Assessment of multi-temporal RADARSAT-2 polarimetric SAR data for crop classification in an urban/rural fringe area

This paper investigated the potential of multi-temporal polarimetric RADARSAT-2 data for crop classification in an urban/rural fringe area. Using five scenes of RADARSAT-2 fine beam Quadpol data acquired during the 2012 growing season, five main crop types (wheat, soybeans, corn, field peas, and forage) in Southwestern Ontario, Canada have been identified. The potential of the RADARSAT-2 data for crop classification was assessed on four aspects: (1) the selection of classifier, (2) the effectiveness of polarimetric parameters, (3) the combination of multi-temporal data, and (4) post-classification processing methods. Pauli decomposition parameters proved to be effective in crop classification using Gaussian based Maximum Likelihood Classifier. With five dates of the images, the five crop types and other four non-crop types were classified at an overall accuracy of 91%. Satisfactory results with an overall accuracy of 87.8% were achieved by using only three dates of data given that the images covering the critical crop growth stages were included. Results demonstrate that polarimetric RADARSAT-2 data are suitable for accurate crop mapping in urban/rural fringe areas.