EpiMob: Interactive Visual Analytics of Citywide Human Mobility Restrictions for Epidemic Control
Chuang Yang, Zhiwen Zhang, Zipei Fan, Renhe Jiang, Quanjun Chen, Xuan Song, Ryosuke Shibasaki
EEpiMob: Interactive Visual Analytics of Citywide Human MobilityRestrictions for Epidemic Control
Zhiwen Zhang*, Chuang Yang*, Zipei Fan*, Renhe Jiang*,Quanjun Chen, Xuan Song † , Ryosuke Shibasaki CB A1A2A3A4A5 A6 A Fig. 1: EpiMob System. It is an interactive visual analytics system for evaluating and simulating the effects of human mobilityrestriction policies for epidemic control. (A) enables the user to specify the mobility restriction policies like “region lockdown” and“telecommuting (work at home), and the simulation result about the transmission and infection situation will be displayed in policyoverview panel (B). Besides, user can do comparative analysis among policies, and the results will be displayed at (C).
Abstract —The coronavirus disease 2019 (COVID-19) outbreak has swept more than 180 countries and territories since late January2020. As worldwide emergency response, governments have taken various measures and policies such as self-quarantine, travelrestriction, work at home, and region lockdown, to control the rapid spread of this epidemic. The common concept of thesecountermeasures is to conduct human mobility restrictions as COVID-19 is a highly contagious disease with human-to-humantransmission. It becomes an urgent request from medical experts and policy makers to effectively evaluate the effects of humanrestriction policies with the aid of big data and information technology. Thus, in this study, based on big human mobility data and city POIdata, we design an interactive visual analytics system called EpiMob (Epidemic Mobility) to intuitively demonstrate and simulate howthe human mobility as well as the the number of infected people will change according to a certain restriction policy or a combination ofpolicies. EpiMob is made up of a set of coupled modules: a data processing module for data cleansing, interpolation, and indexing; asimulation module based on a modified trajectory-based SEIR model; an interaction visualization module to interactively visualize theanalytical results in light of user’s settings. Through multiple case studies for the biggest city of Japan (i.e.,Tokyo) and domain expertinterviews, we demonstrate that our system can be beneficial to give an illustrative insight in measuring and comparing the effects ofdifferent human mobility restriction policies for epidemic control.
Index Terms —human moblity simulation, epidemic control, visual analyitcs, interactive system, big trajectory data • * Equal Contribution; † Corresponding Author.• Z. Zhang, C. Yang, Z. Fan, R. Jiang, Q. Chen, X. Song, R. Shibasaki are withSUSTech-UTokyo Joint Research Center on Super Smart City, SouthernUniversity of Science and Technology; Z. Zhang, C. Yang, Z. Fan, R. Jiang,Q. Chen, X. Song, R. Shibasaki are also with Center for Spatial InformationScience, The University of Tokyo. E-mail: { zhangzhiwen,chuang.yang,jiangrh,songxuan,shiba } @csis.u-tokyo.ac.jp, { fanzipei,chen1990 } @iis.u-tokyo.ac.jp.Manuscript received xx xxx. 201x; accepted xx xxx. 201x. Date of Publicationxx xxx. 201x; date of current version xx xxx. 201x. For information onobtaining reprints of this article, please send e-mail to: [email protected] Object Identifier: xx.xxxx/TVCG.201x.xxxxxxx NTRODUCTION
The coronavirus disease 2019 (COVID-19) outbreak has swept morethan 180 countries and territories since late January 2020, which hascaused significant losses to public health as well as the economy at aworldwide scale. To respond to COVID-19 emergency, governmentshave taken various measures and policies such as self-quarantine, travelrestriction, work at home, event canceling, and region lockdown to curbthe rapid spread. As COVID-19 is a highly contagious disease withhuman-to-human transmission, the core purpose of these countermea-sures is to conduct human mobility restrictions as possible as we can.How effective these policies could be becomes a significant and urgentquestion, especially for medical experts and policy makers. As the-state-of-the-art researches [15, 42], the effects of mobility restrictionpolicies taken at an early stage in China have been revealed. However,everyday the pandemic situation is still rapidly changing in the world,and governments need to flexibly implement different policies accord- a r X i v : . [ c s . H C ] J u l ng to fundamental conditions of their country and gradually adjust theirpolicies with the development of the epidemic. It is always necessaryto demonstrate and simulate the actual effect of a certain restrictionpolicy or a combination of policies in an easy and quick way. Thus,in this study, we design an interactive visual analytics system calledEpiMob (Epidemic Mobility) based on big human mobility data andcity POI data collected in Tokyo, the biggest city of Japan. Our systemmainly focuses on two types of human mobility restriction policies,namely “close a region from community level to citywide level” and “from date A to date B”, which are widely adopted and implemented bynumerous governments in the world. The target is to demonstrate howthe citywide human mobility as well as the number of infected peoplecould change in light of user’s different spatiotemporal settings. Forinstance, through EpiMob, we can easily do the epidemic simulationsas follows: (1) starting to close the Setagaya Ward, one of 23 wards ofTokyo, from 2020-04-01; (2) starting work-from-home policy for theentire 23 wards of Tokyo from 2020-03-01.In order to do so, we extend a normal numerical (Susceptible-Exposed-Infectious-Recovered) to a modified trajectory-based SEIRmodel. The SEIR model is a variant of the SIR (Susceptible-Infectious-Recovered) model, which is seen as one of the most fundamentalcompartmental models in epidemiology. The SEIR model consists offour compartments: S for the number of susceptible, E for the numberof exposed, which means the individuals in an incubation period butnot yet infectious, I for the number of infectious, and R for the numberof recovered or deceased (or immune) individuals. To represent thatthe number of susceptible, infected and recovered individuals may varyover time (even if the total population size remains constant), we makethe precise numbers a function of t (time): S(t), E(t), I(t) and R(t). For aspecific disease in a specific population, these functions may be workedout in order to predict possible outbreaks and bring them under control.The SIR and SEIR model are reasonably predictive for infectious dis-eases that are transmitted from human to human, and where recoveryconfers lasting resistance, such as measles, mumps and rubella [1].Moreover, spatial SIR and SEIR model could be built by meshing anarea into a set of grids and setting the infection propagation from onegrid to its surrounding neighbors. For instance, based on Baidu qianxionline service (China) [2], researchers first collected city-to-city inflowand outflow data (i.e., how many people moved from one city to an-other within one day), then they proposed a city-to-city spatial SEIRmodel to predict the COVID-19 epidemic peaks and sizes in China at anationwide level [52]. The granularity of city-to-city aggregation datais coarse, so that this model is difficult to be applied to fine-grainedsimulation at a citywide level. Therefore, we propose a novel SEIRmodel that assumes that