Exploring Urban Mobility from Taxi Trajectories: A Case Study of

Nanjing, China

Yihong Yuan and Maël Le Noc

Department of Geography, Texas State University, San Marcos, TX, U.S.A.

Keywords: Taxi Trajectories, Time Series Analysis, Urban Dynamics, Big Geodata.

Abstract: Identifying urban mobility patterns is a crucial research topic in geographic information science,

transportation planning, and behavior modeling. Understanding the dynamics of daily mobility patterns is

essential for the management and planning of urban facilities and services. Previous studies have utilized taxi

trajectories collected from the Global Positioning System (GPS) to model various types of urban patterns,

such as identifying urban functional regions and hot spots. However, there is limited research on how the

results of these studies can be used to inform real-world problems in urban planning. This research examines

the development of sub-centers in Nanjing, China based on Taxi GPS trajectories. The results indicate a clear

separation between the urban center and the sub-centers. In addition, we also clustered the time series of taxi

pick-up locations to model dynamic urban movement and identify outlier patterns. The results demonstrate

the importance of considering human mobility patterns in identifying urban functional regions, which

provides valuable input for urban planners and policy makers.

1 INTRODUCTION

In recent decades, modeling human mobility patterns

has become an important research topic in various

fields such as computational physics, urban planning,

Intelligent Transport Systems (ITS), and Geographic

Information Science (GIS). The growing availability

of location-aware devices, such as the Global

Positioning System (GPS) receivers and smart phones

has provided new challenges and opportunities for

planners and policy makers to analyze, model, and

predict human mobility patterns (Chen et al., 2015).

Commonly used datasets include (but are not limited

to) georeferenced mobile phone data (Ahas, 2005;

Calabrese et al., 2013), location-based social media

check-ins (Cao et al., 2015; Lee et al., 2016),

Bluetooth tracking data (Delafontaine et al., 2012),

and floating car GPS locations (Ge et al., 2017, Jiang

et al., 2009). Among these datasets, GPS-enabled

floating cars are particularly effective at capturing

intra-urban mobility patterns across street networks

due to their high spatial precision and sampling

resolution (Jiang and Zhang, 2018; Yang et al., 2018;

Hu et al., 2018; Ge et al., 2017). In practice, floating

cars are often observed as taxis or shared rides in

many cities, where each floating car periodically

records its coordinates via a GPS receiver and stores

the information in a central server (Liu et al., 2012).

Previous studies have used taxi GPS data to

analyze human mobility patterns from both the

individual and urban perspectives (Atmaji and Sig,

2016; Castro et al., 2013; Chen et al., 2017; Cui et al.,

2016; Fanhas and Saptawati, 2016). On the one hand,

many individual-oriented studies focused on the

morphology and internal characteristics of taxi

trajectories and its implication for refining traditional

mobility models, such as modifying the power law

distribution under a given spatio-temporal context

(Jiang et al., 2009). These studies provided

quantitative support to better understand urban

dynamics and to help maintain sustainable mobility in

urban systems. For example, Santi et al., (2014)

quantified the benefit of ridesharing by translating

spatio-temporal sharing problems into a graph-

theoretic framework.

On the other hand, studies also investigated how

taxi trajectories revealed the characteristics of urban

functional regions and the interactions between urban

areas (Mazimpaka and Timpf, 2015; Tang et al.,

2015; Hu et al., 2014). For example, Liu et al., (2016)

used taxi GPS data to analyze the spatial interaction

patterns between parcels, as well as the real-time land

use patterns reflected by the interaction.

Yuan Y. and Le Noc M.

Exploring Urban Mobility from Taxi Trajectories: A Case Study of Nanjing, China.

DOI: 10.5220/0006891203020307

In Proceedings of the 7th International Conference on Data Science, Technology and Applications (DATA 2018), pages 302-307

ISBN: 978-989-758-318-6

Although recent studies have provided valuable

input to advance the theories and methods in

modeling urban mobility patterns from taxi

trajectories, few studies have directly connected their

empirical results to real-world planning problems and

challenges that cities face, nor have they used the

results to validate a planning policy that has actually

been implemented. As such, this study is conducted

from a problem-driven instead of a data-driven

perspective. It is motivated by the 1991-2010

development and planning agenda published by the

local government of Nanjing, China (Nanjing City

Council, 1995), where the city reviewed challenges of

over-crowded central business areas (CBD) and the

associated imbalance in urban resources. The city

proposed to rapidly grow several suburbs of Nanjing

as functional sub-centers for local residents. This

research utilizes one-week taxi GPS data from 2010

to address the following research questions:

 Spatially, are there established functional sub-

centers in Nanjing as proposed in the development

plan? If so, how does this multi-center urban

structure affect the distance and direction

distribution of taxi trips in Nanjing?

