GNSS-Based Estimation of Potential Evapotranspiration from Surface Temperature and Precipitable Water Vapor Observations: Case Study Tehran Station

Evapotranspiration (ET) is a key process in the hydrological cycle, linking land–atmosphere energy exchange and water resource management. Because direct measurement is difficult, potential evapotranspiration (PET) is often used as a proxy. Traditional empirical models, such as Penman–Monteith, estimate PET using multiple meteorological variables, but their reliance on extensive data limits application in regions with scarce observations. This study explores a local PET modeling approach using Global Navigation Satellite System (GNSS) meteorology. For this purpose, Precipitable Water Vapor (PWV) derived from GNSS observations at Tehran station was analyzed over six years. These data were combined with two-meter air temperature and PET values obtained from the evaporation pan at Mehrabad meteorological station. Four regression models: Multiple Linear Regression (MLR), Random Forest (RF), Support Vector Machine (SVM), and Long Short-Term Memory (LSTM) were implemented to estimate PET. Results demonstrated that the LSTM model consistently outperformed the others. It reduced root mean square error (RMSE) by approximately ۱۹%, ۱۶%, and ۱۴% compared with RF, MLR, and SVM, respectively. Seasonal analysis revealed stronger performance in cold seasons, with RMSE reductions of ۱۱%, ۴%, and ۸% relative to RF, MLR, and SVM. In warm seasons, improvements were even greater, at ۲۲%, ۱۹%, and ۱۶%. Overall, the proposed GNSS-based PET estimation model shows promising accuracy and robustness. Its ability to utilize PWV and limited meteorological inputs makes it particularly valuable for both research and operational applications at GNSS stations where conventional meteorological and evaporation-pan data are sparse.

کلیدواژه ها:

evapotranspiration ، Global Navigation Satellite System ، Machine Learning ، Water vapor