Statistical downscaling method of regional climate model results for hydrological modelling

Reasonable and consistent meteorological input data is a crucial factor for modelling the river runoff at the catchment scale. The regional climate models (RCMs) provide sufficient information for the hydrological modelling of impact of expected climate change on the river runoff. However, one must avoid direct usage of RCM data for the forcing of hydrological models without analysing RCM compliance with observations for the reference period. The aim of this study was to provide reasonable meteorological input data for the hydrological models to predict the river runoff changes in the future. We considered the calculations made by the European RCMs organised in a database at the Danish Meteorological Institute under European Commission research project “PRUDENCE” EVK2-CT2001-00132 (prudence.dmi.dk). The spatial resolution of the analysed models is approximately 50 km with temporal resolution of 1 day. The set of 21 model runs was analysed. Each considered model run contained at least calculation for the climatic reference period (1961-1990) and model predictions for climatic scenarios A2 and B2 (2071-2100). The RCMs provide meteorological parameters at each grid point, which we supposed to use as a forcing for hydrological models. We analysed performance of different RCMs by statistically comparing air temperature and precipitation rate with the observations over the Eastern Baltic area for the same period. The penalty function describing the deviation of each of the RCMs from the meteorological observations was constructed, aiming at evaluation of model accuracy in terms of monthly average temperature and precipitation, their monthly and interannual variation, and spatial distribution. Generally, all models reasonably represent the seasonal cycle of temperature, though they overestimate winter precipitation and underestimate summer precipitation in the study area. We proposed a method of RCM data correction, based on shifting the occurrence distribution of an individual daily output (temperature or precipitation). Two cumulative distribution functions - one of the observed data, and one of the RCM data - were constructed for each day-of-the-year, for each parameter in each observation station. The correction function was constructed in a way to have equal probabilities of particular daily parameter for both observed and corrected RCM data. The correction functions were spatially interpolated, giving the possibility to create modified RCM data both for the reference period and future climate scenarios. We analysed the performance of the method by comparing monthly statistical parameters of observed data versus corrected RCM data at a selected station. We show that statistical moments of distribution of temperature and precipitation were corrected by the present method. Interannual variability and temperature/precipitation correlation properties, however, cannot be significantly improved. The proposed approach of RCM data modification allows changing the modelled temperature and precipitation time series for the reference period in such a way that they preserve the characteristics on a small time-scale, and at the same time also having the statistical properties of the observed data. The time series for the future climatic scenarios were obtained assuming that the histogram modification algorithm is the same for present and future climate. The hydrological modelling with the modified meteorological forcing has not been carried out in the present study.