Nonlinear Thresholds of Multifunctional Blue–Green Infrastructure: Balancing Urban Cooling, Habitat Quality, and Green Equity in High-Density Shenzhen

Climate adaptation in high-density cities requires equitable access to cooling and habitat benefits from blue–green infrastructure (BGI). However, the critical thresholds where human activity overwhelms these services remain unknown. Taking Shenzhen as a case study, we quantified habitat quality using the InVEST model and divided the urban landscape into low-, medium-, and high-quality zones. After screening significant drivers via multiple linear regression, we compared six machine learning algorithms and selected XGBoost (for low- and medium-quality zones) and LightGBM (for high-quality zones) with Optuna-tuned hyperparameters, then applied SHAP dependence analysis to identify nonlinear, threshold-driven responses. We identify a systemic thermal threshold of 23.7 °C. Above this temperature, BGI cooling efficiency decouples from habitat quality. We also quantify actionable intervention windows. Low-quality habitats—dominated by dense residential areas—collapse when building density exceeds 0.7 or road density exceeds 0.009. Medium-quality habitats offer a balance window (core area > 2844 m2, building density < 0.361) that enables incidental nature contact during daily travel. High-quality refugia require a core area > 8600 m2 with near-zero disturbance. These thresholds expose a stark green inequity: socioeconomically vulnerable groups in low-quality zones are systematically disconnected from cooling services. We translate these findings into a tiered spatial intervention framework—restoration, accessibility enhancement, and strict protection—to advance climate-resilient and socially equitable urban planning.