Real-Time High-Resolution Global PWV Retrieval Based on Weather Forecast Foundation Models and Cross-Validation With Radiosonde, GNSS, and ERA5

Abstract

High-quality precipitable water vapor (PWV) plays a vital role in climate change and weather prediction studies. This research introduces a novel scheme for retrieving high-resolution surface-domain PWV with real-time and forecasting capabilities with global coverage, utilizing weather forecast foundation models represented by Huawei Cloud Pangu-Weather, Google DeepMind GraphCast, and Shanghai AI Lab FengWu. The accuracy of the new scheme is cross-validated against PWVs from radiosondes, Global Navigation Satellite Systems (GNSSs), and the fifth generation ECMWF reanalysis (ERA5). Results show the new scheme achieves 3.01 mm global root mean square error (RMSE) in real-time, and the value reduce to 2.25 mm when focusing only on land areas, which is more accurate than most existing methods that rely on post-processed surface-domain data. The poor accuracy in low-latitude and mid-latitude ocean regions limits the accuracy of the new scheme and future integration of GNSS PWV data from ocean sources is expected to improve it. Overall, the proposed scheme demonstrates very satisfactory global PWV accuracy and has the potential for further improvement with the development of artificial intelligence.