Abstract

The global circulation models (GCMs) and the regional climate models (RCMs) are the tools that project information on climate change and its impact on water resources by combining hydrological models with climate variables. The present work compares the performance of two bias correction methods - Linear Scaling and Distribution Mapping for hydrologic modelling over the Malaprabha river catchment originating in the Western Ghats of India. The comparison of RCM outputs with observed data over a period of 55 years (1951 to 2005) at similar scales exhibits biases in the magnitude and spatial distribution of climate variables. The performance of the bias correction was assessed by comparing simulated stream flow from ArcSWAT hydrological model using raw CORDEX RCM data and bias corrected climate variables. The raw RCM precipitation when compared with IMD 0.25 gridded rainfall, exhibited a RMSE of 392mm/year along with a low agreement of temporal structure with the observed precipitation. The RMSE after correction of bias was found to be zero (linear scaling) and 20 mm (distribution mapping). The RMSE for maximum and minimum temperatures were found to be insignificant (0.0670 C and 0.0820 C respectively). The distribution mapping method was found to be more suitable in correcting the temporal structure of temperature compared to the linear scaling method. The ArcSWAT model coped well with the Malaprabha river catchment with Nash-Sutcliffe efficiency of 0.87 for calibration period (1977-1990) and 0.77 for the validation period (1991-1996). The simulation of raw RCM precipitation with the ArcSWAT model showed a RMSE of 970 cumecs/year. The precipitation and temperature corrected by linear scaling method improved the RMSE to 856 cumecs/year and the distribution mapping method improved the RMSE to 794 cumecs/year. However, it was found that, the temporal structure of the simulated stream flow was better with the linear scaling method. The results show that, the RCMs are biased in simulating precipitation but the temperatures are fairly simulated. Although both the methods correct bias to a certain extent, the performance of distribution mapping method is better than the linear scaling method in prediction of stream flow using hydrologic models.