Regional maps of forest vegetation and fuels are increasingly needed for assessing fire risk, planning fuel management, and modeling fire behavior and effects. These maps must accurately and consistently portray multiple vegetation and fuels attributes across environmentally heterogeneous, multi-ownership landscapes. We applied the Gradient Nearest-Neighbor (GNN) method to develop predictive maps of forest vegetation and fuels in the Oregon Coastal Province, USA. GNN uses multivariate statistics to link ground data with satellite imagery and digital maps of environmental and socioeconomic variables. These spatial variables are integrated into a raster GIS database, and each raster cell is assigned the vegetation characteristics of the ground plot that most closely matches its spectral, environmental, and socioeconomic characteristics. Spatial predictor variables used to map the Oregon Coastal Province included raw Landsat TM imagery, a classified image of recent disturbances derived from satellite imagery, climate maps, topographic variables derived from digital elevation models, and land ownership. Vegetation and fuel variables were either measured directly or modeled allometrically using data from 800 field plots distributed across the region.

Predicted patterns of forest structure in the Oregon Coastal Province were predominantly a function of the Landsat imagery, whereas predicted gradients in species composition were strongly associated with climate. Prediction accuracy varied widely for different vegetation and fuel attributes. Structural variables derived from the characteristics of overstory trees (e.g. total basal area and stand height) had the highest accuracy whereas variables related to dead trees (e.g. 1000-hour fuels) had the lowest accuracy. Qualitative comparisons of predicted forest patterns with aerial photographs indicated that the spatial configuration of forest overstory characteristics was represented accurately. In the future, we plan to apply these methods to develop vegetation and fuels maps in eastern Washington and northern California.

Although predictive maps based on satellite imagery and GIS data hold promise for improving our understanding of the large-scale configuration of forest fuels, their accuracy is ultimately limited by the information content of the available spatial predictor variables. In assessing the utility of these maps for management and planning, managers and scientists will need to assess the relative importance of accurately predicting the total amount of fuels at a particular location versus characterizing the spatial configuration of fuels at a larger landscape scale.