Understanding Individual and Collective Mobility Patterns from Smart Card Records: a Case Study in Shenzhen

Understanding individual and collective mobility patterns from smart card records: A case study in Shenzhen

The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters.

Citation Liang Liu et al. “Understanding individual and collective mobility patterns from smart card records: A case study in Shenzhen.” Intelligent Transportation Systems, 2009. ITSC '09. 12th International IEEE Conference on. 2009. 1-6. © 2010 Institute of Electrical and Electronics Engineers.

As Published http://dx.doi.org/10.1109/ITSC.2009.5309662

Publisher Institute of Electrical and Electronics Engineers

Version Final published version

Citable link http://hdl.handle.net/1721.1/58794

Terms of Use Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. Proceedings of the 12th International IEEE Conference WeBT4.4 on Intelligent Transportation Systems, St. Louis, MO, USA, October 3-7, 2009 Understanding individual and collective mobility patterns from smart card records: a case study in Shenzhen

Liang Liu, Anyang Hou, Assaf Biderman, Carlo Jun Chen Ratti School of Transportation SENSEable City Lab Tongji University Massachusetts Institute of Technology Shanghai, China Cambridge, MA, USA [email protected] {liuliang, cnhanya,abider,ratti}@mit.edu

Abstract—Understanding the dynamics of the inhabitants’ behavioral data have begun to emerge. For example, daily mobility patterns is essential for the planning and Rattenbury et al. [3] and Girardin et al. [4] used geo- management of urban facilities and services. In this paper, tagging patterns of photographs in Flickr to automatically novel aspects of human mobility patterns are investigated detect interesting real-world events and draw conclusions by means of smart card data. Using extensive smart card about the flow of tourists in a city. In addition, as city-wide records resolved in both time and space, we study the urban infrastructures such as buses, taxis, subways, public mean collective spatial and temporal mobility patterns at utilities, and roads become digitized, other sources of real- large scales and reveal the regularity of these patterns. We world datasets that can be implicitly sensed are becoming available. Ratti et al. [5], Reades et al.[6] and González et also investigate patterns of travel behavior at the al. [7] used cellular network data to study city dynamics individual level and show that the concentricity and and human mobility. McNamara et al. [8] used data regularity of mobility patterns. The analytical collected from an RFID-enabled subway system to predict methodologies to spatially and temporally quantify, co-location patterns amongst mass transit users. Such visualize, and examine urban mobility patterns developed sources of data are ever-expanding and offer large, in this paper could provide decision support for transport underexplored datasets of physically-based interactions planning and management. with the real world. Keywords-mobility pattern; smart card; intelligent transportation system (ITS); visualization; regularity In this paper, we introduce a novel method for understanding human mobility in dense urban area based on millions of smart card records. We show how these data I. INTRODUCTION regarding the position and intensity of digital footprint can The city never sleeps. The human movement be used to infer cultural and geographic aspects of the city constitutes the pulse of the city. Observing and modeling and reveal urban mobility pattern, which corresponds to human movement in urban environments is central to human movement in the city. traffic forecasting, understanding the spread of biological viruses, designing location-based services, and improving In particular, the main contributions of this paper are: urban infrastructure. However, little has changed since (1) demonstrating the potential of using smart card records Whyte [1] observed in his "Street Life Project" that the as data sources to gain insights into city dynamics and actual usage of New York's streets and squares clashed aggregated human behavior; (2) exploring the relationship with the original ideas of architects and city planners. A between spatiotemporal patterns of smart card usage and key difficulty faced by urban planners, virologists, and underlying city behavior and geography; and (3) studying social scientists is that obtaining large, real-world patterns in smart card usage, including an analysis of how observational data of human movement is challenging and factors such as the time of the day affect this prediction. costly [2]. We believe this work not only has direct implications In recent years, the large deployment of pervasive for the design and operation of future urban public technologies in cities has led to a massive increase in the transport systems (e.g., more precise bus/subway volume of records of where people have been and when scheduling, improved service to public transport users), but they were there. These records are the digital footprint of also for urban planning (e.g., for transit oriented urban individual mobility pattern. As websites have evolved to development), traffic forecasting, the social sciences [9]— offer geo-located services, new sources of real-world in particular, studying how people move about a city—and

Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 15:09:57 UTC from IEEE Xplore. Restrictions apply. the development of novel context-based mobile services. journey, we could get the time and location of check-in In addition, we expect that similar types of analyses can be and check-out, from which we could infer the OD feature applied to other sources of urban digital traces such as of trip. Moreover, the transfer information could be those provided by parking management (e.g., San derived from the smart card data. Francisco’s SFpark) and cellular networks [7]. Our work thus emphasizes the increasing role that data mining and Right now Shenzhen Metro Phase 1 is composed by visualization techniques will play to assist the east part of line 1 and south part of line 4. The east part of aforementioned fields in analyzing traces of human line 1 is from Luo Huo Railway station to Shijiezhichuang; behavior. Our work seems to open the way to a new the south of line 4 is from Futian Port to Shaoniangong. approach to the understanding of urban systems, which we The total length of Shenzhen Metro Phase 1 is 21.866 have termed “Urban Mobility Landscapes.” Urban kilometers, and there are 19 subway stops in total (the Mobility Landscape could give new answers to long- detailed information is showed in figure 1. The land use standing questions in urban planning: how to map vehicle information is in table 2). origins and destinations? How to understand the patterns of inhabitant movement? How to highlight critical points in the urban infrastructure? What is the relationship between urban forms and flows? And so on. This paper is organized by the following sequences: in section II, we describe the data sets and the feature extraction process; in section III, we investigate the aggregated spatiotemporal patterns of urban mobility; in Figure 1. Shenzhen subway stop description [11] section IV, the individual mobility pattern is explored; in section V, we draw the conclusion and discuss the future work. TABLE II. LAND USE INFORMATION OF DIFFERENT SUBWAY STOPS

II. DATASETS DESCRIPTION code name Land use The datasets used to describe urban mobility patterns 1 Luohu External transport cover two major public transport modes, i.e. bus, subway. 2 Guomao Business For smart card data, according to the survey conducted by 3 Laojie Recreation the smart card company, there are 55% passengers using 4 Dajuyuan Business (CBD) smart card in bus trip and 61% passengers using smart card in subway trip [10].The smart card data is from 5 million 5 Kexueguan government Business and recreation smart card users’ transit records for one month, through 6 HuaqiangRd December 1st, 2008 to December 31st, 2008. Every day (CBD) there are 1.5 million transit records from the users. The 7 Gangxia residential smart card data description and sample is showed in table 8 HuizhanZhongxin business 1. 9 GouwuGongyuan business 11 Xiangmihu recreation TABLE I. SMART CARD DATA DESCRIPTION 12 Chegongmiao business Field Memo 13 Zhuzilin Ex- transport Date December 5th , 2008 14 Qiaochengdong residential Time 15:02:36 15 Huaqiaocheng residential CardId Anonymous unique user card id 16 Shijiezhichuang transport Tradetype Different transit behavior: 31- bus 17 Futiankouan Ex-transport boarding, 21- subway check-in, 22- 18 Fumin residential subway check-out 19 ShimingZhongxin government terminalid For bus, it represents bus id; for subway, 21 Shaoniangong government it represents the stop id III. AGGREGATED SPATIOTEMPORAL MOBILITY trademoney Final fare payment after discount (cent) PATTERN Tradevalue The full payment/the original fare (cent) transfer Transfer discount (cent) Before exploring the implications of urban mobility landscape to urban planning, we discuss temporal and

spatiotemporal patterns and highlight how these patterns From the smart card data, we could infer two major reflect underlying cultural and spatial characteristics of public transit modes: bus and subway. For bus journey, we Shenzhen. could get the boarding time and travel fare; for the subway

