Evaluation of the CMA-GD(R3): Hourly Precipitation Forecasts in a Tropical Island

Given the intent of the model for use in fine precipitation forecasting with adequate lead time, it is of interest to examine the way in which forecasts of warm season precipitation over Hainan Island behave as a result of the rapidly updated hourly model throughout 24 initializations. Standard metrics were performed on the quantitative precipitation forecasts (QPFs) from CMA-GD(R3), a global/regional assimilation and prediction system, and the cycle of the hourly assimilation and forecast system, in April–September of 2022–2024 over Hainan Island, to understand the way in which the seasonal, diurnal, and spatial variations change in relation to decreasing forecast lead times. The findings indicated that "rain or no rain" QPFs generally exhibited a superior performance at short lead times, with the 1-h forecast being optimal. As the rainfall threshold increased, the forecast performance deteriorated in the first five forecast hours, whereas the nighttime QPF skill often exceeded the daytime performance. Significant instabilities were identified in the mountainous region and the surrounding areas when 24 forecasts that were valid at the same forecast hour were compared. The 1-h model update enhanced the reliability of the 2-m temperature, 2-m dew point, and 10-m wind field across most of the island. However, persistent biases remained, including underestimation of temperature and dew point at night and in central areas, and overestimation of wind speed with strengthened afternoon wind convergence. These biases contributed to uncertainty in the model's prediction of rainfall events over Hainan Island.