Evaluating walkability and vitality in cold-region historic districts of China via street view semantics and SHAP

As historic districts undergo sustainable urban regeneration, enhancing walkability is important for supporting local vitality. However, conventional walkability evaluation frameworks often emphasize openness and may not fully account for the climate-adaptive characteristics of compact spatial configurations in severe cold regions. Moreover, traditional linear models may overlook nonlinear and threshold-like relationships between micro-scale built environments and population vitality. To address these gaps, this study developed an integrated framework combining Street View Imagery, points of interest (POIs), and Location-Based Services (LBS) thermal data to evaluate two typical historic districts, Buzili and Dajingmen, in Zhangjiakou, China. DeepLabV3 + semantic segmentation and K-Means clustering were used to characterize street-view environments and identify spatial patterns, while Random Forest regression coupled with SHAP (SHapley Additive exPlanations) was employed to examine nonlinear associations between built-environment characteristics and population vitality. The prediction model achieved an \:R² of 0.736, with an RMSE of 4.22 and an MAE of 2.94. Global SHAP analysis identified POI Density (mean absolute SHAP = 3.0496) and Functional Mix (2.1150) as the dominant predictors, followed by Walkable Space Ratio (0.9645) and Green View Index (0.6504). The POI-density relationship exhibited a reproducible nonlinear transition around the standardized range of approximately 0.067–0.083, whereas GVI showed a robust J-shaped response pattern across bootstrap models, but with substantial uncertainty in the exact turning-point location. Walkable Space Ratio showed a predominantly negative nonlinear association with population vitality within the observed study area. These findings suggest that openness-oriented walkability criteria may require climate- and context-specific reconsideration in cold-region historic districts, where compact spatial configurations may be compatible with pedestrian vitality. However, because the LBS observations were collected during the National Day period and no direct winter microclimatic measurements were conducted, the proposed mechanisms should be interpreted as associations rather than direct evidence of thermal adaptation. The findings provide context-specific empirical evidence for differentiated historic-district renewal and highlight the need for multi-season and multi-site validation before these transition ranges can be generalized as planning standards.