Context-dependent walkability: built-environment associations and nonlinear contributions to pedestrian space quality–walking behavior mismatch in Xi’an, China

Abstract

In the context of urban regeneration, substantial investments in pedestrian space quality (PSQ) do not always translate into corresponding gains in walking behavior (WB). Existing studies often assume a linear PSQ–WB relationship, which cannot capture nonlinear associations and contribution patterns observed in practice. This study uses Xi’an’s core urban area as a case study and integrates multi-source urban data, including mobile phone signaling data and street-view imagery, with statistical and machine-learning methods to quantify and map the spatial patterns of PSQ–WB mismatch. Ordered logistic regression and XGBoost-SHAP models are employed to examine built-environment factors associated with the mismatch and further reveal their nonlinear contribution patterns and empirical transition points. The results show that: (1) 84.72% of traffic analysis zones in Xi’an’s core urban area exhibit a mismatch between PSQ and WB; (2) floor area ratio, population density, road network density, and public transport stop density are positively associated with the relative position of WB to PSQ, whereas building density and distance to the city center are negatively associated with it; (3) a higher position of WB relative to PSQ is more likely to occur when floor area ratio exceeds 1.0, population density exceeds 10,000 persons/km2, road network density exceeds 6 km/km2, and public transport stop density ranges from 9 to 17 stops/km2. These findings provide new evidence for context-dependent walkability and offer policy implications for shifting urban pedestrian regeneration from undifferentiated investment toward targeted interventions.