Assessment of dependence between SAR data focusing parameters and tundra habitat classification

This study addresses the influence of synthetic aperture radar (SAR) data-focusing parameters on the efficiency of tundra habitat classification. IBM PC/AT-compatible software was developed for synthesizing images from the raw SAR data and the complex SAR images, The focusing algorithm, based on matched filtering allows to control spatial resolution by means of multi-look technique and provides a set of window functions (Rectangle, Hamming, Kaiser-Bessel and etc.) for flexible adjustment of the synthesized gain pattern. The set of ALMAZ and ERS-1 SAR images was generated from a single satellite overpass for a 21 km² study site (flat, wet tundra) at Prudhoe Bay, Alaska. Special spectral classification method was applied to each image and the results were compared with a ground-truth map. The images were classified to best approximate two ground classification schemes: “C” (based on vegetation) and “D” (based on surface form). Results indicated that the number of processing “looks” had a greater influence than windowing function with respect to overall classification agreement with ground classes when texture was not taken into account. Four-look processing was superior to one-look processing for discriminating general landcover types. The choice of windowing function appeared to have greater importance in the detection of specific ground features. The results of classification suggest that an optimal combination of SAR data-focusing parameters may exist-for every ground-derived classification scheme