Decoding urban park use patterns from a day–night disparity perspective: Evidence from Nanjing using machine learning

Urban parks are recognized as important public spaces that support the health and well-being of urban residents. Although georeferenced data have been widely applied in park evaluations, most existing studies address temporal variation by analyzing different time periods separately, rather than adopting a comparative perspective to identify distinct park typologies. This study utilized large-scale mobile phone data from Nanjing, China, to investigate day–night differences in park visitation patterns and their determinants. Using a multi-method analytical approach that integrates principal component analysis, Gaussian mixture model clustering, and random forest models, 169 urban parks were classified into three categories: nighttime local leisure parks, daytime comprehensive leisure parks, and all-day balanced community parks. The results reveal that design and diversity dimensions exerted a consistent influence across park types, ranking among the top two contributors. In contrast, transit accessibility significantly shaped stay duration only in all-day balanced community parks, accounting for 10.64 % of the average relative importance. Moreover, different built-environment elements displayed threshold effects, such as the rapid rise in visiting distance disparities in daytime comprehensive leisure parks when the density of security facilities exceeded 0.4. Conversely, when the number of bus-stops further increased beyond 7, a pronounced accessibility saturation effect emerged, such that additional transit supply no longer influenced day–night stay-duration disparities in all-day balanced community parks. These findings underscore the importance of comparative temporal analysis for characterizing park use and highlight the need for refined, context-sensitive strategies to enhance the effectiveness and inclusiveness of urban park services. • Significant day-night differences were found in park visit volume, stay duration, and visiting distance. • Urban parks were classified into three clusters based on day–night visitation indicators. • Built-environment diversity significantly influenced day–night variation across all park types. • Threshold effects of park size on day–night disparities and size–shape interaction mechanisms. • Our findings support finer temporal monitoring and targeted park management strategies.