Combining agricultural economic and hydrological models with the aid of remote sensing data.

The EU Water Framework Directive (WFD) aims at maintaining and improving the aquatic environment and at establishing the basic principles of a sustainable water policy in the European Community. For this reason the impact of agricultural emissions into the hydrological system is analysed and modeled both for inventory and policy advisory purposes. In the case of nitrogen, regions with a high exposure of eutrophication are characterised on a high precision level and changes in discharge paths are identified. In this study a coupling of the agricultural sector model RAUMIS (Regional Agricultural and Environmental Information System for Germany) with the hydrological model GROWA is illustrated. The area of investigation is the catchment basin of the Ruhr river in the west of North Rhine – Westfalia, Germany. RAUMIS places emphasis on the economic aspects of agriculture and assesses the impact of agricultural-environmental policies. Essential elements are environmental indicators such as the potential diffuse agricultural nitrogen or pesticide emissions into the groundwater. These data are available for crops in an averaged form on the district level. The raster-based model GROWA calculates the main water balance components: the actual evapotranspiration, total discharge, direct runoff and groundwater recharge. One result of the model is an estimation of diffuse agricultural nitrogen emissions into water bodies. However besides soil data the model makes use of CORINE Land Cover which has an insufficient spatial resolution and which does not differentiate between individual crops such as wheat, sugar beet, potatoes, etc. Because of this difference in scale and attribution, the output data of RAUMIS cannot be inserted into GROWA in a simple way. LANDSAT ETM+, SPOT and ASTER data of the Ruhr catchment are used in order to solve this problem. After a panchromatic sharpening and further standard pre-processing methods, a Kalman filter based neural net classifies the images into the specific land use categories. A post-classification filtering with probabilistic label relaxation is availed for correction of misclassified pixels. The images of consecutive years are analysed to define the regional crop rotation for each parcel of land. This precise land cover information then serves as the interface between the two models. Nitrogen balance surpluses estimated by RAUMIS are disaggregated from district scale to a raster level of 15-20m. Furthermore the land cover class probabilities themselves are included as improved input data into GROWA.