Effects of 3D urban morphology on CO2 emissions using machine learning: Towards spatially tailored low-carbon strategies in Central Wuhan, China

Unraveling the effects of urban morphology on CO2 emissions is essential for shaping sustainable and low-carbon urbanization practices. However, few studies have developed spatially tailored mitigation strategies based on fine-grained analysis of 3D urban morphology. This study extracts 3D urban morphology metrics from buildings and streets at a 1 km grid in central Wuhan. Notably, the inter-building obstruction and street topology are taken into account in this field for the first time. Then, Random Forest and interpretive algorithms are used to unravel the effects of urban morphology on CO2 emissions. Ultimately, Geographic Random Forest is adopted to develop spatially tailored mitigation strategies. The main results are: (1) Urban morphology contributes more to CO2 emissions than traditional socioeconomic explanations like population density and land use. (2) Sky view factor significantly influences CO2 emissions, second only to population density. (3) Vertically high-density development leads to higher emissions. (4) Optimal parameters for carbon reduction are observed with the building shape coefficient at 0.68, mean neighbor distance at 85, and Severance at 1.28. (5) Four distinct classes are classified based on local dominant influencing factors, and tailored low-carbon strategies are proposed. This methodological framework can also be applied to global cities undergoing rapid urbanization.