Integrating Multi-source Data for CO₂ Emission Prediction at City and County Level – A Case Study in Shanxi Province, China

Accurately estimating carbon dioxide (CO
2
) emission is a fundamental guarantee for implementing and evaluating regional carbon emission reduction strategies. Taking Shanxi Province as a case study, this paper investigates multi-source data: NPPNIIRS nighttime lights data, socioeconomic statistical data, and meteorological data, to achieve accurate prediction of CO
2
emissions at the city and county level using a prediction model based on Random Forest algorithm. The RF model was trained and tested using the data from 2012 to 2017, and then applied to predict city-county level CO
2
emission for 2018-2020. Comprehensive experiments have been conducted to compare the RF model with linear regression, Gaussian progress regression and support vector regression models, as well as to investigate the effect of multi-source data integration. In addition, the distribution and temporal trends of CO
2
emissions in the cities and counties of Shanxi Province are analyzed. The main findings are: ❨1❩ Integrating multi-source factors could achieve a high prediction accuracy of CO
2
emissions than using a single data source, and combining meteorological data with either nighttime lights indicators or socioeconomic factors can greatly improve the RF model performance; ❨2❩ The RF model with built-in feature importance determination can achieve good performance of CO
2
prediction at the city and county level without feature selection; ❨3❩ From 2012 to 2020, the carbon emissions in Shanxi Province showed a trend of first decreasing and then increasing, and a total growth rate is approximately 15.4%. The city-county analysis of CO
2
emissions can provide reference for the formulation of emission reduction plans in various regions of Shanxi Province.