Application of decision-tree techniques to forest group and basal area mapping using satellite imagery and forest inventory data

Accurate, current, and cost-effective fire fuel data are required by management and fire science communities for use in reducing wildland fire hazards over large areas. In this paper we present results of applying decision-tree techniques to mapping vegetation parameters (such as vegetation types and canopy structure classification) required for fire fuel characterization. Specifically, we present preliminary results of mapping forest types and average basal area by different forest types at 30-meter resolution. Input data into the decision tree model included Landsat-7 ETM+ spring, summer and fall greenness, brightness and wetness of the tasseled cap transformation, topographic data layers such as slope and elevation, and forest variables measured on inventory plots in the Mid-Atlantic region. Using decision-tree models, eight forest types were successfully identified in training cases and mapped for the entire mapping area. Forest basal area per unit area (conifer and deciduous) was estimated as well using regression tree models. Cross-validation conducted for both forest types and basal area showed that discrete forest type estimation error was 35% and continuous basal area relative errors were between 58 and 72%. Accuracy was higher in homogeneous forested lands and lower in areas with fragmented forest cover. The study demonstrated that decision tree and regression tree methods are efficient for large-area vegetation mapping if sufficient large-amount of reference data are available.