Global carbon flux dataset generated by fusing remote sensing and multiple flux networks observation
Qiwang Yuan, Xufeng Wang, Tao Che, Jun Li
Abstract
We developed a global carbon flux dataset, GloFlux, using a machine learning model that integrates in situ observations from FLUXNET, AmeriFlux, ICOS, JapanFlux2024, and HBRFlux with satellite remote sensing and meteorological data. The dataset covers 2000–2023, has a 0.1 ∘ × 0. 1 ∘ spatial resolution, and monthly temporal resolution. It includes three key variables: Gross Primary Productivity (GPP), Net Ecosystem Exchange (NEE), and Ecosystem Respiration (RECO). Validation at independent flux sites not used in model training shows strong performance at the site level, with correlation coefficients of 0.84 for GPP, 0.66 for NEE, and 0.80 for RECO. The spatiotemporal patterns of GloFlux align well with existing datasets such as FLUXCOM and MODIS, supporting the reliability and robustness of the product. GloFlux offers a valuable resource for assessing global vegetation dynamics and understanding ecosystem responses to climate change.