Embed Size (px)
Citation preview
This article was downloaded by: [University of California Santa Barbara]On: 29 April 2015, At: 11:15Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Click for updates
Spatial Cognition & Computation: AnInterdisciplinary JournalPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/hscc20
Spatio-Temporal Analytics for ExploringHuman Mobility Patterns and UrbanDynamics in the Mobile AgeSong Gaoa
a Department of Geography, University of California, Santa Barbara,CA, USAAccepted author version posted online: 18 Nov 2014.
To cite this article: Song Gao (2015) Spatio-Temporal Analytics for Exploring Human Mobility Patternsand Urban Dynamics in the Mobile Age, Spatial Cognition & Computation: An Interdisciplinary Journal,15:2, 86-114, DOI: 10.1080/13875868.2014.984300
To link to this article: http://dx.doi.org/10.1080/13875868.2014.984300
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the“Content”) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sourcesof information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.
This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Spatio-Temporal Analytics forExploring Human MobilityPatterns and Urban Dynamics inthe Mobile Age
Song Gao
Department of Geography, University of California, Santa Barbara, CA, USA
In this research, we present a spatio-temporal analytical framework including spatio-temporal visualization (STV), space-time kernel density estimation (STKDE), andspatio-temporal-autocorrelation-analysis (STAA), to explore human mobility patternsand intra-urban communication dynamics. Experiments were conducted using large-scale detailed records of mobile phone calls in a city. The space-time path, time seriesgraphs, vertical Bezier curves, STKDE, STAA, and related techniques in 3D GIS as wellas statistical tests have been suggested for different spatio-temporal analysis tasks.We also investigated several statistical measures that extend the classic spatialassociation indices for spatio-temporal autocorrelation analysis. The spatial order ofweighted matrix was found to have more significant effects than the temporal neighborson influencing the autocorrelation strength of hourly phone calls.
Keywords: human mobility; mobile phone data; space-time GIS; spatio-temporalanalytics; spatio-temporal visualization; space-time kernel density estimation; spatio-temporal autocorrelation analysis
1. INTRODUCTION
Despite the fact that humans have keen ability to discover patterns hidden insmall-scale data, they may find it difficult to discover large-scale data thatoften vary over both space and time. Researchers have made great effort onspatial data mining and spatio-temporal visual analytics to raise the cognitiveceilings that often prevent the interpretation of large spatio-temporaldatasets (Guo et al., 2006; Shaw, Yu, & Bombom, 2008; Andrienko et al.,
Correspondence concerning this article should be addressed to Song Gao, Department of Geography, University of California, Santa Barbara, 1832 Ellison Hall, Santa Barbara, CA 93106, USA. E-mail: [email protected]
Color versions of one ormore of the figures in the article can be found online at www.tandfonline.com/hscc.
Spatial Cognition & Computation, 15:86–114, 2015
Copyright q Taylor & Francis Group, LLC
ISSN: 1387-5868 print/1542-7633 online
DOI: 10.1080/13875868.2014.984300
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
2010). In the Mobile Age, with the widespread use of location-awarenessdevices, it is possible to collect large-scale location-awareness datasets, suchas mobile phone call data, GPS-enabled taxi trajectories, and social mediadata, to sense complex human movements and human-environmentinteractions.
The study of human mobility using emerging Big Geo-Data has been a hottopic in scientific research and many application domains, including but notlimited to the analysis of social events, tourism, transportation and urbanplanning, land and real estate development, migration and health andepidemic studies (Gonzalez et al., 2008; Calabrese et al., 2010; Song et al.2010; Kang et al., 2012b; Hawelka et al., 2014; Liu et al. 2014; Pindolia et al.,2014). There are increasing mobile network operators in different countriesand cities willing to share their datasets in collaborative projects, such as theOrange “Data for Development Challenge” in the Ivory Coast and Senegal(http://www.d4d.orange.com/en/home). At the same time, it might raise someethical issues and privacy concerns. Several studies have been trying toinvestigate various models and techniques for providing important impli-cations to protect individual location privacy (Ho & Ruan, 2011; de Montjoyeet al., 2013).
With such data, it would be of great significance to explore and understandhow cities function in short-term temporal scales compared with traditionallong-term strategic planning in the new era of Big Data (Batty, 2013). Forexample, although humanmovements and activities may vary over time acrossdifferent regions, the observed activity hotspots and information flow mightexhibit a pattern of spatial dependence. Also, ignoring the temporal dimensionwould not be sufficient to discover underlying urban dynamics. For instance,urban governors might hope to understand patterns of human movements byobserving the neighboring regions in previous time periods. In such space-timeintegration contexts, the spatio-temporal analytics should help to answerquestions such as:
. Where are the spatio-temporal hotspots of human activities?
. Do human/vehicle movement flows exhibit different spatio-temporalpatterns in contrast to overall trips?
. How do these patterns relate to the population distribution and urban land-use structure?
. How to conduct the spatio-temporal autocorrelation analysis?
. What is the impact of spatio-temporal granularity and uncertainty?
To cope with these research questions and problems, spatio-temporal datamining techniques and analytical workflows need to be studied. However,mining big geo-data and discovering knowledge of spatial-temporal relations,
Spatio-Temporal Analytics in the Mobile Age 87
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
patterns and trends about the real world process are not trivial. Issues onspatio-temporal data organization, computation and integration with visualrepresentation and human cognition still challenge researchers to make greatefforts with interdisciplinary knowledge.
There exists a quantity of literature in which space-time integration andquantitative methods for spatio-temporal data analysis have been facilitatedand discussed (Dijst, 2013; Long & Nelson, 2013). After examining sevendistinct spatio-temporal data types and related work, Goodchild (2013) arguedthat space–time geographic information system is unlikely to emerge in thenear future; and instead, attention should focus on the systematic study ofintegration.
Richardson (2013) addressed some trends and research challenges ofdeveloping real-time space-time functions and capabilities, such as temporalscale, space-time models, and frameworks of spatio-temporal analysisoperations to transform geographic data into meaningful information, whichare also main issues to be addressed in this study. Stewart et al. (2013)introduced a spatio-temporal conceptual model for scheduling individualactivities using an ontological approach. More discussions on space-timeintegration in Geography and GIScience can be found in the special forum ofAnnals of the Association of American Geographers [Vol. 103, No. 5, 2013].In short, very few existing studies present systematic spatio-temporalanalytical frameworks and workflows for integrating emerging Big Geo-Datawith computational and statistical approaches to guide urban knowledgediscovery in practice.
To this end, this article aims at proposing a spatio-temporal dataprocessing and analytical framework, which can be applied not only inexploring dynamic mobility and intra-urban flow patterns based on mobiledata, but also in other human and social science research from the emergingbig geospatial datasets and computer techniques. Issues on identifyingappropriate data structures for different spatio-temporal pattern discovery willbe discussed with this framework.
