干旱区生态环境知识资源中心

Data from General Circulation Models (GCMs) are often used to investigate hydrological impacts of climate change. However GCM data are known to have large biases, especially for precipitation. In this study the usefulness of GCM data for hydrological studies, with focus on discharge variability and extremes, was tested by using bias-corrected daily climate data of the 20CM3 control experiment from a selection of twelve GCMs as input to the global hydrological model PCR-GLOBWB. Results of these runs were compared with discharge observations of the GRDC and discharges calculated from model runs based on two meteorological datasets constructed from the observation-based CRU TS2.1 and ERA-40 reanalysis. In the first dataset the CRU TS 2.1 monthly timeseries were downscaled to daily timeseries using the ERA-40 dataset (ERA6190). This dataset served as a best guess of the past climate and was used to analyze the performance of PCR-GLOBWB. The second dataset was created from the ERA-40 timeseries bias-corrected with the CRU TS 2.1 dataset using the same bias-correction method as applied to the GCM datasets (ERACLM). Through this dataset the influence of the bias-correction method was quantified. The bias-correction was limited to monthly mean values of precipitation, potential evaporation and temperature, as our focus was on the reproduction of inter- and intra-annual variability.

After bias-correction the spread in discharge results of the GCM based runs decreased and results were similar to results of the ERA-40 based runs, especially for rivers with a strong seasonal pattern. Overall the bias-correction method resulted in a slight reduction of global runoff and the method performed less well in arid and mountainous regions. However, deviations between GCM results and GRDC statistics did decrease for < span style=’border-top: 1px solid #000; color: #000;’> Q , Q90 and IAV. After bias-correction consistency amongst models was high for mean discharge and timing (Qpeak), but relatively low for inter-annual variability (IAV). This suggests that GCMs can be of use in global hydrological impact studies in which persistence is of less relevance (e.g. in case of flood rather than drought studies). Furthermore, the bias-correction influences mean discharges more than extremes, which has the positive consequence that changes in daily rainfall distribution and subsequent changes in discharge extremes will also be preserved when the bias-correction method is applied to future GCM datasets. However, it also shows that agreement between GCMs remains relatively small for discharge extremes.

Because of the large deviations between observed and simulated discharge, in which both errors in climate forcing, model structure and to a lesser extent observations are accumulated, it is advisable not to work with absolute discharge values for the derivation of future discharge projections, but rather calculate relative changes by dividing the absolute change by the absolute discharge calculated for the control experiment.