Assessment of mountainous ecosystems in Central Asia by using remote sensing based vegetation cover fractions

Mountainous regions are especially vulnerable to climate and land use change. For a monitoring of subtle changes detailed and continuous land cover information is necessary. Therefore vegetation cover fractions are derived for the Naryn river catchment in Kyrgyzstan using a remote sensing multi-scale approach. First a very high resolution QuickBird image is classified into discrete classes using a hybrid approach. This result is aggregated to the spatial resolution of Landsat and used as training data for a random forest and Landsat data to derive sub-pixel fractions of woody vegetation, herbaceous vegetation and bare soil. These results are again spatially aggregated and training pixels are extracted to derive finally vegetation fractions for the whole river catchment using a random forest with moderate resolution MODIS data. The results have a RMSE between 8 and 17 and show that woody cover can be found mainly on northern slopes, while bare soils are predominant on southern