China's county-level monthly CO2 emissions during 2013-2021.

The top-down method is widely used to estimate China's CO₂ emissions at the county level. However, studies have relied on a single indicator of regional total nighttime light brightness as an instrumental variable for prediction, leading to the assumption that there is a positive correlation between CO₂ emissions and total nighttime light brightness in all regions within the same province. This assumption overlooks other heterogeneous relationships and does not correspond to reality. Therefore, this study constructed a dataset of potential feature variables based on multisource data (improved and calibrated nighttime light data, urban and rural human settlement data, and socioeconomic indicator data based on statistical yearbooks). After the main feature variables were identified, a hybrid regression algorithm combining deep neural networks and CatBoost was constructed to generate instrumental variable for predicting CO₂ emissions. Compared with the total nighttime brightness, it has a stronger linear relationship with CO₂ emissions. Using the top-down algorithm, we estimated China's monthly CO₂ emissions at the county level from 2013 to 2021. This dataset provides a solid foundation for predicting the achievement of China's county-level "dual carbon" strategy. The methods used in this study can be generalized to other global regions.