The Impact of COVID-19 on Pedestrian Flow Patterns in Urban POIs - An Example from Beijing

Yihang Li, Liyan XuISPRS Int. J. Geo Inf 2021 · DOI

TL;DR

This paper analyzes how COVID-19 and associated mobility restrictions impacted pedestrian flow patterns in Beijing. By clustering Points of Interest (POIs) based on their temporal pedestrian flow, the study demonstrates that grouping POIs by similar flow patterns, rather than their functional categories, offers a more effective approach for understanding and managing urban mobility during public health crises.

What Problem Does This Paper Address?

The COVID-19 pandemic presented unprecedented challenges for urban management, particularly in controlling the spread of the virus while maintaining essential city functions. A key aspect of this challenge is understanding and influencing human mobility. Previous research often categorized urban areas or Points of Interest (POIs) by their functional type (e.g., commercial, residential) to analyze mobility patterns. However, the authors argue that this functional categorization might not accurately reflect actual pedestrian flow behavior, especially during disruptive events like a pandemic. They question whether mobility restrictions based on functional categories are truly effective and propose that grouping POIs by their actual pedestrian flow patterns could be a more insightful and practical approach for urban planning and crisis management. This study aims to fill this gap by investigating how COVID-19 and its associated policies altered pedestrian flow patterns and whether a pattern-based grouping of POIs is superior to a function-based one.

Methods

The study employs time series clustering to analyze hourly pedestrian flow data from various POIs in Beijing across three distinct periods: before the epidemic, during the strict epidemic prevention and control phase, and during the initial resumption of work.

  1. Data Preprocessing: Hourly crowd density data (people index) was collected for POIs within Beijing's Sixth Ring Road.
  2. Period Division: The data was divided into three stages based on government policies:
  3. Pre-epidemic: January 17, 2020, 00:00 to January 24, 2020, 00:00.
  4. During epidemic: January 24, 2020, 00:00 to February 10, 2020, 00:00.
  5. Resumption of work: February 10, 2020, 0:00 to February 15, 2020, 24:00.
  6. Time Series Clustering: K-means clustering was applied to group POIs with similar temporal pedestrian flow patterns within each period.
  7. Determining Optimal K: To find the optimal number of clusters (K) for each period, the authors utilized a composite clustering index. This index combined three standard metrics:
  8. Silhouette Coefficient: Measures how similar a sample is to its own cluster compared to other clusters.
  9. CH (Calinski-Harabasz) Coefficient: Evaluates the ratio of between-cluster variance to within-cluster variance.
  10. DB (Davies-Bouldin) Index: Measures the average similarity ratio of each cluster with its most similar cluster. These indices were normalized and combined into a "Fixed Clustering Index" to determine the most appropriate number of distinct pedestrian flow patterns for each stage.

Datasets

The study utilized historical multiday sub-hourly grid (200 × 200 m) crowd density data for key urban areas in Beijing. This data represents the "people index," a measure of crowd density. The dataset covers almost all hotspot locations within Beijing's Sixth Ring Road and includes six major categories of POIs (Transit, Medical care, Education, Recreation, Shopping, Sports), further subdivided into 14 subcategories, totaling 997 hotspot locations. The data spans from January 17, 2020, to February 15, 2020.

Key Insights and Ideas

  • Flow Patterns Over Functional Categories: The core insight is that grouping POIs by their pedestrian flow patterns is more effective than using their functional categories for understanding urban mobility dynamics, especially during crises. Functional categories can be misleading, as POIs with similar functions might exhibit vastly different flow behaviors, and vice-versa.
  • Impact of Restrictions: Strict mobility restrictions during the epidemic significantly reduced the number of distinct pedestrian flow patterns, consolidating many POIs into a single, low-activity pattern. This highlights the profound impact of such policies on daily urban life.
  • Resumption Patterns: As restrictions eased, the number of distinct pedestrian flow patterns increased, even though overall visitation numbers might have remained lower than pre-epidemic levels. This suggests a shift towards more differentiated travel needs and potentially increased social distancing.
  • Airport as a Unique Case: Beijing Capital Airport consistently emerged as a distinct cluster across all stages, underscoring its unique role as a major international hub with inherently different mobility characteristics.
  • Divergence in Resumption: During the resumption phase, certain POI categories (e.g., Medical care, Shopping) showed a clearer separation of flow patterns, indicating that necessary trips became more distinct and potentially more localized.

Key Results

  • Number of Patterns: The number of unique pedestrian flow patterns decreased from four before the epidemic to two during the strict control stage, and then increased to six during the initial resumption of work.
  • Clustering Effectiveness: The study found that functional categories of POIs did not consistently group together based on flow patterns. For instance, POIs within the same functional category (e.g., restaurants) could exhibit different flow behaviors depending on their location (e.g., near a mall vs. in a business district).
  • Resumption Phase Dynamics: During the resumption phase, while overall pedestrian numbers were lower than pre-epidemic, the variety of flow patterns increased. This suggests that social restrictions led to changes in POI flow patterns and promoted social distancing.
  • Composite Index Validation: The proposed composite clustering index effectively determined the optimal number of clusters for each stage, aligning with the observed changes in urban mobility due to the pandemic.

Limitations and Future Work

  • Data Limitations: The authors acknowledge that the study's findings on the separation of flow patterns during the resumption period are based on data limitations. They suggest that more geographic data could further explain these phenomena.
  • Phased Impact: The division of epidemic periods into distinct stages simplifies a reality where impacts are often phased and gradual. Future work could explore these step-by-step effects.
  • Visualization: The authors suggest exploring other visualization methods to improve user perception of the data and findings.
  • Generalizability: While Beijing serves as a case study, future research could investigate whether these findings hold true for other cities with different urban structures and policy responses.
  • Policy Implications: The study suggests that mobility restrictions should ideally be based on POI mobility characteristics (flow patterns) rather than just functional categories, offering a more operational approach for future policy implementation.

Summary updated

BibTeX
@article{li2021impact,
  author = {Yihang Li and Liyan Xu},
  title = {{The Impact of COVID-19 on Pedestrian Flow Patterns in Urban POIs - An Example from Beijing}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2021},
  doi = {10.3390/ijgi10070479},
}
Figure