Previous attempts to map land cover at broad scales were based on single year time series and usually on the Normalized Difference Vegetation Index ( NDVI) Single years of data lack statistical representativity and the NDVI is partially driven by short-term climatic characteristics. We investigate two approaches to produce land-cover classifications that are not excessively influenced by short-term climatic variability: (i) averaging a climate-driven variable over several years, and (ii) measuring a more climate-independent variable. We test, compare and combine these two approaches for the African continent using 8 years of AVHRR Global Area Coverage (GAC) data. Our results demonstrate that times series of the ratio between surface temperature and NDVI are less influenced by interannual variations in climatic conditions than NDVI time series and thus produce more stable land-cover classifications. This finding is consistent with the biophysical interpretations of these two variables. When data are averaged over longer periods, NDVI- or Ts/ NDVI-based land-cover classifications display few differences.