A machine learning framework for ERA5-PWV bias calibration

Study region This study focuses on China, and the study area is divided into seven major geographical regions: Northwest, Southwest, South, Central, East, Northeast, and North. Study focus High-precision precipitable water vapor (PWV) data with high spatiotemporal resolution is crucial for hydrometeorological research, as the PWV is a key parameter in the global water cycle and energy balance. In this study, a combined calibration framework that integrates polynomial regression (PR) and gradient boosting machine (GBM) is developed to improve the accuracy of the ERA5-PWV across China. The PR model was first applied to represent the dominant seasonal trends, and the GBM model was subsequently used to capture nonlinear residual errors. New hydrological insights for the region The results show that the PR-3 + GBM model effectively reduces systematic biases in the ERA5-PWV. Compared with the ERA5-PWV, the calibrated product achieves lower RMSEs and MAEs at annual, monthly, and seasonal scales. An analysis of the 2023 North China Doksuri rainfall event demonstrated the hydrometeorological relevance of the calibrated PWV product. In Beijing and Zhengzhou, the PWV increased during the prerainfall period, remained at high levels during continuous precipitation, and decreased after the rainfall weakened, which reflect the accumulation, maintenance, and dissipation, respectively, of atmospheric moisture associated with the rainfall process. These findings indicate that the PR-3 + GBM-PWV product can provide a more reliable representation of regional atmospheric moisture conditions, particularly in regions with extreme precipitation or sparse observations.