Forecasting Capability Verification of the Pangu-Weather and IFS HRES for the 2022 Summer Weather in Jiangsu Province, China

Abstract In 2022, Huawei Cloud developed the Pangu-Weather (Pangu) model and released parts of the forecast model on GitHub in the following year. In this study, we conduct an experiment using the publicly available Pangu models for the 2022 summer weather prediction in Jiangsu Province, China. We assess the forecasting capability of Pangu for both surface variables and upper-level circulation patterns. This study aims to comprehensively evaluate whether the Pangu model outperforms the Integrated Forecasting System High-Resolution Ensemble Prediction System (IFS HRES) in the operational forecasts in Jiangsu Province. Moreover, this evaluation experiment encompasses an assessment of the capacities of the respective models to forecast upper-level circulation patterns. The results show that the Pangu model demonstrates superior performance on the forecasts of 2-m temperature with the IFS HRES, while in the forecast of 10-m wind speed, Pangu shows some degree of improvement over IFS HRES. Regarding upper-level circulation pattern forecasts, the results of the Pangu model are particularly similar to those of the IFS HRES. For instance, both the Pangu and IFS HRES accurately forecast the 588-dagpm contour line and the 500-hPa isobar, without obvious deviations relative to observations. Furthermore, Pangu displays better performance on upper-level wind field forecasts in complex weather situations. Additionally, Pangu-Weather can be initialized every hour, thereby offering more frequent time series in daily operational forecasts, which enhances the performance of objective algorithms. However, Pangu-Weather exhibits higher biases in forecasting extreme events compared to the IFS HRES, highlighting a critical area for future model improvement. Significance Statement The objective of this study is to assess the forecasting performance of the Pangu-Weather (Pangu) model developed based on artificial intelligence big model technology and fifth major global reanalysis produced by ECMWF (ERA5) reanalysis data, utilizing observations from over 1700 automatic meteorological stations established and maintained by the Jiangsu Provincial Meteorological Bureau. Additionally, this study aims to compare the Pangu model with the numerical weather prediction model Integrated Forecasting System High-Resolution Ensemble Prediction System (IFS HRES) from the European Centre for Medium-Range Weather Forecasts (ECMWF) commonly used in the meteorological field, to evaluate the practical value of the Pangu model in operational weather forecasts. The results indicate that the Pangu model provides accurate meteorological variable fields for Jiangsu Province comparable to the IFS HRES forecast products. Furthermore, Pangu can generate hourly rolling forecasts, thereby offering data support and service with higher temporal resolution.