 Temporally, how should the mobility dynamics of

urban functional regions be modeled? Are there

any regions with outlier time series?

2 RESEARCH RESIGN

2.1 Data

The dataset utilized in this research is obtained from

Nanjing City, China. Nanjing is the capital of Jiangsu

province, situated in the Yangtze River Delta Region

in China. With an administrative area of 6,600 km

and a total population of over 8 million, Nanjing is

the second largest city in eastern China and is well

known for being a major cultural, economic, tourist,

and transit center.

The dataset used in this research contains

approximately 135 million GPS locations of 7,194

taxis between June 4, 2010 and June 10, 2010. In

addition to GPS coordinates, the dataset also includes

a unique identifier (ID) for each taxi, a unique ID for

each record, the timestamp for when the location was

logged, driving speed, driving direction, and the

current status of the taxi, where 0 indicates that the

taxi is free to pick up passengers and 1 indicates that

the taxi is currently occupied. The sampling

resolution is between 15 to 30 seconds. On average,

there are 4,666 records per day per taxi. Table 1

provides a sample record of our dataset.

Table 1: A sample taxi record.

Record ID

10000000

Taxi ID

13451852779

Longitude

118.779696

Latitude

32.025413

Speed (km/h)

Direction

(degree)

Status

Figure 1a shows the administrative boundary of

our study area. To facilitate the analysis, we divided

the study area into 1km*1km grid cells. Figure 1b

shows the approximate main urban center in Nanjing

(in red) and two close-by sub-centers (in blue): Pukou

district in the northwest and Dongshan district in the

south.

(a) (b)

Figure 1: Study area. (a) 1km*1km grid cells; (b) Urban

center and sub-centers.

2.2 Methodology

2.2.1 Preprocessing Data and Calculating

Spatial Indicators

Previous studies have used the number of taxi trips to

represent the activeness of an urban area (Ferreira et

al., 2013). In this study, we identify the pick-up and

drop-off points of each trip based on the “occupied”

status field in the original data. Over 1.3 million trips

were extracted. For each trip, we calculate the travel

distance between the pick-up and the drop-off points,

as well the travel direction as defined in Figure 2.

Note that in this study we are interested in how the

origination of taxi trips helps explain the delineation

of sub-centers in Nanjing, so we only consider the

pick-up points in each grid cell.

Figure 2: Calculate travel direction.

The number of pick-ups and the median travel

distance represent the magnitude and the scale of

mobility originated from different urban areas. In

addition, we are also interested in how travel direction

provides informative input for the urban structure in

Nanjing. A direction entropy indicator is defined to

present the randomness of travel directions. The

formula is derived as follows:

(1)

Where p

refers to the percentage of trips going to a

given direction, and N stands for the total number of

directions (Table 2). For simplicity, we divided the

space into eight directions. The direction entropy is

considered an indicator for the consistency of travel

directions.

Table 2: Define eight travel directions.

Direction

Angle

North-Northeast

[0,45)

Northeast-East

[45,90)

East-Southeast

[90;135)

Southeast-South

[135;183)

South-Southwest

[180;225)

Southwest-West

[225;270)

West-Northwest

[270;315)

Northwest-North

[315;360)

2.2.2 Modeling Temporal Patterns

In addition to the spatial heterogeneity of human

mobility, previous studies also explored urban

functional regions based on temporal variations of

activities. The second analysis of this study explores

how the number of pick-ups varies at different times

of the day. We aggregate the number of trips for each

grid cell by hour and use a dynamic time warping

(DTW) algorithm to measure the similarity of hourly

mobility patterns between grid cells. DTW has

proven to be robust to distortion in time series (Yuan

and Raubal, 2012, Zhang et al., 2008), so it allows us

to group similar patterns and identify outlier patterns.

3 PRELIMINARY RESULTS

3.1 Exploring Urban Sub-Centers

Figure 3a shows the density distribution of taxi pick-

ups. As expected, most trips originate from the urban

center; however, the Dongshan sub-center also

demonstrates a cluster of pick-up points. The other

sub-center, Pukou, has substantially lower densities

of pick-up points. Figure 3b shows the distribution of

median trip length by grid cell. As can be seen, the

results map shows three delineated areas, and there is

a clear division between sub-centers and the city

center, where taxi trips get longer for places farther

away from the city center or a sub-center. This

demonstrates an evident separation of urban

functional regions, where residents can take a short

trip (i.e., 0-3km) for their daily needs inside each

center or sub-center.

(a) (b)

Figure 3: Spatial distribution of taxi trips (the dashed lines

delineate the urban center and the sub-centers). (a) Density

of pick-ups; (b) Median travel distance.

