Bias correction of the CCSM4 for improved regional climate modeling of the North American monsoon

Abstract

This study investigates how a form of bias correction using linear regression improves the limitations of the community climate system model (CCSM) version 4 when it is dynamically downscaled with the Weather Research and Forecasting (WRF) model for the North American monsoon (NAM). Long-term biases in the CCSM dataset were removed using the climate forecast system reanalysis (CFSR) dataset as a baseline, from which a physically consistent set of bias-corrected variables were created. To quantitatively identify the effects of CCSM data on the NAM simulations, three 32-year climatologies were generated with WRF driven by (1) CFSR, (2) original CCSM, and (3) bias-corrected CCSM data. The WRF-CFSR simulations serve as a baseline for comparison. With the bias correction, onset dates simulated by WRF bias-corrected CCSM data were generally within a week of the WRF-CFSR climatology, while WRF using the original CCSM data occur up to 3–4 weeks too early over the core of the NAM. Additionally, bias-correction led to improvements in the mature phase of the NAM, reducing August root-mean-square-error values by 26 % over the core of the NAM and 36 % over the northern periphery. Comparison of the CFSR and the bias-corrected CCSM climatologies showed marked consistency in the general evolution of the NAM system. Dry biases in the NAM precipitation existed in each climatology with the original CCSM performing the poorest when compared to observations. The poor performance of the original CCSM simulations stem from biases in the thermodynamic profile supplied to the model through lateral boundary conditions. Bias-correction improved the excessive capping inversions, and mid-level mixing ratio dry biases (2–3 g kg⁻¹) present in the CCSM simulations. Improvements in the bias-corrected CCSM data resulted in greater convective activity and a more representative seasonal distribution of precipitation.