Unraveling how park environmental indicators shape subjective well-being: Multisensory and nonlinear mechanisms in high-density cities

In high-density cities, limited land resources have shifted the focus of urban park planning from expanding green space supply to improving the environmental quality of existing parks. Although previous studies have examined park accessibility or single-sensory attributes, fine-scale indicators explaining how multisensory park environments jointly shape subjective well-being (SWB) remain insufficient. To bridge this gap, this study developed a visual-thermal-acoustic multi-sensory analysis framework, integrating machine learning (ML) and shapley additive explanations (SHAP) method. Taking multi-scale parks in Guangzhou, China as cases, quantify park environmental, examine their nonlinear and heterogeneous relationships with SWB. Results show that: (1)The integrated multisensory model outperformed single-sensory models, while ML model effectively captured nonlinear relationships across park types, with GBDT performing best for Type I parks and random forest for Types II and III, and Physiological Equivalent Temperature (PET) identified as key factors influencing SWB; (2) in large-scale parks, plant diversity and vegetation exert a positive effect beyond 0.42 and 0.58, respectively, whilst Sky View Factor (SVF) and PET turn negative beyond 0.11 and 30.86 °C, respectively. In small-scale parks, acoustic exposure showed a negative effect above 75 dB(A), indicating that micro-environmental constraints are more pronounced; and (3) With the exception of vegetation and PET, children, middle-aged and older adults exhibit a stronger response to shading, acoustic exposure and SVF, whilst young and middle-aged adults respond more strongly to plant diversity, spatial openness and Canny edges. This study translates multisensory park environments into actionable indicators for green space renewal and differentiated park management.