Daily reference evapotranspiration prediction in Iran: A machine learning approach with ERA5-land data

Iran, characterized by diverse climatic conditions, including arid, semi-arid, and humid subtropical regions, where ET₀ dynamics vary significantly due to climatic differences. Reference evapotranspiration (ET₀) is a fundamental component of hydrological modelling and plays a critical role in agricultural water management. Reliable ET₀ predictions are essential for optimizing irrigation systems and estimating water demand. This study evaluates the potential of ERA5-Land reanalysis data, in combination with a Random Forest (RF) machine learning model, to predict daily and 8-day ET₀ across these diverse climatic conditions. Daily ET₀ values were calculated using the FAO-56 Penman-Monteith model and validated against ground-based observations from 50 weather stations (2008–2017). The RF model was trained using ERA5-Land climatic variables (air temperature, relative humidity, and ET₀ from ERA5-Land) along with the day of the year (DOY). Results demonstrated a high correlation between ERA5-Land temperature estimates and observed station data (Pearson correlation coefficient, r = 0.97; Root Mean Square Error, RMSE = 2.77°C), while relative humidity showed a weaker agreement (Normalized Root Mean Square Error, NRMSE = 21 %). The RF model outperformed traditional approaches in arid and semi-arid regions, achieving NRMSE values of 25 % and 28 %, respectively, with a 60 % improvement over humid regions. At the 8-day scale, predictive accuracy improved further (RMSE = 6.05 mm/8 days, r = 0.99). Beyond model performance, this study provides new insights into the spatiotemporal variability of ET₀ across different climatic zones. The findings indicate that temperature is the dominant climatic factor driving ET₀ variability, while relative humidity exhibits higher uncertainty, particularly in humid regions. Seasonal trends highlight notable summer ET₀ peaks exceeding 30 mm/day in arid zones, emphasizing the need for climate-adaptive irrigation strategies. The proposed methodology is computationally efficient, requiring minimal input variables, and demonstrates robust and scalable performance for large-scale ET₀ estimation. These findings provide a cost-effective solution for water resource management, drought monitoring, and climate change adaptation, particularly in data-scarce regions. • Random Forest model accurately predicted 1- and 8-day ET0in Iran using ERA5-Land reanalysis data. • ERA5-Land air temperature (T ERA5-Land ) data correlated strongly with ground-based measurements, unlike relative humidity. • Predicted ET 0 varied seasonally and annually, reflecting climate factors' influence. • Integrating machine learning and climate data improves water resource management and hydrological modeling, especially in data-sparse regions.