Author(s): Garcia Hernandez J.; Jordan F.; Boillat J-L.; Schleiss A.
Linked Author(s): Anton J. Schleiss
Keywords: Flood forecast; Prediction error; Meteorological forecast; Meteorological spatial assimilation
Abstract: During the last two decades, the Upper Rhone River basin in Switzerland has been hit by three flood events. As a consequence, human loss and much damage were deplored in the catchment area. In order to improve the security level for the local population and reduce the probable damage during floods, a new flood forecasting system has been developed. It focuses on the typical difficulties related to mountainous catchments areas, where many hydrological events occur. What is more, numerous hydropower schemes are located in the Upper Rhone River basin, which strongly influence the natural flows. In this framework, a project called MINERVE has been developed, with the purpose of predicting floods three days in advance taking advantage of the existing accumulation reservoirs to reduce the peak flow during floods. A semi-distributed hydrological model of the river basin has been developed with the Routing System II software. Its object-oriented programming features allow the modelling of snow-melt, glacier melt, soil infiltration processes as well as flood routing in rivers including hydropower schemes. A solid calibration and validation procedure was achieved. Furthermore a new data assimilation procedure was developed, in order to allow the operational use of the flood forecasting system. For the computation of flood prediction, the numerical meteorological forecasts provided by MeteoSwiss have to be assimilated. The precipitation and temperature data are available at an hourly rate every 12 hours, for a 72 hours lead time. The spatial resolution is a grid of 7 x 7 km2 and the vertical resolution is about 100 m, depending on the topography. Numerous methods for the spatial assimilation of the meteorological variables by the hydrological model were tested and the results are presented in this contribution. Finally, the dispersion of the forecast is analyzed by comparing the performance of the 12 hours average predicted rainfall over the entire catchment area with the one observed. The influence of lead time is described in order to provide an estimation of the error in the rainfall forecast for decision-makers.