Classification and RadarSAT and Landsat TM Imagery in the Savanna Region of Colombia

摘要

The value of RadarSAT data and textural enhancements in combination with Landsat TM data are evaluated for land use and land cover classification in the Region Andina. The major objective of this study is to develop a remote sensing method using RadarSAT texture derivatives and Landsat TM imagery for classifying forest and other land covers. The method used for classification is a per pixel discriminate analysis, using half the training pixels to develop the model and half to test the model. Second-order texture provide the greatest discrimination of classes, with 52.2% classification accuracy. Using the spectral bands of Landsat TM data alone, a 64.3% accuracy is calculated. The combination of the Landsat TM spectral bands and the second-order RadarSAT texture bands resulted in 87.8% classification accuracy. For the combined data set, the lowest accuracy is 61.5% for the urbano montano class and 100% accuracy for the urbano alto. Using either the Landsat TM or the RadarSAT texture alone produces similar classification results, but with different distributions of confused classes. Combining the RadarSAT texture improves the discrimination between land cover classes on average by 23% over the Landsat TM data alone.