Optimizing regional ecological security patterns based on interpretable machine learning models: A case study of Huzhou, Zhejiang Province

Scientifically delineating ecological source areas surrounding urban environments and constructing ecological resistance surfaces are essential for understanding landscape connectivity and enhancing regional ecological security. They are essential for achieving high-quality, sustainable development in the construction of modern urban ecological civilization. This study selected key indicators that encompass both natural and socio-economic elements as factors contributing to ecological resistance. By employing the CatBoost algorithm with SHAP (Shapley Additive exPlanations), Circuit Theory and a systematic analysis of the ecological security pattern in Huzhou is conducted. The results demonstrate that (1) 20 ecological sources patches were identified, covering a total area of 1117 km²; (2)based on interpretable machine learning results, areas of high ecological resistance in the northeast and central-southern parts of the research area is 500.75 km², accounting for 8.6 % of the total area of the area; (3) circuit theory identified 40 ecological corridors, 348 ecological pinch points, and a total of 11 ecological barrier surfaces. These findings support the development of a spatial framework consisting of ‘two screens, three belts, four corridors, and five zones'. Furthermore, the 3D Realistic Geospatial Landscape Model (3dRGLm) was employed to enhance spatial understanding and provide visually optimized planning insights within ecologically critical areas. This research provides policy guidance insights to promote high-quality and sustainable regional development, offering significant practical implications for ecological planning and management.