843

Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 15:09:57 UTC from IEEE Xplore. Restrictions apply. A. Temporal patterns of public transit passengers B. Spatiotemporal patterns of subway passengers Through statistics of smart card records of the bus and Through detailed analysis of daily check-in and check- subway in different day, their temporal mobility patterns out in different subway stops, we could get the could be inferred. The temporal pattern of bus trip and spatiotemporal patterns of subway travel. The proportion subway trip is showed in figure 1. of check-in and check-out in different day is showed in Figure 3. Figure 2(a) and 2(b) shows that during weekday there are obvious peak hours for bus trips. AM peak begins from Figure 3 (a) and 3 (b) is basically the same, which 7am and reaches the peak at 8am; PM peak begins from means that the check-in and check-outs is almost similar. 17pm and reaches peak at 18pm. This rhythm reflects the The proportion remains the same during different date and daily life patterns of citizens in Shenzhen, most of different bus stops, which means they have great similarity. government institution and enterprises begin their work The largest proportion subway stop is Dajuyuan, Huaqiang around 9am and finish their work around 18pm. The Road and Shijiezhichuang. The proportion of Saturday and morning peak proportion is 26%, the evening peak Sunday are similar, the three largest proportion data is proportion is 20%, the sum of two peaks achieves 46%, Laojie, Huaqiang Road and Shijiezhichuang. and the trips in peak hours almost occupy the half of daily travel demand. On Sunday, the peak hour is 10am, 14pm From the proportion change of from weekday to and 17pm, but in total is smooth, because the inhabitants weekend, Laojie Subway station(Dongmen Pedestrian can choose the free travel time on Sunday. The night Street) has largest increase, then is Huaqiang Road, Luohu activity is more frequent during night on weekend than on Station and Futian Port station. Meanwhile, the proportion weekday, which means more citizens choose to go out for decreases in Guomao, Dajuyuan, Gouwu Gongyuan and recreation. The morning peak of passengers on Saturday is Chegongmiao. This describes the life rhythms of citizens’ still big, it is 10.2%, it means on Saturday a large lives. The working place activity decrease during the proportion of citizens work during Saturday (From the weekend, while the two biggest recreation center- Laojie proportion, it contribute half of daily working population), and Huaqiang Road increase the activity. this reflects the corporation composition of Shenzhen, 16.0 16 because most of HongKong, Taiwan and Japan enterprises 14.0 14 12.0 12 ask employees to work half day on Saturday, their daily 10.0 10 星期日 SUN travel cause the peak hour on Saturday, and the small noon 8.0 8 星期一 MON 星期二 TUE peak. In some sense, the travel feature of Saturday is the 6.0 6 星期三 WED 星期四 THR 4.0 4 星期五 FRI combination of weekday and Sunday, in the morning it 星期六 SAT 2.0 2 likes the weekday and after the noon it is the same like 0.0 0 Sunday. However, the Saturday peak hour is 18PM.

Figure 2(c) and 2(d) shows the temporal mobility pattern for subway. During weekday, the subway trip Figure 3. Check-in(check-out) proportion in different stops in different mobility pattern is almost the same with bus trip mobility day pattern. The only difference is that subway AM peak hour is one hour after bus AM peak hour, which means subway This smart card datasets open a new way to measure is more reliable than bus. The inhabitants spend less travel the mobility pulse in the city, which give us the intuitive time in subway than in bus. sense about the cultural and life character of the city.

190,000 14.0 180,000 13.0 170,000 160,000 12.0 150,000 11.0 C. Daily check-in and check-out of different subway 140,000 10.0 130,000 120,000 9.0 stops 110,000 8.0 100,000 星期日 SUN 90,000 7.0 80,000 6.0 工作日 WKD From figure 3 we could know the unique mobility 70,000 5.0 星期六 60,000 SAT 50,000 4.0 40,000 3.0 pattern for individual subway stop. however, what we are 30,000 2.0 20,000 10,000 1.0 more concerned is the connection between different 0 0.0 67891011121314151617181920212223 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 subway stops. While the most efficient way to measure the connections between different subway stops is deriving 30,000 16.0

14.0 OD matrix. According to check-in and check-out 25,000 12.0 timestamps and locations for each individual passenger, 20,000 10.0 星期日 SUN we can inter the OD matrix for arbitrary time period, 15,000 8.0 工作日 WKD 6.0 which is a 19*19 matrix. 10,000 星期六 SAT 4.0 5,000 2.0 From the analysis results, we find the OD pair between 0 0.0 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 67891011121314151617181920212223 different two points is in the same magnitude and there is little difference between them. So we sum the OD pair, and display them through GIS platform to compare Figure 2. public transit temporal patterns in different day directly. The summarized OD pair is showed in Figure 4.

