Decoding the spatiotemporal dynamics and driving mechanisms of ecological resilience in the Beijing-Tianjin-Hebei urban agglomeration: A deep learning approach

Abstract

Urban agglomerations face escalating ecological challenges due to rapid urbanization and climate change, yet the dynamic spatiotemporal patterns and drivers of ecological resilience remain underexplored. This study examines the ecological resilience of the Beijing-Tianjin-Hebei (BTH) region from 2010 to 2020, integrating a deep learning approach using the Transformer-based TSAR-SHAP model with spatiotemporal analysis of a 1 km × 1 km grid dataset. Ecological resilience is assessed from morphology, density, and coordination dimension, alongside socio-economic, environmental, and climatic factors. The findings reveal a marked decline in ecological resilience levels between 2010 and 2015, particularly in urban cores like Beijing and Tianjin, driven by urban sprawl, PM2.5 pollution, and CO₂ emissions. A partial recovery from 2015 to 2020 reflects the positive impact of coordinated environmental policies, including air pollution control and ecological restoration initiatives. Spatially, urban centers exhibited persistent ecological stress due to high population density and built-up area expansion, while rural areas in northern Hebei displayed higher resilience supported by natural ecosystems and favorable climatic conditions. The TSAR-SHAP model captured the temporal shift from anthropogenic to climatic drivers and revealed significant spatial heterogeneity. These findings highlight the need for spatiotemporal differentiated strategies to balance urban growth with ecological preservation and provide actionable insights for sustainable regional development in rapidly urbanizing areas.