Estimation of carbon emissions in various clustered regions of China based on OCO-2 satellite XCO2 data and random forest modelling

Abstract

Atmospheric carbon dioxide (CO₂) stands as one of the most important greenhouse gasses, with steadily increasing concentrations attributable to human activities. In the pursuit of reaching peak carbon and carbon neutrality goals, it is essential to quantify carbon emissions and evaluate carbon reduction strategies. To establish a high-precision observation with full time series and spatial coverage, a spatio-temporal interpolation method was developed to obtain XCO₂ data over mainland China at a resolution of 0.5° × 0.5° for the years 2015–2021. An east-west gradient, higher levels in the east and lower levels in the west, was observed, exhibiting a seasonal pattern of elevation in spring and reduction in summer. Subsequently, the research area is classified into seven clusters based on time-series XCO₂ anomalies (ΔXCO₂) and ODIAC (Open Source Data Inventory of Anthropogenic Carbon Dioxide) carbon emission data. This classification aims to emphasize the differentiation of spatial heterogeneity in carbon emissions and the results highlight that regions with high ΔXCO₂ reflect higher carbon emission. Finally, the carbon emissions of each cluster were estimated by using a random forest model individually yielding an R² of approximately 0.6. For assessing the variables influencing carbon emission predictions, the importance of each variable was calculated. Specifically, NightTime Lighting data (NTL), representing human production activities, emerged as a crucial variable influencing carbon emission predictions in most clusters. In comparison, Gross Primary Productivity (GPP) is considered a more critical variable in Southwest China (SWC), primarily owing to the intricate vegetation carbon sink system in this region. Temperature (T) emerges as a key variable influencing the estimation of carbon emissions in certain developed cities in Eastern China (EC), driven by the urban heat island effect which amplifies energy consumption, modifies land use, and impacts urban systems, influencing the spatial patterns of carbon emissions. Carbon emissions in different characteristic regions was quantified by establishing machine learning models with remote sensing data, which can provide new insights and support for refined carbon monitoring and management strategy.