Assessing carbon emissions from multiple transport modes and prediction via an interpretable machine learning framework in China's urban agglomerations: Evidence from the Yangtze River Delta

Accurately assessing and predicting transportation carbon emissions (TCE) is essential for targeted mitigation. However, the lack of city-level historical emission inventories covering multiple transport modes over consecutive years has constrained TCE assessment and prediction at the urban agglomeration scale. Existing prediction models offer limited interpretability, focus mainly on socioeconomic factors, and rarely include spatial compactness or freight gravity. In response, we assess the carbon emissions from multiple transport modes and propose an interpretable machine learning framework for emission prediction, using the Yangtze River Delta urban agglomeration in China as a case study. The framework incorporates spatial compactness and freight gravity as innovative explanatory variables to capture intercity freight interactions and urban spatial structure. Our findings indicate that TCE increased by 40.98% from 2010 to 2021, while Shanghai's share fell from 37.46% to 29.36%; emissions from private transport and waterways also increased markedly. The rising emissions in private transport and waterways are particularly concerning. The CatBoost model (R 2 = 98.42%, RMSE = 1.3406) outperformed Linear Regression, Random Forest, XGBoost, and LightGBM, providing accurate insights into TCE dynamics. Based on SHAP interpretability, socioeconomic factors (GDP, population, and urbanization) dominate. Besides, higher urban spatial compactness exhibits a strong predictive association with lower emissions. Cities with strong highway and railway freight gravitational attractions witness a notable rise in emissions, reflecting the carbon pressure from intensive multimodal freight activities. These findings inform decarbonization planning through interpretable machine learning and point to cleaner waterway transport, scaled railway hubs, compact urban development, and less private car use.