Applications of Machine Learning in Enhancing Evaporation Estimation for Small Reservoirs: A Case Study in Semi-Arid South Texas

To develop a multi-reservoir machine learning (ML) modeling framework to enhance daily evaporation estimation for small reservoirs in South Texas. ➢ To benchmark and compare conventional empirical evaporation models. ➢ To validate ML modeling framework for accuracy, generalizability, and transferability using independent datasets. Evaporation is a dominant pathway of water loss in arid and semiarid regions, significantly reducing storage in lakes, reservoirs, and irrigation canals. Small reservoirs in South Texas experience high evaporative losses (Wurbs & Ayala, 2014), but are seldom monitored at daily scales, which limits water management decisions. To address this gap, we developed a multi-reservoir machine learning (ML) framework trained with datasets from the Texas A&M Daily Lake Evaporation Model (DLEM) to enhance daily evaporation estimation. Predictors combine gridded meteorological data from gridMET with reservoirspecific attributes. ML models are evaluated for accuracy, generalizability, and cross-reservoir transferability using empirical model outputs, station observations, and standard performance metrics. The findings contribute to the development of improved evaporation monitoring strategies for data-scarce semi-arid regions.