At what thresholds do blue-green spaces promote walking for older adults? A GeoAI inquiry into nonlinear thresholds and spatial heterogeneity.

Despite broad consensus on age-friendly built environments, how specific attributes shape older adults' walking behavior through nonlinear and spatially varying influence pathways remains insufficiently understood. We develop an explainable GeoAI framework integrating route-level PPGIS mapping, perception surveys, and high-resolution blue-green metrics in Chengdu, China. Using multinomial logistic regression, XGBoost-SHAP, and multiscale geographically weighted regression (MGWR), we identify threshold-like and spatially heterogeneous associations for walking frequency and duration. PDPs suggest an inflection around intersection density (∼300/km2), beyond which the predicted probability of walking ≥3 times/day increases markedly. Green-open space shows an indicative threshold near ∼10% for ≥3 h/day walking. Bivariate PDPs further indicate that moderate-high cultural facility density (≈80-120/km2) combined with moderate-high transit-stop density (≈100-140/km2) is associated with elevated probabilities of frequent walking (in some ranges approaching ∼80%), with diminishing returns at very high densities. SHAP assigns slightly greater overall importance to perceived comfort/enjoyment than to objective attributes, and MGWR shows stronger duration associations in central districts. These findings provide empirically grounded threshold ranges and spatially differentiated evidence to inform age-friendly, walkable community planning.