each person has a certain probability to getinfected by the other persons inside the same grid at the same times-tamp. Given the human trajectories of one city and a set of parameters,the epidemic simulation could be dynamically and continuously exe-cuted by calculating and updating our new SEIR model at a fixed timefrequency (i.e., every 5 minutes in our case).EpiMob is a system that can interactively manipulate the modifiedSEIR model (i.e., setting inputs) and intuitively demonstrate the simu-lation results of the SEIR model (i.e., getting outputs). Each simulationessentially involves a set of human trajectories and model parameters.Given different human mobility restriction policies, we first generate anew set of restricted human trajectories via a mobility generative model,then feed the new trajectories plus new parameters to our SEIR modelto trigger a new simulation. The main user interface of our EpiMob sys-tem is shown in Fig.1. Users can specify the mobility restriction policeslike “region lockdown” and “telecommuting (work at home)” throughthe console panel in the left, and the simulation result about how thetransmission and infection situation will change will be displayed in theright. Users can also make spatial interactions like “close a polygonalregion” or temporal interactions like “set time interval” through theright panel. Through multiple case studies for the biggest city of Japan(i.e.,Tokyo) and domain expert interviews, we demonstrate that oursystem can be beneficial to give an illustrative insight in measuring andcomparing the effects of different human mobility restriction policiesfor epidemic control. To the best of our knowledge, EpiMob is the first interactive visual analytics system that can provide epidemic controlpolicy simulation at fine-grained spatiotemporal granularity by utilizingcitywide human mobility data and city POI data as inputs. The majorcontributions of our study are summarized as follows:• We propose a novel trajectory-based SEIR model to simulate theepidemic spreading based on real-world big human trajectory dataand city POI data.• We design a “trajectory replacement” simulation strategy to han-dle the different settings on human mobility restrictions. Based onthis strategy, we implement an online web system with a delicateand efficient system architecture.• We provide a comfortable user interface with a set of new visu-alization techniques to visually and interactively help end usermeasure and compare the quantitative effects of different humanmobility restriction policies.• We evaluate our system through multiple case studies as wellas interviews of domain experts and demonstrate the superiorperformance, functionality, and usability. ELATED W ORK
To help prevent the spread of COVID-19, researchers have done epi-demic data analytics as well as developed the epidemic predictionmodel. For instance, an investigation of transmission control measureswas conducted for the first 50 days of outbreak in China [42]. As one ofthe most effective mobility restriction policies, travel restriction and itseffects were analyzed in [15]. [49] proposed a mathematical predictionmodel to nowcast and forecast the spread of COVID-19 inside andoutside of China. Using Baidu Qianxi data (i.e., city-to-city inflow andoutflow number) [2], modified SEIR and AI prediction model wereproposed to predict the COVID-19 epidemic peaks and sizes [52]. [38]was specifically proposed to estimate the epidemiological parametersof prediction model. A trajectory-based algorithm has been proposedto efficiently detect suspected infected person from a large crowd ofpeople [21]. Before the recent COVID-19 studies, epidemic-relatedstudies have already been done in our visualization community. Forexample, VoroGraph [17] integrated a set of visualization tools for epi-demic analysis; Epiviz [10] was an interactive visual analytics systemdesigned for functional genomics data. However, these systems are notwell tailored to our human-mobility-based epidemic control problem.
Human mobility has been analyzed through big trajectory data such asmobile phone GPS data or taxi GPS data. Researchers have proposeda family of algorithms to efficiently find different patterns from bigtrajectory database, such as “Convoy Pattern” [24], “Swarm Pattern”[28], and “Gathering Pattern” [55]. A robust and fast algorithm wasproposed to discover similar trajectories given on query trajectory [11].Meanwhile, mobility simulation at a citywide level has been a bigchallenge since the last decade. Brinkhoff generator [8] was proposedto generate and simulate network-based trajectories given the roadnetwork of a city. MNTG [33] extends [8] into a web-based trafficgenerator. SUMO [7] can simulate human mobility in a big urbanarea. MATSim (Multi-Agent Transport Simulation) [22] has beenseen as the state-of-the-art solution for citywide traffic simulation inthe filed of transportation engineering. By employing the machinelearning technologies, data-driven methodologies have been proposedfor mobility simulation or generation for a large urban area [6, 27, 34,53]. Besides, a human mobility simulator was specifically developedfor disaster situation (i.e., 2011 Great East Japan Earthquake) [40].Simulating or generating the citywide human mobility under differentrestriction policies is still seen as a relatively underexplored topic. .3 Visual Analytics on Spatiotemporal Data
Recent advances and challenges in information visualization are sum-marized in [32]. Especially, researchers have given an overview ofvisualization techniques on classic trajectory [5], traffic flow [14, 39],and urban computing [56]. As representative applications of visualanalytics on spatiotemporal data, SmartAdP [30] was a visual analyticssystem for selecting billboard locations with trajectory data; AirVis [16]and [37] focused on visual analytics for air pollution problem. In par-ticular, based on trajectory data, [45] did visual analysis for traffic jamproblem; [4] extracted significant places; [31] explored route diversity;Trajgraph [23] studied urban network centralities. From a data per-spective, visual analytics was done on new york city taxi data [18],sparse bus trajectory data [36], trajectory attribute data [43], and origin-destination trajectory data [57]. Moreover, as spatiotemporal data arehigh complex, many studies aimed to utilize visualization techniques tobetter understand big trajectory data or discover trajectory pattern. Mo-bilitygraphs [44] utilized graph and clustering to visually understandmass mobility dynamics; [13] focused on pattern discovering fromgeo-tagged social media data; Telcovis [50] explored co-occurrencepattern based on telco data; [20] aimed to provide pandemic deci-sion support using spatial interaction patterns. Lastly, visualizationtechniques on spatiotemporal data are proposed for some special pur-poses. [41] embedded spatiotemporal information to map; [51] enrichedgeographical information to mobility flow; Location2vec [58] proposeda situation-aware visual representation of urban locations; R-Map [12]designed a map metaphor to understand reposting process in socialmedia; Srvis [46] focused on ranking visualization for spatial informa-tion; Homefinder [47] aimed to find ideal home via visual analytics;SmartCube [29] was proposed for the real-time visualization of spa-tiotemporal data. Our study emphasizes on fitting the spatiotemporalvisualization and interaction well into city-scale epidemic simulator.
