Probabilistic 2-meter surface temperature forecasting over Xinjiang based on Bayesian model averaging

Abstract

Based on Bayesian model averaging (BMA), the suitability and characteristics of the BMA model for forecasting 2-m temperature in Xinjiang of China were analyzed by using the forecast results of the Desert Oasis Gobi Regional Analysis Forecast System (DOGRAFS) and Rapid-refresh Multiscale Analysis and Prediction System (RMAPS) developed by the Urumqi Institute of Desert Meteorology of the China Meteorological Administration, China Meteorological Administration–Global Forecast System (CMA-GFS) developed by the China Meteorological Administration, and the European Center for Medium-Range Weather Forecasts (ECMWF) developed by the European Center. The results showed that (1) the weight of ECMWF to the 2-m temperature forecast is maintained at about 0.6–0.7 under different lengths of training periods, and the weight of other model products is below 0.15. (2) The forecasts of each model at the four representative stations are quite different, and the maximum forecast error reaches 6.9°C. However, the maximum error of the BMA forecast is only about 2°C. In addition, the forecast uncertainty in southern Xinjiang is greater than that in northern Xinjiang. (3) Compared with multi-model ensembles, the overall prediction performance of the BMA method is more consistent in spatial distribution. Additionally, the standard deviation and correlation coefficient between the BMA forecast and observation were greater than 0.98, and the RMSE decreased significantly. It is feasible to use the BMA method to correct the accuracy of the 2-m temperature forecast in Xinjiang.