'Meuse Model Moulding': On the effect of spatial resolution

The main objective of this research is the determination of the effect of spatial resolution on hydrological model performance. In order to do so, a model of the Meuse basin was setup at the appropriate scales of Booij (2001) in the existing HBV structure and compared with a model of lower resolution (HBV-15). This comparison was made in three steps, namely the calibration, the validation and the simulations under current and changed climate conditions. On the basis of a few logical criteria the Meuse basin was divided into 118 subbasins (HBV-118). This map of 118 subbasins was the basis of the setup of the model in the HBV-structure. One of the main problems was the lack of calibration data. Only discharge series of 5 stations were available, covering only a small part of the 118 subbasins. Therefore, in order to determine the parameters values of each individual subbasin, use was made of a regionalisation indicator. This indicator is a measure of the physical catchment characteristics slope, soil texture and lithology. For each of these characteristics a classification on a range of fast reacting to slow reacting areas v/as made, after which these three characteristics were integrated into one regionalisation indicator. On the basis of the value of this indicator for a certain subbasin, the parameters of that subbasin were increased or decreased, depending on the relationship between the indicator and the parameter. The model simulations were assessed using three different criteria: the Nash-Sutcliffe coefficient, the relative volume error and the relative return value error. In general, both the HBV-15 and the HBV-118 performed good for both calibration and validation. Only small differences could be determined, but the uncertainties were too large to determine which model performed better. Also for the climate scenario's the models showed almost equal results. It was concluded that the HBV-118 has no advantage over the HBV-15 model when using the current available data. Therefore, the near future focus should be on more and better data, in order to decrease the model uncertainties