MidwestET-500: Upscaled Daily Evapotranspiration Estimates for the U.S. Midwest (2019–2024)

Abstract

MidwestET-500 is a daily 500 m resolution evapotranspiration (ET) dataset covering the U.S. Upper Midwest (36°N–49°N, −104°W to −82°W) from 2019 to 2024. It was generated using a knowledge-guided LightGBM model trained on eddy covariance data, meteorology, and remote sensing inputs. The model integrates physically informed features, including Penman–Monteith components, and was validated via stratified GroupKFold to ensure robust generalization across space and time. MidwestET-500 shows strong agreement with OpenET ensemble data (r = 0.95) and independent Penman–Monteith estimates (r = 0.90). The dataset is released in NetCDF4 format and is intended to support applications in irrigation planning, drought monitoring, and regional water modeling. All model code and preprocessing workflows are openly available to promote reproducibility.