Spatial models for estimating fuel loads in the Black Hills, South Dakota, USA

Abstract

Fire suppression has increased fuel loadings and fuel continuity in many forested ecosystems, resulting in forest structures that are vulnerable to catastrophic fire. This paper describes the statistical properties of models developed to describe the spatial variability in forest fuels on the Black Hills National Forest, South Dakota. Forest fuel loadings (tonnes/ha) are modeled to a 30 m resolution using a combination of trend surface models to describe the coarse-scale variability in forest fuel, and binary regression trees to describe the fine-scale variability associated with site-specific variability in forest fuels. Independent variables used in the models included various Landsat TM bands, forest class, elevation, slope, and aspect. The models accounted for 55% to 72% of the variability in forest fuels. In spite of having highly skewed distributions, cross-validation showed the models to have nominal prediction bias. This paper also evaluates the feasibility of using the estimation error variance to explain estimation uncertainty. The models are allowing us to study the influence of small-scale disturbances on forest fuel loadings and diversity of resident and migratory birds on the Black Hills National Forest.