Master Thesis
Water Availability in the Aral Sea Basin: Derivation of Fractional Vegetation Covers from Multi-scale Remote Sensing Data for Hydrological Modeling in Central Asia
Sarah Asam (07/2009-06/2010)
Support: Christopher Conrad, Carl Beierkuhnlein
As Central Asia faces various water-related problems, a sustainable and transnational water management is needed, in particular for the regions specialized in the production of water-intensive crops, such as cotton. Regional hydrological models can provide an insight into water cycles and projections of the water availability under changing climatic conditions. Land cover information that describes the vegetation cover provides important input for the parameterization of such hydrologic models. The percentage cover of woody and herbaceous life forms as well as bare soil fractions would meet this high information requirement. The aim of this study is therefore to describe in detail the small-scale vegetation patterns of a high mountain ecosystem in Central Asia.
For mapping fractions of vegetation cover, multi-scale remote sensing data is analyzed in conjunction with vegetation field data. Vegetation data was collected on a training site in the upper reaches of the Naryn River, Kyrgyzstan. The ground truth samples were used in a hybrid classification scheme (object- and pixel-based) applied to spatially very high resolution (0.6 m) satellite data, a QuickBird scene of 80 km2. A map for each single cover type is derived from this classification. By aggregating the results to the Landsat TM spatial resolution of 30 m, continuous cover fractions for each cover type are gained. Using the regression tree ensemble method ‘random forest’, this land cover fraction information is extrapolated to the whole 180 x 190 km2 area of the Landsat TM tile. The accuracy of the resulting land cover product is between 85 and 92 % for the various cover types. The continuous cover fractions of woody, herbaceous and uncovered soil can be used as input parameters for hydrological modeling. Besides modeling, the continuous cover can be applied for biodiversity research or vegetation pattern analysis.