Influence of the Mosaicking Algorithm on Sentinel-1 Land Cover Classification Over Rough Terrain

The aim of this study was to assess different strategies for wall to wall SAR-based (Sentinel-1) land cover mapping over mountainous areas. The first strategy was to classify by-orbit and combine the resulting classifications. The second strategy derived land cover from ascending-descending SAR mosaics. In both cases, the inverse of the scattering area was used for weighting. Slightly more accurate results were obtained using by-orbit classification, with larger omission errors for urban areas and forest being observed when classifying the SAR mosaics. Such errors were caused by mismatches between orbits caused by image normalization errors and/or directional effects.