Assessing Urban Travel Patterns: An Analysis of Traffic Analysis Zone-Based Mobility Patterns

Yanyan Chen, Zheng Zhang, Tianwen LiangSustainability 2019 · DOI

TL;DR

This paper introduces a novel, multi-step methodology to analyze urban mobility patterns at the Traffic Analysis Zone (TAZ) level using mobile phone data. By clustering TAZs based on their temporal mobility profiles and correlating these clusters with land use characteristics derived from Points of Interest (POI) data, the study provides a deeper understanding of how urban form influences travel behavior, offering valuable insights for urban planning and transportation management.

What Problem Does This Paper Address?

Understanding urban mobility patterns is crucial for effective transportation planning and demand management. While existing research has explored individual travel behaviors and aggregate mobility characteristics, there's a notable gap in analyzing regional, TAZ-based mobility patterns and quantitatively assessing their relationship with local land use characteristics. This lack of understanding hinders the development of targeted, region-specific transportation strategies. The paper aims to bridge this gap by providing a framework to identify distinct TAZ-level mobility patterns and link them to the underlying urban land use, moving beyond individual trip analysis to a more macro-level perspective.

Methods

The study employs a three-pronged methodological approach:

  1. TAZ-Based Mobility Pattern Identification (Clustering):
    • Data Source: Anonymized mobile phone Call Detail Records (CDRs) are used to extract hourly departure and arrival trips for each TAZ.
  2. Feature Engineering: Each TAZ is represented by a vector of 48 features (24 hours x 2 types of trips: arrival/departure). These features are normalized to account for varying land use characteristics.
  3. Clustering Algorithm: K-means clustering is applied to group TAZs with similar temporal mobility patterns. The optimal number of clusters (K=4) was determined using the Davies-Bouldin and Silhouette indices.
  4. Land Use Characteristics Representation (POI Data):
    • Data Source: Points of Interest (POI) data (e.g., from Google Places API) are used to characterize land use within each TAZ.
  5. Feature Engineering: A "land use vector" for each TAZ is constructed using the Term Frequency-Inverse Document Frequency (TF-IDF) method applied to POI categories. This captures the relative importance of different land use types within a TAZ.
  6. Relationship Analysis (Multinomial Logit Model):
    • Model: A multinomial logit model is used to investigate the co-occurrence probabilities between the identified TAZ-based mobility pattern clusters (dependent variable) and the land use characteristics (independent variables).
  7. Purpose: This model helps to understand how specific land use indicators influence the likelihood of a TAZ belonging to a particular mobility pattern cluster.

Datasets

  • Mobile Phone CDRs: A large-scale dataset of anonymized mobile phone records from a major operator in Beijing, China, covering the period from June 1 to June 7, 2015. This dataset was used to extract hourly trip generation and attraction data for each TAZ.
  • Points of Interest (POI) Data: Data collected from Google Places API was used to characterize land use within each TAZ.

Key Insights and Ideas

  • TAZ-level patterns are distinct and meaningful: The study demonstrates that TAZs can be meaningfully clustered into distinct mobility pattern groups, revealing significant differences in travel intensity and temporal distribution. This moves beyond aggregate city-wide patterns to localized behaviors.
  • Land use is a strong predictor of mobility patterns: The research quantifies the relationship between land use types (e.g., workplace, residential, commercial) and TAZ mobility clusters. This provides empirical evidence for how urban form shapes travel.
  • Imbalanced land use drives imbalanced travel: The findings suggest that an imbalanced distribution of land uses within a TAZ can lead to significant differences between arrival and departure trips, particularly during peak hours. This has direct implications for traffic management.
  • A transferable methodology: The proposed multi-step approach (clustering + POI analysis + logit modeling) is presented as a generalizable framework applicable to other cities.

Key Results

  • Four distinct TAZ mobility clusters were identified:
    • Cluster 1 (Significant Travel Intensity): Characterized by high hourly trip numbers, with a notable imbalance between morning arrival and afternoon departure, suggesting a strong workplace-oriented profile.
  • Cluster 2 (Normal Travel Intensity): Exhibited balanced departure and arrival peaks, indicating residence-oriented TAZs.
  • Cluster 3 (Moderate Travel Intensity): Showed moderate travel intensity with balanced departure/arrival patterns, often located near residential and workplace areas, suggesting a mixed-use or transitional function.
  • Cluster 4 (Low Travel Intensity): Characterized by low travel intensity and more stochastic movements, often found in less-developed areas or green spaces, indicating public facility-oriented or less active zones.
  • Land use correlations:
    • Workplace-dominated TAZs strongly correlated with Cluster 1 (high travel intensity, workplace-oriented).
  • Residence-dominated TAZs correlated with Cluster 2 (normal travel intensity, residence-oriented).
  • Commercial and business areas significantly influenced mobility patterns, particularly in Cluster 1.
  • Transportation facilities and green areas showed some influence, especially in Cluster 3.
  • The multinomial logit model provided coefficients indicating the likelihood of a TAZ belonging to a specific cluster based on its land use composition.

Limitations and Future Work

  • Short trip insensitivity: The authors acknowledge that mobile phone data may not accurately capture very short trips, potentially affecting the origin-destination identification.
  • Future Work Suggestions:
    • Temporal Scope: Investigate TAZ-based mobility patterns across different time periods (weekdays, weekends, seasons) for a more comprehensive understanding.
  • Cross-validation: Conduct more cross-validation studies, potentially incorporating other data sources like Automatic Fare Collection (AFC) and GPS data, to validate findings and improve short-trip detection.
  • Comparative Studies: Perform comparative and empirical studies across different cities to enrich the understanding of TAZ-based mobility patterns globally.

Summary updated

Abstract

Information and communication technology development has yielded large-scale spatiotemporal datasets, such as mobile phone, automatic collection system, and car-hailing data, which have resulted in new opportunities to investigate urban transportation systems. However, few studies have focused on regional mobility patterns. This study presents a multistep method for exploring traffic analysis zone (TAZ)-based mobility patterns and the corresponding relations with local land use characteristics. Based on a large-scale mobile phone dataset from a major mobile phone operator in Beijing, we applied the K-means clustering algorithm to the hourly aggregated trip data to create clusters of TAZs with similar temporal mobility patterns. Land use characteristics were then derived and correlated with the temporal TAZ-based mobility patterns. Four clusters of TAZs with the similar patterns and intensities of urban activities during given time windows were identified. Land use indicators, such as residence and commercial and business area indicators, were correlated with specific temporal TAZ-based mobility patterns. The proposed multistep method could be applied in other cities to enrich relevant analyses and improve urban design and transportation planning.
BibTeX
@inproceedings{chen2019assessing,
  author = {Yanyan Chen and Zheng Zhang and Tianwen Liang},
  title = {{Assessing Urban Travel Patterns: An Analysis of Traffic Analysis Zone-Based Mobility Patterns}},
  booktitle = {Sustainability},
  year = {2019},
  doi = {10.3390/su11195452},
}
Figure