This study focuses on the use of coarse spatial resolution (CR, pixel size about 1 km 2) remote sensing data for land cover change detection and estimation. Since, in the presence of some changes, both the multitemporal class features and the pixel composition in terms of classes are unknown, the proposed algorithm is based on the iterative alternate estimation of each unknown variable: class features, and pixel composition. Final estimation of the pixel composition is constrained using a Markovian chain model, introducing the previous land cover map as a 'memory' term. This approach has been validated both using simulated data and actual data (SPOT/VGT and NOAA/AVHRR). The thematic application was the study of the evolution of an agricultural watershed during the last two decades.