Carbon-water flux datasets of Eurasian meteorological stations

Mining carbon-water flux information from more widely distributed meteorological stations has great potential to deal with the challenge of insufficient observational data on these fluxes. We constructed a new framework to assess the transferability of Eurasian carbon-water flux simulation models (random forest model, RFM) to meteorological stations. We used a combination of the determination coefficient (R²) and Euclidean distance to evaluate the match between these models and meteorological station data. Not all meteorological stations could be matched with suitable RFM. Results showed that 79.3% and 98.6% of the meteorological stations (N=4478) satisfying the transfer condition (R² ≥ 0.5) could produce the carbon and water fluxes when using remote sensing (RS) variables in RFM construction, respectively. Without the use of RS variables, 68.8% and 98.9% of the meteorological stations (N=6860) satisfying the transfer condition (R² ≥ 0.5) could produce the carbon and water fluxes, respectively. The 4 spatio-temporal carbon-water flux datasets with quasi-observational characteristics that we generated at meteorological stations have great potential to improve the accuracy of assessments of ecosystem carbon-water dynamics at regional and global scales (comparison reveals possible overestimation or underestimation in datasets, e.g., FLUXCOM, GOSAT, MODIS, etc.)