Explainable second-dimension spatial association between park sentiment and built-social environments

Understanding human–environment interactions is crucial for sustainable urban development and well-being. Urban parks are key urban environments affecting residential emotions. However, existing studies with fixed neighborhood ranges and mean-based indicators limit the examination of nonlinear effects across spatial scales and value ranges. This study develops an explainable second dimension of spatial association (ESDA) model to identify spatially nonlinear drivers of park sentiment. The sentiment is quantified using a RoBERTa model with volunteered geographic information for parks in Xiamen, China. ESDA constructs second dimension variables by searching ranges ( b ) and probability parameters ( τ ), and fits an SDA-based random forest with SHapley Additive exPlanations (SHAP) for interpretation. Leave-one-out cross-validation shows that SDA (R 2 = 0.459) improved modeling accuracy compared with the fixed-neighborhood, mean-based baseline (R 2 = 0.357). SHAP indicates nonlinear, scale-dependent, and sign-reversing effects on sentiment. For instance, high near-neighborhood road density and baseline population are associated with lower sentiment, whereas street-scale service aggregation is associated with higher sentiment, and the effects of building height and nighttime light vary with scale and background conditions. The findings suggest prioritizing parks in macro-scale areas with high accessibility and activity levels, while avoiding parcels with overly dense nearby roads and high existing light levels or population. In conclusion, the spatial nonlinearity is diagnosed through the ESDA model that constructs critical b – τ combinations and integrates SHAP for layered park siting and surrounding-environment optimization.