Stacking machine learning model for precipitable water vapor vertical adjustment using GNSS networks and radio occultation data

Abstract

This study presents a stacking machine learning (SML) model for vertical adjustment of precipitable water vapor (PWV), addressing missing water vapor information near the surface in radio occultation (RO) profiles and enhancing the accuracy of PWV estimation from RO data. The model is trained and validated using more than 1500 ground-based Global Navigation Satellite System (GNSS) stations and more than 320,000 RO profiles of the Constellation Observing System for Meteorology, Ionosphere, and Climate-2 (COSMIC-2) for two regions of the Northern hemisphere from January 2020 to December 2023. Results show that in the North American region, the SML model reduces the root-mean-square error (RMSE) of PWV estimates by 58.05 %, 36.99 %, and 33.05 % compared to conventional linear, exponential, and global PWV vertical adjustment (GPWV-H) models, respectively. In the region of China and Southeast Asia, the RMSE of PWV estimates is reduced by more than 42.9 %. External validation reveals that the SML-adjusted RO-PWV is in close agreement with PWV estimated from radiosondes and other RO products. Notably, the SML model outperforms conventional models across various latitudes and longitudes, making it well-suited for complex terrain and different climatic conditions. This study also examines the SML model performance for different climate types and extreme weather and proposes incorporating these factors in future work to improve model adaptability. Overall, the SML model excels in PWV vertical adjustment, providing a high-accuracy, fast solution for global PWV estimation, water vapor monitoring and weather forecasting.