Improved polarimetric SAR classification by application of terrain azimuth slope corrections

In polarimetric SAR imagery, along-track or azimuth slopes cause a shift in the polarization orientation angles. By estimating the orientation angle shift, Schuler et al. have been able to generate topographic models in good agreement with DEMs produced by other means. An improved method for estimating the orientation angle was developed by Lee et al. along with a unified analysis of available techniques. Another way to view this work is as a means of correcting the polarimetric SAR imagery for the terrain-induced slope effects in order to improve the accuracy of geophysical parameters derived from the SAR imagery. In this paper the techniques of Lee et al. are used to correct C and L-band polarimetric AIRSAR imagery taken in 1996 over the flanks of Mt Tongariro, New Zealand. We then use the corrected and uncorrected data in classification schemes and compare the results to demonstrate the effects of the azimuth slope corrections. The volcanic Mt Tongariro provides a rich variety of surface slopes, and a variety of vegetation is present including exotic forest, native forest, scrub, grasses and bare ground. Class statistics extracted from fore- and back-slopes (in the azimuth direction) are observed to cluster more tightly than for uncorrected data. This effect is more pronounced for L-band than for C-band. The largest effects were seen in targets of low polarisation entropy such as exotic forest. These corrections result in a modest improvement in the overall classification accuracies.