Source-linked AI summary
Revealing travel patterns and city structure with taxi trip data
Xi Liu, Li Gong, Yongxi Gong, Yu Liu
TL;DR
The paper asks how spatial interactions from travel data can reveal urban structure beyond physical environments and resources. It analyzes Shanghai taxi GPS trips with complex network methods and community detection, finding a hierarchical polycentric structure whose sub-regions and centers have transportation-planning relevance. The authors note that taxi data represent only part of intra-urban travel, with long-distance commuting poorly reflected.
Problem
The paper addresses limited city-level analyses that systematically use intra-city flows to reveal regional urban structure rather than treating trip origins and destinations separately.
Method
The study applies complex network science and community detection to Shanghai taxi GPS trip data, emphasizing spatial interactions and different trip lengths.
Results
The study reveals a two-level hierarchical polycentric structure of Shanghai from spatial interactions represented by taxi trips.
Takeaways & Limitations
The identified sub-regions and borders can inform urban and transportation planning, help determine planning boundaries, and support validation of urban management policies.
Takeaways & Limitations
Taxi trips represent only part of intra-urban travel, and long-distance commuting is hardly reflected by taxi trajectories.
Abstract
from arXiv · showhide
Detecting regional spatial structures based on spatial interactions is crucial in applications ranging from urban planning to traffic control. In the big data era, various movement trajectories are available for studying spatial structures. This research uses large scale Shanghai taxi trip data extracted from GPS-enabled taxi trajectories to reveal traffic flow patterns and urban structure of the city. Using the network science methods, 15 temporally stable regions reflecting the scope of people's daily travels are found using community detection method on the network built from short trips, which represent residents' daily intra-urban travels and exhibit a clear pattern. In each region, taxi traffic flows are dominated by a few 'hubs' and 'hubs' in suburbs impact more trips than 'hubs' in urban areas. Land use conditions in urban regions are different from those in suburban areas. Additionally, 'hubs' in urban area associate with office buildings and commercial areas more, whereas residential land use is more common in suburban hubs. The taxi flow structures and land uses reveal the polycentric and layered concentric structure of Shanghai. Finally, according to the temporal variations of taxi flows and the diversity levels of taxi trip lengths, we explore the total taxi traffic properties of each region and proved the city structure we find. External trips across regions also take large proportion of the total traffic in each region, especially in suburbs. The results could help transportation policy making and shed light on the way to reveal urban structures with big data.
1. Introduction
The paper addresses how travel flows can reveal dynamic urban structure, responding to limited flow-based analyses that often separate trip origins and destinations. It uses taxi GPS data and complex network methods to analyze intra-city flows and travel patterns in Shanghai.
- Motivation: Urban structure is expressed through spatial interactions among places, with people and freight flows connecting physical resources into an integrated system.Flow systems reveal how regions interact and how city centers relate to their vicinities.
- Motivation: Big geospatial data from sources including taxi trajectories provide abundant, accurate, objective, and accessible movement locations for studying cities.These data offer advantages over traditional travel surveys for describing people’s movements.
- Research gap: Many city-level studies treat trip origins and destinations as unrelated activities, limiting systematic analyses of intra-city flows.Related work examined partitioned travel patterns or subway-based structure, but did not fully capture urban dynamics.
- Data rationale: Taxi GPS trajectories capture collective intra-city mobility with accurate positions and timestamps, making trips between consecutive activities easier to extract.They do not represent continuous displacements of specific people, but they support collective mobility analysis.
- Approach: The study introduces complex network science methods to analyze Shanghai taxi data and reveal intra-city flow structure across different trip lengths.It presents a bottom-up view of city structure based on residents’ travel flows and connects the analysis to transportation applications.
2. Study area and data preparation
The study examines Shanghai using taxi GPS trajectories covering the city, with emphasis on weekday travel and a defined urban–suburban study-area context. Taxis are presented as an important supplement to buses and metros for intra-city travel.
- Study area: Shanghai covers more than 6,000 km2 and comprises 16 districts and 1 county; the study excludes Chongming County.The core urban area includes eight Puxi districts and Lujiazui in Pudong.
- Travel context: Taxis produced 19.3% of intra-urban trips by public transportation in Shanghai in 2010.Public transportation served 34% of Shanghai travel and 47% in the core urban area.
- Study area: The study-area map distinguishes core urban districts from suburban and rural districts using red and purple regions, respectively.Chongming is shown as not included in the study area.
