The sensitivity of simulated maxima of 10-day precipitation to the set of historical base years used for resampling has been assessed. Eight subsets of 33 years were extracted from the base period 1930-1998 (with the exclusion of 1940). The fraction of days with 10 mm of precipitation or more in the winter half-year and the mean winter precipitation in these subsets were both either high or low, near the edge of the 95%-confidence region of their expected values. Furthermore, the subsets differed with respect to the inclusion or exclusion of the years 1995 or 1984. With each subset a 20000-year simulation was performed and the quantiles of the simulated 10-day winter maxima were considered. In addition to this, a 20000-year simulation was run in which the entire base period was used. This simulation serves as a reference and is assumed to be comparable with earlier simulations. It was found that for return periods up to five years the influence of 1995 or 1984 on the 10-day maxima was negligible compared to that of the mean winter precipitation and the fraction of winter days with 10 mm of precipitation or more. For longer return periods, especially the influence of the winter of 1995 was noticeable. The 10-day precipitation with a return period of 1250 year in the eight simulations spanned a range from 165 to 210 mm (approx. 24%). The influence of the repetition of historical days in the simulation was investigated, using a modified algorithm in which such repetition was excluded. This resulted in slightly lower 10-day maxima (for a return period of 1250 years, the differences are within 5%).
R Leander, TA Buishand. Rainfall generator for the Meuse basin; Description of 20000-year simulations
KNMI number: PUBL-196-IV, Year: 2008, Pages: 15