This article is structured as follows: In Section 2, we briefly discuss somerelated work and propose a spatio-temporal analytical framework thatincludes spatio-temporal visualization (STV), space-time kernel densityestimation (STKDE), and spatio-temporal autocorrelation-analysis (STAA)for exploring human mobility and urban dynamic patterns. The methodology,technical implementation of these analytics will be presented in detail. Thenwe apply the framework to analyze amounts of geo-referenced mobile phonecall records in a city to reveal the spatio-temporal patterns hidden in such biggeo-data and further, to understand the complex urban dynamics. The dataprocessing, experiments, main findings, and discussions are presented inSection 3 and 4. We conclude with a summary and directions for furtherresearch in Section 5.
88 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
2. SPATIO-TEMPORAL ANALYTICAL FRAMEWORK
Modelling human mobility patterns and understanding dynamic urbanstructures based on a large amount of GPS sensors, mobile devices, persons,vehicles, and street networks have become a hot topic in many fields such asurban planning, transportation, GIScience and computer science (Jiang &Claramunt, 2004; Ratti et al., 2006; Gonzalez et al., 2008; Kang et al., 2012a,2012b; Liu et al., 2012; Yuan et al., 2012; Gao et al., 2013b, 2013c; Shen et al.,2013; Yue et al., 2014). In general, the mining and analyzing processes of suchspatio-temporal data require combined qualitative-quantitative approachesthat involve data extraction and analytics, statistical inference, andgeovisualization.
Recently, Long and Nelson (2013) conducted a thorough review onquantitative methods for analyzing movement data which is one prominenttype of spatio-temporal data. They classified those quantitative methods intoseven groups: (1) time geography, (2) path descriptors, (3) similarity indices,(4) pattern and cluster methods, (5) individual–group dynamics, (6) spatialfield methods, and (7) spatial range methods. Because of special data structurerequirements and computational complexity, not all of the mentioned methodscan be directly applied into different types of spatio-temporal datasets inpractice. Identifying or creating appropriate space-time data structures forspecified quantitative methods is still challenging.
In this research, we present a spatio-temporal analytical framework(Figure 1), which combines STV, STKDE, and STAA for understandingspatial-temporal patterns (both individual and aggregated) hidden in urbanBig Geo-Data. As discussed before, there are not only these three types ofspatio-temporal analytical methods available for exploring patterns.However, these three provide relatively simple and efficient ways to extendspatial analysis functions and spatial statistics from 2D space into 3D spaceand could be integrated well with GIS environments. Each of theproposed analytics has different characteristics and data-format require-ments. In the processing, the raw data were converted into appropriate datastructures for various analytical purposes. In the following part, we willdiscuss the roles of different spatio-temporal analytics for the presentedresearch.
2.1. Spatio-Temporal Visualization Techniques for Trajectory and FlowBy utilizing the power of human vision, previous studies have demonstratedthe effectiveness of geovisualization in spatial data exploration andknowledge discovery (MacEachren & Kraak, 2001; Kwan, 2004; Guo et al.2005; Andrienko et al., 2008). The understanding of urban spatial structurescan benefit the studies of visualizing individual space-time behaviors (Kwan,
Spatio-Temporal Analytics in the Mobile Age 89
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
2000; Chai, 2013). We can use both individual-based visualization andaggregation-based visualization to explore dynamic patterns in urban studies.However, the representation of dynamic human activities and movements overboth space and time is one of the major challenges in geocomputation andgeovisualization.
Hagerstrand’s time-geography conceptual framework provides an excel-lent integrated representation of human movements in space and time(Hagerstrand, 1970). But the space-time cube idea was not applied so widelyuntil the development of GIS-based implementations and analytical discus-sions about space-time relationships, interactions and uncertainties (Miller,2005; Shaw, Yu, & Bombom, 2008; Chen et al., 2011; Nakaya, 2013) movedforward, as well as the opportunity to explore potential human activities inboth physical and virtual spaces (Yu & Shaw, 2008).
For the individual-based movement representation, a space-time path (3Dpolyline) was created to connect time-ordered sequence of locations, X, Y, T .
of one person in a 3D-GIS environment which consists of a two-dimensionalhorizontal geographic plane and one vertical dimension of time (Figure 2).It can be used for visual exploration of continuous spatial and temporalmovement patterns. Several analytical models and measurements have beendeveloped by Miller (2005) such as space-time prism, composite path-prisms,stations, bundling and intersections to further analyze the complex spatio-
Figure 1: A spatio-temporal analytical framework for identifying human mobility patterns andurban dynamics.
90 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
temporal relationships among human activities and interactions under specialspace-time constraints.
But when numerous trajectories were collected in the datasets, it washard to visualize and interpret. Different approaches for generalization andaggregation of massive movement data have been introduced such as traffic-oriented view and trajectory-oriented view (Andrienko & Andrienko, 2008).Spatial clustering approaches have been developed for discovering places ofinterest and similar routes from human movement trajectories (Palma et al.,2008; Giannotti et al., 2011). In addition, sequence alignment method, whichwas originally employed by biochemists to analyze DNA sequences, has beenapplied in analyzing the space-time sequential aspects of human activities(Wilson, 1998; Shoval & Isaacson, 2007; Mavoa et al., 2011; Kwan et al., 2014).It could help the identification of individual-based spatio-temporal recurringtrends and group-based similarity patterns.
In the urban context, the aggregation of massive human movementtrajectories by origins and destinations (OD) can be utilized to understand thedynamic OD-flow patterns among traffic analysis zones (TAZs) or otherpolygonal divisions of region in different temporal scales. Traditional flowmapping is used for representing the amount and the direction (with arrowsymbol) of from-tomovements of human or things among regions in a 2D space,such as migration and goods trade (Tobler, 1987). Some graph layoutoptimization algorithms and aggregation strategies have been suggested tominimize the edge crossings between flow symbols (Phan et al., 2005;Andrienko & Andrienko, 2011).
Here we introduce another approach of using vertical Bezier curves in 3D-GIS environment for interactive visual exploration of information ormovement
Figure 2: Space-time path of an individual’s movement in a week.
Spatio-Temporal Analytics in the Mobile Age 91
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
flows between places. The main advantages of such an approach lie in theintegration of 3D visualization techniques which support interactions between3D geometry objects and OD-flow values in multiple time snapshots or in acontinuous animation.