To verify this finding, we also calculated the

direction entropy for each grid cell. As discussed in

Section 2.2.1, direction entropy represents the

randomness of movement direction. A lower entropy

for a given cell indicates that trips from this cell

follow a more unified moving direction. Similar to

Figure 3b, Figure 4 also demonstrates a clear division

between the urban center and sub-centers, where trips

originate from the central area are likely to go to

different directions (i.e., with a high direction

entropy), but trips that start from the border of urban

centers follow a more unified direction, as they are

mostly going into the urban center/sub-center.

Figure 4: The distribution of direction entropy.

3.2 Temporal Variations of Urban

Trips

As discussed in Section 2.2.2, we aggregate the

number of pick-ups by hour for each grid cell, so each

cell is associated with a 24-hour time series showing

the temporal pattern of taxi pick-ups. To ensure data

quality, here we only consider cells that average more

than ten pick-ups per hour. For each cell, the hourly

pick-ups values are divided by the maximum of the

24 hourly values. This standardizes the magnitude of

data and helps with investigating the internal structure

of each time series (Yuan and Raubal, 2012). Based

on the DTW algorithm described in Section 2.2.2, we

construct the distance matrix for the relative time

series associated with each grid cell. The output is a

distance matrix D, in which D

represents the DTW

distance between cell i and j. Based on the DTW

matrix, we conducted a hierarchical clustering

analysis to identify outlier time series. There are

several methods to set the number of clusters in

hierarchical clustering. As an example analysis, here

we adopt the criteria discussed in Mardia et al., (1979),

where numCluster = max(2; sqrt(n/2)), n is the

number of cells. Here outliers are defined as clusters

with fewer than 3 cells.

Figure 5 shows the clustering results, where red

indicates outliers and blue are regular cells with more

than 10 pick-ups per hour. Figure 6a shows the time

series of pick-ups from an example outlier cell

(highlighted in the inset map of Figure 5). As can be

seen, the number of taxi pick-ups increases during

night hours. The background imagery in Figure 6b

shows that this cell covers a residental neiborghood,

and the increasing night hour pick-ups are not

consistent with common sense, where more residents

should arrive instead of depart home during night

hours. The result demonstrates that more detailed

activity patterns need to be investigated in follow-up

studies. For example, it is possible that many

residents in this area work during night hours, or there

may be recereational events that attract people to go

out at night. This example demonstrates the

effectiveness of our methods in capturing fine-scale

activity patterns that may not be reflected by basic

land use satelite imagery.

Figure 5: Outlier cells.

(a)

(b)

Figure 6: An example outlier cell. (a) 24-hour time series of

pick-ups; (b) Satellite imagery.

4 CONCLUSIONS

This study examined the urban mobility patterns in

Nanjing, China based on taxi GPS trajectories. We

analyzed the distribution of pick-up locations, travel

distance, and travel direction. All three indicators

demonstrate a clear division between the city center

of Nanjing and the sub-centers. The result can be used

to support and examine the development agenda of

Nanjing City Council, where city officials proposed

to further grow sub-centers to release the pressure

from the central business area. In addition, we also

explored the temporal dynamics of different urban

regions based on a DTW algorithm. The extracted

outliers demonstrated the importance of

incorporating human mobility and activity data to

refine small-area land use classification. Further

research can focus on incorporating more indicators

from the taxi trajectories to improve the accuracy of

the analysis. It is also important to cross-validate the

results with other public data, such as census and

urban demographic data. Also, further research may

involve extending the study period to analyze

seasonal time series patterns when the data becomes

available. The methodology discussed in this paper

can be applied to other cities to identify urban

functional regions and provide useful input for policy

makers.

REFERENCES

Ahas, R. 2005. Mobile phones and geography: Social

Positioning Method. Power over Time-Space:

Inaugural Nordic Geographers Meeting. Lund,

Sweden.

Atmaji, F. T. D. & Sig, K. Y. 2016. Mining the GPS big

data to optimize the taxi dispatching management. 2016

4th International Conference on Information and

Communication Technology (ICoICT).

Calabrese, F., Diao, M., Di Lorenzo, G., Ferreira, J. & Ratti,

C. 2013. Understanding individual mobility patterns

from urban sensing data: A mobile phone trace

example. Transportation Research Part C-Emerging

Technologies, 26, 301-313.

Cao, G., Wang, S., Hwang, M., Padmanabhan, A., Zhang,

Z. & Soltani, K. 2015. A scalable framework for

spatiotemporal analysis of location-based social media

data. Computers, Environment and Urban Systems, 51,

70-82.

Castro, P. S., Zhang, D. Q., Chen, C., Li, S. J. & Pan, G.

2013. From Taxi GPS Traces to Social and Community

Dynamics: A Survey. ACM Computing Surveys, 46.

Chen, B. Y., Yuan, H., Li, Q., Shaw, S.-L., Lam, W. H. K.