- Data preparation: More than 6,600 taxis supplied GPS trajectories covering the entire city, with relatively higher trip volumes in the central area.The analysis uses four consecutive weekdays from Monday to Thursday; Friday data were excluded because of nighttime entertainment travel.
3. Revealing the two-Level hierarchical polycentric city
The study models taxi trips as a weighted, directed network and progressively applies community detection to identify a two-level hierarchical city structure. Short trips define stable local zones, while longer trips merge these zones into broader regions whose centers and land uses reveal Shanghai’s polycentric organization.
- Network construction and community detection: Taxi trips between 1 km × 1 km cells form a weighted, directed network whose communities represent intensely interactive sub-regions.Each cell is a node, each origin–destination linkage is a directed edge, and edge weight equals the number of trips.
- Network construction and community detection: Short-trip networks are more spatially stable than long-trip networks, supporting more meaningful detection of local city structure.Although the networks contain approximately the same number of trips, 78% of long-trip edges aggregate three trips or fewer and are more random.
- Hierarchical city structure: Progressively adding longer trips produces a two-level hierarchy: stable L1Zs from N5 and broader L2Zs from N_all.N4–N6 yield similar communities, whereas later additions mainly merge existing communities, especially in urban areas.
- Hierarchical city structure: Most L1Z and L2Z boundaries differ from administrative districts, providing a bottom-up delineation based on residents’ intra-city travel flows.The extracted hierarchy emphasizes the different meanings of short- and long-distance travel rather than administrative organization.
- Properties and centers of sub-regions: The 15 L1Zs are densely connected and contain many low-volume cells alongside a few high-volume hubs.Betweenness and closeness centrality increase with node strength, indicating that high-strength hubs directly connect to more nodes and lie closer to others within an L1Z.
- Properties and centers of sub-regions: L2Zs retain some L1Z centers and add long-distance attractors, producing a hierarchical polycentric structure with distinct urban and suburban hub functions.Urban L1Z centers are mainly commercial and business areas, whereas suburban centers favor mixed residential areas and metro stations; transport terminals can become L2Z centers.
4. Discussion and conclusions
Using taxi-derived spatial interactions, the study identifies a two-level hierarchical polycentric structure of Shanghai and proposes sub-regional boundaries that reflect travel connectivity. These findings inform transportation and urban planning while acknowledging limitations in taxi-data coverage and representativeness.
- Taxi data represent only part of intra-urban travel, underrepresent some long-distance commuting, and sample passengers unevenly.The authors suggest combining multiple transportation data sources to obtain more comprehensive urban-interaction patterns.
- The study reveals Shanghai’s two-level hierarchical polycentric structure from spatial interactions represented by taxi trips.
- Sub-regional borders based on connectivity strength differ from administrative boundaries and may provide alternative boundaries for urban planning.
- The identified sub-regions can help validate whether built facilities or functional regions serve residents as planned.
- The hierarchical structure and land-use mechanisms can guide policies to improve accessibility, reduce travel, and adjust transit services to areas where demand exceeds current levels.The study discusses bus-route modifications, metro extensions, and land-use changes as possible planning responses.
- Future research could combine private-car, bus, and metro data and use longer time series to examine changing urban structures and policy effects.
Appendix
The appendix details how center functions and hierarchical regions are interpreted from taxi-flow timing, land use, and transportation roles. Urban centers tend to be commercial or business-oriented, whereas suburban centers more often combine residential uses with metro or public-transport functions.
- Center functions are inferred partly from temporal variations in taxi pick-up and drop-off points because mixed land uses make direct identification difficult.
- The listed L1Z and L2Z centers include commercial, residential, mixed-use, industrial, free-trade, railway, airport, and metro-station functions.
- Urban L1Z centers 1–6 are mainly commercial or business areas, with daytime taxi demand and evening departures reflecting daily office and commercial activity.
- Residential land use mixed with small commercial areas produces more balanced taxi arrivals and departures throughout the day in L1Z 7.
- Suburban centers are more strongly influenced by residential land use and metro stations than most urban centers.
- Metro-centered suburban regions show morning drop-off peaks as residents transfer to cheaper long-distance transit, followed by evening taxi pick-up peaks.
- Shanghai’s railway station and Hongqiao International Airport function as higher-level centers because they attract and generate strong long-distance flows.