Abstract:Based on the European Center for Medium-range Weather forecast (ECMWF) model and the Regional Mesoscale Model for South China (GZ 3 km), clear or rain daily forecast in South China in 2019 was analyzed, the performance of the model prediction was discussed. With the use of TS evaluation method, the characteristics of the model rainfall forecast and the statistical law of the model rain or clear forecast are found in comparison with simple extrapolation forecast. For instance, the case that two models forecast clear simultaneously, or model predict rainy and the predicted precipitation exceed a certain threshold can reach very high accuracy. However, there are some uncertainties when the two models do not agree, or when the model predict rain with small rainfall. Based on the above research, the method of using improved model-based observation extrapolation and using frequency matching between observation and model to improve model prediction is proposed.