Understanding taxi travel patterns

https://doi.org/10.1016/j.physa.2016.03.047Get rights and content

Highlights

  • Holistic analysis of taxi travel patterns including occupied and unoccupied trips.

  • Trip displacement of short trips follows power-law distribution.

  • Trip displacement of long trips follows exponential distribution.

  • With increased characteristic travel distances, taxis can have dual high-probability locations.

Abstract

Taxis play important roles in modern urban transportation systems, especially in mega cities. While providing necessary amenities, taxis also significantly contribute to traffic congestion, urban energy consumption, and air pollution. Understanding the travel patterns of taxis is thus important for addressing many urban sustainability challenges. Previous research has primarily focused on examining the statistical properties of passenger trips, which include only taxi trips occupied with passengers. However, unoccupied trips are also important for urban sustainability issues because they represent potential opportunities to improve the efficiency of the transportation system. Therefore, we need to understand the travel patterns of taxis as an integrated system, instead of focusing only on the occupied trips. In this study we examine GPS trajectory data of 11,880 taxis in Beijing, China for a period of three weeks. Our results show that taxi travel patterns share similar traits with travel patterns of individuals but also exhibit differences. Trip displacement distribution of taxi travels is statistically greater than the exponential distribution and smaller than the truncated power-law distribution. The distribution of short trips (less than 30 miles) can be best fitted with power-law while long trips follow exponential decay. We use radius of gyration to characterize individual taxi’s travel distance and find that it does not follow a truncated power-law as observed in previous studies. Spatial and temporal regularities exist in taxi travels. However, with increasing spatial coverage, taxi trips can exhibit dual high probability density centers.

Introduction

Understanding the dynamics of human mobility is critical to many fields such as epidemics  [1], [2], transportation [3], [4], urban planning  [5], and genetics  [6]. Recent development of information and communications technology (ICT) has significantly improved our ability to collect, store, and analyze large-scale datasets, which enables studying human mobility at a wide range of spatial and temporal scales  [7]. Analyzing statistical properties of bank notes circulation as a proxy of human movement, Brockmann et al.  [8] observed that human travel distance exhibits a power law distribution and human travel trajectories may be approximated as Lévy flights (heavy tailed random walk)  [8]. This observation was confirmed by Rhee et al. [9] using Global Positioning System (GPS) traces collected from volunteers, showing non-negligible probability of high displacement trips and long pause-time between trips  [9]. Despite the randomness indicated by Lévy flight models, using cell phone traces, González et al. [10] discovered that strong spatial and temporal regularities exist for human trajectories—individuals have a tendency to frequently revisit a few locations  [10]. To explain the observed scaling properties and regularities, Song et al.  [11] proposed a combination of two mechanisms, exploration and preferential return, for modeling human mobility  [11]. It is worth noting that data used in these studies include trajectories generated from different transportation modes (walking, driving, public transportation, and flying etc.). Yan et al.  [12] argued that transportation modes affect aggregated travel patterns and displacements from single mode transportation should follow an exponential distribution instead of power law  [12].

Vehicular travel is one important transportation mode that has implications to many societal challenges  [13]. Due to the privacy concerns related to private vehicle travel data, most studies on examining vehicle travel patterns so far are based on taxi trajectories captured by GPS. Despite having different travel needs compared to private vehicles, taxis can be viewed as floating sensors in a city that provide invaluable information on mobility dynamics, traffic conditions, and epidemic spreading risks. In addition, compared to other datasets used in studying human mobility dynamics such as the movement of bank notes  [8], smart card data  [14], and mobile phone records  [10], GPS traces of taxis usually provide more detailed information with finer spatial and temporal resolutions and larger sampling size.

Previous studies on taxi travel patterns have been focused on extracting passenger trips to understand passenger travel patterns  [3], [15], [16], [17]. Peng et al.  [16] characterized taxi passenger trips for different travel needs  [16]. Jiang et al.  [3] verified the scaling properties of taxi passenger-trip length and suggested that such property is attributed to the underlying street topology  [3]. Liang et al.  [17] found that passenger trip displacements of taxis can be better approximated with exponential distributions instead of power law, possibly due to high costs associated with long taxi trips  [17]. However, passenger trips are not representative of how taxis travel. In many cities, taxis always cruise around between passenger trips  [18], [19]. In some countries such as China, taxi drivers possess the vehicle throughout the day, sharing similar commuting patterns with private vehicles (i.e., returning home after work and then driving out for business the next day). To date, limited attentions have been paid to the patterns of taxi travels as a whole.

Understanding taxi travel patterns can also contribute to addressing several urban sustainability challenges. In particular, taxis are important components of the public transportation sector  [13], providing flexible and convenient mobility solutions for urban residents. Because they are on the road all the time, taxis also contribute significantly to traffic congestion, energy consumption, and air pollution in the cities  [20]. In addition, taxis in many countries are likely to be early adopters of emerging technologies (e.g., electric vehicles, connected and autonomous vehicles) and business models (e.g., Uber) [21], [22]. Therefore, it is important to understand the travel patterns of taxis, taking into account the taxi travels both with and without passengers. This research aims to fill this gap by examining taxi trajectory data holistically to understand how taxi travel patterns are similar or different from those of individuals and private vehicles. Results of this research can help guide traffic modeling, transportation planning, urban planning, and infrastructure development.

Section snippets

Methods

This study examines vehicle trajectory data of taxis in Beijing to study taxi travel patterns. The data are analyzed from the perspective of characteristic travel length (radius of gyration), distribution of trip distance, and spatial density distribution in each taxi’s intrinsic reference frame.

Characteristic travel distance and mass center

Strong spatiotemporal regularities exist for taxi trajectories. On a daily basis, taxi drivers leave home in the morning for business and return home at late night. This routine could be easily identified from the segments of flat lines in Fig. 1 showing the periods of time when this example taxi is parked at its driver’s home (taxi drivers in Beijing normally take the vehicle home after work, please refer to the Method section for more background information about the taxi fleet in Beijing).

Conclusions

In this paper, we examine the travel patterns of taxis from a holistic vehicle travel perspective, which considers not only the trips to satisfy passenger needs, but also the movement of taxis to satisfy the travel needs of the drivers (e.g. leaving from home for business and returning to home after work, cruising along streets for customers, personal travels). Our results show that taxi travel patterns share similar traits with travel patterns of individuals, but also exhibit specific

Acknowledgments

This material is based upon work partially supported by the Department of Energy under Award Number DE-PI0000012. H.C. also acknowledges the support from the ERM Group Foundation.

References (27)

  • L. Hufnagel et al.

    Forecast and control of epidemics in a globalized world

    Proc. Natl. Acad. Sci. USA

    (2004)
  • B. Jiang et al.

    Characterizing the human mobility pattern in a large street network

    Phys. Rev. E

    (2009)
  • R. Kitamura et al.

    Micro-simulation of daily activity-travel patterns for travel demand forecasting

    Transportation

    (2000)
  • Cited by (0)

    1

    Tel.: +1 765 494 7701; fax: +1 765 494 7704.

    2

    Tel.: +1 734 763 8644; fax: +1 734 936 2195.

    View full text