844

Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 15:09:57 UTC from IEEE Xplore. Restrictions apply. From figure 4(a), it is easy to understand that the D. Spatiotemopral patterns during the AM and PM peak largest connections in weekday are Guangxia and hours Huaqiang Road, Shijiezhichuang and Huaqiang Road, Through temporal analysis we can know the AM and Shijiezhichuang and Dajuyuan, Huangqiang Road and PM peak mobility pattern. But how do they distribute in Laojie. The first three shows the working travel, the last spatial scale and do they have obvious directions? Through one is the recreation. The working travel is more than the spatial mapping of AM and PM peak flow, we can answer recreation travel. these research questions. Figure 4(c) shows the largest connections in Sunday Generally, the check-in during AM peak hour is more are Laojie and Huaqiang Road, Shijiezhichuang and close to the residential center, and the check-out is more Huaqiang Road, Huaqiang Road and Gangxia. The close to the working zones. On the contrary, the check-in recreation trips are more than working trips. during PM peak is more close to the working zones and the Figure 4(b) indicates the mobility pattern of Saturday is check-out is more close to the residential area. Based on the mix of weekday and weekend, work travel and these assumptions, we could detect working zones and recreation travel. Though Saturday travel is less than residential zones. weekday, but the connections are more concentrated, i.e. Through analyzing different proportion of check-in and several stops have more connections than weekday. check-out in AM peak 8am and PM peak 18pm, we could use simple figure to detect residential area and working area. Check-in(check-out) proportion in different stops in peak hour is showed in Figure 5. From Figure 5, obviously Shijiezhichuang and Gangxia represents the residential centers, the two biggest communities, Shijiezhichuang represents western communities, including Nanshan district and Xin’an and Xi’xiang community in Bao’an district; Gangxia represents northern communities, including Meilin, longhua communities. While Guomao, Dajuyuan, Huaqiang Road, Gouwu Gongyuan, ChegongMiao are the center of working area.

24.0 26.0 24.0 22.0 24.0 22.0 20.0 22.0 20.0 18.0 20.0 18.0 16.0 18.0 16.0 14.0 16.0 14.0 12.0 14.0 12.0 12.0 10.0 10.0 10.0 8.0 8.0 8.0 6.0 6.0 8AM IN 6.08AM IN 8AM IN 4.0 4.0 4.0 2.0 2.0 18PM OUT 2.018PM OUT 18PM OUT 0.0 0.0 0.0 Laojie Laojie Laojie Luohu Luohu Fumin Fumin Fumin Luohu Zhuzilin Zhuzilin Zhuzilin Gangxia Gangxia Gangxia Guomao Guomao Guomao Dajuyuan Dajuyuan Dajuyuan Xiangmihu Xiangmihu Xiangmihu Kexueguan Kexueguan Kexueguan HuaqiangRd HuaqiangRd HuaqiangRd Futiankouan Futiankouan Futiankouan Shaoniangong Shaoniangong Shaoniangong Huaqiaocheng Huaqiaocheng Huaqiaocheng Shijiezhichuang Shijiezhichuang Shijiezhichuang Qiaochengdong Qiaochengdong Qiaochengdong Chonggongmiao Chonggongmiao GouwuZhongxin GouwuZhongxin Chonggongmiao GouwuZhongxin Shimingzhongxin Shimingzhongxin Shimingzhongxin HuizhanZhongxin HuizhanZhongxin HuizhanZhongxin

18.0 24.0 22.0 16.0 22.0 20.0 20.0 14.0 18.0 18.0 16.0 12.0 16.0 14.0 10.0 14.0 12.0 12.0 8.0 10.0 10.0 6.0 8.0 8.0 6.0 4.0 6.0 8AM IN 8AM OUT 8AM OUT 4.0 4.0 2.0 2.0 18PM OUT 2.0 18PM IN 18PM IN 0.0 0.0 0.0 Laojie Laojie Laojie Fumin Fumin Fumin Luohu Luohu Luohu Zhuzilin Zhuzilin Zhuzilin Gangxia Gangxia Gangxia Guomao Guomao Guomao Dajuyuan Dajuyuan Dajuyuan Xiangmihu Xiangmihu Xiangmihu Kexueguan Kexueguan Kexueguan HuaqiangRd HuaqiangRd HuaqiangRd Futiankouan Futiankouan Futiankouan Shaoniangong Shaoniangong Shaoniangong Huaqiaocheng Huaqiaocheng Huaqiaocheng Shijiezhichuang Shijiezhichuang Shijiezhichuang Qiaochengdong Qiaochengdong Qiaochengdong Chonggongmiao Chonggongmiao Chonggongmiao GouwuZhongxin GouwuZhongxin GouwuZhongxin Shimingzhongxin Shimingzhongxin Shimingzhongxin HuizhanZhongxin HuizhanZhongxin HuizhanZhongxin

