Reducing biases in regional climate downscaling by applying Bayesian model averaging on large-scale forcing

Reduction of uncertainty in large-scale lateral-boundary forcing in regional climate modeling is a critical issue for improving the performance of regional climate downscaling. Numerical simulations of 1998 East Asian summer monsoon were conducted using the Weather Research and Forecast model forced by four different reanalysis datasets, their equal-weight ensemble, and Bayesian model averaging (BMA) ensemble means. Large discrepancies were found among experiments forced by the four individual reanalysis datasets mainly due to the uncertainties in the moisture field of large-scale forcing over ocean. We used satellite water-vapor-path data as observed truth-and-training data to determine the posterior probability (weight) for each forcing dataset using the BMA method. The experiment forced by the equal-weight ensemble reduced the circulation biases significantly but reduced the precipitation biases only moderately. However, the experiment forced by the BMA ensemble outperformed not only the experiments forced by individual reanalysis datasets but also the equal-weight ensemble experiment in simulating the seasonal mean circulation and precipitation. These results suggest that the BMA ensemble method is an effective method for reducing the uncertainties in lateral-boundary forcing and improving model performance in regional climate downscaling.