An image processing chain for land-cover classification using multitemporal ERS-1 data

Given the frequent cloud cover, regular updating of land-cover maps of tropical areas using optical satellite data is problematic. As weather-independent ERS images are available on a regular basis, the question is raised as to whether they can be used to produce a land-cover map. A processing chain has been developed: it consists of calibration, resampling, filtering segmentation, principal component transformation, and supervised classification of multitemporal radar images. As input to this processing chain, any number of SRR.PRI images or derived texture images can be used. The output consists of a land-cover map and an accuracy assessment. The procedure hers been applied to a series of four SAR images, taken over northeast Costa Rica, which yielded a map with an overall accuracy of 76 percent. The high precision with which the large banana plantations can be mapped is most interesting both for its economic importance and for environmental monitoring.