RELIMINARY
In this section, we first introduce some basic concepts, describe thedata source used in our study, and give the task analyses based on thediscussion with domain experts.
To elaborate the problem, we formally define some terms related tocitywide human mobility.•
Human Trajectory:
The human trajectory collected for anindividual person essentially comprises a 3-tuple sequence:( timestamp , latitude , longitude ), which can indicate a person’slocation according to a captured timestamp. It can be furtherdenoted as a sequence of ( t , l )-pair attached with user ID uid bysimplifying timestamp as t and ( latitude , longitude ) as l . tra j = { uid , ( t , l ) , ( t , l ) ,..., ( t n , l n ) } • Citywide Human Mobility:
Citywide human mobility Γ refersto a large group of user trajectories in a given urban area. Given ause ID uid , we can retrieve his/her personal trajectory from Γ asfollows: Γ uid = { ( t , l ) , ( t , l ) ,..., ( t n , l n ) } • Grid-Mapped Interpolated Human Trajectory:
The samplingrate of raw human trajectory data is usually unconstant. Afterapplying a typical prepocessing method proposed in [25, 26], wecan get interpolated human trajectory with a constant samplingrate ∆ τ as follows: Γ uid = { ( t , l ) , ( t , l ) ,..., ( t n , l n ) } , ∀ i ∈ [ , n ) , | t i + − t i | = ∆ τ where ∆ τ is set to 5 minutes in this study. Furthermore, we mapthe interpolated human trajectory onto mesh-grid as follows: Γ uid = { ( t , g ) , ( t , g ) ,..., ( t n , g n ) } , ∀ i [ , n ] , l i ∈ g i To this end, we have done preprocessing including interpolationand grid-based mapping to citywide human mobility data. Next, weformally illustrate how to run the epidemic simulation for differentmobility restriction policies.•
Trajectory-Based Epidemic Simulation:
We can do epidemicsimulation with trajectory-based SEIR model F SEIR as follows: E sim = F SEIR ( Γ ; Θ ) where Γ is the given citywide human mobility, Θ refers to theparameters of SEIR model, and E s im denotes the simulation re-sults including the infected trajectories and the infection number.Every time we give a set of ( Γ , Θ ) to the trajectory-based SEIRmodel, we could run the epidemic simulation in a new round.• Mobility Restriction Policy:
Our study focuses on evaluatingand measuring the effects of different restriction policies. Specifi-cally, we list some restriction policy terms as follows: – Detection refers to setting up a infection detection point ina specific location like roadside or station to detect whetherthis person is infectious or not. – Telecommuting is a corporate policy that allows employ-ees to work from home using information and communica-tions technologies instead of commuting to the office. – Region Lockdown is a government policy that implementsmandatory geographic quarantine to all of the citizens livingin a specific region (city or ward).•
Restricted Mobility Generation:
Given one mobility restrictionpolicy or a combination of several policies Φ , citywide humanmobility will forcibly change due to the given Φ . In our study,we utilize a mobility replacement model denoted as F MOB togenerate the restricted human mobility Γ (cid:48) w.r.t Φ as follows: Γ (cid:48) = F MOB ( Γ ; Φ ) • Epidemic Simulation with Restricted Mobility:
Given the re-stricted citywide human mobility Γ (cid:48) w.r.t Φ and a set of newparameters Θ (cid:48) , epidemic simulation for the restriction policy set-tings E (cid:48) sim could be implemented as follows: E (cid:48) sim = F SEIR ( Γ (cid:48) ; Θ (cid:48) ) This reflects the key idea of our simulation strategy: (1) generatingnew citywide human trajectories for the given restriction policysettings; (2) applying the trajectory-based SEIR model on the newgenerated trajectories.
To model real-world human mobility used for epidemic simulation, wecollected a GPS log dataset anonymously from about 1.6 million realmobile-phone users in Japan over a three-year period (from August 1,2010 to July 31, 2013). This dataset contains about 30 billion GPSrecords, more than 1.5 TeraBytes. The data collection was conductedby a mobile operator (i.e., NTT DoCoMo, Inc.) and private company(i.e., ZENRIN DataCom Co., Ltd.) under the consent of mobile phoneusers. These data were processed collectively and statistically in orderto conceal private information such as gender or age. By default, thepositioning function on the users’ mobile phones is activated every5 minutes, so their positioning data (i.e., latitude and longitude) areuploaded onto the server. However, the data acquisition is affected byseveral factors such as loss of signal or low battery power. In addition,when a mobile phone user stops at a location, the positioning functionof his/ is automatically turned off to save power. In this study, weselect Greater Tokyo Area (including Tokyo City, Kanagawa Prefecture,Chiba Prefecture, and Saitama Prefecture) as the target area of epidemicimulation. The user ID will be selected as our experimental data if 80%of user’s trajectory points locate in Greater Tokyo Area. After this, wecan obtain 145507 users’ trajectories in total that covers approximately1% of the real-world population.
The distribution of POI (point of interest) has a strong relationship withthe parameter settings of our SEIR simulation model. A quantitativecharacterization of the POI effects plays an important role in conduct-ing real-world simulation. Therefore, we collected the Telepoint PackDB of POI data in February 2011 provided by ZENRIN DataCom Co.,Ltd [3]. In the original database, each record is a registered land-linetelephone number with coordinates (latitude, longitude) and industrycategory information included. We treated each “telepoint” as onespecific POI. All the POIs were classified into 40 categories. The totalnumbers of POIs for Tokyo is 281,400. We manually deleted someof unrelated categories retained five categories that are very relevantwith epidemic simulation, namely “entertainment”, “restaurant”, “su-permarket and shopping mall”, ”public place”, and “subway and busstation”.
