Legend Development Process for Forest Structural Mapping of Wide Areas using Alos PalSAR Data
Quinones, Marcela1; Hoekman, Dirk2
1SarVision, NETHERLANDS; 2Wageningen University, NETHERLANDS
For the whole of Borneo and Suriname, Alos PalSAR strip data was processed for two consecutive years. All data was calibrated, orthorectified and mosaiqued in order to minimize differences in calibration due to incidence angle, rain events and moisture conditions. Unsupervised and supervised classification techniques were used for the classification of the radar images. Intensive Statistical sampling of the images was done by visual interpretation in combination to the available ecological maps, and field data. In total 480 polygons were delineated over the composite of Suriname and 550 polygon over the composite of Borneo and use for the classification. The resultant vegetation structural type (VST) maps for both areas were analyses in terms of vegetation structure and soil conditions. From 20 to 18 different classes could be extracted consistently for both study sites. Spectral analysis of classes is presented and explained in the frame of the necessary post processing to achieve higher accuracies. The type of information extracted from the radar data and information from the field give the basis for the development of an structural legend defined by the LCCS land cover classification system. Statistical differences between classes are shown and explained in the frame of a consistent legend that could be applied for different tropical regions in the world. Validation of the final VST map is done using field geo-referenced photographs taken for the study period (2007-2008) in both study sites and high resolution images in Google earth. Final accuracy for the vegetation structural map reaches up to 92%.when structural are limited to basic classes. Maps for these two different Eco-regions are presented and discussed.