From small-scale modelling of alpine catchments towards probabilistic flood forecasting in the Rhine basin

Abstract

Recent floods in Central Europe caused damages in the order of several billions of Euros and numerous casualties. To mitigate the destructive consequences of severe flooding, society requires reliable fore casts of these events with a sufficient lead time. Hence, this study uses a coupled atmospheric-hydrologic forecasting System and explores the feasibility of exploiting recent advances in numerical weather prediction (e.g. Limited-area Ensemble Prediction System) for probabilistic flood fore casting. Because runoff formation in mountainous regions is mainly dominated by snow and glacier melt, chapter 2 presents a detailed analysis of melt processes in alpine catchments.The spatially distributed WaSiM-ETH model is applied to three Swiss high-alpine river catchments with different portions of glacierized areas for the period 1981-2000 with a spatial resolution of 100x100 m2 and an hourly time-step. To improve the calculation of glacier runoff, a seasonal varying radiation factor has been implemented in the glacier melt equation. Results show a strong dependency of the melt processes on season, altitude and exposition. Annual snow melt amount shows a maximum at around 2900 m a.s.l. in all investigated catchments. Besides natural processes, anthropogenic water storages and releases by hydropower stations can also have a considerable impact on the discharges and water cycle. In chapter3, these mechanisms have therefore been incorporated into hydrologic modelling of the Swiss Alpine Rhine basin (4838 km2). The hydrologic model was not able to simulate the diurnal and weekly cycle in the release mechanisms. By reconstructing hourly storage and release volumes from reservoir level data, a clear increase in the model Performance was reached, especially during the release periods and smaller floods. Additionally, an analysis of possible consequences of land-use changes is performed,showing that, although urbanization may have an impact on local hydrologic processes, its effect appears negligible in larger catchments. After the analyses of small-scale hydrologic processes, the next goal was the setup of an off-line coupled atmospheric-hydrologic modelling system (chapter 4) for runoff fore casting in the Rhine basin down to the gauge Rheinfelden (34550 km2). Due to the complex topography, the highly variable meteorology and the presence of lakes, it was necessary to divide the basin into several sub catchments. The conceptual PREVAH model was applied with a spatial resolution of 500x500 m2. The calibration of the catchments was done with the help of a semi-automatic parameter calibration program. The model correctly reproduces the relevant by drologic processes and properly captures the extreme runoff peaks. To validate the numerical weather prediction (NWP) model (Lokal Modell, LM), 6 years of 19-42 h precipitation and 2m temperature forecasts are used as input in the hydrologic model. Consequences for runoff forecasts, arising out of errors in precipitation forecasts, are most pronounced for events with high precipitation intensities, while the coupled modelling System in most cases properly predicts smaller flood events. Because of the hydrologic model sensitivity to atmospheric forcing, it would be desirable to quantify forecast uncertainties. A high-resolution atmospheric ensemble forecasting system based on 51 runs of the LM, which runs in a horizontal resolution of 10 km, has therefore been used to make probabilistic runoff forecasts for the Rhine basin down to Rheinfelden (chapter 5). The 5-day forecasts of the LM are used to drive the PREVAH model for two flood events. The case studies investigated are the spring 1999 flood event in the Rhine basin and the November 2002 flood in the Alpine Rhine catchment. The Limited-area Ensemble Prediction System (LEPS) allows to quantify hydrologic forecast uncertainties. Although the deterministic simulations yield large forecast failures, the coupled atmospheric-hydrologic LEPS provides appropriate forecast guidance with proper uncertainty intervals. The use of the clustering technique did not reduce ensemble spread compared to the 51 predictions. Furthermore, it was shown that the inclusion of horizontal advection of precipitation may be crucial for flood forecasts in alpine catchments.