The global carbon dioxide (CO₂) concentration has shown a consistent and substantial increase over the years, representing the dominant component of greenhouse gases (GHGs). Hence, there is an urgent demand to accurately quantify a broad spectrum of CO₂ concentration at a fine-scale level to aid policymakers in making informed decisions. Consequently, we present a novel method aimed at addressing the scarcity of ground-based data, enabling the generation of a globally large-scale, continuous CO₂ concentration data product.

In consideration of the requirements for temporal and spatial coverage of remote sensing imagery, we opt for the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the Terra satellite, which provides daily surface reflectance of MODIS bands 1 to 7 at resolutions of 500m and 1km.

Carbon satellites have developed rapidly and performed well in retrieving the vertically integrated atmospheric column CO₂(XCO₂) concentrations, which can provide independent top-down CO₂ concentration evaluations. Here, the new generated Orbiting Carbon Observatory 3 (OCO-3) with 1.6 km×2.2 km (across × along track) resolution is added three Near Infrared (NIR) wavelength bands, which guarantees a higher accuracy of XCO₂ than OCO-2.

In this study, we propose a regression model-based method that leverages MODIS data and OCO-3 XCO₂ data for training regression models and predicting CO₂ concentrations. The proposed method enables rapid establishment of the relationship between MODIS surface reflectivity and CO₂ concentration, facilitating the generation of continuous CO₂ concentration maps over a large geographical area. Moreover, it offers reliable information for regions lacking ground-based CO₂ measurements, such as suburban areas.

Additionally, to validate the accuracy of the generated XCO2 data product, we utilize the Total Carbon Column Observing Network (TCCON) as an essential validation source. Upon evaluation, it was observed that the relative errors for each month of the year 2020 at the respective TCCON  sites consistently remained below 2%.  This finding suggests that the proposed method possesses the potential for expansion to additional geographical regions and temporal spans, whilst sustaining a high level of precision.