Revisiting the impact of environmental factors on outdoor public space use: A machine learning analysis of social media data

Abstract

Urban outdoor public spaces have become increasingly vital for physical and mental well-being in high-density cities under mounting environmental stress. Since the Covid-19 pandemic, their use has grown more frequent and widespread. However, intensifying climate change and persistent air pollution continue to pose significant health risks, underscoring the need for new approaches to monitor and assess environmental impacts on outdoor public space visitation. Drawing on environment-behavior theory, this study systematically examines how environmental factors and social rhythms jointly influence the use of different types of public spaces, with a focus on nonlinear and threshold effects. Using Shanghai as a case study, city-scale social media check-in data for 2024 were integrated with official environmental monitoring data, and machine learning (CatBoost) with SHAP was applied to quantify the relative importance, thresholds, and interactions of these factors on visitation patterns. The results show that there is a significant nonlinear relationship between the heat index, air temperature, and relative humidity and the outdoor public space visitation. Among thermal indicators, the heat index outperforms single variables such as air temperature or relative humidity in capturing behavioral responses: outdoor visitation declines markedly once the heat index exceeds 24 °C. Mean air temperature thresholds associated with minimum visitation differ substantially across space types, occurring at approximately 26.5 °C in parks, 22.3 °C in squares, and 31.1 °C in waterfront spaces. In addition, relative humidity above 70% consistently suppresses visitation across all space types. In contrast, air quality indicators exhibit relatively limited direct effects on visitation, suggesting that air pollution risks are often underestimated in everyday activity decisions. Interaction effects further indicate that outdoor visitation increases under thermally comfortable conditions even when air pollution levels are elevated, reflecting a behavioral preference for immediately perceptible thermal comfort over less salient pollution risks. Meanwhile, holidays show clear context-dependent effects: they suppress outdoor activity under extreme heat conditions, yet amplify population exposure during high pollution episodes by intensifying outdoor activity. These findings reveal complex environment–behavior mechanisms in outdoor public space use and provide actionable insights for urban planners to implement heat index-based monitoring, differentiated environmental management, and holiday-specific exposure interventions to enhance safety and comfort.