Polarimetric signatures and classification of tropical land covers

Polarimetric signatures for different tropical land covers were extracted from RADARSAT-2 data. Subsequently, the data were classified. The objective of this work was to assess the potential of RADARSAT-2 polarimetric C band data on land cover mapping. RADARSAT-2 data were acquired over Tapajos National Forest, a tropical forest reserve in Brazil, and surroundings, in September 2008. A field campaign was conducted during the same week of the SAR data recording. Polarimetric signatures for the different land covers were extracted for co- and cross-polarised bands and results indicated the variety of scattering mechanisms in the study area. Following that, the coherence and covariance matrices were used for the Freeman-Durden target decomposition, which decomposed the image targets in new bands representing the main scattering mechanism in the resolution cells - corner reflection, volumetric and superficial. Data were later classified by a k-means-Wishart classifier. The bands representing volumetric and superficial scattering helped discriminating vegetated and non-vegetated areas. Classification accuracy reached around 80% for forest and pasture/bare soil classes. For the remaining classes, the classification accuracy results did not reach 50%.