AbstractRemote‐sensing data have proved to be a valuable source of land cover information, especially when looking at larger areas. In this study, multi‐temporal NOAA‐AVHRR data are used together with a spectral unmixing approach in order to overcome the restrictions of low spatial resolution and to derive subscale land cover of the upper Danube watershed. New data‐processing techniques account for topography‐dependent distortions and hence allow the inclusion of mountainous subwatersheds, such as the River Inn. To increase the number of separable land cover classes, geographical information system and fuzzy logic methodologies are applied to preselect possible endmembers according to their possibility. The results of spectral unmixing were validated against a LANDSAT‐TM classification and the CORINE land cover database, which both showed acceptable results. Copyright © 2005 John Wiley & Sons, Ltd.