A Bezier curve is defined by a set of control points P0 through Pn, where nrepresents called its order (n ¼ 1 for linear, n ¼ 2 for quadratic which is used inour work, etc.). Bezier curves have been widely used in computer graphics andgeometry designs (Farin, 1996). We develop an algorithm to approximate thequadratic Bezier curves. As shown in Figure 3, the first point P0 and the lastpoint P2 are used to represent the centroids of two regions (i.e., the origin andthe destination of each flow) and the intermediate control points areinterpolated by the functions as follows:
BðtÞlongitude ¼ P0 þ tðP2 2 P0Þ ð1Þ
BðtÞlatitude ¼ SlopeðP0; P2Þ*BðtÞlongitude þ b ð2Þ
BðtÞx ¼ 2ð12 tÞt*DistðP0;P2Þ=2þ t2*DistðP0; P2Þ ð3Þ
BðtÞy ¼ 2ð12 tÞt*H ð4Þ
Figure 3: Drawing and visualizing vertical Bezier curves in 3D-GIS environment (ArcScene).
92 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
BðtÞh ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiBðtÞ2x þ BðtÞ2y
qð5Þ
t ¼ k=Njk [ ð1; 2; . . . . . . ;N 2 1Þ ð6Þ
The parameter t is determined by the step k and the number of interpolatepoints N to be inserted between each OD pair. To control the shape of a curve,we took a point P1 on the perpendicular bisector of the chord linking P0 and P2.The height H represents the amount of the flow. Dist(P0,P2) is the Euclidiandistance between the two points. After getting all the control points B(t), weconnected all of them together to draw a vertical Bezier curve and then projecta flow between regions in the 3D-GIS environment. We wrote a Python script togenerate all Bezier curve controlling points based on the OD-flow matrix tableand linked them in Esri’s ArcScene software for further interactive explorationof information flow or physical movement flow patterns.
2.2. Space-Time Kernel Density EstimationAsdiscussed above, the temporal information ofmovements ingeographic space isimportant to detect the spatio-temporal trends of underlying humanmobility. Butwith the increasing number of aggregated human/vehicle trajectories in urbanspace, the space-time path representationmodel will be hard to interpret becauseof the overlapping and cluttering issues. To solve this problem, an extension ofkernel density estimation (KDE) (Silverman, 1986) was suggested. The KDE hasbeenwidely used in spatial analysis to characterize a smooth density surface thatshows the geographic clustering of point or line features in 2D space.
To incorporate the time information, the space-time kernel densityestimation (STKDE) can be taken as a generalization approach of the 2D-space KDE into the 3D space-time cube which can support the exploration ofspatio-temporal patterns, clusters and changes. Such STKDE techniques havebeen used in several studies, such as crime clustering analysis (Brunsdon et al.,2007; Nakaya and Yano, 2010), trajectory data mining (Demsar & Virrantaus,2010), publication citation analysis (Gao et al., 2013a), and dengue feverpattern discovery (Delmelle et al., 2014). The STKDE value of each voxel(volumetric pixel) in the three-dimensional space-time cube is estimated as:
Dðx; y; tÞ ¼ 1
nh2sht
Xi
Ksx2 xihs
;y2 yihs
� �Kt
t2 tiht
� �ð7Þ
where D (x, y, t) is the density estimation of each voxel based on the data inneighboring volumetric pixels; n is the number of point events, and hs and ht
are the spatial and temporal neighboring bandwidths.
Spatio-Temporal Analytics in the Mobile Age 93
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Each point in the neighboring pixels is weighted based on the proximity inboth space and time to the voxel using kernel functions (Ks and Kt). In thisstudy, the Epanechnikov kernel was used for multivariate probability densityestimation within the bandwidths (Epanechnikov, 1969). Similar to the 2Dspatial KDE, larger bandwidths may result in smooth surface, yet smallerbandwidths may result in the lack of trending patterns, so we needed tocalibrate both spatial and temporal bandwidths of STKDE based on theexperiments with actual datasets. In practice, the data-driven approach couldbe applied by considering the least squares cross-validation for kernel-bandwidth selection (Sheather & Jones, 1991). More discussions on bandwidthselection methods for kernel density estimation were presented in a review byJones et al. (1996).
The results of STKDE are volume data, i.e., 3D-grids. Direct visualizationof such STKDE would require four-dimensional space because of theirvolumetric data structure consisting of 2D geographic space, time and anotherone for the density estimation scalar. Such volume visualization is not verycommon in GIS but very popular in geophysics, geology, medical science, and incomputer graphics (Kaufman, 1990). The three main approaches for volumevisualization were discussed by Demsar & Virrantaus (2010): (1) direct volumerendering by assigning color and transparency to voxels; (2) isosurface that isthe equivalent of isoline connecting points of equal value on a two-dimensionalmap; and (3) volume slicing by planes. We apply the volume slicing approachwith color schema and transparency to the voxels regarding the consistency ofKDE visualization in GIS.
2.3. Spatio-Temporal Autocorrelation AnalysisAnalyzing spatio-temporal autocorrelation structures of human activitieswould be helpful to understand the urban dynamic patterns in space and timesimultaneously. In statistics, autocorrelation can be taken as the correlation ofa variable with a lagged specification of itself (Box et al., 2008). The temporalautocorrelation can be defined as the correlation of the same variable Xbetween values at different time s and t.
Rðs; tÞ ¼ E½ðXt 2 mÞðXs 2 mÞ�s2
ð8Þ
E is the expected value operator, m is the mean of the observation values and s 2
is the variance. The temporal autocorrelation can be used to explore the time-series autocorrelation patterns.
With regard to the spatial dependence, spatial autocorrelation (association)statistics have been used to analyze the degree of dependency amongobservations in a geographic space (Cliff & Ord, 1973). These measurements
94 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
can be divided into two categories: global indices and local indices. Classicglobal indices of spatial autocorrelation include Moran’s I (1950), Geary’s C(1954), and Getis–Ord’s General G (1992), while local indices of spatialassociation (LISA) can be established by transforming the global indices intocorresponding local measurements based on different measures of similarity(Anselin, 1995).
All of these spatial autocorrelation statistics require a spatial weightsmatrix that reflects the intensity of the geographic relationship betweenobservations and their neighbors, e.g., the distance-to-neighbor matrix or thebinarymatrix in which the element value is 0 or 1 determined by whether thereis a shared boundary between the observation location and neighbors.As suggested by Hardisty and Klippel (2010), adding the temporal neighborsinto the weights matrix would be one approach to extend the traditional spatialautocorrelation measurements. Griffith (2010) gave an overview of spatio-temporal modelling techniques including autoregressive-integrated-movingaverage models, space-time autoregressive models, geostatistical models, andpanel data models, as well as proposed spatial filtering models. All of thosemodels have been motivated by considering the spatio-temporal associationssimultaneously. Cheng et al. (2012) further extended both global and localspatio-temporal autocorrelation analysis onto the road network data forcomplex traffic analysis.
