Evapotranspiration estimation at different land surface scales in semi-arid areas using gene expression programming and the FAO56 Penman-Monteith model

This study was conducted at the Agula Eco-hydrological Experimental Station, located in the southeastern part of the Horqin Sandy Land, Tongliao City, Inner Mongolia, North China (43°18’48’’–43°21’2’’N, 122°32’30’’–122°41’00’’E). To enhance ET prediction, this study modifies the FAO56 Penman-Monteith model by integrating Gene Expression Programming (GEP), developing the GEP-PM model. The model improves ET estimation by recombining the radiation and aerodynamic terms and optimizing parameters for different surface types. The study focuses on two contrasting land surfaces—sand dunes and meadows—where the model is calibrated using a cost function minimization approach to capture dynamic ET variations at a 30-minute time scale. The model’s performance was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), the coefficient of determination (R²), and Nash-Sutcliffe efficiency coefficient (NSE). Results demonstrate that the GEP-PM model provides accurate ET predictions across both sand dunes and meadows, with R² values ranging from 0.71 to 0.83, RMSE from 0.04 to 0.07 mm/h, MAE from 0.02 to 0.05 mm/h, and NSE from 0.72 to 0.86. ET rates were observed between 0 and 0.8 mm/day on meadows and 0 to 0.6 mm/day on sand dunes, highlighting the model’s adaptability to heterogeneous land surface conditions. These findings provide valuable insights into semi-arid water balance dynamics, supporting sustainable water resource management, precision irrigation planning, and hydrological modeling in dryland ecosystems. • Optimized the FAO56 Penman-Monteith model using GEP to improve evapotranspiration estimation in semi-arid regions. • Developed separate GEP-PM models for different surfaces, incorporating distinct landscape community characteristics. • Built models based on 30-minute time steps to capture fine evapotranspiration changes, enhancing model accuracy.