Fuzzy rule-based downscaling of precipitation

A fuzzy rule-based methodology for downscaling local hydrological variables from large-scale atmospheric circulation is presented. The method is used to estimate the frequency distribution of daily precipitation conditioned on daily geopotential fields. The task is accomplished in two steps. First, the exceedence probabilities corresponding to selected precipitation thresholds are estimated by fuzzy rules defined between geopotential fields (premises) and exceedence events (response). Then a continuous probability distribution is constructed from the discrete exceedence probabilities and the observed behaviour of precipitation. The methodology is applied to precipitation measured at Essen, a location in the Ruhr catchment, Germany. Ten years of precipitation data (1970-1979) were used for training and another ten years (1980-1989) for validation. The 700hPa geopotential fields are used to characterise large-scale circulation. The application example demonstrates that this direct downscaling method is able to capture the relationship between premises and the response; namely both the estimated exceedence probabilities and the frequency distribution reproduce the empirical data observed in the validation period.