By discussing with the experts in the form of structured interviews, wecompiled a list of analytical tasks.
R.1
Epidemic Transmission Visualization: How do the citywide hu-man trajectories distribute at a citywide level? How does theepidemic transmission process look? These visualizations helpusers better understand the epidemic situation of a big city from aperspective of human mobility.
R.2
Epidemic Control Policy: How to do the epidemic simulationby setting one specific policy (e.g., region lockdown, detection,telecommuting) or selecting several policies as a combination?These require our system not only to provide an easy-using fron-tend UI, but also to design a robust and flexible backend architec-ture for multi-policy simulation.
R.3
Spatiotemporal Setting: How to select a specific region to imple-ment lockdown policy or telecommuting policy? How to set aninfection detection point at a specific location? How to set thestart date and the end date of one specific policy? User shouldbe able to do these spatiotemporal settings with interactive visualassistance.
R.4
Basic Parameter Setting: How to set the basic parameters such as β , β , and r for the epidemic simulation model SEIR? R.5
Advanced Parameter Setting: How to do the advanced parame-ter adjustments for the epidemic simulation model SEIR? Forinstance, human-to-human transmission probability r could varyfrom one type of POI to another. User should be able to adjustthe transmission probability r according to POI distribution. R.6
Policy Evaluation & Comparison: How to intuitively demonstrateand compare the evaluation results? User requests us to providemultiple visualization analytics results in a well-organized, user-friendly, and highly-informative layout.
YSTEM A RCHITECTURE
EpiMob is a web application with frontend backend separation archi-tecture. The frontend is implemented by React.js (for building userinterfaces) and DECK.GL (for visual analysis of large-scale spatialdata). The backend is designed as a Restful API, implemented byPython. A set of coupled modules are utilized to construct our EpiMobsystem, the architecture of which is depicted as Fig.2.• The visual module can show: (1) the movement and heatmap ofinputed human trajectories; (2) time-series plots of infection num-ber, which is the output of our SEIR epidemic model. Multiple
Data Preprocessing Module
CleansingInterpolationIndex BuildingStorage (Date,Grid)-Uids Index(Date,Uid)-Traj Index (Uid,Grid)-Dates Index A Restricted Mobility AParameter ANon-Restricted MobilityDefault Parameter B Restricted Mobility BParameter B
Simulation ModuleVisual Module
Policy Setting & Parameter Setting Results Visualization & AnalysisAffected PopulationExtraction
Query Processing Module
Trajectory Replacement A Interactive Module
Trajectory DataPOI Data
Fig. 2: System Architecture.analytics results are well displayed to help user compare differentpolicies intuitively.• The interactive module can specify: (1) the epidemic control pol-icy such as “close region”, “telecommuting (work-from-home)”,or the combinations; (2) the spatiotemporal settings of the se-lected policy such as the specific polygonal area and the start/enddata of policy implementation period; (3) the basic parametersettings including β (the rate of transmission for the susceptibleto infected), β (the rate of transmission for the susceptible toexposed), and r (the number of contacts per person per day); (4)the advanced parameter setting like POI risk factor r POI . It isused to adjust the original r based on the POI distributions as theregion containing more restaurants and bars is supposed to havea higher infectious risk. Here, (1) and (2) are combined as thehuman trajectories with specified restrictions. (3) and (4) are thebasic and advanced parameters. Furthermore, user can set (1) ∼ (4)simultaneously as a combination of restriction policies.• The query processing module can respond to the user settings ondifferent restriction policies, also called as “queries”, delivered bythe interactive module. It first extracts the people who are affectedby the given policy, then generates a substitution trajectory foreach of those affected people.• The simulation module can simulate the epidemic spreading withour trajectory-based SEIR model.• The data preprocessing module can do data cleansing, interpo-lation, and indexing for citywide human mobility data and cityPOI data. LevelDB is used as the key-value database to efficientlystore and retrieve trajectory data. ODEL
In this study, we used modified SEIR-equation to account for a dynamicSusceptible[S] and Exposed [E] population state, which was introducedby infectious disease experts Nanshan Zhong for predicting epidemicstrend of COVID-19 [52]. The latent [E] population is asymptomatic butinfectious, and [I] refers to the symptomatic and infectious population.Here, we modified this model by replacing inflow/outflow rate withlarge-scale real GPS trajectory data, as Fig.3. To construct a peopleflow for epidemic simulation from a raw GPS record dataset, in the firsttep, we need to discretize the time and coordinates. We selected 5 minas the time interval therefore divided one day into 288 time-slices. Inaddition, we meshed the city into hexagon mesh by H3 grid system [9].The grid are generated by H3 Hexagonal mesh of Level 8. And weconducted the epidemic simulation in every hexagon mesh as time-sliceincreases, respectively. Our modified model is given by: S [ t + ] = S [ t ] − β × r [ t ] × I [ t ] × S [ t ] N [ t ] − β × r [ t ] × E [ t ] × S [ t ] N [ t ] E [ t + ] = E [ t ] + β × r [ t ] × I [ t ] × S [ t ] N [ t ] + β × r [ t ] × E [ t ] × S [ t ] N [ t ] − σ E [ t ] I [ t + ] = σ E [ t ] + I [ t ] − γ I [ t ] R [ t + ] = γ I [ t ] + R [ t ] (1)Here, S ( t ) denotes the number of susceptible people in a hexagonFig. 3: Our proposed SEIR model for epidemic simulation based ongrid-mapped interpolated human trajectory.mesh, N ( t ) denotes the total population, and E ( t ) denotes the numberof exposed number. I ( t ) denotes the number of infected people. β denotes the rate of transmission for the susceptible to infected, β denotes the rate of transmission for the susceptible to exposed, and r ( t ) denotes the number of contacts per person per day, related to controlpolicies. σ is the incubation rate which is the rate of latent individualsbecoming symptomatic (average duration of incubation is 1/ σ ), and γ is the average rate of recovery or death in infected populations. Forbasic epidemic parameter setting, we also set above parameter β , β , r ( t ) , σ and γ as the estimated trends of COVID-19 (coronavirus disease2019) transmission [52]. Basic epidemic parameter setting can be listedas follows: r =
