A study on AI-based real-time estimation of precipitable water vapor using GNSS and satellite brightness temperatures

Abstract

Precipitable water vapor (PWV) is a key variable for precipitation forecasting in both numerical and data-driven AI-based weather prediction. Real-time operation of data-driven AI forecasts requires rapid assimilation of observational data to update spatial at-mospheric fields. In this study, we propose a method to assimilate PWV derived from GNSS observations and Himawari satellite brightness temperatures into a data-driven forecasting framework based on the Adaptive Fourier Neural Operator (AFNO), and eval-uate its effectiveness. Both GNSS- and satellite-derived PWV can be obtained within 30 minutes, enabling real-time operation. Results show that GNSS-derived PWV exhibits smaller observational errors relative to model and background errors, whereas PWV es-timated from satellite brightness temperatures shows larger errors, particularly over oce-anic regions. Reducing satellite-derived PWV errors is therefore essential for improving real-time AFNO-based atmospheric prediction.