KNMI contribution to the European project WRINCLE: downscaling relationships for precipitation for several European sites

Statistical downscaling of precipitation refers to statistical techniques that have been used to obtain precipitation data with the required spatial resolution for climate-change impact studies. It is widely acknowledged that the direct precipitation output of climate change simulations from General Circulation Models (GCMs) is inadequate for such studies. Statistical downscaling techniques make use of fitted relationships between observed precipitation and other meteorological variables that can be extracted from GCM simulations. The recent availability of reanalysis data from numerical weather prediction models offers new opportunities to improve the meteorological basis of statistical downscaling models. In this report the statistical linkage of daily precipitation to NCEP reanalysis data is described for eleven stations across Europe: De Bilt and Maastricht (the Netherlands), Hamburg, Hanover and Berlin (Germany), Vienna (Austria), Berne, Neuchâtel and Payerne (Switzerland), and Salto de Bolarque and Munera (Spain). Daily data for the period 1968-1997 were considered. The work forms the KNMI contribution to the European project WRINCLE (Water Resources: the INfluence of CLimate change in Europe).

Two separate statistical models were used to describe the daily precipitation at a particular site: an additive logistic model for rainfall occurrence (1 for wet days and 0 for dry days) and a generalised additive model for wet-day rainfall. Both models are extensions of the standard linear regression model for data from a normal distribution. Typical predictor variables were the u-velocity (westerly flow), the v-velocity (southerly flow), the relative vorticity, the 1000-500 hPa thickness, the baroclinicity and atmospheric moisture. With respect to the latter two options for rainfall amount modelling were compared: (i) the use of the specific humidity at 700 hPa, and (ii) the use of both the relative humidity at 700 hPa and precipitable water. For rainfall occurrence modelling the relative humidity at 700 hPa was considered as moisture variable. The 1000-500 hPa thickness was only included in the model for rainfall occurrence.

For all stations the moisture variable appears to be one of the most significant predictors both for rainfall occurrence and rainfall amount. The u-velocity is an important predictor for the northern stations De Bilt, Maastricht, Hamburg and Hanover. Although precipitation usually has a strong link to relative vorticity, this is not the case for Vienna and the Swiss stations. The baroclinicity is an important predictor for these middle European stations. For the Spanish sites it was beneficial to derive the velocity components and the relative vorticity from the 850 hPa heights instead of the 1000 hPa heights or sea level pressure as was done for the other sites. Although the downscaling relationships for rainfall occurrence and wet-day rainfall are assumed constant over the year, they can reasonably reproduce the seasonal cycles in the probability of rain and mean wet-day rainfall.

An application is given with data from a time-dependent greenhouse gas forcing experiment using the coupled ECHAM4/OPYC3 atmosphere-ocean GCM for the periods 1968 - 1997 and 2070 - 2099. The fitted statistical relationships were used to estimate the changes in the mean number of wet days and mean rainfall amounts for the winter and summer halves of the year at De Bilt, Hanover, Berlin, Berne and Salto de Bolarque. For all these stations a decrease in the mean number of wet days was found. This decrease is mainly due to the larger 1000-500 hPa thickness in the future climate. At Berlin this thickness effect is somewhat reduced by the changes in the velocity components. A marked influence of the velocity components was also observed at Salto de Bolarque, but here the effect of the change in the u-velocity is counterbalanced by that in the v-velocity. The largest decrease in the number of wet days occurs in the summer period. The mean wet-day precipitation amounts are sensitive to the larger atmospheric water content in the future climate (increase in specific humidity and precipitable water). This leads to an increase in the mean winter rainfall amounts at De Bilt, Berlin and Berne despite the decrease in the mean number of wet days. The estimated increase in mean winter rainfall for these stations is comparable with that in the simulated rainfall of the ECHAM4/OPYC3 model. The use of the rainfall amount model with specific humidity results in a larger increase than the model with precipitable water and relative humidity. The estimated change in mean winter rainfall at Hanover differs from that at De Bilt and Berlin, and from the direct climate model output. This can be attributed to an anomalous decrease in the relative vorticity near Hanover in the ECHAM4/OPYC3 experiment and a rather strong impact of vorticity on daily rainfall at this site. Consistent with the simulated rainfall of ECHAM4/OPYC3 a decrease in the mean winter rainfall amouts at Salto de Bolarque was found. The use of a rainfall amount model with specific humidity results in a slight increase in the mean summer rainfall amounts at De Bilt, Hanover and Berlin. In the other cases there is a decrease in mean summer rainfall, quite often comparable with that from the direct climate model output. Except for the changes in winter rainfall at Hanover and the number of wet days in Berlin and Salto de Bolarque, the changes in the circulation variables have little effect on the mean number of wet days and the mean rainfall amounts. Moreover, the changes in circulation variables are generally small compared with their bias. The anomalous change in vorticity during winter at Hanover is in fact mainly due to biases in the simulated sea level pressure.

The use of the downscaling models for scenario production was examined for Berne. Scenarios of daily precipitation for the 2070 - 2099 period were obtained by perturbation of the observed rainfall record and by stochastic time series simulation. In the case of time series perturbation the standard method of scaling the observed rainfall amounts was extended to allow for a decrease in the number of wet days. The 90th percentile of the distributions of the N-day annual maximum precipitation amounts for N = 1, 3, 10 and 30 was considered to compare the resulting scenarios. The representation of the coefficient of variation of the wet-day precipitation amounts in the stochastic model for time series simulation strongly influences the reproduction of this extreme-value characteristic and its changes. Besides difficulties with the description of the coefficient of variation, the model for wet-day rainfall often overpredicts the mean rainfall amounts in situations where extreme rainfall could be expected. Interaction between predictor variables has to be incorporated to reduce this bias.