The escalating global greenhouse gas emissions pose a significant threat to climate stability and sustainable development. Given China’s status as one of the largest anthropogenic CO₂ emitters globally, regional emission projections are crucial for informed policy-making and the successful realization of carbon neutrality goals. Focusing on Jiangsu Province, an economic powerhouse in China, this study proposes a dynamic weighted ensemble model GNNWR-GBDT that integrates Geographically Neural Network Weighted Regression (GNNWR) and Gradient Boosting Decision Tree (GBDT). GNNWR addresses spatial heterogeneity by learning spatially varying coefficients via a neural network architecture, while GBDT models nonlinear socio-economic relationships. WeightNet is employed to dynamically reallocate sub-model weights according to input features, thus improving the precision and resilience of predictions. The model demonstrates exceptional performance in carbon emission predictions for Jiangsu Province, achieving an R2 of 0.908 and surpassing traditional methods. It reveals a ‘‘high-density economy, high carbon emissions’’ pattern in Southern Jiangsu, particularly in the Nanjing-Suzhou-Wuxi area. Scenario analyses underscore the critical role of GDP and population density as emission drivers, offering valuable insights for urban carbon reduction strategies. This study advances carbon emission prediction methodologies and provides practical guidance for balancing economic growth with environmental sustainability in Jiangsu and beyond.