Source-linked AI summary
The scaling of human mobility by taxis is exponential
Xiao Liang, Xudong Zheng, Weifeng Lv, Tongyu Zhu, Ke Xu
TL;DR
The paper examines human mobility using fine-grained taxi trajectories, addressing limitations in earlier mobility datasets. It models 20 million trajectories from more than 10,000 Beijing taxis and finds exponential displacement and elapsed-time distributions, alongside bursty interevent activity.
Problem
Existing human-mobility datasets have limitations including coarse granularity, indirect observations, and deviations from actual movements, motivating fine-grained urban trajectory analysis.
Method
The paper analyzes fine-grained trajectories collected from more than 10,000 Beijing taxis and compares power-law and exponential models using AIC-based selection.
Results
20 million taxi trajectories show exponential distributions for urban displacements and elapsed time, while interevent-time analysis indicates bursty human mobility.
Takeaways & Limitations
Urban taxi mobility in this study differs from commonly reported power-law patterns, and interevent times resemble bursty activity found in other human activities.
Takeaways & Limitations
The quality of mobility data depends heavily on elapsed-time characteristics and is dataset dependent.
Abstract
from arXiv · showhide
As a significant factor in urban planning, traffic forecasting and prediction of epidemics, modeling patterns of human mobility draws intensive attention from researchers for decades. Power-law distribution and its variations are observed from quite a few real-world human mobility datasets such as the movements of banking notes, trackings of cell phone users' locations and trajectories of vehicles. In this paper, we build models for 20 million trajectories with fine granularity collected from more than 10 thousand taxis in Beijing. In contrast to most models observed in human mobility data, the taxis' traveling displacements in urban areas tend to follow an exponential distribution instead of a power-law. Similarly, the elapsed time can also be well approximated by an exponential distribution. Worth mentioning, analysis of the interevent time indicates the bursty nature of human mobility, similar to many other human activities.
1. Introduction
Human mobility research uses location traces to model reproducible travel patterns relevant to urban planning, traffic forecasting, and epidemic prediction. This paper addresses limitations in existing datasets by analyzing fine-grained taxi trajectories from Beijing.
- Location tracking and location-based services provide large-scale data for studying human mobility patterns.These data support research in urban planning, traffic forecasting, marketing, epidemic prediction, and mobile-network protocol design.
- Studies of cell-phone, GPS, wireless-network, location-based-service, and banking-note data report reproducible mobility patterns.
- Power-law and Lévy-walk or Lévy-flight models have frequently been used to describe observed human and animal travel behavior.
- Earlier datasets can have coarse granularity, indirect observations, long gaps, and locations that deviate substantially from actual movements.Examples include banking-note transfers and cell-phone records captured during trips rather than at their endpoints.
- 20 millions of trajectories from more than 10,000 taxis in urban Beijing provide fine-grained mobility data and may reveal effects of the urban traffic network.
2. Preliminary
The paper uses model selection to compare power-law and exponential fits while accounting for both data fit and model complexity. Maximum likelihood estimates, AIC scores, and Akaike weights determine the preferred model.
- Model selection compares candidate models using both goodness of fit and model complexity.AIC and BIC are identified as commonly used model-selection criteria.
- The analysis mainly compares the power-law y = Ax^−α with the exponential y = Be^−λx.A and B are normalization constants.
- Parameters are first estimated by maximum likelihood, followed by calculation of each model’s AIC score.
- AIC combines the model likelihood with a penalty based on the number of parameters.Li is the likelihood at the estimated parameters, and Ki is the number of parameters in model i.
- Akaike weights represent relative likelihoods of being the best model for the observed data.The model with the largest Akaike weight is selected.
3. Data description
The study constructs taxi trajectories from two Beijing GPS datasets collected at approximately one-minute intervals, retaining operational status to identify passenger trips. It derives displacement, elapsed time, and passenger-free interevent time after excluding implausible trips.
- Two Beijing taxi GPS datasets cover October–December 2010 and October–November 2008, with records collected at about one-minute intervals.Each record includes taxi identity, location, velocity, operational status, and time.
- Operational status identifies where and when customers entered and exited the same taxi, enabling trajectory construction.A trajectory is represented by taxi, origin, origin time, destination, and destination time.
- The extracted trajectories support calculation of displacement and elapsed travel time, while passenger-free duration defines the interevent time.
- Trips shorter than 1 minute or longer than 120 minutes are excluded as invalid.
- 12,028,929 trajectories are derived for D1 and 9,942,697 for D2.The main analyses use D1, while D2 is used to compare changes in trends.
- Table 1 reports the numbers of trajectories and taxis in both datasets.
4. Statistical results and explanations
The study analyzes taxi mobility at fine spatial and temporal granularity, finding exponential patterns for urban displacements and elapsed times, alongside bursty interevent dynamics. These distributions vary with urban context, traffic conditions, and passenger demand.
- Displacement: AIC model selection indicates that displacement distributions in both distance ranges are better fitted by exponential distributions with exponential cutoffs than by power laws.The conclusion is supported by Wexp ≫ Wpow and by the distinct exponential fits observed in Figure 1a.
- Displacement: Displacement patterns are broadly stable across the two datasets, whose CCDFs nearly coincide and whose maximum-likelihood estimates are very close.The authors interpret this as indicating no obvious changes in human travel patterns in recent years.
- Displacement: Displacement distributions are generally exponential across sampled areas, except BCIA, where trips are longer and differ because it lies in Beijing’s suburb.Most trips originating in these areas are shorter than 20 km.
- Elapsed time: Elapsed-time distributions are approximately exponential for urban trips, with about 98.9% of D1 and 99.5% of D2 trips lasting under 60 minutes.The longer-time portions incline toward exponential behavior in D1 and power-law behavior in D2.
- Displacement and elapsed time: For elapsed times of at most 40 minutes, mean displacement increases approximately linearly with time as ∆l = µ∆T, with µ = 0.3326.The growth rate slows at larger elapsed times, which the authors associate with congestion and lower average speed; the relation is only a numerical approximation.
- Displacement and elapsed time: The fitted exponential exponents are λl = 0.2329 and λT = 0.0797, while 0.2329 × 0.3326 ≈ 0.0775 closely matches the elapsed-time estimate.This provides an approximate relationship between the displacement and elapsed-time distributions.
- Interevent time: Taxi inactivity durations are close to an inverse-square power law, whereas passenger-occupied activity durations are well approximated by an exponential.Interevent times also indicate non-Poisson, bursty travel demand, including frequent events followed by long inactive periods.
5. Conclusion and future work
Using 20 million trajectories from 10,000 Beijing taxis, the paper finds exponential models for urban displacements and elapsed times, alongside bursty interevent activity. Future work examines the roles of movement range, economic effects, data limitations, and cross-dataset relationships.
- 20 million trajectories from 10,000 Beijing taxis underpin models of urban human mobility.The trajectories were collected in urban areas of Beijing.
- Urban taxi displacements tend to follow an exponential distribution rather than a power-law, while elapsed times are also well approximated exponentially.The paper attributes the displacement pattern possibly to movement range and economic effects.
- Bursty interevent activity between trajectories supports the bursty nature of human daily activities.The paper presents this as a contribution of its interevent-time modeling.
- Future work will further examine how movement range and economic effects affect taxi trajectories.These two factors are identified as affecting the trajectories.
- The authors will study how data limitations influence displacement-distribution modeling and how taxi-based mobility patterns relate to patterns in other datasets.The proposed comparisons extend beyond taxi trajectories.