Abstract:Based on the numerical forecast products of ECMWF, CMA and JMA gathered in the TIGGE data, the multi-model ensemble forecasting experiment about temperature at 2 m above ground in most parts of China at the 24 h, 48 h and 72 h forecasting lead time is carried out using some linear ensemble methods such as weighted ensemble, regression ensemble and bias-removed ensemble as well as BP neural network ensemble optimized by genetic algorithm (GABP). Through the test for forecast from January to June in 2013, results show that forecasting result of GABP ensemble has been greatly improved, whose mean square error is obviously less than that of any single model. The error distribution of GABP ensemble forecast shows relatively larger mean square error in Xinjiang and North China than in other parts, but in terms of forecast result improvement, the error of GABP ensemble has reduced more obviously compared with that of single-model forecast in western China. When several multi-model ensemble experiments are being carried out, the forecasting result of GABP ensemble is more accurate than that of linear ensemble methods. For forecast about a case of synoptic process, GABP ensemble can reveal variation trend of the predict, and its forecasting result is better than that of single-model forecast and linear-ensemble forecast. No matter for long-time or short-time synoptic process, GABP ensemble plays the best role in improving forecasting results.