DEM) data, point data from the forest's fire occurrence database, a habitat type model, the forest's aquatic survey database, and a landtype layer depicting units of land with similar soil characteristics.
      Joe Frost, GIS analyst for the BNF and GIS manager for the project said, "The team relied upon the best data available and made it as reliable as possible. We verified as much as we could."
      The BNF subscribed to this philosophy during the development of their GIS data layers. For example, the BNF strived to ensure the accuracy of the vegetative cover type layer. In 1993, Bill Kramber, senior remote sensing analyst for the Idaho Department of Water Resources (the designated lead agency for remote sensing work in Idaho) began work with the BNF on vegetation cover classification. Kramber selected June 20, 1992 Landsat TM data since this was the most recent summer scene with less than 10 percent cloud cover and the highest sun angle.
      Kramber used the tasseled cap algorithm to separate brightness and greenness from minor components of the Landsat TM data. He classified the first three tasseled cap components using a hybrid supervised/unsupervised approach. DEM data was added to improve the spectral classification scheme. In 1993 foresters compared the classification to on-the-ground conditions from a helicopter. In 1994 foresters compared the generated cover types to aerial photography. After both comparisons, the classification was adjusted to better reflect conditions on the ground.

Five Submodels
The Hazard/Risk model links five GIS submodels. Submodels assess vegetation risk, ignition risk, wildlife habitat persistence, watershed hazard, and fisheries condition. Submodel output provides managers specific information about the effects of fire on each of these resources.
      Joined, the submodels identify Ponderosa pine ecosystems where uncharacteristically dense vegetation creates a high fire risk and where catastrophic fire could cause severe damage to other resources. In other words, the model focuses on areas where conditions fall outside the range of historic variability. These areas have the highest risk for severe, large wildfires. The model highlights the effects of fire exclusion in Ponderosa pine-dominated ecosystems over the entire BNF.
      The ID team developed each submodel using a combination of reasoned judgment and computer analysis. Using professional knowledge and experience, managers identified critical conditions and likely sequences of events. Then, as GIS manager, Frost queried the appropriate data layers for each submodel using the ARC/INFO Version 7 Grid module to find where these conditions occurred.
      Where data resolution differed between layers, as they did between the fire occurrence data base and other layers, Frost assigned a composite number for each section of land. Then he took this number down to the 30-meter cell resolution, giving each raster in the section the same attribute. This effectively brought all data to the same scale for the analysis.
      Working with Frost, managers developed correlations between query results and hazard ratings. Then Frost developed ARC Macro Language or AML's that enabled the model to assign hazard ratings to each of the Forest's 378 subwatersheds. Each subwatershed in each submodel was assigned a score of 1 to 5, with 5 always having the highest hazard.
      Next, Frost had the model generate a composite rating for each subwatershed by using the combine function in the ARC/INFO Grid module and AML routines that operated on the associated INFO data base. If the subwatershed received moderate or higher hazard ratings from all five submodels it received a high risk rating.
      Finally, the model assigned overall ratings to each of the 82 watersheds on the BNF. Composed of subwatersheds, watersheds encompass up to 30,000 acres. If at least one subwatershed within the watershed received an overall high risk rating, the model assigned a high risk rating to the entire watershed. The team imposed this condition because observations over the last decade show that fires burning under uncharacteristically severe conditions are likely to burn across subwatersheds. The resulting overall hazard rating for each watershed helps identify its relative risk for severe fire. Team leader Weiland said, "We can't say that a high-risk watershed will surely burn, and burn with uncharacteristic intensity. But we can say that it's at highest risk, compared to other watersheds on the BNF."

Validation
Because resource specialists developed each submodel based upon their best scientific and professional judgment, the BNF sent the model to other National Forests and to other resource specialists for review. The team used reviewer comments to revise and simplify the fisheries condition submodel. The team reran the adjusted model in January 1996.
      Perhaps the toughest test occurred with the 1994 Boise River Fire. Frost said that as the team worked, they noticed a direct correlation between the preliminary model results and the area burned by the fire. The Boise River Fire started outside a high risk watershed. Fire crews had the fire almost contained, but unfavorable weather conditions pushed the fire into a watershed the team had rated as high risk. The fire quickly grew to 180,000 acres. The fire perimeter coincides closely with areas identified as having high fire risk in the model.

What Next?
Had she to do the process over, Weiland would make few changes. She emphasized the importance of spending time up-front to identify the specific questions the analysis needs to address.
      By doing this, Weiland said, "We fought the impulse to include other stuff that was not pertinent to a fire-based assessment. If we couldn't say how information helped answer our pivotal questions-where are the forest ecosystems most at risk to uncharacteristic fire and what resources are at risk-we didn't include it."
      Weiland also emphasized the value of involving Dr. Neuenschwander from the University of Idaho. She said, "Having Leon as our coach and mentor was a great decision. He brought so much expertise in model development and kept us focused on the questions we needed to answer. "
      Frost concurred, saying, "We knew what we wanted to do, but we didn't know how to get it done. Dr. Neuenschwander came up with the design."
      The model tells managers where to take a closer look on the ground. After field verification, if land managers agree with the model's assessment of fire risk, it helps them identify the resources within those areas that are most at risk to fire damage. In this way, the model serves as a bridge between broad forest-level planning and site-specific project planning.
      Weiland cautioned, "It's important to remember that this assessment provides information at a large scale. Data and conclusions from our assessment can't be used at the site-specific level without verification from folks that know the ground best."
      After field verification, managers will start on the process of prescribing treatments that may help to lower the risk of catastrophic fire. The hazard/risk assessment model may help managers adjust forest management practices in a way that benefits all resources.

About the Author:
Ruth Carapella writes about forestry and natural resources for Pen Craft Writing & Editing Services in Harrison, Idaho. She may be reached at 800-689-9235 by phone or fax.

Back