Assessment of operational radar satellite for monitoring land cover change in Amazo/spl circ/nia

The objective of this paper is to verify the viability of using existing radar satellite (ERS and RADARSAT), as an operational tool for monitoring land cover in Amazo/spl circ/nia. It is well known that C band radar data are not adequate for land applications, but as cloud cover in Amazo/spl circ/nia is a constant problem, particularly in certain areas, radar data can help, as complimentary information, for change monitoring. In this paper ERS and RADARSAT images are classified using texture measures, in several classes of land use, and then the adequacy of using these classes for change detection is analyzed. Progressive sequential feature selection, using the Kappa coefficient of agreement as a selection criterion, chooses a subset of the texture layers that maximizes that coefficient. It was observed that even for the best feature set, the Kappa coefficient was considered too low and unsuitable to be used for change detection. However, it is shown that this coefficient progressively increases when classes are merged sequentially. When only two classes are considered, identified as forest/non-forest, the overall accuracy is higher than 85%, which was considered adequate for change detection. The classifications of the 1992, 1993 and 1996 ERS1/2 images over the Tapajo/spl acute/s National Forest, Brazil, were performed using the iterative contextual mode (ICM) classifier. Deforestation was detected for those points changing from forest in one year to non-forest in other year, with very good agreement with the results obtained with optical imagery sequence. Similar results were obtained using RADARSAT imagery for the year 1996.