15 as initial contact number per person per day, and r = β , β , σ and γ are set as 0.15747, 0.78735, 0.154 (95% confidence interval) and 1/7(incubation period of seven days). When it comes to the restricted mobility model, we need extract thesignificant location places particularly home and work places to im-plement control policy. Most human activities are routine and peopletend to spend time in the same places in their daily life. To extract thesignificant places particularly home and work places, we applied staypoint extraction algorithm [35] based on the spatial and temporal valuesof points. We detected stay points for each individual in a trajectory ofone month(from July 1st to July 31st, 2012). Stay points are detectedwhen the individual spends at least one hour within a distance of 500meters from a given trajectory point. Every stay points coordinates arethe median latitude and longitude values of the points found withinthe specified distance. Finally, we compute the mesh id of these staypoints’ coordinates also according to level 8 of H3 grid system, whichis used for getting mobile phone users home and work mesh to gen-erate replacement-based restricted mobility model. By analyzing andclassifying time duration of each mesh that corresponds to stay pointsof mobile user in the day of 24 − h period, home and work places can be possibly derived. We used periods from 00:00 to 06:00 for nighttime and 11:0017:00 for day time [48]. Some period was omitted dueto the high possibility of being commuting time. Then, the percentagewas calculated comparing the sum of all values in every mesh. Finally,we determine home and work places where mobile phone users spentmore than 80% of their total stay time during the period of night andmorning time, respectively. According to this standard, we select 11985mobile phone users form 145507 users whose activities mainly locatedin Greater Tokyo area, in order to conduct epidemic simulation underdifferent control policy. For example, stay hours detection of a mobilephone user who has determined home and work location in main staymesh can be shown in Table 1. Mesh ID Hour 0 1 2 3 4 5 6 11 12 13 14 15 16 17235375 17.3 20.2 22.6 23.8 24 25 26.5 5.1 5 4.8 3.8 3.2 5.0 4235737 3.8 2.0 2.0 2.0 2.0 2.0 1.4 22.4 23.7 24.3 23.6 22.0 23.1 23.1235198 4.0 3.9 3 3 3 1.95 1 1 1 0.2 0 0 0 0235197 1.8 1 1 1 1 1 1 0 0 0 0 0 0 0236274 1 1 1 1 1 1 1 0 0 0 0 0 0 0
Table 1: Stay hours of a mobile phone user who has determined homeand work location in main stay mesh. For example, the hour 0 denotestotal stay hours of 0 ∼ Detection is a common control policy due to its flexibility and economy.However, where to set up detection point is a crucial problem forinfection detection. At the same time, the setting of detection pointsis closely related to the distribution of POI. For example, governmenthealth management department often sets up detection points in publicgathering areas such as subway stations or large shopping malls. AsFig.4 shows, users can set up reasonable trajectory-based detectionpoints by exploring the distribution of concerned POI. And for proposedtrajectory-based epidemic simulation under detection policy, after usersselect grid-based detection points (i.e. selected meshes), we assumethat temperature detection is performed in the selected meshes in entireepidemic process, and the probability of detecting an infected personfrom the infected group is 87.9%, which is set according to the newestresearch of COVID-19 [19]. Once these infected persons are detected,they will be quarantined(i.e., cut off subsequent trajectories) and notinfect others.
Telecommuting is a work arrangement in which employees do not com-mute or travel to a central place of work, such as an office building,warehouse, or store. In this system, we try to simulate the spread ofinfectious diseases under telecommuting restriction from the view ofGPS trajectory. And this government control policy often relates tocertain administrative district. By analyzing activity pattern of mobilephone users’ GPS trajectory, we can acquire home and workplaces ofmobile phone users and identify their workplaces belonging to whichcity or prefecture in Greater Tokyo area, and we can detect how manypeople work in every city or prefecture in Greater Tokyo area as well.Therefore, our system allows users to select administrative districtswhere people stay at home and work remotely, and users can set thetelecommuting ratio of selected administrative districts. For proposedepidemic simulation under telecommuting restriction policy, we ran-domly select mobile phone users in selected administrative districts,and the number of selected users meets user-defined telecommutingratio. We detect selected mobile phone users’ grid-trajectory of onemonth, and determine if they go to work mesh day by day. And wecombine their work days with the period of policy implementation toeplace the grid-trajectory of working day with home mesh during theperiod of policy implementation, i.e. making selected users stay athome all day during influenced work days.
Region lockdown is a very urgent infectious disease control policy,which may bring huge loses to the society and economy. On 23 January2020, China imposed a lockdown in Wuhan and other cities in Hubeiprovince in an effort to quarantine the center of an outbreak of COVID-19. Aside from locking down the Greater Wuhan area, Hubei residentswere dissuaded from returning to their workplace. The effectivenessand necessity of such undertakings have been proved. For example,Wu et al. [49], predicted that without control measures the epidemicsize in Wuhan would reach 75,000 infections by January 25 and theepidemic would peak in April. Similarly, Read et al. [38], predicteda peak of 190,000 cases by February 4 without control measures. Forproposed epidemic stimulation under region lockdown, initially, oursystem allows users to circle polygons of any shapes as a blocked area.So we can got all the mesh numbers within the closed area based onH3 grid system. As is shown as query processing module of Fig.2,once users determine the blocked area and blocked time period, theaffected mobile phone userss trajectory simulation can be divided intofollowing two situations. For mobile phone users who stayed in theblocked area at the beginning of the blocked time period, their grid-trajectory remained unchanged throughout the blocked period. And forthose mobile phone users who visited the blocked area after the regionlockdown, we query their one-month historical trajectory database, andrandomly find the trajectories that have not passed the blocked areafor a day to replace the trajectories that passed the blocked area thatday for every mobile phone user. Noted that we set r = ISUAL D ESIGN
In this section, we present the design goals from the perspective of userrequirements (6.1). Then introduce in detail the visualization views ofEpiMob for interactive policy setting and results analysis (6.2,6.3,6.4).
G1: Flexible basic parameters settings of the propagation model.
