Bayesian Model Averaging Prediction Of The Circulation In East Asia Using CMIP5 Data

Using the CMIP5 runs for global climate system decadal hindcast of the 500hPa geopotential height and 850hPa temperature, with the NCEP/NCAR monthly reanalysis data as’observation’,the capabilities of Bayesian model average (BMA) method to hindcast the atmospheric circulation over East Asia have been evaluated employing assessment indexes as anomaly correlation coefficient (ACC)、root mean square error (RMSE)、continuous ranked probability score (CRPS) and 90% prediction interval Width (90%PW).As 500hPa geopotential height and 850hPa temperature can mainly represent the atmospheric circulation situation, this paper studies the BMA hindcasting effects of atmospheric circulation situation in East Asia based on the two factors. BMA is a multi-model statistical way of postprocessing the forecast ensembles to create calibrated probability density function. Both the training period length and the model number involved in the BMA model can affect the BMA hindcasting results. The forecasting periods of hindcasting research is from 1996 to 2005.After confirming the forecasting periods, the training period length and the model number involved in the BMA should be determined first. Thus, the optimal training period length is 25 and the best model number is 13 for 500hPa geopotential height; while for 850hPa temperature, the optimal training period length is 23 and the best model number is 11. Afterwards, the BMA of the two factors is conducted, also compared with the models involved in the BMA and multi-model ensemble mean (EMN).Results show that, all the climate models involved in the integration、BMA and EMN all have some hindcasting ability for 500hPa geopotential height and 850hPa temperature in East Asia. And the BMA’s skill is the best, no matter in time series or in spatial characteristics; This is also the case for inter-annul variability of subtropical high over West Pacific (WPSH) and East Asia deep trough; Besides, from the PDFs, we can see that the hindcasting variance of BMA probability forecast is smaller than EMN, that is to say, the uncertainty is smaller, BMA can create calibrated probability density function and contains the quantitative estimations of the uncertainty of the BMA hindcast, which can effectively improve the hindcasting effects of 500hPa geopotential height and 850hPa temperature.On the basis of above analysis conclusions, in the RCP4.5 future scenario, the prediction of 500hPa geopotential height and 850hPa temperature in East Asia from 2006 to 2035 is proceeded by the use of BMA.The results indicate that, the 500hPa geopotential height and 850hPa temperature in East Asia both increases in total relative to the climate average, the increase in low latitudes is smaller, and in high latitudes is bigger. These can provide basis for predicting the change of the atmospheric circulation in East Asia in the future.