Interpretable Ecological Resistance Modeling and Biodiversity-Based Ecological Consistency Assessment for Tropical Forest Corridor Construction in Hainan Tropical Rainforest National Park, China

Tropical forest national parks are increasingly threatened by habitat fragmentation and declining ecological connectivity. Ecological network studies often rely on subjectively weighted resistance surfaces and lack adequate model consistency assessment and interpretability. This study developed an interpretable machine learning framework to optimize ecological security patterns by integrating ecosystem service assessment, ecological resistance modeling using extreme gradient boosting (XGBoost), interpretation using Shapley additive explanations (SHAP), biodiversity-based ecological consistency assessment, and circuit theory. Taking Hainan Tropical Rainforest National Park, China, as a case study, four ecosystem services were assessed in 2000, 2010, and 2020 using the Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model, and ecological sources were identified using Morphological Spatial Pattern Analysis (MSPA). Proxy resistance surfaces constructed using the analytic hierarchy process (AHP) and XGBoost proxy resistance surfaces were compared using a biodiversity-based ecological consistency assessment. XGBoost was treated as an interpretable nonlinear candidate, rather than being selected on the basis of superior classification performance. Ecological corridors, pinch points, and barrier areas were subsequently identified. The results show the following: (1) Ecosystem services declined from 2000 to 2010 and recovered by 2020. Ecological resistance derived from the XGBoost model initially increased and then decreased. SHAP analysis identified elevation, the normalized difference vegetation index (NDVI), and human footprint as the dominant predictors of high ecological resistance. (2) Biodiversity-based ecological consistency assessment based on species richness showed that the Spearman correlation coefficients between XGBoost-derived resistance and species richness were −0.657, −0.656, and −0.626 in 2000, 2010, and 2020, respectively, and were consistently stronger than those obtained using the conventional AHP-weighted resistance surface. Based on the XGBoost-derived ecological resistance surfaces, key ecological corridors, pinch points, and barrier areas were identified, revealing an ecological connectivity network centered on Wuzhishan, Jianfengling, Yinggeling, and Bawangling. (3) A spatial ecological security pattern comprising “two cores, two ecological barrier belts, three zones, and multiple corridors” was proposed. This study provides an interpretable and transferable framework for forest landscape connectivity conservation and ecological restoration in tropical forest protected areas.