For the drainage basins of the rivers Rhine and Meuse long-duration sequences of multi-site daily precipitation have to be generated as part of a new methodology to reduce the uncertainties in the design water levels for flood protection. Stochastic rainfall generators have been developed for this purpose, one for the Rhine basin upstream of the Netherlands (165 000 km2) and one for the French and Belgian parts of the Meuse basin (21 000 km2). Besides precipitation, daily temperatures are also generated in order to account for evaporation, snow accumulation and snowmelt. The rainfall generator forms the weather component of the GRADE (Generator of Rainfall And Discharge Extremes) instrument for generating extreme river flows. This instrument is developed in cooperation with Rijkswaterstaat Centre for Water Management and Delft Hydraulics.

Our weather generators make use of a non-parametric nearest-neighbour resampling scheme. Nearest-neighbour resampling has been introduced in 1996 in the hydrological literature by U. Lall, B. Rajagopalan and A. Sharma (Water Resources Research, 32, 679-693). The main advantage of a non-parametric resampling technique is that it preserves the spatial association of daily rainfall over the drainage basin and the dependence between daily rainfall and temperature. To reproduce the autocorrelation structure of the data, a new day is resampled from the historical data by conditioning on the generated values for the previous day. Only the k nearest neighbours of the latter are considered for resampling. Summary statistics of the daily precipitation and temperature fields as well as atmospheric circulation indices have been used to find the nearest neighbours in the historical data. The choice of k turns out to be rather crucial for the reproduction of autocorrelation coefficients and properties of extreme multi-day rainfall. For coupling with hydrological (HBV) and hydraulic models daily precipitation and temperature are generated for 134 sub-catchments in the Rhine basin. The Meuse basin has been partioned into 15 sub-basins. Apart from resampling historical observed daily precipitation and temperature, resampling of the output of Regional Climate Models (RCMs) has also been considered to study the impact of future climate change on extreme river flows. Figure shows the spatial distribution of 10-day rainfall for the German part of the Rhine basin (105 000 km2) for the event that gives the largest 10-day area-average rainfall (141 mm) in the winter half-year (October-March) in a 1000-year simulation. It is this kind of multi-day events in the winter season that give rise to extreme river flows in the Netherlands. For the notable flood events of December 1993 and January 1995, the area-average rainfall in the German part of the Rhine basin was 106 mm and 101 mm, respectively.

A large source of uncertainty is the choice of the historical precipitation and temperature data used for resampling. For the Meuse basin, eight 20 000-year simulations were conducted with different 33-year sets of historical data. The 1250-year return value of the 10-day winter maximum precipitation varied between 165 mm and 210 mm in these simulations. The first simulations using climate model output were done for the Meuse basin (PhD thesis Robert Leander, Utrecht University, 2009). For the Rhine basin such simulations were conducted in the framework of the RheinBlick2050 project, an international project initiated and coordinated by theinternational project initiated and coordinated by the International Commission for the Hydrology of the Rhine basin (CHR). Figure 2 presents the 10-day winter maximum precipitation amounts in 3000-year resampled series based on the output from a RCM simulation for the periods 1961-1990 and 2071-2100, and the corresponding observed 10-day maxima. The RCM output has been corrected for systematic differences between the observations.