[Paper Review] Urban Explorations: Analysis of Public Park Usage using Mobile GPS Data

Wenfei Xu|arXiv (Cornell University)|Jan 5, 2018

Human Mobility and Location-Based Analysis20 references3 citations

TL;DR

This study uses mobile GPS data from 10 million daily users to analyze spatio-temporal activity patterns in Central Park, employing hierarchical clustering and trip analysis to quantify shared experiences. It finds that established amenities act as 'beacons' by significantly increasing the percentage of shared user experiences, indicating higher social interaction potential in these zones.

ABSTRACT

This study analyzes mobile phone data derived from 10 million daily active users across the United States to better understand the spatio-temporal activity patterns of users in Central Park, New York. The aim of this initial investigation is to create quantifiable measures for understanding public space usage in regions of the city that have no natural data source for measuring activity. We analyze the trip behaviors of users across time and different regions in the park to find patterns of co-location and shared time and, thus, potential social interaction. We find that regions with established amenities and points of interest exhibit a higher percentage of shared experiences, indicating that institutional amenities act as 'beacons' for users' experiences in the park.

Motivation & Objective

• To develop a scalable, data-driven method for quantifying public space usage where natural data sources are absent.
• To understand the spatio-temporal dynamics of user activity in Central Park using mobile phone GPS traces.
• To measure the average percentage of shared user experiences as a proxy for potential social interaction.
• To investigate how spatial and temporal characteristics of park amenities influence user congregation and interaction.
• To evaluate the effectiveness of hierarchical clustering (HDBSCAN) in detecting multi-scale activity clusters in urban parks.

Proposed method

• The study uses anonymized mobile GPS traces from 556,385 unique users within New York City parks, focusing on Central Park in May 2017.
• Trips are identified using a 1-dimensional range-search algorithm to detect user movement within spatial and temporal windows.
• Spatial clustering is performed using HDBSCAN, which dynamically adjusts to varying densities without requiring fixed parameters like ε in DBSCAN.
• Clusters are formed based on user trace density, enabling detection of activity zones at multiple spatial scales, such as the Loeb Boathouse or the Zoo.
• The average percentage of shared experience per cluster is calculated using set intersection of user visit times across trips.
• A log-log regression model is applied to examine the relationship between cluster size (number of unique trips) and average shared time.

Experimental results

Research questions

• RQ1 What is the average percentage of shared user experience in different regions of Central Park, and how does it vary by location?
• RQ2 How do established amenities and points of interest influence the likelihood of social interaction among park users?
• RQ3 To what extent does the size of a user cluster correlate with the level of shared experience?
• RQ4 Can hierarchical clustering (HDBSCAN) effectively detect multi-scale activity patterns in urban park environments?
• RQ5 How do spatio-temporal mobility patterns differ across regions with varying levels of infrastructure and amenities?

Key findings

• Regions with established amenities—such as the Loeb Boathouse, the Zoo, and the Skating Rink—exhibit the highest average percentages of shared user experience.
• The average percentage of shared time increases with cluster size, but at a sublinear rate, with a scaling slope of approximately 0.058 on a log-log scale.
• The top 10 regions by average shared experience are predominantly associated with specific amenities, indicating that institutional features act as social 'beacons'.
• HDBSCAN successfully identifies multi-scale activity clusters, including both localized hotspots (e.g., Huddlestone Arch) and larger zones (e.g., Metropolitan Museum area).
• The distribution of average dwell time across clusters is roughly uniform, with the exception of the North Meadows baseball field, which shows an outlier pattern.
• Larger clusters such as Columbus Circle and the Metropolitan Museum are excluded from shared experience calculations due to their high expected shared time, but their behavior is assumed to follow the same trend.