The living habits, living environment, and public health conditions ofdifferent cities are different, which leads to the variation of β , β , r .The system should allow the user to set the parameters of spread basedon the objective condition of the target city. Besides, the traditionalspread model treats the whole city as a homogeneous region, whichmeans the spreading parameters are all the same across the whole city.However, depending on the functional division of the city, some regionshave more entertainment facilities/shopping malls than others like thecentral business district. It makes people more likely to be exposedto other people and causes a higher r value. To address this issue, thedesign should support a more fine-grained setting(i.e., each grid has itsown r value). G2: Interactive spatial-temporal restriction setting.
A specific re-striction policy must have concrete spatio-temporal information(i.e.,the implementation period and regions). However, the setting of spatial-temporal attributes is quite complicated for users due to the vast selec-tion space. From the perspective of users, they want to set time rangeand regions reasonably, intuitively, and conveniently. To satisfy thisrequirement, we need to provide some prior knowledge to assist theuser setting, and the prior knowledge should be displayed intuitively.Besides, users can preview, adjust their settings before submitting tothe simulation module.
G3: Comparative visual analysis of different control policies.
Aftersubmitting different policies, it is inevitable to compare the advantagesand disadvantages of different policies. However, due to the diversityof potential geographical selection space, how to help users distinguishdifferent policies has also become a challenge. For example, A userlaunched two policies, both including a variety of regions, and thetwo regions set intersect a lot. It is hard to automatically generate a name code for them so that they can be easily distinguished. In orderto facilitate comparison, our interactive design also needs to boot andallow users to set a name code for policy by themselves. Further, withthe help of distinguished name codes, we also need to support usersto select different policies for comparison comfortably and save theresults for further analysis.
Users are desired to set basic epidemic parameters and POI risk toconduct trajectory-based epidemic simulation and acquire the solutionviews for the infectious results.
Different cities have different actual conditions. As shown in the Fig.1-A4, we allow experts to set the basic spread parameters(i.e., β , β , r )based on the concrete situation. Also, the selection of simulationperiods is supported, which could help experts to explore the impactof periodical human behavior changes on epidemic control(e.g., thehuman mobility behavior in winter and summer is quite different [54]). In order to achieve the design objectives mentioned in G1, we designedthe POI Risk Adjustment panel to reduce the difficulty of fine-grained r setting(Fig.1-A5), which utilizes the POI information in a grid to setits r value. As there are many kinds of POI scattered in a grid, ourplain idea is that experts could assign a new r value to each kind ofPOI based on their experience. The specific setting method is to adjustthe r-value change rate of each POI type based on the r base value, alsocalled risk adjustment in EpiMob. Then, according to the proportion ofeach type of POI in the grid, weighted summation to calculate a new rvalue ( r new ). The concrete calculation method of r new for each kind ofPOI is shown in Equation (2). r new = r base ∗ n ∑ i = p i ∗ ( + ∆ i ) (2) Users can obtain sufficient and effective prior knowledge such as trafficflow and POI distribution through interactive restriction setting view toformulate control policy that meets users’ expectations.
Fig. 4: Spatial distribution of the Entertainment POI. The markersrepresent the fever detection points during simulation.To help users discover potential detection points, we design a de-tection view, which shows the geographic distribution of various typesof POIs (Fig.4). In actual life, most of the detection points set at theentrance and exit of POI, such as the entrances and exits of stations andlarge shopping malls. The user may prefer to select the locations wheresome kinds of POIs are denser than others(e.g., In Fig.4 the user selectsthe places which have a higher entertainment POIs density than others).In this view, users can tick one or several types of POIs to observe thedistribution.The method of adding detection points is straightforward.We supply two types of adding methods in the detection control panelFig.1-A2): draw selection areas directly on the map or drag mark to thelocation of interest. After successfully added, a mark will be generatedat the selected grid, indicating that the detection will be performedhere. In the subsequent simulation, all passing people will be detectedaccording to the detection model in Section 5.2.1.
To help the user discover areas which administrative district implementtelecommuting policy is necessary, we design the telecommuting viewas Fig.1-A3 shows. Users could set a series of regions to executetelecommuting. According to the region’s conditions, users can controlthe policy executive strength by setting the start date, reduction rate,and duration ( e.g., Reduction to 90% means that 90% of people work inthat region will work at home). To supply more prior knowledge for theregion selection, we integrate a heat map of all person’s workplaces onthis view (Fig.5), when the mouse places over an area, the informationof the area is displayed. Users could select the regions where has alarger daily commuting flow to reduce the spread risk. The switchlocates on the right up corner of the telecommuting panel sets thevisibility of heatmap.Fig. 5: Workplace Heatmap. The darker color represents more peopleworking at there.
In order to help users select potential lockdown areas, we propose theregion lockdown view, which provides the traffic flow information ofthe city. Fig.1-A1 shows the control panel of this view, where userscould set the lockdown start date and duration. By opening the switchon the control panel, users could get the prior traffic flow information(Fig.1-A6). Generally speaking, a large-scale epidemic spread hasmore possibility to happen at the place with heavy traffic flow. Wemap the flow value to a color schema(i.e., dark blue to dark red) inwhich the deeper color represents the larger flow (for each grid, thetraffic flow refers to the number of move in and move out trips during aperiod). Besides, we provide two auxiliary subviews containing moredetailed information.
Flow Delta sub-view : users can get the flowchange information, which is calculated based on the data of the sameperiod last week. this sub-view also can help to explore abnormalflow. For example, in Fig.6, we find a significant event, the SumidaRiver Firework Festival. Further, based on the area where a largeamount of abnormal traffic found, the user can block the area during theabnormal period, perform a customized propagation simulation, andanalyze the impact of stopping the large-scale activity on the spread ofthe virus.
OD Analysis sub-view : users could analyze the in-outflowdistribution of the target area on the map by this one. With the aboveprior knowledge supplied in our region lockdown view, users couldfind potential blocked areas effectively.
Users can acquire the policy results according to their basic epidemicparameter setting and the selected control policy, including not onlysingle policy result view, but also the comparative analysis view.
Single Policy Result View.
