Green in isolation: An interpretable spatial machine learning approach to understanding green satisfaction disparities in segregated neighborhoods

Urban green spaces are crucial for the health and well-being of residents. However, green planning in isolated urban informal neighborhoods might exacerbate spatial inequalities, leading to green gentrification and stigmatization. Understanding the nonlinear and continuous relationships between green environments and residents' satisfaction within segregated areas remains challenging. This study established an interpretable spatial machine learning framework, integrating the MGW_LightGBM and SHAP models to address nonlinearity, spatial heterogeneity, and interpretability issues. Utilizing multisource data and deep learning technology, the study provides global and local explanations of residents' green satisfaction relative to their green environment. The findings indicated that: (1) 3D green morphologies play a more significant role in influencing green satisfaction compared to 2D green landscape patterns, especially the contributions from tree height and canopy volume; (2) All 2D and 3D green variables, including contagion, Shannon diversity index, and tree height, exhibit distinct nonlinear and threshold effects on green satisfaction; (3) In informal neighborhoods, due to spatial and economic constraints, residents' attitudes towards green spaces show marked pragmatism and spatial adaptability; (4) Compared to existing models, the MGW_LightGBM model demonstrates superior performance in modeling and predicting green satisfaction. These findings are vital for differentiated green planning in neighborhoods, alleviating residential segregation, and enhancing social integration.