Forest type classification with combination of advanced polarimetric decompositions and textures of L-band synthetic aperture radar data

Abstract

Information on distribution of forest types and land cover classes is essential for decision making and significant in climate regulation, biodiversity conservation, and societal issues. An approach for the combination of advanced polarimetric decompositions and textures of Advanced Land Observing Satellite Phased Array L-band Synthetic Aperture Radar full polarimetric data for the purpose of forest type classification is proposed. Using a support vector machine (SVM) classifier, we classified forest types over a selected Indian region. Further, we tested the classification performance of the Wishart method for the same forest types. The classified results were assessed with confusion matrix-based statistics. The results suggest that incorporation of various polarimetric decompositions features into gray-level co-occurrence matrix textures refines the SVM classification overall accuracy (OA) from 73.82% (k=0.69) to 76.34% (k=0.72). The Wishart supervised classification algorithm has the OA of 73.38% (kappa=0.68). We observed that integration of polarimetric information with textures can give complimentary information in forest type discrimination and produce high accuracy maps. Further, this approach overcomes the limitations of optical remote sensing data in continuous cloud coverage areas.