Here, we present three global measures of spatio-temporal associationregarding the Moran’s I, Geary’s C, and Getis-Ord’s General G:
Ist ¼PN
i¼1
PNj¼1 wij½ziðtÞ2 �zt�½zjðtþ tÞ2 �ztþt�ststþt
PNi¼1
PNj¼1 wij
ð9Þ
Cst ¼PN
i¼1
PNj¼1 wij½ziðtÞ2 zjðtþ tÞ�2
2ststþt
PNi¼1
PNj¼1 wij
ð10Þ
Gst ¼PN
i¼1
PNj¼1 wijzjðtÞzjðtþ tÞPN
i¼1
PNj¼1 ziðtÞzjðtþ tÞ ð11Þ
Where Ist, Cst, and Gst can be taken as different formats of space-time cross-correlation (or cross-product) models (Getis, 1991); Z is the target variable ofinterest; i and j are indices of total N spatial units; wij is an element of the k-order-neighbor spatial weighted matrix (1st, 2nd, . . . . . . , kth); �zt and �ztþt are themeans of variable Z within a time lag, while st and stþt are the variances. Thelocal measures of spatio-temporal autocorrelation can be derived bydecomposing a global measure into particular spatial neighboring units.
Spatio-Temporal Analytics in the Mobile Age 95
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
In the experimental section, we evaluate how different spatial andtemporal neighbors (lags) affect the results of three spatio-temporalautocorrelation measures for the mobile phone call activities in a city.
3. DATA PROCESSING
In this research, the dataset contains a week of about 74,000,000 anonymizedmobile phone call detail records (CDR) in a large city from a Chinesetelecommunication operating company. The CDR data lists the information ofcaller, receiver, mobile base stations, date, time, duration et al. (Table 1).As shown in Figure 4, every time when a user (caller/receiver) made a call, he/shewas geo-referenced to a correspondingmobile base station that has a uniquelongitude/latitude position. The coverage area of eachmobile base station can beexpressed as aVoronoi polygon for call activity analysis and termed as a “cell.” In
Figure 4: Spatial distribution of mobile base stations and an illustration of phone call interactionsamong cells.
Table 1: Data format of mobile phone call detail records.
Caller Receiver Date Start Time End TimeDuration(seconds)
BaseStation Long Lat
serveNub1 oppNub1 2007-07-23 09:25:10 09:28:20 190 A 127.495 50.243serveNub1 oppNub2 2007-07-23 12:15:32 12:15:52 20 B 127.502 50.241
96 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
this Voronoi partition, all phone calls within a given polygon are closer to thecorresponding mobile base station than any other station. Generally, urbancentral regions have a higher density of mobile cells (the coverage area of eachcell is smaller) than the outer suburb regions. Different data subsets have beenextracted and processed for various research purposes such as human mobilitymodelling (Kang et al., 2012b), travel behavior studies (Yuan et al., 2012),population estimation (Kang et al., 2012a), and urban community networkstructure analysis (Gao et al., 2013b).
In this work, on one hand, for individual spatio-temporal mobility patternanalysis, the corresponding approximate movement trajectories of each mobilesubscriber were created by connecting his/her series of geo-referenced callrecords in space and time. On the other hand, for urban dynamic structureanalysis, phone call activities and phone users’ physical movement flows wereaggregated based on the Voronoi polygons. More detailed spatio-temporalanalysis and results are presented next.
4. EXPERIMENTS
4.1. Individual Spatial-Temporal Movement PatternsIn urban studies, the understanding of human movement patterns over time isvery important in transportation planning and city management. Figure 5illustrates the 3D STV of a person’s mobile tracking trajectories in space overone week. Each color represents a single mobile subscriber. Different shapes of3D polylines reveal different space-time behaviors for specified users. Forinstance, the regular patterns at fixed locations (e.g., home and job places) withrespect to the daily/weekly cycles can be indicated by the length of verticalsegments of the space-time paths (Figure 5a). In addition, irregular movementpatterns have been also found for certain type of individuals (Figure 5b). One
Figure 5: Space-time visualization of mobile phone users’ trajectories: (a) regular movementpatterns of three individuals; (b) irregular movement pattern of an individual.
Spatio-Temporal Analytics in the Mobile Age 97
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
might guess that the user is a delivery employee or working for relatedprofessions.
The STV technique can help to simultaneously visualize the mobiletrajectory patterns in space and time with an intuitive manner but may needthe statistical analysis to extract more meaningful personal places of interests(POIs). Figure 6a displays the spatial distribution of a mobile user’s call-activity movement in a week. Avisit was recorded when the user made a call ona cell. The visit probability was calculated as the ratio of visiting frequency in agiven cell to the total number of visits across the whole cell network. It tellsthat this person has a higher probability (Cell A: 0.31, Cell B: 0.22; Cell C: 0.14)to frequently visit or stay in a few places, while the visiting probability of mostcells is less than 0.05.The interpretation of time series graphs of phone calls onthese visited places by the user offers the possibility of identifying the personalPOIs based on their temporal signatures.
As shown in Figure 6b, the user often made phone calls at Cell A and Cell Bduring 19:00–24:00 throughout the whole week. One can infer that his/herhome might be located nearby or inside the boundary of the two cells. Also, thetemporal signature of Cell C may indicate a working place since the user ispresent there only during office hours (9:00–18:00) on weekdays but not onweekends. However, the detailed information about personal characterizedPOIs and actual frequent visited places may need further investigation ondaily-activity trips of the mobile subscribers or the study of geographicalcontexts, since the CDR-based trajectories are only abstract presentations ofthe actual trajectories of individuals at the cell-tower coverage scale. Generally,urban core areas have a higher density of cell-tower coverage (the distancebetween two mobile base stations is about 100–500 meters), whereas it isapproximately 2 kilometers or larger in the suburbs.
To identify the spatio-temporal hot-spots of individual phone call locations,we calculated the space-time kernel density estimation (STKDE) of each user’scall activities. Figure 7a shows an individual’s mobile phone call activities overa week in a space-time cube. The STKDE of this user’s call activities wascalculated in a spatial resolution of 500 meters and a temporal resolution of100 minutes. The different combinations of spatio-temporal bandwidths resultin various visual representations. The selection of bandwidths needs severalrounds of calibration by adjusting both spatial and temporal bandwidths to findan optimization that can help to uncover hidden patterns.