Figure 5. Check-in(check-out) proportion in different stops in peak hour

Through analysis of different proportion during the peak hours, we can distinguish the residential area and working area, but we can not tell the directions of the travel. Thus we must calculate the related OD matrix and show it on GIS platform, from which we could obtain the direction of morning and evening direction. AM and PM peak OD in different day is showed in Figure 6.

Figure 4. Connections between different stops in different days From Figure 6 we could understand that the AM peak in weekday represents very obvious rules: from residential area to working area is unidirectional flow, and the main flow direction is from west to east. The center of trip generation is Shijiezhichuang and Gangxia, which represents the western and northern residential center.

845

Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 15:09:57 UTC from IEEE Xplore. Restrictions apply. The spatiotemporal pattern is also very obvious during = (1.0471)− x PM peak in weekday, which is from working area to yx( ) 6.1246exp (1) residential area, and the main flow direction is from east to From the result we get previously, we can know that west. The center of the trip generation is Huaqiang Road the transit system is in steady condition. The mean value of and Dajuyuan. number of reaching stops in different people is a constant value. According to the maximum entropy principle, we can judge that it belongs to exponential distribution. It also means that most people just access to a small part of areas under the physical constrain.

Figure 6. AM and PM peak OD in different day

In sum, if we summarize the travel OD in peak hours, we could find the daily mobility patterns is simple and clear, every morning the inhabitants move from residential area to working area, while in the evening they travel from Figure 7. the different user percentage in different stops working zone back to residential area, some of them choose to go to recreation area. B. percentage in different stops which passengers visit In all of the subway stations, Shijiezhichuang and For each of the smart card user, in his different travel Gangxia are the center of residential area, representing stops, the percentage of different stops is different. Thus, western and northern inhabitants; Guomao, Dajuyuan, we calculate the percentage of different stops above 5 Kexueguan, Huaqiang Road, Gouwu Park and stops. The result is showed in figure 8. In this graph, we Chegongmiao are the center of working zone; Laojie and only list the stops which are less than 10 stops. Huaqiang Road are the center of shopping and recreation. Inside of these, Huaqiang Road has special statuses, which is the center of both working area and recreation area. The mobility patterns of peak hours show the uni-CBD model in Shenzhen, which causes the clock pendulum movement of urban citizens.

IV. THE INDIVIDUAL MOBILITY PATTERN OF SUBWAY TRANSIT PASSENGER According to the subway trip record, we could Figure 8. the different user percentage in different stops calculate the each user’s subway stops in one week sampling period and the frequency that each user shows up. Most of the people are concentrated on the first two These patterns will help us understand the travel pattern of stops, the sum of top 2 reaches 52%, which means most of different individual inhabitants. the travels are concentrated in a small part of areas, home and office. If we use the logarithmic coordinate again to A. passengers reaching different number of stops redraw the data in Figure 8(a), we can get Figure 8(b). It is almost a linear line. The regression result shows that the According to the smart card record, we summarize the power law is nearly -1, the same as Zipf law. The equation stops that different subway users reach, and the result is is shown in equation (2): show in figure 7. − From figure 7 we could know that, most of the users yx( )= 0.4144 x1.0954 only commute between two stops, the percentage is 61.5%, (2) while the stops is inside the 4 stops is above 94%, which We know that if people have a constant proportion describe most of people are only active in several points. If among different stations in a macro flow system, its we use logarithmic coordinate, we can get figure 7(b). It distribution should be exponential distribution. Due to the shows that users and the number of stops have linear physical constrains, most people will have a high relationship (equation 5.3). proportion activity in a small area.

