Recent studies showed that soil fertility properties can be predicted from soil spectral reflectance data and in a second step can be combined successfully with information from satellite imagery for rapid assessment of soil quality over large areas. This approach shall be adapted for a test area in the Loess zone of Tajikistan in order to assess the impact of land use on soil fertility. The groundtruth data collected confirms that widespread land use changes have taken place since 1992 (30 % of the area formerly used as grazing land has been cultivated since 1992). The newly cultivated areas are situated on steep slopes (the average slope is 20 %) and show visible signs of water erosion in 60 % of the cases observed. Also 48 % of the plots recorded from grazing land showed signs of water erosion. VIS-NIR measurements of soil samples collected from each sampling plot have been explored for relations between soil reflectance data and commonly used indicators of soil fertility in the study area. First results show that reflectance wavebands are strongly relating to CaCO3 and soil colour. Regression tree modelling has been carried out successfully to calibrate total nitrogen contents determined by chemical analysis against reflectance wavebands (validation r2 for regression was 0.71). A classification tree model predicting areas with water erosion shows the potential of decision tree modelling when combining different datasets. Hierarchical structures can be revealed and thresholds for mapping purposes using raster datasets available (DEM and Landsat 7 satellite imagery) can be determined. Prediction success determined by 10 fold cross-validation was 72 % and 61% for the classes erosion and no erosion respectively.