Reducing uncertainty in hydrological modelling in a data sparse region : Integrated Water Resources Management Under Different Hydrological. Climatic and Socio-Economic Conditions

One subproject within the project IWAS aims at the identification of effective and sustainable measures to improve the water quality of the river basin Western Bug in the Ukraine in preparation of changes of climate, land use and socio-economy. An important part of the system analysis is the quantification of water balance components. Rare and uncertain data impede hydrological modelling in the region. Approaches to reduce the modelling uncertainty are needed. In this work, we focus on the reduction of uncertainty that results from precipitation observations and parameter estimation during the calibration process. The main aim is to set up a well performing model for water balance simulations. The semi-physically based model Soil and Water Assessment Tool (SWAT) was applied. A calibration and uncertainty reduction strategy was set up that consists of time- and spatial-scale dependencies as well as alternative precipitation inputs. The single models that were set up and calibrated with alternative precipitation inputs are treated as an ensemble and were averaged with different methods. Nine parameters were chosen from the completed sensitivity analysis for all calibration approaches. The calibration strategy revealed the benefit of applying a complex bottom-up calibration that starts with sub-basins and ends with the entire basin. Not only performance improved significantly, but also water balance gaps were identified and henceforth reduced. The usage of daily runoff data in the calibration procedure did not enhance simulations in comparison to monthly data. Simulations improved and parameter uncertainty was reduced applying the SWAT model variants, where precipitation was included in calibration and by the ensemble averaging of the different models. Nevertheless, chosen uncertainty measures are not optimal and indicate that parameter uncertainty is still high. This is attributed to the low density of precipitation stations in the region, their low representativity, and scarce data situation in general. Differences between modelled and observed runoff are large in some occasions and can not be balanced completely by the applied methods. Although this approach reached its limit, it was demonstrated that simulations can be improved and modelling uncertainty can be reduced with an appropriate calibration strategy, with extended precipitation information and the application of an ensemble approach.