Multi-Lead Forecasting of Precipitable Water Vapor over Xinjiang Using Fengyun-4B Satellite and GNSS Observations

Abstract

Atmospheric water vapor is a key driver of weather variability, and accurate forecasts of precipitable water vapor (PWV) can support quantitative precipitation forecasting. Existing PWV forecasting studies have relied mainly on Global Navigation Satellite System (GNSS) observations, which provide high temporal resolution but limited spatial coverage. This study develops a regional multi-lead PWV forecasting framework for Xinjiang, China, by combining Fengyun-4B (FY-4B) Advanced Geosynchronous Radiation Imager (AGRI) observations with ground-based GNSS PWV measurements. FY-4B gridded observations were collocated with 34 GNSS stations using a KD-tree search followed by a Haversine distance calculation, yielding 332,115 valid hourly samples from June 2023 to August 2024. Predictors included geographic and calendar variables, AGRI thermal-infrared brightness temperatures, Level-2 cloud products, physically derived water-vapor features, and multiscale lagged variables. Twenty-four independent LightGBM regressors were trained for lead times from 1 to 72 h, avoiding the error accumulation inherent in autoregressive forecasting. At T + 1 h, the model achieved a mean absolute error (MAE) of 0.72 mm and a coefficient of determination (R2) of 0.92. At T + 12 h, MAE increased to 2.45 mm while R2 remained 0.47. Forecast skill declined further at T + 24 h (MAE = 5.10 mm; R2 = 0.23) and was limited at T + 72 h (MAE = 6.22 mm; R2 = 0.04). The results demonstrate that satellite-driven machine learning can provide useful very-short-range PWV guidance, particularly within the first 12 h.