Resource information
Land cover is a key parameter in geosciences and a linkage between many aspects of the physical and human environments. Savannas belong to the biomes where land cover mapping with remote sensing faces the most difficulties and several studies already addressed the challenging definition of savanna land cover classes. With the aim to standardize ongoing mapping activities, the Land Cover Classification System (LCCS) was initiated in 1993. The classification scheme has been repeatedly utilized for global approaches with coarse resolution remote sensing data, while local or regional applications are still limited in number. This study systematically explored the potential of Terra-ASTER data for LCCS classification in Burkina Faso using 502 field sites. Due to the small-scale landscape heterogeneity, pixel-based classifiers were applied and training data were clustered according to their spectral signatures. Overall classification accuracy decreased from 95.6% over 88.4% to 78.5% when considering the LCCS dichotomous-phase classifiers presence of vegetation, edaphic conditions, and artificiality of cover, respectively. For 16 classes, an overall accuracy of 61.4% was achieved. Altogether, this study is a step towards the systematic combination of standardized LCCS legends with continuously available remote sensing data – one of the core challenges for land cover mapping in the future.