Identifying Carbon Emission Drivers and Regional Mitigation Priorities at the County Level in China

This paper aims to estimate county-level CO2 emissions across China and identify the socioeconomic and spatial factors associated with their variation. Current research lacks a consistent national county-level dataset after 2017. To address the lack of consistent national data after 2017, we developed a unified dataset covering 2735 county-level units from 1997 to 2022. Seven machine learning models were evaluated using 21,197 observations for 1997–2017, after which the best-performing model was used to estimate emissions for 2018–2022. LightGBM achieved a test R2 value of 0.9529 and an MAPE of 15.32%, while SHAP identified secondary-sector value added as the most influential predictor, followed by provincial area and GDP. Estimated emissions exceeded 12,000 Mt in 2022, whereas carbon intensity declined by approximately 10% from 2018 to 2022 despite rising per-capita emissions. These findings indicate improving carbon efficiency alongside persistent regional disparities. Industrial and resource-dependent counties should prioritize structural upgrading and cleaner energy substitution, while developed regions should consolidate efficiency gains and control absolute emissions.