In practice, we tried different combinations and finally chose twice thespatial resolution (1 km) as the spatial bandwidth for both horizontaldimensions, and 500 minutes as the temporal bandwidth for the verticaldimension, which gave us a better visual representation for identifyingpatterns in the case study. The resulting density volume of 50*50*100 voxelsfor a user’s call activities is visualized in Figure 7b. One can find that this useris more likely to make calls at a fixed spatial region (red and orange voxels)
98 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Figure 6: Spatial distribution and temporal signatures of individual frequently visitedmobile cells:(a) Spatial display of cell visiting probability for a user (the larger circle represents higherprobability); (b) The time series graphs show the temporal signatures of the user’s frequentvisited places (mobile cells) based on the number of hourly calls over a week.
Spatio-Temporal Analytics in the Mobile Age 99
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Figure 7:Visualizing georeferencedphone call activities and space-time density: (a) phone callevents in space-time cube; (b) STKDE results for a specifiedmobile user in aweek and the kerneldensity value is a normalized value.
100 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
across time (including Monday night, Thursday afternoon and evening,Saturday, and Sunday evening). This example shows that it is much easier tovisually identify the spatio-temporal hot-spot patterns of call activities byusing the STKDE approach.
4.2. Aggregated Phone-Call Interaction PatternsOne of the prominent characteristics of phone call data lies in its indication ofhuman communications and spatial interactions among different places. Theaggregated phone call interactions among mobile cells in different timerepresent the dynamic intra-urban communication landscape that cannot becaptured using other traditional activity survey data. As shown in Figure 8, thevertical Bezier curves were created to represent the hourly phone-call flowacross cells. The height of the arcs represents the relative volume of phonecalls. The tall and narrow arcs show strong call communication within a nearbyintra-urban space; yet, some long-distance curves across nonadjacent cellsindicate strong call interactions among these spatially separated regions.
It is apparent to observe some patterns: (1) there are only several long-distance interactions atmidnight whilemore communications are concentratedin the urban center; (2) very few call interactions exist from 3 a.m.–6 a.m.;(3) there are increasing urban central interactions in the period 6 a.m.–9 a.m.and keeping high-volume interactions across the core region till night; (4) thecall interactions decrease after 9 p.m. Such information flow patterns arestrongly related daily human activity rhythms and working-social connections,as well as geographical contexts including urban land-use types and spatialdistributions of home-job locations (Ratti et al., 2006; Gao et al., 2013b).
To understand the dynamic “source–sink” structures of informationlandscape (Liu et al., 2012), we calculated the phone call net-balance-flow foreach mobile cell by subtracting the outgoing call volume from the incoming callvolume in each hour. Figure 9 shows the time-series plot of phone-call net flowamong all cells. Each line represents the net flow pattern for a specific Voronoimobile coverage area. In Figure 10, it is clear to see dynamic spatialdistributions of the “source” areas (red color), which have more outgoing phonecalls, and the “sink” areas (blue color), which have more incoming phone callsin different hours. The yellow cells mean that the net-balance call flow in thathour is zero. One can interactively interpret the call flow patterns in the 3D-GIS environment or sense the dynamic urban phone-call landscape under theanimation mode.
4.3. The Spatio-Temporal Autocorrelation Patterns of Phone CallsThe study of spatio-temporal autocorrelation structure of mobile phone calls inurban space can help to understand the citizens’ mobile communication
Spatio-Temporal Analytics in the Mobile Age 101
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Figure 8: Vertical Bezier curve 3D visualization of hourly phone-call flow patterns across cells in aday. (Green dots are locations of mobile base stations; each arc represents an OD flow linkingtwo mobile cells).
102 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
24232221201918171615141311 1210987
100
50
0
–50
–100
150
6543
Net
Flo
w o
f Cal
l Int
erac
tions
Hour
210
Figure 9: Time-series plot of phone-call net flow among all mobile cells (each line represents adifferent mobile cell).
Figure 10: Visualizing the dynamic phone-call “source” areas (red), “sink” areas (blue) and“zero-balance” areas (yellow) in 3D-GIS environment.
Spatio-Temporal Analytics in the Mobile Age 103
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
patterns and urban structures. In order to investigate how the spatialautocorrelation structure changes throughout the day in this city, the phone-call volume was aggregated into the Voronoi cells by hour at first. Then, theMoran’s I local indicator of spatial association (LISA) (Anselin, 1995) wascalculated for each cell every three hours (see Figure 11). The value of Moran’s Ihas been standardized to lie in [21, 1]. If the index is larger than 0, the cellshows positive spatial autocorrelation with its neighbors; yet, it indicates thenegative spatial autocorrelation of phone call patterns if the index is smallerthan 0. The closer the value is approaching to 1 (or 21), the stronger thepositive (or negative) spatial autocorrelation is. Examining the spatialstructure of LISA in different time periods, we can clearly see that the spatialautocorrelation patterns of phone calls across all cells are very dynamic and
Figure 11: Local Moran’s indicators of spatial autocorrelation (LISA) analysis on phone calls indifferent time periods of a day (using the 1st-order-neighbor spatial weighted matrix).
104 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
heterogeneous. The central region (small cells) shows more diverse patternsthan the outer suburb areas, where most spatially adjacent cells show similarvalues in the whole day. It might reflect the mixture land-use types of urbancentral areas and human’s convergence and divergence in this place withdifferent frequencies and purposes of phone call behaviors in different timeperiods.
To identify a more stable autocorrelation structure, we apply the spatialstatistic test of running 10,000 simulations of randomized permutations ofneighboring cells to find the local significant spatial autocorrelation patterns(Anselin, 1995). As shown in Figure 12, we generated both Moran’s scatter plotand spatial distribution of the labelled mobile cells with IDs to visually andinteractively identify the statistically significant local association cells (with a0.05 significance level). By comparing the phone call volume of each cell with itsfirst-order adjacent neighbors, there are four types of associations identified:(1) HH: observations in both the target cell and neighbors are high; (2) HL: highcall volume in the target cell with low volume in neighbors; (3) LH: low callvolume in the target cell with high volume in neighbors; (4) LL: the call volumein both the target cell and neighbors are low. Figure 12a illustrates the resultsof significant Moran’s I LISA in the period 3 a.m.–6 a.m.
There are continuous significant LL spatial associations in the urbancentral region and HH association structure in the southwest and northeastsuburb regions which contain large residential housing referring to the GoogleEarth imagery in this city (The imagery with labels is not shown here asrequired by the data provider). But such spatial autocorrelation structurechanges over time, for instance, the central area tends to have mixturepatterns of all HH, HL, LH, LL association types in the period 9 a.m.–12 p.m.(See Figure 12b). The findings help the understanding of local citizens’ mobilecommunications at the spatial and temporal dimensions simultaneously.In addition, the spatio-temporal autocorrelation structure helps the evaluationof current land-use in the study area, e.g., which cells are “the hot regions” allthe day or at a given time-period. It provides important implications for realestate development and business locators. The findings could help the localgovernment and planning agency to make future strategic planning andsupport their decision making.
