Data-driven identification of economic drivers and regional disparities of ecological resilience in china’s urban agglomerations: an explainable deep learning framework

Abstract

As key spatial carriers for regional economic growth and ecological protection, urban agglomerations’ ecological resilience is critical to regional risk adaptation and sustainable development. This study examines 92 prefecture-level and above cities in China’s five major urban agglomerations from 2007 to 2021. An ecological resilience evaluation system is constructed under the resistance–adaptability–recovery framework and measured using the CRITIC method. The CNN-BiLSTM-Attention model combined with the SHAP explainable framework is then employed to explore the economic drivers and their heterogeneous impacts, and its predictive performance is benchmarked against seven established baseline models. Results show that ecological resilience presents a distinct “core-periphery” spatial pattern with strong path dependence, and central-city polarization dominates over diffusion effects. The proposed model achieves the lowest prediction errors and the smallest training-to-testing performance gap among all compared models, confirming both its accuracy and generalization. Per capita GDP, population density, and urbanization rate are core drivers: per capita GDP has the highest explanatory power but an overall negative effect, while population density and urbanization rate mostly show positive contributions. The SHAP interaction analysis further reveals a pronounced inhibitory coupling between per capita GDP and local fiscal expenditure, wherein their joint high levels compound rather than mitigate ecological pressure. Drivers feature nonlinearity, thresholds, and interactions, and the dominant predictors differ across agglomerations in ways that map onto their distinct structural characteristics. These findings connect urban ecological resilience to the global Sustainable Development Goals agenda and offer transferable lessons for developing countries facing development–conservation tensions, supporting differentiated governance and policy optimization for high-quality and resilient urban development.