After the user launches a mobility restric-tion policy, the result of that policy will be displayed in the policy Fig. 6: Flow Detla View. The peak shows the abnormal flow caused bythe Sumida River Firework Festival at 2012/07/28.Fig. 7: The simulation result for implementing region lockdown for thecentral area of Tokyo since July 8th.results overview panel(Fig.1-B). Fig.7 shows a policy result, named“lockdown tokyo center from 0708”, referring to perform lockdownin Tokyo center part since July 8th. As mentioned in G3, in order tofacilitate subsequent comparisons, the policy name is set by the users.The clips under title represent a preview of the basic settings, whichcould deliver an intuitive message of the policy. In the chart section,the blue curve represents the cumulative number of infections, and thepurple area represents the 95% confidence interval. When the mousehovers on the corresponding position of the curve, specific details ofthat position will be displayed. Besides, there is a checkbox in the upperright corner of view, which designed for users to conduct comparativeanalysis conveniently. The user checks the target policies first, thenclicks the compare button (the bottom right corner of Fig.1-B), andthe corresponding analysis result will be displayed in the ComparativeAnalysis View.
Comparative Analysis View.
After the user selects several singlepolicies to compare, a new comparative view will be generated, whichwill put multiple curves together for comparative analysis (Fig.1-C).Similarly, the user can customize the name of the analysis result.
VALUATION
In this section, we conducted several case studies to validate proposedtrajectory-based epidemic model and EpiMob system. We first analyzedthe epidemic simulation of large public events to verify if our proposedtrajectory-based epidemic simulation can actually reflect the spread ofinfectious disease. Then we explored the containment of infectiousdiseases under different control policy, i.e., detection, telecommutingand region lockdown. Furthermore, we also explored the effect of multi-policy combination for epidemic spread. Finally, our EpiMob systemis evaluated by the experts in the fields of immunology, computationengineering and urban computing. .1 Case Study (a) GPS trajectories in Taito Ward ofTokyo at 20:00 PM, on July 28th. (b) GPS trajectories in Taito Ward ofTokyo at 23:00 PM, on July 28th.(c) Origin-Destination (OD) of the fire-work festival gathering. (d) Origin-Destination (OD) of the fire-work festival dispersing.
Fig. 8: Sumida river firework festival of epidemic simulation.
Stopping large public events including sports fixtures and concertsplays a crucial role in curbing the spread of infectious diseases. In thisscenario, we aim to verify our system by doing the epidemic simulationon a public gathering event, namely Japanese Sumida River FireworkFestival. It is a firework festival with a long history, being a successorto the “Ryogoku Kawabiraki Firework” festival that began in 1733, andit is a signature Tokyo summer event enjoyed by many Japanese people.First we detect stay points of GPS trajectory of mobile phone users todetermine if user attended the Japanese Sumida River Firework Festivalin Taito Ward of Tokyo during the period from 19:00 to 21:00 on July28th, 2012. Then we extract 1319 users from raw GPS log dataset. AsFig.8(a) and Fig.8(b) show, we can clearly observe that mobile phoneusers who attend Firework Festival gathered around Sumida River at20:00 PM, on July 28th, 2012. And people had dispersed to leave fromthe Sumida River Firework Festival at 23:00 PM. Furthermore, weconducted trajectory-based epidemic simulation for these people whoattended the Firework Festival on July 28th. We randomly select 10people as initial infected persons, as Fig.8(c) shows, OD from their laststay point to Sumida River of overall crowd who attended the FireworkFestival, red OD lines among them denote that initial infected personsattend this festival. The epidemic result can be shown in Fig.9, the curveFig. 9: Epidemic simulation on 1319 mobile phone users who attendedSumida River Firework Festival on July 28th, 2012.of number of accumulative infection Intuitively proves that activitiesof large public gatherings significantly increased the risk of infection.Increasing infected number per hour on July 28th obviously rises from 17:00 PM when the crowd started to gather, and the increasing numberdeclines as the Firework Festival ended. And we can find the spread ofinfection by the crowds after the event whose OD from Sumida river tonext stay points can spread the virus to other places in Fig.8(d). Theseanalysis results prove our proposed trajectory-based simulation cantruthfully reveal the cluster spread of infectious diseases, and dangerousconsequences of the spread of viruses caused by large public events.
This case is aimed to demonstrate the effectiveness of the spread ofinfectious disease under different control policy based on proposedtrajectory-based epidemic simulation. We apply 11985 mobile users’trajectory of one month (from July 1st to July 31st, 2012) as experi-mental GPS trajectory dataset. In order to ensure the validity of thesystem solutions of epidemic simulation under different control policy,we applied MCMC (Markov Chain Monte Carlo) to conduct repeatrandom sampling for initial random 10 infected people, and we alsodrew the 95% CI (confidence interval) for the simulation results inthe solution view. Noted that we applied parallel computing for theabove sampling to ensure computational efficiency. Finally, all of thesimulation results will be reviewed by domain experts.
Detection.
The experts can easily find the location of desired detec-tion point according to the distribution of various POIs on the map bydetection view (Interactive Detection Station Placement) of EpiMob.The experts set up detection points in the clusters of entertainment,supermarket and public place, respectively, and compare their effec-tiveness of the spread of infectious diseases. As a result, the expertsalso can easily acquire the result of epidemic simulation (Fig.10-A2)under different distribution of detection points (Fig.10-A1). We canfind that setting up temperature detection point in public places don’thave obvious effects than entertainment and supermarket.
Telecommuting.
Also, the experts can easily view heat maps of work-places in various cities or prefectures in Greater Tokyo area by telecom-muting view (Workplace Distribution) of system. As observed in Fig.10-B Solution View, the experts acquired the solution for implementingtelecommuting at different telecommuting ratios in Tokyo. And theexperts can conclude that telecommuting policy is very effective in theearly stages of the spread of infectious disease, and higher telecommut-ing ratio is more effective, which is in line with the true expectations ofthis policy. However, in the middle and late stages of the spreading, theproportion of telecommuting has little effect on the final results.
Region Lockdown.
With the help of region-lockdown view, whichprovides the traffic flow information of the city including flow deltasub-view and OD-analysis sub-view, the experts can easily find thepotential region-lockdown area by these sufficient prior knowledge. Asobserved in Fig.10-C Solution View, the results of epidemic simulationunder region lockdown also coincides with the projections of the expert.The sooner the closure policy is implemented, the better the spread ofthe epidemic can be effectively controlled.
Multi-policy Comparison.
