Estimation of urban land use implication on energy-related carbon emissions based on machine learning methods

Understanding the impact of urban land use patterns on energy-related carbon emissions is critical for developing effective climate change mitigation strategies. This study employed machine learning techniques to model the relationship between multidimensional urban land use characteristics and city-scale carbon emissions. Urban land use was characterized across four dimensions: scale, structure, mixture, and intensity. Machine learning algorithms, including CART, Random Forest, and XGBoost, were trained to quantify the relative importance of these land use features in predicting carbon emissions. The machine learning models demonstrated strong predictive performance, outperforming traditional linear regression. The feature importance analysis revealed that urban land use indicators collectively account for over one-quarter of the models' predictive power, with land use scale, structure, and intensity features exhibiting greater importance than socioeconomic variables. These findings underscore the value of data-driven, nonparametric modeling approaches in elucidating the complex, multifaceted links between urban form and greenhouse gas emissions.