Spatio-temporal random forest-based estimation of monthly gridded carbon emissions in China (2019–2022) using multisource remote sensing data

Accurate quantification of anthropogenic CO₂ emissions at fine scales is essential for effective climate mitigation but remains challenging because of spatial–temporal heterogeneity and data limitations. This study presents a spatio-temporal random forest (STRF) model as an inventory enhanced framework, integrating multisource remote sensing datasets to estimate monthly gridded anthropogenic CO₂ emissions across China at 0.1° × 0.1° resolution from 2019 to 2022. Compared with conventional models, the STRF model explicitly incorporates spatial autocorrelation and temporal continuity through spatio-temporal weighting and dynamic feature selection, achieving superior accuracy. A comparison against existing emission inventories reveals strong agreement with the Multi-resolution Emission Inventory for China (MEIC), highlighting the model’s reliability. Feature importance further identifies nighttime light, tropospheric NO₂ and CO concentrations, and XCO₂ anomalies as the dominant predictors. The results reveal that high-emission hotspots are consistently concentrated in industrial and urban agglomerations. Temporally, emissions display distinct seasonal variability, with peaks in winter (driven by heating demand) and summer (fuelled by cooling energy needs). The COVID-19 pandemic temporarily reduced emissions by approximately 30% in 2020, followed by a rapid rebound thereafter. These findings underscore the ability of STRF model to provide high-resolution, dynamic CO₂ emission estimates via multisource remote sensing data, offering valuable insights for targeted, season-specific mitigation strategies aligned with China’s dual carbon goals.