Evaluating Aster Satellite Imagery And Gradient Modeling For Mapping And Characterizing Wildland Fire Fuels

Land managers need cost-effective methods for mapping and characterizing fire fuels quickly and accurately. The advent of sensors with increased spatial resolution may improve the accuracy and reduce the cost of fuels mapping. The objective of this research is to evaluate the accuracy and utility of imagery from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) satellite and gradient modeling for mapping fuel layers for fire behavior modeling within FARSITE. An empirical model, based upon field data and spectral information from an ASTER image, was employed to test the efficacy of ASTER for mapping and characterizing canopy closure and crown bulk density. Surface fuel models (NFFL 1-13) were mapped using a classification tree based upon three gradient layers; potential vegetation type, cover type, and structural stage. INTRODUCTION AND BACKGROUND Wildland fire is an important issue facing local and regional land managers in the United States. Fires occurring in many parts of the western United States today are far more severe than fires that occurred before the suppression era (Arno and Brown, 1989; Hessburg et al. 2000). Increased fire size and severity coupled with an increase in the number of people living in the wildland-urban interface has resulted in millions of dollars of damage to property and loss of life throughout the western United States in recent years. In 2002, federal agencies spent an estimated $1.6 billion on fire suppression (National Interagency Fire Center, 2003). As human populations move closer to the edges of wildlands, their lives and property become increasingly threatened by wildfire. In order to reduce fire risk to people and their homes, land managers must prioritize areas for fire mitigation and hazardous fuels reduction. In 2000, the US Department of Agriculture teamed with the Department of Interior and the National Association of State Foresters to develop the National Fire Plan (www.fireplan.gov). Along with post-fire rehabilitation and maintaining firefighting preparedness, the goals of the National Fire Plan include reducing fuels in at-risk areas, particularly in and around the wildland urban interface (Bisson et al., 2003). Each year, the National Fire Plan provides funds to local fire districts to increase fire suppression capabilities and implement fuels reduction projects (USDA, 2000). In order to utilize monies from the National Fire Plan efficiently, land managers need cost-effective methods for mapping and characterizing fire fuels quickly and accurately. Some of the most potentially useful approaches for accomplishing this involve the integration of remote sensing (RS), Geographic Information System’s (GIS), field data and gradient modeling. Such analyses could provide consistent maps of fire fuel conditions across a diversity of land ownerships. ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession Fuel Mapping One of the most important factors influencing fire hazard and fire risk is the type, composition, and distribution of fuels (Chuvieco and Congalton, 1989). Wildland fuels are typically divided into three strata: ground fuels, surface fuels, and crown fuels (Pyne et al., 1996). Ground fuels consist of roots, duff, and buried woody debris. Fires burning in this stratum usually exhibit slow rates of spread. Surface fuels are composed of leaf litter, coarse woody debris, seedlings, saplings, and herbaceous vegetation. Most wildland fires start in, and are carried by, the surface fuel strata. Overstory trees and shrubs comprise the crown fuel strata. Fires burning in the crown fuel strata are often extremely intense and nearly impossible to control (Pyne et al., 1996). Since fuel stratum relationships are extremely complex, fire managers often describe fuels by grouping vegetation communities, based upon similar potential fire behavior, into fuel types (Riano et al., 2002) or fuel models (Anderson, 1982). However, since the distribution and accumulation of fuels is highly variable (Brown, 1979) and, in forested areas, highly dependent upon vegetation type as well as stand history (Keane et al., 2001; Brandis and Jacobson, 2003) fuel quantity and distribution are not directly related to fuel types (Pyne et al., 1996). Field Mapping of Fuels. Prior to the development of remote sensing technologies, fuels were typically mapped through extensive field inventory (Miller et al., 2003). Although these technologies were successful, the development of remote sensing technologies could potentially reduce the cost and time required to map fuels on the ground (Keane et al., 2001). Remote sensing technology also has the potential to update fuel maps quickly in areas where conditions are dynamic due to logging, fire, or other changes. Remote Sensing of Crown Fuels. Traditionally, interpretation of aerial photography coupled with field data was the primary method used to map fire-related canopy variables (Riano et al., 2003) such as crown bulk density, crown closure, and canopy height. More recently empirical methods, which are less labor intensive, have been used to estimate these variables from Landsat TM and SPOT (Systeme Probatoire D'Observation de la Terre) HRV (high resolution visible) data (Riano