Abstract:The NJU 4 km forecasting system (WRF_NJU) produce 48 h real-time forecasts at a 4 km grid spacing over China twice daily for the summer seasons at Nanjing University since 2013. The precipitation forecast from June to August in the summer of 2016-2018 in North and South Xinjiang from this model, the European Center for medium range weather forecast (EC), China Meteorological Administration (CMA) and Japan Meteorological Agency (JMA)were compared with the observational data of automatic weather stations. Results show that: generally, Gilbert Skill Score (GSS) of EC is the highest, WRF_NJU is second and JMA is the lowest. In terms of spatial distribution, the forecast in North and South Xinjiang from EC is the closest to observation, and the average precipitation forecast from WRF_NJU is lower, while CMA and JMA are obviously higher. The diurnal variations of precipitation forecast from WRF_NJU is consistent with the observed variations. The peak period is basically consistent, with a slight deviation of 1-2 hours. Forecast is generally higher than observation, especially in peak period. In the aspect of precipitation episodes and propagation, WRF_NJU can simulate most precipitation processes well. After studying selected cases, it is found that the precipitation forecast is directly affected by the forecast of upper-level jet stream and lower-level water vapor transport.