Résumé
Urban areas are becoming better equipped with dense rainfall measurement networks and there is a growing demand for local specific rainfall-frequency curves. However, the time series from such networks are generally short, often less than 20 yr, and spatially correlated (1 station per 5 or 10 km 2) even at small time steps, usually less than 1 h. It is therefore necessary to use a method which takes the available information as much as possible into account in order to minimize the confidence intervals of the estimates. The French Department of Seine-Saint-Denis, covering an area of 236 km 2 in the eastern part of the Paris area, manages a 17-year-old rainfall data-base which contains about 250 station-years of checked data. These data are considered within a discussion of the issues of homogeneity of correlated time-series on an urban area, and the amount of information such a network includes at small time steps. Local rainfall frequency curves are developed with a regional approach and put into a non-dimensional form. They are compared with those obtained from a long-term series (more than 30 yr) available in Paris and with the curves obtained by the station-year method on 21 independent stations in the north of France.