Evapotranspiration (ET), defined as the process where water moves from land to the atmosphere through evaporation and plant transpiration, is crucial for understanding climate and ecosystem health. Many models have been developed to predict ET, but it’s difficult to compare results across studies because the models are deployed on different datasets with varying locations, climates, and time periods.

To address this, we propose FluxBench, a benchmark dataset and evaluation protocol developed to compare ET model performances. FluxBench uses high-quality data from flux tower sites and emphasizes transparent, well-documented preprocessing steps. We train and test a range of statistical and physics-based models, using consistent evaluation metrics to assess how well each predicts both average ET and its variability. Our results offer insights into which models perform best under different climate conditions and identify key environmental factors that drive ET.