846

Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 15:09:57 UTC from IEEE Xplore. Restrictions apply. C. First paggege Time human mobility pattern. Through the analysis, we show The concept FPT (First Passage Time) means the time the regularity of the mobility patterns. The strong interval between two consecutive appearances in one relationship between mobility pattern and land use location. In this research we calculate the time interval properties is investigated in this study, which could help us between the first appearance and the following appearance. better plan the related facilities and services. We also Firstly we do not distinguish the arrivals and departures. investigated mobility patterns at the individual level. We The calculation result is shown in figure 9. To further observed that the individual mobility pattern is extreme illustrate the mobility pattern, we then calculate the FPT of concentric and regular, in both spatial and temporal scale. departure behavior, which is shown in figure 10. The case study in Shenzhen using smart card records to From the figure 9 we can see that FPT is very regular. understand urban mobility pattern is carry out and analysis The proportion is decreasing from 1 to 8 hours. The large results show that it is of great importance to understand the proportion of 1-hour FPT means that the duration for an functioning of metropolitan mobility systems in order to activity in one location is less than one hour. Beyond the enhance life quality, to protect the environment and to 8-hour, we can see that a peak shows in 10-hour. This is achieve sustainable development. the time interval between morning peak and evening peak In the future, we would like to incorporate contextual of working days (18:00-8:00 = 10 hours. It means people features into our urban mobility landscape system, such as go to office at 8:00 and leave office at 18:00). Other high weather, season, special events (e.g. concerts or soccer FPT peaks show on the multiples of 24. Obviously, people matches), public transportation schedules and locations, have a regular mobility pattern in the location and the time. and data from additional urban infrastructure (e.g., cellular networks). We also plan to fuse the different data sources to derive high level understanding of urban dynamics. The most important aspect of the research is to understand how these urban monitoring systems can become a tool for urban planning and policy making.

REFERENCES [1] Whyte, W. H. The social life of small urban spaces. Washington, Figure 9. Proportion of FPT (no distinguish between arrival and D.C.: Conservation Foundation, 1980. departure) [2] Brockmann, D. D., Hufnagel, L. & Geisel, T. The scaling laws of human travel. Nature 439, 2006, pp. 462–465 [3] Rattenbury, T., Good, N., & Naaman, M. Towards automatic extraction of event and place semantics from flickr tags. Proceed. ACM SIGIR ‘07. Amsterdam, July 23-27, 2007, pp. 103-110. [4] Girardin, F., Calabrese, F., Dal Fiore, F. , Ratti, C., and Blat, J.. Digital footprinting: Uncovering tourists with user-generated content. IEEE Pervasive Computing, 2008, 7(4): pp. 36–43. [5] Ratti, C., Pulselli, R. M., Williams, S., & Frenchman, D. (2006). Mobile landscapes: Using location data from cell-phones for urban analysis. Environment & Planning. 33(5), pp. 727–748. [6] Reades, J., Calabrese, F., Sevtsuk, A. , & Ratti, C.(2007). Cellular Figure 10. Proportion of FPT (arrival) census: Explorations in urban data collection. IEEE Pervasive Computing, 6(3):30-38. Figure 10 further illustrates that most people have a [7] González, M. C., Hidalgo, C. A., and Barabási, A. L. (2008). clear mobility pattern. They always appear in one location Understanding individual human mobility patterns. Nature 453, pp. at the same time every day. 779-782. [8] McNamara, L., Mascolo, C., & Capra, L(2008). Media sharing The individual travel behavior of public transit shows based on colocation prediction in urban transport. Proceed. ACM strong spatial-temporal regularity, such as a lot of trips are MobiCom '08. San Francisco, CA, Sept. 14 - 19, 2008, pp. 58-69. carried out between home and office, and the 24-hour [9] Latour, B(2007). Beware, your imagination leaves digital traces, Column for Times Higher Education Supplement, 6th of April cycle of activity. Planning and operation should consider 2007. http://www.brunolatour.fr/poparticles/poparticle/P-129- these characteristics in transportation facility and service THES-GB.doc design. [10] Shenzhentong Company (2008). Shenzhentong card usage survey. Internal report. V. CONCLUSION [11] Shenzhen Metro (2008). Shenzhen Metro Map and description. Accessed December 18th, 2008. Novel aspects of human mobility patterns were http://www.szmc.net/10station/index.jsp addressed by means of smart card data with time and space resolution. This allowed us to study the mean collective behavior at large scales and focus on the regularity of

847

Authorized licensed use limited to: MIT Libraries. Downloaded on April 21,2010 at 15:09:57 UTC from IEEE Xplore. Restrictions apply.