Abstract:

Based on seven kinds of numerical weather prediction (NWP) products, the binary test and frequency distribution test methods were used to test and evaluate the 24 h precipitation forecast and the 3 h heavy precipitation forecast with more than 20 mm precipitation in Liaoning province from May to September of 2020.The results show that most NWP models have the error characteristic of excessive precipitation forecast frequency, especially the global model, which is more obvious for the excessive forecast of the weak precipitation process.The prediction effect of the numerical model is generally good for the 24 h moderate rainfall, while the global model is poor for heavy rainfall.The forecast intensity of extreme precipitation in the SHANGHAI mesoscale model is higher than the observations.The ECMWF (European Centre for Medium-Range Weather Forecasts) model has the balanced performance of different grades precipitation forecasting.It also has some ability to forecast heavy precipitation.The mesoscale model has a good forecast effect on heavy precipitation during the daytime, but the forecast effect becomes worse at 6-9 h after the initial time.The NWP models have a good forecasting effect on the large-scale heavy precipitation process and the small-scale stable heavy precipitation process, but a poor forecasting effect on the small-scale heavy precipitation process behind the subtropical high.

Key words: Global model, Meso-scale model, Precipitation forecasting, Assessment of models