Daily Evapotranspiration Estimation Across Turkiye with Limited Sensor Data and Comparative Analysis of Regions Using Machine Learning

Accurate estimation of reference evapotranspiration (ET0) is crucial for efficient irrigation scheduling and sustainable water management. While the FAO-56 Penman-Monteith equation remains the standard method, it requires a full suite of meteorological inputs, which are often unavailable in many regions. This study proposes a data-driven framework using the Light Gradient Boosting Machine algorithm to identify and exploit the most relevant meteorological features for ET0 estimation. Leveraging a large-scale dataset collected from 444 stations across Türkiye between 2014 and 2019, we evaluate model performance using both complete and reduced feature sets. Our findings show that accurate ET0 predictions can be achieved using a minimal subset of commonly available variables such as temperature, humidity, wind speed, and temporal features. Moreover, we assess regional variability by comparing global and region-specific models, revealing that a globally trained model performs competitively across diverse climatic zones. This approach enables practical ET0 estimation in sensor-limited environments, offering scalable support for data-driven agricultural planning and resource-efficient irrigation.