Multi-scale carbon emission characterization and prediction based on land use and interpretable machine learning model: A case study of the Yangtze River Delta Region, China

Carbon emissions are a significant factor contributing to global climate change, and their characterization and prediction are of great significance for regional sustainable development. This study proposes a novel carbon emission characterization and prediction model based on interpretable machine learning and land use. It does not rely on socio-economic indicators, thus enabling carbon emission predictions after the decoupling effect. It can also reflect spatial distribution characteristics of carbon emissions, and demonstrates high accuracy and interpretability. The Yangtze River Delta (YRD) region serves as the application case for the model. Utilizing GIS-Kernel Density for land-use subdivision and Optimized Extra Tree Regression, the model achieves high precision (R² = 0.99 for training, R² = 0.86 for testing). Shapley Additive exPlanations (SHAP) model was employed to interpret the model, revealing the impact curves of different land areas on carbon emissions. Optimized Land Expansion Analysis Strategy (Opti-LEAS) and Cellular Automaton based on Multiple Random Seeds (CARS) models simulated land use under baseline scenarios, confirming an overall accuracy exceeding 85%. The total carbon emissions in the YRD in 2030 are projected to reach 1580.70 million tons, with Shanghai leading at 223.84 million tons, followed by Suzhou at 172.20 million tons. County-level carbon emissions were characterized, and a spatial econometrics model was employed to reveal the spatial distribution characteristics of future carbon emissions, indicating a clustering effect (Moran's I = 0.6076). As industrial land disperses, clustering shifts towards regional centers, with areas like Wuzhong District identified as 99% confident carbon emission hotspots.