Das Projekt "Goldersbach - Hochwasservorhersagemodell für das Einzugsgebiet des Goldersbachs" wird vom Umweltbundesamt gefördert und von Universität Stuttgart, Institut für Wasserbau durchgeführt. Although draining only an area of 75 km2 the Goldersbach caused severe flooding in the town of Tuebingen-Lustnau several times in the second half of the last century. Because of local constrictions the city decided after several feasibility studies to set up a flood management system based on three columns: flood forecasting, partial storage in retention reservoirs and local protective measures. The goal was to develop a reliable operational flood-forecasting system, but due to the small catchment size, the anticipated lead time of 3.5 hours could not be achieved by gauge observations only. The principal approach was then to develop a weather radar-based, short-term rainfall forecasting system, valid for roughly 2 hours lead time, and to use its forecasts in combination with real-time observations in a rainfall-runoff model to gain the desired lead time. A gauge system in the Goldersbach catchment was established, along with a data transmittal and storage system to retrieve and store measurements from rain-gauges, river-gauges and a Doppler weather radar. In case of exceeding certain rainfall and discharge threshold values based on synthetic rainfall time series, the local authorities are alarmed. Following a predefined alarm plan preparatory procedures are initiated and if necessary the retention reservoirs filled and local protective measures implemented. The gauging system is currently optimized and the forecast system tested under operation. The retention reservoirs are going to be planned next year. As especially for short-term rainfall forecasting, knowledge of the current rainfield advection is crucial, two estimation techniques are applied: one based on the Doppler effect, the other on covariance maximization. Based on the advection estimates, a short-term, auto-regressive forecast model was developed. To combine the advantages of the available sources of rainfall observation, namely radar and rain-gauges, a new method termed 'Merging' is tested. It preserves both the mean rainfall field estimated by the rain-gauges and the spatial variability of the radar image. For short-term rainfall forecasting, a new model named 'SCM model', short for 'Spectrum-Corrected Markov chain' is used. Based on radar data, it follows a two-step hierarchical approach. A bi-variate, auto-regressive process is forecasting the large-scale development of rainfall in a radar image. The individual development of each grid-cell in the image is forecasted by a Markov chain approach. Finally, two rainfall-runoff models are used for short-term flood forecasting. The first, FGMOD/LARSIM, is an event-based model, the second, HBV-IWS, is a continuous water balance model. Both rainfall-runoff models, in combination with the rainfall forecast, allow reasonable discharge estimates for up to 3 hours.