Using Machine Learning Method to Estimate Evapotranspiration (Case Study: Semnan Province)

Document Type : Research Paper


1 Department of Desertification, Faculty of Desert Studies, Semnan University, Semnan, Iran.

2 Department of Desertification, Faculty of Desert Studies, Semnan University, Semnan, Iran


Evapotranspiration (ET) plays a crucial role in arid and semi-arid regions, and its precise estimation of ET is essential for effective irrigation planning and management. Semnan province faces challenges due to a scarcity of synoptic and evaporation stations, making spatial estimation of ET difficult. This study utilized the evapotranspiration product from the ERA5-Land reanalysis dataset, in conjunction with auxiliary variables such as elevation and temperature, to estimate ET in the study area. The Random Forest (RF) model was employed to establish the relationship between auxiliary variables and ET data, resulting in the creation of an ET map using the RF model. The accuracy of the RF model in estimating ET was assessed against observational data using four statistical criteria: R², BIAS, RMSE, and KGE. The validation results demonstrated the high efficiency of the RF model (R² = 0.95, BIAS = -4.1, RMSE = 98.6, and KGE = 0.92). It was observed that the RF model's performance in ET estimation is influenced by the relationship between model error (BIAS) and topography, with elevation playing a significant role in ET estimation accuracy. This study underscores the effectiveness of utilizing data mining and processing techniques within the R programming environment to accurately estimate ET in regions with limited weather stations, particularly in arid and semi-arid areas. By leveraging these methods, it becomes possible to enhance the estimation of evapotranspiration in such challenging environments.


Main Subjects