Modèle de prévision et de gestion des crues

Floods are not only due to fatality. It is the responsibility of the human being to protect himself from heavy damages due to natural elements, and he sometimes has the chance to profit from a given situation to influence events. One of these possibilities is to manage floods by using existing dams and reservoirs, which can be efficiently operated to control the downstream discharges. Moreover, their retention capabilities may be enhanced by preventive gate or turbine operations, based on a flood forecast. The objectives of this research were to develop a new model for flood prediction and management for the Rhone river basin upstream from Lake of Geneva. Also, the underlaying objective was to create an operational discharge prediction and decision making tool, taking advantage of 72 hours ahead of the weather forecast provided by MeteoSwiss. A new hydrological model was developed, as well as another optimization tool for the preventive turbine and gate operations of large hydropower reservoirs. The new hydrological forecasting tool is based on a concept developed by the HYDRAM-EPFL. This concept allows the model to integrate tri-dimensional rainfall, temperature and evapo-transpiration fields and to simulate multiple hydrological processes. Indeed, the model is able to simulate glacier melt, snow pack constitution and melt, soil infiltration and runoff. The altimetric temperature gradient is considered by subdividing each basin into elevation bands, which allows segregating rainfalls and snowfalls. A new software called Routing System II was used and improved for modelling the catchment area. This object-oriented modelling tool permits the integration of flood routing in rivers as well as hydraulic structures such as river water intakes, reservoirs, turbines, gates and regulated systems. All these new products were used to build a flood prediction model, including the 10 major hydropower plants of the Rhone river basin upstream from Lake of Geneva. It has been calibrated and validated over a 60 months' period for an one hour time step continuous simulation. It also integrates all hydroelectricity production data of the existing hydropower plants. Its performance allows to correctly represent all hydrological cycles, as well as the observed floods, whose phases and intensities were perfectly simulated. Moreover, a new procedure for real-time data assimilation was developed in the model, in order to automatically adjust the initial conditions before starting a new hydrological forecast. The new optimization tool uses the hydrological forecast, especially the inflow forecasts in the reservoirs and the hydrographs at the numerous control points in the river network. This tool takes into account the mentioned data in order to provide an operational decision about any necessary preventive turbine or gate operations. It also allows the decision maker to obtain an indication on the cost of the decisions and of the non-decisions, as well as of the cost of an inappropriate decision due to an error in the flood forecast. This algorithm was validated against full simulation and against another optimization evolutionary algorithm called MOO , developed in the LENI-EPFL. It is able to provide a similar efficiency without excessive computation time, and to forward these results automatically into the simulation model for validation. The operational performance of the flood prediction and management model was evaluated by simulation of two major flood events occurred in the Rhone river basin in September 1993 and October 2000, as well as by a first operational use in September 2006. The obtained results indicated, that if the decisions relative to the operations of the 10 considered hydropower plants had been in accordance with the new decision tool, the decision would have been similar to a posteriori optimal solution. The application of the model to the Rhone River catchment area demonstrates the possibility to widely increase the protection effect due to the existing reservoirs. In fact, the observed reduction of the peak flow in the Rhone basin outlet was 12% in 1993 and 10% in 2000. By using the new model, the reduction would have reached 26% in 1993 and 21% in 2000. In this case, important damages could have been avoided near the riverside. This research project offers the possibility to highlight the potential of an increased flood routing into the existing accumulation reservoirs. It leads to a new theoretical optimization model, which can be used at a planification stage to evaluate the optimal flood routing influence of a multireservoir system. Finally, a new flood forecast and management model was developed for the Rhone River catchment area. This tool is in operation and its aim is now to optimize the existing flood protection potential. Thanks to its new performing forecast, data assimilation and optimization tools, the model is user-friendly and can be used in real-time. Although the system is now operational and convincing, it would be more profitable to explore the possibilities of dealing with uncertainty in the whole information path of the system. As a main goal, the system should be capable of providing the decision maker with an optimal decision and its associated objective risks.