Multi-Objective Optimization Methods for Bias Correction of Statistically Downscaled Precipitation

Statistical downscaling modeling-based results often depict biases and are incapable of reproducing the variances and approximating the distributions of the predictands. This paper proposes a multi-objective optimization framework to reduce biases resulting from the limitations of the transfer function methods adopted under statistical downscaling procedures. The proposed formulations use a combination of several error measures including Linear Error in Probability Space (LEPS) to reduce any systematic biases. Also, different optimization algorithms both linear and nonlinear are evaluated for this purpose. The optimization formulations are applied and tested at 18 different locations for Support Vector Machine (SVM)-based statistical downscaling model results obtained from CGCM3 simulations for precipitation. Results indicate that the multi-objective optimization framework helps in reducing the biases of the statistically downscaled model results.