Mass-conservation datasets of global terrestrial gross primary production, net ecosystem exchange and ecosystem respiration (2001-2020)

Gross primary production (GPP), net ecosystem exchange (NEE), and ecosystem respiration (RECO) are critical components of the carbon cycle in global terrestrial ecosystems. Satellite remote sensing provides a unique means to retrieve these three carbon fluxes at regional and global scales. However, accurately reproducing global terrestrial ecosystem carbon fluxes while meeting carbon budget balance remains a significant challenge. We generated the first mass conservation-constrained data-driven datasets of global 0.05° daily GPP, NEE and RECO from 2001 to 2020 using our developed Co-Ecocarbon model based on the multivariate random forest (MRF) algorithm, remote-sensing data (e.g., GLASS, MODIS, GOSIF), ERA5-Land reanalysis datasets, and in-situ observations from the FLUXNET and AmeriFlux networks. The resulting flux estimates are internally consistent and strictly preserve carbon mass balance, satisfying the principle GPP+NEE−RECO=0. Model performance was evaluated against in-situ observations from 356 globally distributed sites in the FLUXNET2015 and AmeriFlux FLUXNET datasets by a spatial 10-fold cross-validation scheme, yielding root mean square errors (RMSE) of 2.21 gC·m⁻²·d⁻¹, 1.93 gC·m⁻²·d⁻¹, and 1.79 gC·m⁻²·d⁻¹ for GPP, NEE, and RECO, respectively. The improved carbon flux estimates under the constraint of mass conservation can not only enhance the understanding of terrestrial carbon dynamics but also provide a more robust foundation for carbon accounting and carbon trading, which is essential for tracking ecosystem carbon budget and informing carbon management strategies.