Performance assessment of different bias correction methods in statistical downscaling of precipitation

Abstract

Global circulation models (GCMs) are widely used for the modeling and assessing the impacts of climate change. These models do not always accurately simulate climate variables due to the risk of considerable biases. Several bias correction methods have been proposed and applied so far. The selection and application of appropriate bias correction can improve accuracy and reduce uncertainty in downscaled precipitation in arid regions. In this study, initially multilayer perceptron (MLP) neural network was applied to downscale the mean monthly precipitation. The MLP model was calibrated by using National Center for environmental prediction (NCEP) reanalysis dataset and monthly precipitation observations located in selected hyper-arid, arid and semi-arid regions. Later, the performance of four bias correction methods namely, power transformation, simple multiplicative, variance inflation and quantile mapping were evaluated by comparing the mean and standard deviation of observed and downscaled precipitation. It has been found that the power transformation method is the most reliable and suitable method for downscaling precipitation in the arid region.