Capabilities of ERS sensor for Mediterranean vegetation detection using multitemporal data

摘要

The objective of the present study is to evaluate the performances of a series of SAR ERS images for a land cover classification of a Mediterranean landscape, focusing on the discrimination of vegetation types. We tested the contribution of multitemporal data and contextual methods of classification with and without filtering for land cover discrimination. An index of temporal change was developed to characterise the stability of land covers, this index is based on the mean normalised difference between consecutive dates. This study shows the importance of time series of ERS sensor and of the vectorial MMSE filter based on segmentation, for land cover classification. Fifteen land cover classes, where eight of them concern to different vegetation types, have been classified obtaining a 80.1% of mean producer's accuracy for 1998 series, and 70.6% for 1994. These results are comparable with those from two-date SPOT images (85.3% of mean producer's accuracy).