The impact of green infrastructure on ecosystem quality based on explainable machine learning: a case study of Shanxi Province, China.

Green infrastructure (GI) is a critical strategy for maintaining ecological security and sustainable development amidst rapid urbanization. This study examines changes in GI within the ecologically vulnerable Loess Plateau region, particularly in response to complex human activities, including ecological restoration projects. Utilizing Explainable Machine Learning methods, we explore the impact of GI's coverage-feature-form on ecosystem quality. The findings indicate that over the past two decades (2000-2022), there has been a general improvement in ecosystem quality within the study area due to large-scale ecological restoration efforts. While core GI areas expanded, the reduction of other morphological types led to a more fragmented landscape with significant spatial heterogeneity. Crucially, our XGBoost models demonstrate that morphologically minor components, such as bridge and islet types, exert a disproportionately strong influence on ecosystem quality. These results highlight that ecosystem health is determined not only by the amount of green space but also by its specific features and spatial arrangement. We therefore advocate for integrating the coverage-feature-form framework into future urban planning and ecological restoration to optimize GI performance and enhance ecosystem resilience.