Using Geostationary Satellite Observations and Machine Learning Models to Estimate Ecosystem Carbon Uptake and Respiration at Half Hourly Time Steps at Eddy Covariance Sites

Abstract

Abstract Polar‐orbiting satellites have significantly improved our understanding of the terrestrial carbon cycle, yet they are not designed to observe sub‐daily dynamics that can provide unique insight into carbon cycle processes. Geostationary satellites offer remote sensing capabilities at temporal resolutions of 5‐min, or even less. This study explores the use of geostationary satellite data acquired by the Geostationary Operational Environmental Satellite—R Series (GOES‐R) to estimate terrestrial gross primary productivity (GPP) and ecosystem respiration (RECO) using machine learning. We collected and processed data from 126 AmeriFlux eddy covariance towers in the Contiguous United States synchronized with imagery from the GOES‐R Advanced Baseline Imager (ABI) from 2017 to 2022 to develop ML models and assess their performance. Tree‐based ensemble regressions showed promising performance for predicting GPP (R2 of 0.70 ± 0.11 and RMSE of 4.04 ± 1.65 μmol m−2 s−1) and RECO (R2 of 0.77 ± 0.10 and RMSE of 0.90 ± 0.49 μmol m−2 s−1) on a half‐hourly time step using GOES‐R surface products and top‐of‐atmosphere observations. Our findings align with global efforts to utilize geostationary satellites to improve carbon flux estimation and provide insight into how to estimate terrestrial carbon dioxide fluxes in near‐real time.

Keywords

• carbon cycle
• ecosystem respiration
• geostationary satellite
• gross primary productivity
• machine learning