et al., 2003). Franklin et al. (2003) mapped various stand attribute classes, including canopy height and crown closure, through the classification of spectral and textural information derived from Landsat 5 data. Miller et al. (2003) successfully mapped structural stage classes in Arizona by running Landsat TM data through a clustering algorithm. Remote Sensing of Surface Fuels. The inability of optical sensors, such as Landsat TM and MSS, to penetrate the forest canopy (Miller et al., 2003) limits their utility for mapping surface fuels (Keane et al., 2002). As a result, most studies using remote sensing to characterize surface fuels first classify an image into vegetation categories and assign fuel types or fuel models to each category (Keane et al., 2001). Chuvieco and Salas (1996) characterized fuel types through the classification of Landsat Thematic Mapper (Landsat TM) data. Chuviceo and Congalton (1989) and Castro and Chuvieco (1998) used similar methods to map fuel types in Spain and Chile, respectively. Wilson et al. (1994) applied maximum likelihood decision rules to a Landsat Multi-Spectral Scanner (Landsat MSS) image to directly classify fuel types across Wood Buffalo National Park, Canada. Riano et al. (2002) improved a fuel type classification by incorporating two seasonal Landsat TM images, to account for phenological differences in vegetation, into a classification algorithm. Hyperspectral remote sensing has also been used to map fuel types and vegetation moisture content for a chaparral community in Southern California (Roberts et al., 1998). Gradient Modeling of Fuels. Gradient modeling refers to the use of environmental gradients (topographical, biogeochemical, biophysical, and vegetational) to model the occurrence of natural phenomena (Keane et al., 2002). This approach has been used with moderate success in estimating fuel types and fuel loading. Environmental gradients such as topography, moisture, and time since last burn have a large impact on fuel loading (Kessell, 1979). High fuel loading, for example, can be partially explained by lower decomposition rates (characterized by moisture and temperature gradients) and a long time interval since the last fire (Keane et al., 2001). Gradient modeling has been used to model fuel characteristics in Glacier National Park, Montana (Kessel, 1979). Integrated Fuels Mapping. The integration of remote sensing and gradient modeling may also increase the accuracy of fuels mapping projects. For example, Keane et al. (2002) integrated remote sensing and gradient modeling to map fuels across the Gila National Forest in New Mexico. This approach, termed the ‘vegetation triplet’, incorporates three layers: potential vegetation type (PVT), cover type (CT), and structural stage (SS). PVT is a site classification based upon the climax vegetation that would be found on a site in the absence of disturbance (Keane et al., 2002; Smith et al., 2003). CT describes the dominant species found on a site, and SS refers to the current canopy structure of a site. PVT is directly related to the biophysical setting of a site, which ultimately determines the site’s productivity and decomposition rates, and therefore has a large impact on fuel characteristics (Keane et al., 2002). CT is important for fuels mapping because dead woody debris and litter are directly related to the dominant tree species found on the site (Keane et al., 2002). The potential of a surface fire spreading to the crown is highly dependent upon the vertical structure of the stand, which is described by SS. The triplet approach has been used to assess the hazard of forest disease outbreak, and vulnerability to fire in the Columbia basin ASPRS Annual Conference Proceedings May 2004 * Denver, Colorado ASPRS – 70 years of service to the profession (Hessburg et al., 2000); it has been used in the Gila National Forest and the Selway-Bitteroot Wilderness, to map fuels and input layers required to run FARSITE (Keane et al., 2001; Keane et al., 2002). Future of Fuels Mapping. Remote sensing based fuels mapping has typically employed one of the Landsat sensors (MSS, TM, or ETM+) to map fuels characteristics (Riano et al., 2003). Although these sensors are effective, and are widely applicable to many environmental mapping and monitoring situations, the advent of new sensors with improved spatial and spectral resolutions may improve the accuracy (Chuviceo and Congalton, 1989) and reduce the cost (Zhu and Blumberg, 2001) of forest fire fuel mapping. ASTER, a sensor aboard NASA’s Terra platform (see specifications (Table 1), has untested potential for characterizing and mapping forest fire fuels. The visible and near-infrared telescope (VNIR), which collects data with a spatial resolution of 15 m in the green (0.52 – 0.60 μm), red (0.63-0.69 μm), and near infrared (0.76 0.86 μm) portions of the electromagnetic spectrum, should be particularly useful for obtaining information about vegetation (Rowan and Mars, 2003), and may prove successful in mapping fuel characteristics. Table 1. ASTER specifications (adapted from Abrams, 2003) Spectral Region Spatial Resolution (m) Channel Bandwidth (μm) VNIR telescope 15 1 0.52 0.60 15 2 0.63 0.69 15 3 0.76 0.86 SWIR telescope 30 4 1.60 1.70 30 5 2.145 – 2.185 30 6 2.185 – 2.225 30 7 2.235 – 2.285 30 8 2.295 – 2.369