The spatial weighted matrix plays an important role in spatialautocorrelation analysis (Getis & Aldstadt, 2010). As shown in Figure 13, alarger-order of spatial adjacency tends to have a larger number of neighbors.Another important factor for identifying spatio-temporal autocorrelationstructure is time granularity (e.g., half an hour, per hour, two hours andothers). It inspired us to examine how the different combinations of spatialweights and temporal neighbors affect the STAA results.
Using the methodology introduced in Section 2.3, we implement the globalMoran’s I like statistic of STAA with different spatial lags and time lags for
Spatio-Temporal Analytics in the Mobile Age 105
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Figure 12: Moran’s scatter plot and spatial display of mobile cells with statistically significantlocal associations (HH: High-High; HL:High-Low; LH: Low-High; LL: Low-Low): (a) period3 a.m.–6 a.m.; (b) period 9 a.m.–12 p.m.
106 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
hourly phone-call patterns across all cells. Examining the results reveals twokey findings (See Figure 14a). First, the strength of global Moran’s I, such asspatio-temporal autocorrelation measure (Ist), for hourly phone calls istemporally dynamic and there is a positive-association peak between 6 a.m.–7 a.m.. Second, the Ist measure is more sensitive to the spatial order than thetemporal neighbors. A higher order of spatial weights generally results inhigher strength of spatio-temporal autocorrelation structure.
In addition, we implement the global Geary’s C like STAA measure (Cst)and Getis-Ord’s G like STAA measure (Gst) to identify the spatio-temporalautocorrelation of hourly phone-call patterns in different hours. Note that theCst statistic indicates a positive autocorrelation structure when the value lie in
Figure 13: The 1st, 2nd and 3rd-order of spatial adjacency matrix (a dot means the weightbetween the two cells is 1; otherwise is 0) and the corresponding distributions of neighbors foreach mobile cell.
Spatio-Temporal Analytics in the Mobile Age 107
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Figure 14: Three global measures of spatio-temporal association with different combinations ofspatial weights (spatial orders) and temporal neighbors (1 time-lag: 1 hour; 2 time-lag: 2 hours; 3time-lag: 3 hours) for hourly phone-call patterns: (a) Ist; (b) Cst; and (c) Gst measures.
108 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
(0–1); yet, it is a negative autocorrelation when the value lie in (1–2). It isfound that the hourly autocorrelation trends of Cst measures are more similarto the Ist measures (Figure 14b). But using the Gst measure did not reveal thetemporal dynamics of autocorrelation strength in our datasets. The Gst
statistic is also sensitive to the spatial-order of weighted matrix (Figure 14c).
5. CONCLUSIONS AND FUTURE WORK
In this article, we introduced a spatio-temporal analytical framework forexploring human mobility patterns and urban dynamics. The integration ofspatial-temporal visualization, space-time density estimation and spatio-temporal autocorrelation analysis can not only help to represent spatio-temporaldata visually and interactively but also offer quantitative analytics to identify thespatio-temporal patterns (such as spatio-temporal hotspots) in the mobile phonedata. Our experiments have demonstrated that different spatio-temporaltechniques have their potential advantages but also limitations. Expeditions onthe study of spatio-temporal analytical techniques can contribute to the futuredevelopment of space-time GIS.
For instance, the space-time path is good for visual exploration or anoverview of individual regular (or irregular) movement patterns but might notbe suitable for massive trajectories because of overlapping and clutteringproblems in the space-time cube. The study also demonstrates that the user’s“home” and “working place” can be inferred at the cell-tower coverage scalebased on statistical information of frequently visited mobile cells with thecharacteristics of personal temporal signatures on these places. However,because of the mobile phone data limitations on the spatial scale which heavilydepends on the coverage size of a cell tower, it is difficult to tell which POIs at afiner spatial resolution are actually visited or frequented by a particularindividual. It might need the integration of phone-call data with survey orinterviews on these mobile subscribers for a holistic view of personal activities.In addition, we implement the STKDE to analyze the continuous spatio-temporal density structure for the georeferenced phone call activities and thistechnique can facilitate the identification of spatio-temporal patternssimultaneously.
Furthermore, to understand the phone-call dynamic spatial interactionstructure, we introduce a novel 3D flow visualization approach of generatingvertical Bezier curves in the 3D-GIS environment which also supportinteractive analysis of spatial information and thematic attributes. Moreover,we have investigated different statistical measures (Ist, Cst and Gst) whichextended the classic spatial association indices for the spatio-temporalautocorrelation analysis. The spatial order of weighted matrix was found tohave more significant effects than the temporal neighbors on influencing the
Spatio-Temporal Analytics in the Mobile Age 109
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
autocorrelation strength of hourly phone-call volume across the whole studyarea.
The spatio-temporal analytical framework introduced in this paper can bealso applied in other spatio-temporal datasets (e.g., infectious diseases, crimes,and GPS tracks) for facilitating knowledge discovery and decision support inurban informatics and social sciences. However, there are several relatedresearch issues that still require further work: (1) Although the integration ofspace-time visualization and spatio-temporal statistics can help understandthe individual human mobility patterns and identify limited types of personalplace of interests (e.g., home and working place), more fruitful semanticinformation of locational activities need further investigation on the socio-economic background of the mobile subscribers and geographical contextanalysis in this region. (2) The results of STKDE computation are sensitive tothe selection of spatial and temporal bandwidths. More work is needed to betterunderstand the impact of both factors and may find an optimal solution. (3) Thespatio-temporal uncertainty issues in both STKDE and STAA should bestudied. (4) Last but not least, the effectiveness of spatio-temporal visualanalytics should be evaluated by human participants to find whichrepresentations are more appropriate with respect to certain types of spatio-temporal data for pattern identification and knowledge discovery.
ACKNOWLEDGMENTS
The author would like to thank Dr. Yu Liu at Peking University for providingdata support and thank colleagues Dr. Krzysztof Janowicz, Dr. PhaedonC. Kyriakidis and Yingjie Hu at UCSB Geography for insightful discussionsand suggestions on this work. Thanks to guest editors Dr. ChristopheClaramunt and Dr. Kathleen Stewart and reviewers for their comments thathelped to improve this manuscript.
REFERENCES
Andrienko, G., & Andrienko, N. (2008). Spatio-temporal aggregation for visual analysisof movements. In D. Ebert & T. Ertl (Eds.),Visual Analytics Science and Technology,2008. VAST’08. IEEE Symposium (pp. 51–58). Piscataway, NJ: IEEE.
Andrienko, G., Andrienko, N., Demsar, U., Dransch, D., Dykes, J., Fabrikant, S. I., Jern,M., Kraak, M. J., Schumann, H., & Tominski, C. (2010). Space, time and visualanalytics. International Journal of Geographical Information Science, 24(10),1577–1600.
