High-accuracy, real-time precipitable water vapor (PWV) products with high spatiotemporal resolution are of great significance for precipitation forecasting, extreme weather monitoring and early warning. Global reanalysis data and Global Navigation Satellite System (GNSS), face limitations in providing real-time, high spatiotemporal resolution PWV products owing to the latency of reanalysis data and the uneven distribution of GNSS stations. To overcome these challenges, a novel PWV retrieval approach was developed. The Huawei Cloud Pangu-Weather system was introduced to predict hourly real-time PWV globally, namely Pangu PWV, with a horizontal resolution of 0.25° from 2018 to 2020. The estimation result of Pangu PWV was evaluated using PWV dataset from 540 radiosonde stations and 12,292 GPS stations worldwide. In this work, the gridded PWV product from MERRA-2 was incorporated for comparison. To mitigate evaluation uncertainties, the previously developed Global PWV Vertical Correction Model (GPWV-H) was employed to adjust the gridded PWV for both products. The mean bias and root mean square error (RMSE) for MERRA-2 and Pangu PWV are -2.59/0.59 mm and 3.67/3.67 mm against radiosonde records, respectively, and mean bias and RMSE are -2.20/-0.01 mm and 3.17/3.37 mm against GPS PWV data, respectively. In addition, Pangu PWV has close performance to MERRA-2 PWV, further demonstrating that Pangu PWV is well-suited for meteorological applications. Besides, the performance of the two products was tested under extreme weather conditions in a specific region. The mean bias and RMSE for MERRA-2 and Pangu PWV are -6.14/0.19 mm and 7.05/3.76 mm, respectively, indicating that the Pangu PWV still yield stable performance during extreme weather events. Thus, the Pangu PWV, exhibiting the merits such as real-time, great precision, and high spatiotemporal resolution on a global scale, holds great potential for applications in extreme weather monitoring and precipitation forecasting.

Global Navigation Satellite Systems (GNSS) provide an effective means for remote sensing of precipitable water vapor (PWV). However, conventional GNSS‐based PWV retrieval approaches rely heavily on auxiliary meteorological parameters, which are frequently unavailable in real time, while most direct retrieval methods use station data aggregated over large regions, overlooking local characteristics. To address these issues, this study proposes three innovative window‐augmented machine learning models—RF‐G, BP‐S‐G, and BP‐D‐G—that directly estimate PWV from GNSS signals without dependence on meteorological data. They were trained using zenith tropospheric delay (ZTD) and PWV data from the ERA5 reanalysis, requiring only GNSS station coordinates, observation times, and ZTD values as inputs. Validation was performed using PWV derived from GNSS and meteorological observations (GNSS‐M PWV) and radiosonde observations in China. Compared with the previously developed D‐PWV model, the proposed models achieved superior accuracy. Using GNSS‐M PWV as reference, the RMSEs for the D‐PWV, RF‐G, BP‐S‐G, and BP‐D‐G models were 2.00 mm, 1.67 mm, 1.88 mm, and 1.74 mm, representing improvements of 16.5%, 6%, and 13%, respectively. Against radiosonde data, the RMSEs were 3.22 mm, 2.62 mm, 3.07 mm, and 2.92 mm, corresponding to improvements of 18.6%, 4.7%, and 9%. To further examine spatial transferability, the three models were additionally evaluated over Australia using GNSS‐M PWV as reference, yielding RMSEs of 2.60 mm, 3.49 mm, and 2.92 mm. Furthermore, the models maintained stable performance during extreme weather events. These results demonstrate that the proposed models enable accurate GNSS PWV retrieval while eliminating dependence on meteorological parameters.

Accurate prediction of near-surface meteorological variables is important for weather services and coastal risk management. However, the station-level performance of global artificial intelligence (AI) weather models remains insufficiently characterized in complex coastal environments. This study evaluated Pangu-Weather, FengWu, FuXi, and the Global Forecast System (GFS) against observations from 210 stations in eastern coastal China from July to December 2022. The assessment focused specifically on 2 m temperature, surface pressure, 10 m wind speed, and wind direction across forecast lead times, stations, and routine and typhoon conditions. FuXi had the lowest temperature RMSE (1.70 °C), whereas FengWu had the lowest pressure and wind-speed RMSE values (0.89 hPa and 1.17 m/s, respectively). The models showed distinct spatial error patterns, and wind-speed errors were concentrated at several northern coastal and transition-zone stations. During Typhoon Muifa, errors increased for all models, with the largest deterioration occurring for wind speed. FengWu retained the lowest typhoon-period wind-speed RMSE (1.87 m/s), whereas GFS had the largest value (3.03 m/s). Wind-direction distributions remained difficult for all models to reproduce. These results support variable-specific model selection, but they should not be interpreted as a general ranking of atmospheric forecast systems because the validation is limited to near-surface station data and a six-month period.