Abstract: As an important greenhouse gas, CO2 has a significant impact on the global climate due to its concentration changes. The continuous, stable, and large-scale characteristics of satellite remote sensing make it an effective tool for monitoring atmospheric CO2. However, due to the influence of satellite payload settings and factors such as clouds and aerosols in the atmosphere, it is currently difficult for a single carbon satellite to obtain continuous high-resolution global CO2 concentration distribution information. Therefore, in order to better determine the multi-source satellite CO2 fusion method, it is necessary to analyze the uncertainty of different satellite products. This paper utilizes ground-based Total Carbon Column Observing Network (TCCON) data from 2019 to 2021 to conduct an uncertainty analysis of CO2 retrieval accuracy for the GOSAT, OCO-2, and GOSAT-2 satellites. Based on the analysis results, a global multi-source CO2 fusion model was established using the error inverse distance weighting method incorporating unit weight principles and the Kriging interpolation method. The spatiotemporal distribution patterns of the fused CO2 were then further analyzed. The analysis results show that the uncertainty of OCO-2 is the lowest, with a root mean square error ERMS of 1.10 × 10-6, followed by GOSAT with an ERMS of 1.88 × 10-6, and GOSAT2 has the highest uncertainty, with an ERMS of 3.02 × 10-6. The fusion model established has good accuracy, with a mean absolute error of 0.91 × 10-6 and a mean absolute error percentage of 0.22%. In terms of CO2 spatial distribution, it is found that the concentration of CO2 in the northern hemisphere is higher than that in the southern hemisphere, with high-value areas appearing in some regions. While in terms of seasonal changes, the CO2 concentration is higher in spring and winter than in summer and autumn, with the highest concentration in spring.

Key words: atmospheric CO2, multi-source satellite remote sensing, uncertainty analysis, fusion simulation, spatial-temporal distribution characteristics