Search for a command to run...
Urban parks are fundamental to building sustainable and inclusive cities, yet the mechanisms linking their spatial configuration to human activities and social functions remain insufficiently understood. A scalable multimodal framework is developed to quantify how spatial visibility is associated with patterns of park use and the provision of social ecosystem services. A total of 94,635 geo-tagged user-generated images from 148 parks in Vancouver, Canada, were analyzed using the Contrastive Language-Image Pretraining (CLIP) model to classify user activities into six behavioral categories. Concurrently, airborne LiDAR data and space syntax analysis were used to derive three visibility metrics—Mean Isovist Area (MIA), reflecting internal openness; Mean Visual Integration (MVI), indicating visual connectivity within the park interior; and External Isovist Ratio (EIR), representing edge openness and boundary visibility. The results indicate that EIR exhibits the strongest and most consistent relationships with user activity patterns, positively associated with family recreation, social vibrancy, and physical activity, while negatively linked to nature immersion and quiet relaxation. MIA shows moderate associations with socially interactive and child-oriented activities, whereas MVI contributes little explanatory power compared to localized visibility conditions. These findings highlight spatial visibility as a critical design attribute that is closely associated with human–forest interactions. By illustrating that moderate visual openness and edge permeability are associated with more inclusive and multifunctional patterns of park use, actionable insights are provided for urban park planning and design, and the promotion of social sustainability.