Golkar et al. (2026) Diagnosing and correcting openet satellite evapotranspiration across us climate zones: Error decomposition, explainable AI, and uncertainty quantification

Identification

• Journal: Agricultural and Forest Meteorology
• Year: 2026
• Date: 2026-10-13
• Authors: Mohammad Kian Golkar, Luciano Alves de Oliveira
• DOI: 10.1016/j.agrformet.2026.111524

Research Groups

• Department of Agricultural and Biological Engineering, Grainger College of Engineering, University of Illinois at Urbana-Champaign

Short Summary

This study develops an XGBoost correction framework that integrates OpenET satellite outputs with ground meteorological observations to improve satellite evapotranspiration (ET) accuracy across US climate zones. The framework achieves a 41.9% RMSE improvement over OpenET and provides mechanistic understanding of the correction mechanisms.

Objective

• Develop and validate an XGBoost correction model for OpenET satellite ET using 22 AmeriFlux eddy covariance sites spanning six K¨oppen climate zones.
• Decompose the improvement in mean squared error into bias and variance components across climate zones and seasons to understand the correction mechanisms.

Study Configuration

• Spatial Scale: Field scale (30 m) with a focus on croplands
• Temporal Scale: Daily ET estimates from 2020-2025

Methodology and Data

• Models used: XGBoost, LightGBM, CatBoost, Gradient Boosting, Random Forest, feedforward neural networks, and LSTM architectures
• Data sources: OpenET satellite retrievals, AmeriFlux eddy covariance measurements, MODIS NDVI, gridMET precipitation, and reference ET

Main Results

• The XGBoost correction framework achieves a 41.9% RMSE improvement over OpenET.
• Error decomposition reveals that the dominant correction mechanism shifts seasonally: summer corrections are driven primarily by bias removal (61.3%), while winter corrections are driven by variance reduction (96.8%).
• SHAP analysis shows that different features drive corrections in humid versus water limited zones and across seasons.

Contributions

• This study provides a mechanistic understanding of the correction mechanisms for satellite ET.
• The XGBoost correction framework can be used to improve satellite ET accuracy across US climate zones.

Funding

• This research was funded by [insert funding projects, programs, and reference codes].

Citation

@article{Golkar2026Diagnosing,
  author = {Golkar, Mohammad Kian and Oliveira, Luciano Alves de},
  title = {Diagnosing and correcting openet satellite evapotranspiration across us climate zones: Error decomposition, explainable AI, and uncertainty quantification},
  journal = {Agricultural and Forest Meteorology},
  year = {2026},
  doi = {10.1016/j.agrformet.2026.111524},
  url = {https://doi.org/10.1016/j.agrformet.2026.111524}
}