A Data Fusion Framework to Gap Fill and Improve Satellite-based Evapotranspiration: Integrating Eddy Covariance and CubeSat Observations with Machine Learning

This repository contains the datasets, source code, and supporting materials developed for the manuscript "Improving Daily Evapotranspiration Estimates in Dryland Agricultural Systems Using Multi-Sensor Machine Learning and Data Fusion." The study evaluates the performance of six OpenET evapotranspiration (ET) models using eddy covariance (EC) observations collected from eight flux tower sites across dryland agricultural systems in the United States. The analysis demonstrates that OpenET products tend to overestimate daily ET, particularly during the growing season and under water-limited conditions. To address these limitations, the study integrates high-resolution remote sensing data from Landsat 8/9, Sentinel-2, and PlanetScope with DAYMET climate variables and tests multiple machine learning models, including Linear Regression, Random Forest, Support Vector Machine, XGBoost, and Artificial Neural Networks, to improve daily ET estimation. A hierarchical multi-sensor data fusion framework was subsequently developed to gap-fill Landsat-based OpenET estimates, producing more accurate and temporally continuous daily ET time series. The repository includes all datasets and materials required to reproduce the analyses presented in the manuscript, including Python notebooks, Excel datasets, processed machine learning inputs and outputs, and all figures and tables generated for the manuscript. The repository is intended to support transparency, reproducibility, and future research on satellite-based evapotranspiration estimation, machine learning, and agricultural water management. Acknowledgements The first author acknowledges the US-PAK Knowledge Corridor Funding provided by the Higher Education Commission (HEC), Pakistan. Funding from the United States Bureau of Reclamation (USBR Award R22AP00238-00), USDA-ARS (Project No. 3070-21500-001-000D), and the USDA’s Long-Term Agroecosystem Research (LTAR) network is gratefully acknowledged.