Enhancing reference evapotranspiration estimate using regression kriging in semi-arid climate

Study region Using weather data from 340 meteorological stations in Iran spanning 2011–2021, we conducted a comprehensive mapping of reference evapotranspiration (ET₀) across the country. This extensive network of stations spans diverse climatic and environmental conditions, enabling us to evaluate the influence of meteorological and geographic drivers on the precision of ET₀ estimates across various regions of Iran. Study focus Reference evapotranspiration (ET₀) is a crucial variable in hydrologic, environmental, ecological, and meteorological modeling, and its large-scale mapping is essential for agricultural optimization and water resources management. Although most studies map ET₀ at large scales based on point-scale measurements, this study integrates regression–kriging (RK) methods with meteorological and coordinate data to produce more accurate ET₀ maps. Three RK variants are developed, including generalized linear model–ordinary kriging (GLM+OK), generalized additive model–ordinary kriging (GAM+OK), and random forest–ordinary kriging (RF+OK) for the first time for large-scale ET₀ mapping. The ReliefF algorithm is used to identify influential input variables, and different strategies are defined based on geographical coordinates, meteorological variables, and point-scale ET₀ measurements. The proposed models are then compared with standalone GLM, GAM, RF, and OK ones. New hydrological insights for the region The GAM+OK hybrid model achieved the highest accuracy (testing R² = 0.932) by explicitly capturing nonlinear spatial patterns and residual autocorrelation across Iran’s hyper-arid interior, semi-arid plateaus, and humid coastal zones. ReliefF-guided feature selection demonstrated that omitting wind speed markedly degrades ET₀ estimates in coastal and island regions, where sea-breeze-driven advection dominates evaporative demand, whereas coordinate information (latitude and elevation) is indispensable for inland orographic gradients. These findings provided a high-resolution, uncertainty-quantified framework for refining regional water-balance assessments, optimizing irrigation scheduling in water-scarce agricultural basins, and enhancing drought-early-warning systems under climate variability, offering hydrological insights for sustainable water-resources management in arid regions.