Integrated use of multi-mode and multi-angle SAR data for land cover identification in tropics

Abstract

C- and L-band SAR backscatter data have predominantly been used for land-cover identification. Adding coherence data improved the accuracy. However, coherence data are not readily available due to technical constraints but the integrated use of multi-band backscatter data from different satellites is readily feasible. According to previous studies, the application of multi-polarimetry and multi-squint-angle data on L-band may result in identical accuracies as the combination of backscatter and coherence data on C-band. Further improvements may be achieved by adding coherence data to the combination of multi-polarimetry and multi-squint-angle data and it is expected that a combination of all data may yield to the same order of accuracy as with optical sensors. In our study we used ALOS PALSAR L-band data in different polarizations and different incident angles to analyze their information content for land cover monitoring in the humid tropics. Using these datasets we wanted to examine to what extent backscatter data with multi-polarimetry or multi-squint angles can improve the accuracy in land-cover identification in the tropics. Our study area is situated in Central-Sumatra because it shows a wide variety of tropical land-cover types. However, this region is also famous for its fast land cover changes due to human impact. Thus, the capabilities of the different sensor-mode combinations could be scrutinized. Applying Maximum-Likelihood, all data combinations were classified using the same training areas derived by high resolution SPOT 4 reference imagery. The accuracies of all classification results were analyzed to evaluate the best performing data combination.