Assessing human-scale green equity in the 15-minute city using street-view-based visual landscape indicators and...

Against the backdrop of rapid urbanization, identifying disparities in green environmental exposure among different socioeconomic groups is of great importance. Taking the central urban area of Chengdu, China, as a case study, this study assesses street-level green environmental equity within a 15-minute walking distance using Baidu Street View images.

A Mask2Former semantic segmentation model was employed to extract seven street-view visual landscape indicators, including Green View Index (GVI), Tree View Index (TVI), Sky View Factor (SVF), Building View Index (BVI), Hard Paving View Index (HVI), naturalness, and wildness. Second-hand housing prices were used as a proxy for socioeconomic stratification. Getis-Ord Gi*, global Moran's I, local Moran's I, bivariate local Moran's I, random forest, Shapley Additive Explanations (SHAP), and partial dependence plots (PDPs) were applied to reveal the spatial relationships between housing prices and visual landscape indicators, as well as their nonlinear effects.

The results show that (1) housing prices in Chengdu's central urban area exhibit significant positive spatial autocorrelation, with both high-price and low-price areas showing clear clustering patterns; (2) high-price areas generally correspond to higher TVI, GVI, and SVF values, indicating significant differences in street-view landscape exposure across housing-price groups; and (3) among the seven indicators, SVF shows strong explanatory power across all housing-price groups, whereas TVI plays a more prominent role in the high-price group.

This study deepens our understanding of green environmental equity from the perspective of street-view visual perception and provides a reference for the fine-scale optimization of urban green space allocation.