Electromagnetic Land Surface Classification Through Integration of Optical and Radar Remote Sensing Data

摘要

We present a nonhierarchical electromagnetic (EM) land surface classification method through the integration of satellite multispectral high-resolution optical and polarized radar images of central Alberta near the Saskatchewan border. We implement a conventional supervised land surface classification method and a principal component analysis to a QuickBird image. The EM properties are then assigned to the classified surfaces to produce hierarchical EM land classification maps. To further classify a hierarchical EM surface (i.e., dielectric constant), we calculate the root-mean-square surface height with a Shuttle Radar Topography Mission 3-arc-second digital elevation model and the temperatures from a thermal band of a Landsat-5 Thematic Mapper image. We also calculate the backscattering coefficients from the Advanced Land Observing Satellite Phased Array L-band Synthetic Aperture Radar image. Using these estimated values, we calculate the intrinsic weighting factors with the Dubois (1995) model for less vegetated land areas and the Ulaby (1986) model for open water areas. By applying these weighting factors to the hierarchical EM surface, we generate a nonhierarchical higher resolution EM surface map of the study area.