Multimodal urban park service accessibility and spatial distributional inequality through a multiscale assessment

Urban park planning requires attention to both service accessibility and its spatial distribution. This study assessed 420 parks and 488 residential communities in Nanjing using Multi-G3SFCA, population-weighted Gini coefficients, sensitivity analysis, and spatially validated RF–SHAP. Travel impedance was calibrated using 320 route records obtained from the Amap Route Planning API. Mean Overall accessibility was 17.3005 hm² per 100 persons, with a Gini coefficient of 0.4931. Comprehensive parks contributed 87.67% of mean Overall accessibility, whereas Community parks had the lowest mean accessibility and the highest Gini coefficient of 0.9603. Under the baseline 10-min threshold, 32 communities, representing 6.56% of the sample and 5.83% of estimated residential demand, had zero Community park accessibility. Overall accessibility rankings and the identification of low-accessibility communities remained comparatively stable under changes in travel speeds, service thresholds, modal weights, population demand, and entrance-allocation rules. RF achieved an R 2 of 0.5701 under conventional ten-fold validation and 0.2016 under leave-one-district-out validation, indicating weaker transferability to districts excluded from model training. Spatially cross-fitted SHAP identified older-adult proportion and district-level income context as the leading predictors.