Calibration of WRF model parameters using multiobjective adaptive surrogate model-based optimization to improve the prediction of the Indian summer monsoon

Sensitive parameters of a numerical weather prediction model substantially influence the model prediction. Weather research and forecasting (WRF) model parameters are assigned default values based on theoretical and experimental analysis by the scheme developers. Calibrating the sensitive parameters of the model has the potential to improve model prediction. The objective of this study is to improve the prediction of the Indian summer monsoon by calibrating the WRF model parameters. A multiobjective adaptive surrogate model-based optimization (MO-ASMO) method is used to calibrate nine sensitive parameters from five physics parameterization schemes. Normalized root-mean-square error values corresponding to four meteorological variables precipitation, surface air temperature, surface air pressure, and wind speed are minimized by calibrating the WRF model sensitive parameters for high-intensity precipitation events of the Indian summer monsoon (ISM). Twelve high-intensity four-day precipitation events of ISM during the years 2015-2017 over the monsoon core region in India are considered to calibrate the model parameters. MO-ASMO method outputs a set of nondominated solutions for the model parameters that reduce the model prediction error. A decision analysis method is used to identify the best solution among the nondominated solutions, which contains the calibrated values of the parameters. A comparison of the default and calibrated parameter values across various precipitation events, driving data, and physical processes in the monsoon core region are carried out. Eighteen high-intensity four-day precipitation events of ISM during the years 2014-2018 are chosen to validate the robustness of the calibrated parameters. The WRF model is run with two different boundary data to verify the effectiveness of the calibrated parameters against the default parameters. The model calibrated parameters obtained using the MO-ASMO method are superior to the default parameters across various precipitation events and boundary data over the monsoon core region during the Indian summer monsoon.