How Can Satellites Map Carbon Emissions in Cities?

Carbon‑observing satellites act like eyes in space, measuring CO₂ concentrations worldwide. To capture what really happens in cities, finer detail is needed. Researchers built a model that merges satellite data with ground‑level information such as points of interest, road networks, and population distribution. Using K‑means clustering, they linked XCO₂ anomalies (the amount of CO₂ in the atmospheric column) to these urban variables. The outcome is a high‑resolution map with 1 km × 1 km cells, highlighting emission hotspots. In Urumqi, 2020 saw average XCO₂ levels between 408 and 415 ppm, fluctuating from –1.58 to +1.13 ppm. The model estimated roughly 58.3 million tonnes of CO₂, matching EDGAR data with monthly errors within ±10 %. Compared with the ODIAC dataset, the correlation was low (r = 0.192), yet the most polluted zones aligned with industrial parks and commercial centers, providing a more realistic picture of urban emissions. This approach shows that satellites, when combined with local data, can become powerful tools for monitoring and managing city‑scale carbon emissions.

The proposed model hinges on merging top‑down (satellite) and bottom‑up (urban proxies) data. Satellite XCO₂ measurements come from missions like OCO‑2 and GOSAT, which capture the total atmospheric CO₂ column with precision of a few ppm. However, their spatial resolution—typically 10‑25 km—is too coarse to resolve emissions at the neighborhood level. To bridge this gap, researchers introduced high‑density variables: points of interest (POI) representing commercial, industrial, and service activities; road networks as a proxy for vehicular traffic; and population distribution indicating residential energy demand. K‑means clustering was applied to a matrix combining XCO₂ anomalies with these proxies, identifying homogeneous emission behavior groups. For each cluster, a local scaling factor was calibrated by comparing satellite estimates with emission inventories (e.g., EDGAR). Emissions were then redistributed onto 1 km² pixels using weighting based on proxy density, ensuring that areas with high POI or traffic density received a larger share. Validation showed monthly errors within ±10 % relative to EDGAR, yet a low correlation with ODIAC (r = 0.192) suggests that the latter, based on energy consumption data, does not adequately capture localized industrial emissions. Thus, the method underscores the importance of integrating satellite observations with socio‑economic data for more accurate urban emission mapping, paving the way for targeted mitigation policies and near‑real‑time monitoring systems.