Andrienko, G., Andrienko, N., Dykes, J., Fabrikant, S., & Wachowicz, M. (2008).Geovisualization of dynamics, movement and change: key issues and developingapproaches in visualization research. Information Visualization, 7(3/4), 173–180.
Adrienko, N., & Adrienko, G. (2011). Spatial generalization and aggregation of massivemovement data. IEEE Transactions on Visualization and Computer Graphics,17(2), 205–219.
110 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Anselin, L. (1995). Local indicators of spatial association—LISA.Geographical Analysis,27(2), 93–115.
Batty, M. (2013). Big data, smart cities and city planning. Dialogues in HumanGeography, 3(3), 274–279.
Box, G. E., Jenkins, G. M., & Reinsel, G. C. (2008). Time series analysis: Forecasting andcontrol. New York, NY: John Wiley & Sons.
Brunsdon, C., Corcoran, J., & Higgs, G. (2007). Visualizing space and time in crimepatterns: A comparison of methods. Computers, Environment and Urban Systems,31(1), 52–75.
Calabrese, F., Pereira, F. C., Di Lorenzo, G., Liu, L., & Ratti, C. (2010). The geography oftaste: analyzing cell-phone mobility and social events. In P. Floreen, A. Kruger, &M. Spasojevic (Eds.), Pervasive computing (pp. 22–37). Berlin/Heidelberg:Springer.
Chai, Y. (2013). Space–time behavior research in China: Recent development andfuture prospect. Annals of the Association of American Geographers, 103(5),1093–1099.
Chen, J., Shaw, S. L., Yu, H., Lu, F., Chai, Y., & Jia, Q. (2011). Exploratory data analysisof activity diary data: a space–time GIS approach. Journal of Transport Geography,19(3), 394–404.
Cheng, T., Haworth, J., & Wang, J. (2012). Spatio-temporal autocorrelation of roadnetwork data. Journal of Geographical Systems, 14(4), 389–413.
Cliff, A. D., & Ord, J. K. (1973). Spatial autocorrelation (Vol. 5). London: Pion.
de Montjoye, Y. A., Hidalgo, C. A., Verleysen, M., & Blondel, V. D. (2013). Unique in thecrowd: The privacy bounds of human mobility. Scientific Reports, 3, 1376.
Delmelle, E., Dony, C., Casas, I., Jia, M., & Tang, W. (2014). Visualizing the impact ofspace-time uncertainties on dengue fever patterns. International Journal ofGeographical Information Science, 28(5), 1107–1127.
Demsar, U., & Virrantaus, K. (2010). Space–time density of trajectories: exploringspatio-temporal patterns in movement data. International Journal of GeographicalInformation Science, 24(10), 1527–1542.
Dijst, M. (2013). Space–time integration in a dynamic urbanizing world: Current statusand future prospects in geography and GIScience: Space–time integration ingeography and GIScience. Annals of the Association of American Geographers,103(5), 1058–1061.
Epanechnikov, V. A. (1969). Non-parametric estimation of a multivariate probabilitydensity. Theory of Probability & its Applications, 14(1), 153–158.
Farin, G. E. (1996). Curves and surfaces for computer-aided geometric design:A practical code. San Diego, CA: Academic Press, Inc.
Gao, S., Hu, Y., Janowicz, K., & McKenzie, G. (2013a). A spatiotemporal scientometricsframework for exploring the citation impact of publications and scientists.In Proceedings of the 21st ACM SIGSPATIAL International Conference onAdvances in Geographic Information Systems (pp. 204–213). New York, NY: ACM.
Gao, S., Liu, Y., Wang, Y., &Ma, X. (2013b). Discovering spatial interaction communitiesfrom mobile phone data. Transactions in GIS, 17(3), 463–481.
Gao, S., Wang, Y., Gao, Y., & Liu, Y. (2013c). Understanding urban traffic-flowcharacteristics: a rethinking of betweenness centrality. Environment and PlanningB: Planning and Design, 40(1), 135–153.
Spatio-Temporal Analytics in the Mobile Age 111
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Geary, R. C. (1954). The contiguity ratio and statistical mapping. The incorporatedstatistician, 5(3), 115–146.
Getis, A. (1991). Spatial interaction and spatial autocorrelation: a cross-productapproach. Environment and Planning A, 23(9), 1269–1277.
Getis, A., & Aldstadt, J. (2010). Constructing the spatial weights matrix using a localstatistic. In Perspectives on spatial data analysis (pp. 147–163). Berlin/Heidelberg:Springer.
Getis, A., & Ord, J. K. (1992). The analysis of spatial association by use of distancestatistics. Geographical Analysis, 24(3), 189–206.
Giannotti, F., Nanni, M., Pedreschi, D., Pinelli, F., Renso, C., Rinzivillo, S., &Trasarti, R. (2011). Unveiling the complexity of human mobility by querying andmining massive trajectory data. The VLDB Journal—The International Journal onVery Large Data Bases, 20(5), 695–719.
Gonzalez, M. C., Hidalgo, C. A., & Barabasi, A. L. (2008). Understanding individualhuman mobility patterns. Nature, 453(7196), 779–782.
Goodchild, M. F. (2013). Prospects for a space–time GIS: Space–time integration ingeography and GIScience. Annals of the Association of American Geographers,103(5), 1072–1077.
Griffith, D. A. (2010). Modeling spatio-temporal relationships: retrospect and prospect.Journal of Geographical Systems, 12(2), 111–123.
Guo, D., Chen, J., MacEachren, A. M., & Liao, K. (2006). A visualization system forspace-time and multivariate patterns (vis-stamp). Visualization and ComputerGraphics, IEEE Transactions, 12(6), 1461–1474.
Guo, D., Gahegan, M., MacEachren, A. M., & Zhou, B. (2005). Multivariate analysis andgeovisualization with an integrated geographic knowledge discovery approach.Cartography and Geographic Information Science, 32(2), 113–132.
Hagerstrand, T. (1970). What about people in regional science? Papers in RegionalScience, 24(1), 7–24.
Hardisty, F., & Klippel, A. (2010). Analysing spatio-temporal autocorrelation withLISTA-Viz. International Journal of Geographical Information Science, 24(10),1515–1526.
Hawelka, B., Sitko, I., Beinat, E., Sobolevsky, S., Kazakopoulos, P., & Ratti, C. (2014).Geo-located Twitter as proxy for global mobility patterns. Cartography andGeographic Information Science, 41(3), 260–271.
Ho, S. S., & Ruan, S. (2011, November). Differential privacy for location pattern mining.In Proceedings of the 4th ACM SIGSPATIAL International Workshop on Securityand Privacy in GIS and LBS (pp. 17–24). New York, NY: ACM.
