Decoding drivers of multi-level ecological networks through key bird species integration: A machine learning interpretability framework for biodiversity conservation

Global urbanization has intensified habitat fragmentation, creating an urgent necessity to develop effective ecological networks to alleviate the risk of biodiversity loss to regional ecological security. Conventional approaches for constructing ecological networks overlook the synergy between species’ requirements and landscape functions, failing to elucidate the nonlinear responses of driving processes, which consequently limit the conservation efficacy of ecological networks. This study innovatively combines bird key species data with an interpretable machine learning framework to develop a multilevel ecological network, utilizing the Harbin-Changchun urban agglomeration as a case study. It employs the Maximum entropy modeling (MaxEnt), Morphological spatial pattern analysis (MSPA), and circuit theory, in conjunction with XGBoost-SHAP, to examine the social-ecological driving mechanisms that influence ecological network performance. The findings indicated that: (1) The incorporation of species data markedly enhanced network functionality, while birds’ vital habitat can be consistently overlooked by landscape morphology approaches alone; (2) The evaluation of ecological network efficacy demonstrated that the bird network established an effective conservation continuity in the western region of the Songnen Plain, whereas the eco-space network exhibited superior performance in the southeastern area; (3) The SHAP model identified sensitive intervals for climatic variables, while land use and anthropogenic activities displayed distinct threshold effects. The findings suggest a spatial optimization tool for harmonizing urban development and biodiversity preservation, providing insights for species conservation in rapidly urbanizing areas worldwide.