A long-term global XCO2 dataset: Ensemble of satellite products

Many satellites such as SCIAMACHY (scanning imaging absorption spectrometer for atmospheric cartography), GOSAT-1/2 (greenhouse gases observing satellite), OCO-2/3 (orbiting carbon observatory) and TanSat (CarbonSat, Tan means "carbon" in Chinese) provide key observations of atmospheric CO2 concentration and different XCO2 retrieval algorithms have been developed for these satellite measurements. However, limited by the revisit period and scanning swath of the satellites, the effective daily observation coverage of satellites mentioned above is very small (<1%), which is a great challenge for high-resolution mapping of global XCO2. In this paper, we re-evaluated the uncertainty of each XCO2 pixel with 0.5 degrees x 0.5 degrees spatial resolution in the global based on season, AOD (aerosol optical depth), surface albedo, uncertainty parameters of SCIAMACHY, GOSAT-1/ 2, OCO-2/3 and TanSat XCO2 products by TCCON (Total Carbon Column Observing Network) data. Then, a 30day time window was used to smooth XCO2 datasets from satellite observations and filled in the missing values according to CarbonTracker. Finally, we ensembled global XCO2 datasets using maximum likelihood estimation (MLE) method and optimal interpolation (OI) based on the re-evaluated uncertainty. The ensemble XCO2 dataset at 0.5 degrees. x 0.5 degrees spatial resolution for every three hours from January 2003 to August 2020 was generated. Compared to TCCON and WDCGG (World Data Centre for Greenhouse Gases) measurements separately on the global scale, we obtained the correlation coefficient R = 0.96, the Root Mean Square Error (RMSE) of 2.62 ppm for TCCON and R = 0.82, the RMSE of 6.75 ppm for WDCGG, respectively. The RMSE of the ensemble dataset was 6 ppm lower than that of the satellites' XCO2 used for fusion compared with WDCGG and the coverage is greatly improved at a higher spatial and temporal resolution, proving the feasibility and high precision of the method.