Using a rainfall stochastic generator to detect trends in extreme rainfall

An original approach is proposed to estimate the impacts of climate change on extreme events using an hourly rainfall stochastic generator. The considered generator relies on three parameters. These parameters are estimated by average, not by extreme, values of daily climatic characteristics. Since climate changes should result in parameters instability in time, the paper focuses on testing the presence of linear trends in the generator parameters. Maximum likelihood tests are used under a Poisson-Pareto-Peak-Over-Threshold model. A general regionalization procedure is also proposed which offers the possibility to work on both local and regional scales. From the daily information of 139 rain gauge stations between 1960 and 2003, changes in heavy precipitations in France and their impacts on quantile predictions are investigated. It appears that significant changes occur mainly between December and May for the rainfall occurrence which increased during the four last decades, except in the Mediterranean area. Using the trend estimates, one can deduced that these changes, up to now, do not affect quantile estimations.