Jiang, B., & Claramunt, C. (2004). Topological analysis of urban street networks.Environment and Planning B, 31(1), 151–162.
Jones, M. C., Marron, J. S., & Sheather, S. J. (1996). A brief survey of bandwidthselection for density estimation. Journal of the American Statistical Association,91(433), 401–407.
Kang, C., Liu, Y., Ma, X., & Wu, L. (2012a). Towards estimating urban populationdistributions from mobile call data. Journal of Urban Technology, 19(4), 3–21.
Kang, C., Ma, X., Tong, D., & Liu, Y. (2012b). Intra-urban human mobility patterns: Anurban morphology perspective. Physica A: Statistical Mechanics and itsApplications, 391(4), 1702–1717.
112 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Kaufman, A. (1990). Volume visualization. The Visual Computer, 6(1), 1.
Kwan, M. P. (2000). Interactive geovisualization of activity-travel patterns using three-dimensional geographical information systems: a methodological exploration with alarge data set. Transportation Research Part C: Emerging Technologies, 8(1),185–203.
Kwan, M. P. (2004). GIS methods in time-geographic research: Geocomputation andgeovisualization of human activity patterns.Geografiska Annaler: Series B, HumanGeography, 86(4), 267–280.
Kwan, M. P., Xiao, N., & Ding, G. (2014). Assessing activity pattern similarity withmultidimensional sequence alignment based on a multiobjective optimizationevolutionary algorithm. Geographical Analysis, 46(3), 297–320.
Liu, Y., Wang, F., Xiao, Y., & Gao, S. (2012). Urban land uses and traffic ‘source-sinkareas’: Evidence from GPS-enabled taxi data in Shanghai. Landscape and UrbanPlanning, 106(1), 73–87.
Liu, Y., Sui, Z., Kang, C., & Gao, Y. (2014). Uncovering patterns of inter-urban trip andspatial interaction from social media check-in data. PloS One, 9(1), e86026.
Long, J. A., & Nelson, T. A. (2013). A review of quantitative methods for movement data.International Journal of Geographical Information Science, 27(2), 292–318.
MacEachren, A. M., & Kraak, M. J. (2001). Research challenges in geovisualization.Cartography and Geographic Information Science, 28(1), 3–12.
Mavoa, S., Oliver, M., Witten, K., & Badland, H. M. (2011). Linking GPS and traveldiary data using sequence alignment in a study of children’s independent mobility.International Journal of Health Geographics, 10(64), 1–10.
Miller, H. J. (2005). A measurement theory for time geography. Geographical Analysis,37(1), 17–45.
Moran, P. A. (1950). Notes on continuous stochastic phenomena. Biometrika, 37(1–2),17–23.
Nakaya, T. (2013). Analytical data transformations in space–time region: Three storiesof space–time cube. Annals of the Association of American Geographers, 103(5),1093–1099.
Nakaya, T., & Yano, K. (2010). Visualizing crime clusters in a space-time cube: Anexploratory data-analysis approach using space-time kernel density estimation andscan statistics. Transactions in GIS, 14(3), 223–239.
Palma, A. T., Bogorny, V., Kuijpers, B., & Alvares, L. O. (2008, March). A clustering-based approach for discovering interesting places in trajectories. In Proceedings ofthe 2008 ACM symposium on Applied computing (pp. 863–868). New York, NY:ACM.
Phan, D., Xiao, L., Yeh, R., & Hanrahan, P. (2005). Flow map layout. In J. Stasko &M. Ward (Eds.), IEEE Symposium In Information Visualization (INFOVIS)(pp. 219–224). MN, USA. October 23-25.
Pindolia, D. K., Garcia, A. J., Huang, Z., Fik, T., Smith, D. L., & Tatem, A. J.(2014). Quantifying cross-border movements and migrations for guidingthe strategic planning of malaria control and elimination. Malaria Journal,13(169), 1–11.
Ratti, C., Williams, S., Frenchman, D., & Pulselli, R. M. (2006). Mobile landscapes: usinglocation data from cell phones for urban analysis. Environment and Planning BPlanning and Design, 33(5), 727.
Spatio-Temporal Analytics in the Mobile Age 113
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
Richardson, D. B. (2013). Real-time space–time integration in GIScience andgeography: Space–time integration in geography and GIScience. Annals of theAssociation of American Geographers, 103(5), 1062–1071.
Shaw, S. L., Yu, H., & Bombom, L. S. (2008). A space-time GIS approach to exploringlarge individual-based spatiotemporal datasets. Transactions in GIS, 12(4),425–441.
Sheather, S. J., & Jones, M. C. (1991). A reliable data-based bandwidth selection methodfor kernel density estimation. Journal of the Royal Statistical Society. Series B(Methodological), 53(3), 683–690.
Shen, Y., Kwan, M. P., & Chai, Y. (2013). Investigating commuting flexibility with GPSdata and 3D geovisualization: A case study of Beijing, China. Journal of TransportGeography, 32, 1–11.
Shoval, N., & Isaacson, M. (2007). Sequence alignment as a method for human activityanalysis in space and time. Annals of the Association of American Geographers,97(2), 282–297.
Silverman, B. W. (1986).Density estimation for statistics and data analysis. Boca Raton,FL: CRC Press.
Song, C., Qu, Z., Blumm, N., & Barabasi, A. L. (2010). Limits of predictability in humanmobility. Science, 327(5968), 1018–1021.
Stewart, K., Fan, J., & White, E. (2013). Thinking about space-time connections:spatiotemporal scheduling of individual activities. Transactions in GIS, 17(6),791–807.
Tobler, W. R. (1987). Experiments in migration mapping by computer. The AmericanCartographer, 14(2), 155–163.
Wilson, W. C. (1998). Activity pattern analysis by means of sequence-alignmentmethods. Environment and Planning A, 30(6), 1017–1038.
Yu, H., & Shaw, S. L. (2008). Exploring potential human activities in physical andvirtual spaces: a spatio-temporal GIS approach. International Journal ofGeographical Information Science, 22(4), 409–430.
Yuan, Y., Raubal, M., & Liu, Y. (2012). Correlating mobile phone usage and travelbehavior—A case study of Harbin, China. Computers, Environment and UrbanSystems, 36(2), 118–130.
Yue, Y., Lan, T., Yeh, A. G., & Li, Q. Q. (2014). Zooming into individuals to understandthe collective: A review of trajectory-based travel behavior studies. Travel Behaviorand Society, 1(2), 69–78.
114 S. Gao
Dow
nloa
ded
by [
Uni
vers
ity o
f C
alif
orni
a Sa
nta
Bar
bara
] at
11:
15 2
9 A
pril
2015
