An advanced regional integrated water vapor estimation model utilizing least squares support vector machine for the upper Rhine graben region

Integrated Water Vapor (IWV) is crucial in environmental research, offering insights into atmospheric dynamics. Direct IWV measurement is challenging, necessitating alternative estimation technologies. Existing methods including Navigation Satellite System (GNSS), radiosondes, water vapor radiometers (WVR), satellite remote sensing, and numerical weather models (NWM), have specific limitations. GNSS and WVR provide high precision and temporal resolution (e.g., 5 min) but are limited to specific locations. Radiosondes, while accurate, have sparse spatial distribution and low temporal resolution (e.g., twice daily). Satellite remote sensing offers broad spatial resolution but lower temporal resolution (hours to days) and reduced accuracy under cloudy conditions and due to satellite tracks. NWMs provide global hourly products but their accuracy depends on meteorological data and model precision. This study introduces a regional IWV predictive model using Machine Learning to address these challenges. Utilizing IWV data from GNSS stations, the study develops a predictive model based on least squares support vector machine, which autonomously determines optimal parameters to enhance performance. ... mehrThe model enables accurate IWV estimation at any location within a region, using inputs such as latitude, longitude, altitude, and temperature, achieving an average root mean square error of 0.95 mm. The model’s performance varies across seasons and terrains, showing adaptability to diverse conditions. The model’s reliability is validated by comparing its predictions with the conventional ERA5 IWV method, showing a 61% improvement rate. This refined IWV estimation model is applied for regional climate analysis, demonstrating its practical utility in environmental research, specifically for the Upper Rhine Graben Region.