Atmospheric precipitable water vapor (PWV) is a key variable for weather forecast and climate research. Various techniques (e.g., radiosondes, global navigation satellite system, satellite remote sensing and reanalysis products by data assimilation) can be used to measure (or retrieve) PWV. However, gathering PWV data with high spatial and temporal resolutions remains a challenge. In this study, we propose a new empirical PWV grid model (called ASV-PWV) using the zenith wet delay from the Askne model and improved by the spherical harmonic function and vertical correction. Our method is convenient and enables the user to gain PWV data with only four input parameters (e.g., the longitude and latitude, time, and atmospheric pressure of the desired position). Profiles of 20 radiosonde stations in Qinghai Tibet Plateau, China, along with the latest publicly available C-PWVC2 model are used to validate the local performance. The PWV data from ASV-PWV and C-PWVC2 is generally consistent with radiosonde (the average annual bias is −0.44 mm for ASV-PWV and −1.36 mm for C-PWVC2, the root mean square error (RMSE) is 3.44 mm for ASV-PWV and 2.51 mm for C-PWVC2, respectively). Our ASV-PWV performs better than C-PWVC2 in terms of seasonal characteristics. In general, a sound consistency exists between PWV values of ASV-PWV and the fifth generation of European Centre for Medium-Range Weather Forecasts Atmospheric Reanalysis (ERA5) (total 7381 grid points in 2020). The average annual bias and RMSE are −0.73 mm and 4.28 mm, respectively. ASV-PWV has a similar performance as ERA5 reanalysis products, indicating that ASV-PWV is a potentially alternative option for rapidly gaining PWV.

Water vapor is a critical parameter in the Earth's climate system, affecting precipitation and global warming. Precipitable Water Vapor (PWV) is a measure of atmospheric water vapor content that can be obtained by multiple methods. In this study, we construct two deep learning models and four other models to obtain high precision and spatial resolution PWV fusion product, namely convolutional neural networks (CNN), multilayer perceptron (MLP), random forest (RF), Gradient Boosting Regression (GBR), elastic network regression (ENR), and multiple linear regression (MLR). The fusion data sources are mainly from FY3D/MERSI PWV and FY4A/AGRI PWV. Moreover, we use PWV derived from 207 GNSS stations in mainland China to help with model training and testing. The experimental duration lasts two years, from May 2019 to April 2021. The results show that the CNN PWV has the best consistency with the GNSS PWV, with a correlation coefficient of 0.98, Root Mean Square Error (RMSE) of 3.01 mm, and Mean Bias (MB) of 1.00 mm. Regarding the RMSE, CNN PWV shows an improvement of 81.5% and 27.3% when compared to FY3D/MERSI PWV and FY4A/AGRI PWV, respectively. Furthermore, the water vapor maps produced by the CNN model exhibit more precise details than the original PWV products. We also find that the performance of six models has spatiotemporal characteristics, such as winter being better modelled than in summer, and eastern China being better modelled than in western China.