& Chen, X. 2015. Spatiotemporal data model for

network time geographic analysis in the era of big data.

International Journal of Geographical Information

Science, 1-31.

Chen, Z. L., Gong, X. & Xie, Z. 2017. An analysis of

movement patterns between zones using taxi GPS data.

Transactions in GIS, 21, 1341-1363.

Cui, J. X., Liu, F., Janssens, D., An, S., Wets, G. & Cools,

M. 2016. Detecting urban road network accessibility

problems using taxi GPS data. Journal of Transport

Geography, 51, 147-157.

Delafontaine, M., Versichele, M., Neutens, T. & Van De

Weghe, N. 2012. Analysing spatiotemporal sequences

in Bluetooth tracking data. Applied Geography, 34,

659-668.

Fanhas, R. S. & Saptawati, G. a. P. 2016. Discovering

Frequent Origin-Destination Flow from Taxi GPS Data.

Proceedings of 2016 International Conference on Data

and Software Engineering (ICoDSE).

Ferreira, N., Poco, J., Vo, H. T., Freire, J. & Silva, C. T.

2013. Visual Exploration of Big Spatio-Temporal

Urban Data: A Study of New York City Taxi Trips. Ieee

Transactions on Visualization and Computer Graphics,

19, 2149-2158.

Ge, W., Shao, D., Xue, M., Zhu, H. & Cheng, J. 2017.

Urban Taxi Ridership Analysis in the Emerging

Metropolis: Case Study in Shanghai. Transportation

Research Procedia, 25, 4916-4927.

Hu, X. W., An, S. & Wang, J. 2014. Exploring Urban Taxi

Drivers' Activity Distribution Based on GPS Data.

Mathematical Problems in Engineering, 708482.

Hu, X. W., An, S. & Wang, J. 2018. Taxi Driver's Operation

Behavior and Passengers' Demand Analysis Based on

GPS Data. Journal of Advanced Transportation,

6197549.

Jiang, B., Yin, J. J. & Zhao, S. J. 2009. Characterizing the

human mobility pattern in a large street network.

Physical Review E, 80, 021136-1 - 021136-11.

Jiang, W. W. & Zhang, L. 2018. The Impact of the

Transportation Network Companies on the Taxi

Industry: Evidence from Beijing's GPS Taxi Trajectory

Data. IEEE Access, 6, 12438-12450.

Lee, J. H., Davis, A. W., Yoon, S. Y. & Goulias, K. G.

2016. Activity space estimation with longitudinal

observations of social media data. Transportation, 43,

955-977.

Liu, X., Kang, C. G., Gong, L. & Liu, Y. 2016.

Incorporating spatial interaction patterns in classifying

and understanding urban land use. International

Journal of Geographical Information Science, 30, 334-

350.

Liu, Y., Kang, C. G., Gao, S., Xiao, Y. & Tian, Y. 2012.

Understanding intra-urban trip patterns from taxi

trajectory data. Journal of Geographical Systems, 14,

463-483.

Mardia, K. V., Kent, J. T. & Bibby, J. M. 1979. Multivariate

analysis, London & New York, Academic Press.

Mazimpaka, J. D. & Timpf, S. 2015. Exploring the

Potential of Combining Taxi GPS and Flickr Data for

Discovering Functional Regions. AGILE 2015:

Geographic Information Science as an Enabler of

Smarter Cities and Communities, 3-18.

Nanjing City Council. 1995. The development agenga of

Nanjing: 1991-2010 [Online]. Available:

http://www.njghj.gov.cn [Accessed].

Santi, P., Resta, G., Szell, M., Sobolevsky, S., Strogatz, S.

H. & Ratti, C. 2014. Quantifying the benefits of vehicle

pooling with shareability networks. Proceedings of the

National Academy of Sciences of the United States of

America, 111, 13290-13294.

Tang, J. J., Liu, F., Wang, Y. H. & Wang, H. 2015.

Uncovering urban human mobility from large scale taxi

GPS data. Physica A-Statistical Mechanics and Its

Applications, 438, 140-153.

Yang, Z., Franz, M. L., Zhu, S. J., Mahmoudi, J., Nasri, A.

& Zhang, L. 2018. Analysis of Washington, DC taxi

demand using GPS and land-use data. Journal of

Transport Geography, 66, 35-44.

Yuan, Y. & Raubal, M. Extracting dynamic urban mobility

patterns from mobile phone data. Geographic

Information Science - 7th International Conference,

2012 Columbus, USA. Lecture Notes in Computer

Science, Springer, 354-367.

Zhang, B., Yan, K., Jiang, S. & Wei;, D. 2008. Walking

Stability Analysis by Age Based on Dynamic Time

Warping. IEEE International Conference on Computer

and Information Technology.