Urban Pulse: Capturing the Rhythm of Cities

Kai Zhao, Fabio Miranda, Harish Doraiswamy, Marcos Lage, Kai Zhao 0011, Bruno Gonçalves, Luc Wilson, Mondrian Hsieh, Cláudio T. SilvaarXiv 2016 · arXiv

Abstract

Cities are dynamic environments with distinct temporal patterns, yet traditional urban planning tools often treat them as static maps of land use. This paper introduces "Urban Pulse," a framework that leverages topological data analysis to characterize the "rhythm" of city locations across multiple time scales. By defining urban activity as a set of "beats" (hourly, daily, and monthly), the system enables researchers to identify functional analogs—locations in different cities that share behavioral patterns despite having disparate physical characteristics.

Introduction

Urban environments are living organisms characterized by complex temporal rhythms. While modern data collection—including GPS pings, social media check-ins, and taxi trips—provides a wealth of spatio-temporal information, existing analysis techniques often focus on isolated events or individual mobility. There remains a significant gap in "rhythmic" analysis: the ability to capture the essence of a location's activity across hours, days, and months simultaneously. Without this holistic view, architects and planners struggle to find "precedents"—existing locations that function similarly to a site under design, particularly when those precedents are located in different cities.

Methodology

The Urban Pulse framework defines a location's pulse through a set of temporal "beats." The process involves several key steps:

  1. Scalar Function Modeling: Urban data is modeled as a time-varying density function. Discrete data points, such as geo-located social media posts, are transformed into a continuous "activity surface" using Gaussian kernels.
Activity density surface of New York City at 11:00 AM, modeled using Gaussian kernels from discrete data points.
  1. Topological Persistence: To distinguish meaningful activity from noise, the framework employs topological persistence. This technique identifies "prominent" peaks in the activity surface that stand significantly above their surrounding areas.
  2. Defining the Beats: Three types of beats are extracted for every location:
  3. Significant Beats ($B^s$): Binary sequences indicating if a location was a prominent peak at a specific time.
  4. Maxima Beats ($B^m$): Binary sequences indicating any local peak.
  5. Function Beats ($B^f$): Raw activity values over time.
  6. Pulse Comparison: By treating these beats as feature vectors, the system calculates the mathematical "distance" between locations, allowing for cross-city queries.

Key Insights: Functional Analogs and Cultural Geography

A primary finding of this research is that physical similarity does not necessarily imply functional similarity. The Urban Pulse framework allows for the identification of "functional analogs"—places that behave the same way over time.

Comparison of urban pulses for Rockefeller Center (NYC) and Alcatraz Island (SF) at hourly and monthly resolutions, demonstrating their shared 'tourist rhythm.'

As illustrated above, the pulse of Rockefeller Center in New York is remarkably similar to Alcatraz Island in San Francisco. Despite their physical differences, both are driven by a "tourist rhythm" that is consistent across months but highly specific to certain hours of the day. Conversely, the system reveals that broad categories like "Park" are insufficient; Bryant Park and Union Square in NYC exhibit vastly different pulses, with the former driven by seasonal events and the latter by a steady heartbeat of local users.

Furthermore, by segmenting Twitter data by language, the authors identified a "pulse divide" in NYC. English-speaking activity peaks late at night in Lower Manhattan, while Spanish-speaking activity peaks at night in Harlem and the Bronx, reflecting distinct residential demographics and social rhythms.

Results and Expert Validation

The framework's utility was demonstrated through several key results:

  • Cross-City Matching: The system identified Mission Dolores Park (SF) as a functional analog to Bryant Park (NYC) based on shared weekend and evening activity spikes.
  • Computational Efficiency: Processing the entire NYC Flickr dataset—creating all scalar functions across multiple resolutions—took only 18 minutes on a standard consumer laptop.
  • Professional Application: Architects from Kohn Pedersen Fox (KPF) utilized the tool to identify "unclassified regional assets"—emergent hubs of activity that traditional planning might overlook because they lack official landmark status.

Limitations and Future Work

While robust, the current framework primarily analyzes single data streams (e.g., Flickr or Twitter separately). Future iterations could incorporate "multi-modal pulses" combining social media, taxi data, and weather. Additionally, the model currently focuses on activity peaks; incorporating "valleys" (low activity) or "saddles" (transitions) could provide a more nuanced understanding of urban flow. Finally, the authors acknowledge potential demographic biases inherent in social media data, which may not perfectly represent the entire urban population.

Summary updated

BibTeX
@misc{zhao2016urban,
  author = {Kai Zhao and Fabio Miranda and Harish Doraiswamy and Marcos Lage and Kai Zhao and Bruno Gonçalves and Luc Wilson and Mondrian Hsieh and Cláudio T. Silva},
  title = {{Urban Pulse: Capturing the Rhythm of Cities}},
  year = {2016},
  eprint = {1608.06949},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1608.06949},
}
Figure