Data-Driven Approaches for Improved Evapotranspiration Modelling with Limited Data

Abstract

This study uses data-driven methods to estimate FAO Penman-Monteith Reference Evapotranspiration (ETo) using only temperature data. Reference evapotranspiration, as an important variable for estimating actual evapotranspiration, is crucial in various water management tasks. However, some data for the Penman-Monteith equation is often unavailable. Thus, the need to use alternative methods emerges. The research shows DDM's effectiveness particularly when feature engineering was used. The study tested standard equations (Hargreaves Samani) and a proposed CatBOOST model with feature engineering to model ETo. The CatBOOST model achieved a higher R2 of 0.94 than the standard equations' R2 of 0.86. This result underscores DDM’s potential to refine evapotranspiration modelling for wide applications in water resource management, irrigation, and agriculture.