Mesoscale water balance modelling in the Upper Danube watershed using sub-scale land cover information derived from NOAA-AVHRR imagery and GIS-techniques

Earth observation from space provides unique data to obtain up-to-date information on the rapidly changing state of the environment. While imagery from high spatial resolution sensors are still inadequate to derive consistent land use information for mesoscale areas, fine spatial resolution of land use information is essential for the description of hydrological processes at the landscape level, such as runoff generation and evapotranspiration. The study presents a procedure to overcome existing limitations by using coarse spatial resolution NOAA-AVHRR (Advanced Very High Resolution Radiometer) data within a framework of combined multitemporal imagery and fuzzy-logic based geospatial data analysis. The spectral unmixing methodology determines fractional land cover data for each raster cell in the watershed. It assumes that the spectrum of a surface is linearly composed of the area-weighted spectra of its known components (endmembers). In extension to existing unmixing approaches, each "spectrum" refers to a multitemporal spectral profile of a pixel, which consists of the temporal development of the pixel's spectral behaviour over an entire vegetation period. In order to minimise classification errors, geographical expert knowledge is utilised to evaluate the geofactors elevation, slope, soil and precipitation in a fuzzy-logic approach to priorily determine a valid set of possible endmembers for each raster cell. The final unmixing results are validated against both a reference classification from LANDSAT-TM imagery and the CORINE land cover classification. The method is employed for the Upper Danube watershed (76.653 km(2)) to provide subscale land use information, which is used as an input for the physically based and raster-oriented SVAT model PROMET (J. Hydrol. 212-213 (1998) 250; J. Hydrol. 254 (2001) 199). The model is operated in hourly time steps on a 1-km(2) grid, each raster cell comprising the various land cover classes, to simulate the spatial and temporal course of evapotranspiration, soil moisture, snow and runoff formation. Sensitivity analysis of event-based modellings as well as annual water balance simulations exhibit a quantitative improvement in the spatial representation of hydrological processes in the Upper Danube watershed and a considerable reduction of uncertainty, when the information of sub-scale heterogeneity is taken into account. (C) 2003 Elsevier Ltd. All rights reserved.