The experts finally conduct a multi-policycomparison that not only a certain one policy is implemented and ana-lyzed, in order to find a relatively optimal solution to control the spreadof epidemic. As compared in Fig.10-D Solution View, the expertscompare five strategies under different control policy or their combina-tion. These control policy are telecommuting (From July 8th, 90% oftelecommuting ratio), region lockdown (From July 8th), the combina-tion of telecommuting (From July 8th, 90% of telecommuting ratio) andregion lockdown (From July 8th), the combination of telecommuting(From July 8th, 90% of telecommuting ratio) and detection (Set updetection point in the cluster of entertainment), detection (Set up detec-tion points in the cluster of entertainment), respectively. According tothese infectious results of five strategies, the experts think it’s a veryinteresting and reasonable finding that the detection in the cluster ofentertainment works better than any other.
Epidemic simulation and control policy analysis is a multidisciplinaryresearch problem that involves immunology, computation engineering, ntertainment Supermarket & Shopping mall
A1 A2 C
Public Space
B D
Fig. 10: The comparison of epidemic results under different control policies: (A) various detection strategies; (B) different telecommutingpolicies. (C) different lockdown policies; (D) a series policies with different restriction type. Especially, the outline of each accumulative infectioncurve under different policy represents 95% confidence interval.and urban computing. Therefore, we ask three experts in those corre-sponding domains to evaluate our EpiMob system. Specifically, thefirst (EA) is an expert in pathogenic microbiology and immunology,the second (EB) is an expert of high performance computing, and thethird (EC) is a senior researcher in urban computing.
Reliability of Epidemic Simulation and Policy Evaluation.
EA con-firmed that our trajectory-based epidemic model is based on a modifiedSEIR equation with dynamic susceptible and exposed population asvariables proposed in [52]. EA commented, “this epidemic modelutilizes real GPS record data to conduct epidemic simulation, whichcan be used for small sites, such as homes, workplaces, and a specificgathering. Based on this, the corresponding policies and regulationscan be formulated, and the action tracking (individual trajectory) can beused to set up detection points, remote offices, and city blockades. Andthis system allows for an intuitive preview and a comparative analysisof the selected strategies. Moreover, the system can assess the impactof a combination of multiple strategies (i.e. detection, telecommutingand region lockdown) on the spread of infectious diseases.” However,he mentioned that it would be better if this system could further unsealand predict the likely risks of travel.
Visual Design and Usability.
The experts confirmed that our EpiMobsystem has a clear and friendly UI, which provides rich interactionsto conduct reasonable settings for epidemic simulation. EB praisedthe easy interactions, and thought that it’s easy for normal people tounderstand the spread of infectious disease and the government controlpolicy. At the same time, EB confirmed the strategy including par-allel computing and MCMC (Markov Chain Monte Carlo) samplingimproved the computation efficiency, and could bring a credible re-sult. However, he mentioned that this system need apply distributedcomputation framework to simulate more population in the future.
Rationality of Restricted Human Mobility Model.
EC confirmedthat human mobility is highly related to the spread of epidemic, andtrajectory-based epidemic model is a very good and interesting applica-tion by building connection between human mobility and epidemics.EC commented, “the replacement-based restricted mobility model isvery promising in simulating the restricted human mobility under differ-ent policies. This is a relatively underexplored but highly challenging direction in the urban computing community. In particular, publicpolicy is quite complex and multifaceted, so visual interaction for hu-man mobility simulation under public policy is very necessary andreasonable. Their interaction design for human mobility and epidemicsimulation complements the current most human mobility studies inthe sense of receiving more complex and in-time input from users aswell as visualizing key information for decision making.”
ONCLUSION
In this study, we design an interactive visual analytics system calledEpiMob to effectively measure and evaluate different human mobil-ity restrictions (i.e., detection, telecommuting, region lockdown) forepidemic control. First, a novel trajectory-based SEIR model is pro-posed to simulate the epidemic spreading based on real-world humanmobility data and city POI data. Then we design a “trajectory replace-ment” strategy to generate a new set of human trajectories accordingto the mobility restrictions. The new generated trajectories will befed into our trajectory-based SEIR model to trigger a new round ofsimulation. Through EpiMob, user can easily select one policy ora combination of policies as the simulation target and set the spa-tiotemporal settings as well as the epidemic parameter settings withinteractive visual assistance. By employing the advanced visualizationtechniques, those simulation results could be confirmed and comparedin a well-organized, user-friendly, and highly-informative layout. Thefunctionality and usability of our system are validated through multiplecase studies and domain-expert interviews. In the future, we will con-tinue to improve our system from the following aspects: (1) integratingmore epidemic control policies and the interactive parameter settings;(2) enhancing the trajectory processing capability for bigger trajectorydata; (3) modifying the user interface for better user experience. A CKNOWLEDGMENTS
This work was partially supported by Leading Initiative for ExcellentYoung Researchers (LEADER) Program and Grant in-Aid for Scien-tific Research B (17H01784) of Japans Ministry of Education, Culture,Sports, Science, and Technology(MEXT); and JST, Strategic Interna-tional Collaborative Research Program (SICORP).
EFERENCES [1] https://en.wikipedia.org/wiki/Compartmental_models_in_epidemiology .[2] https://qianxi.baidu.com/ .[3] https://joras.csis.u-tokyo.ac.jp/dataset/show/id/14000201400 .[4] G. Andrienko, N. Andrienko, C. Hurter, S. Rinzivillo, and S. Wrobel.Scalable analysis of movement data for extracting and exploring signifi-cant places.
IEEE transactions on visualization and computer graphics ,19(7):1078–1094, 2012.[5] N. Andrienko and G. Andrienko. Visual analytics of movement: Anoverview of methods, tools and procedures.
Information Visualization ,12(1):3–24, 2013.[6] M. Baratchi, N. Meratnia, P. J. Havinga, A. K. Skidmore, and B. A. Tox-opeus. A hierarchical hidden semi-markov model for modeling mobilitydata. In
Proceedings of the 2014 ACM International Joint Conference onPervasive and Ubiquitous Computing , pp. 401–412, 2014.[7] M. Behrisch, L. Bieker, J. Erdmann, and D. Krajzewicz. Sumo–simulationof urban mobility: an overview. In
Proceedings of SIMUL 2011, The ThirdInternational Conference on Advances in System Simulation . ThinkMind,2011.[8] T. Brinkhoff. A framework for generating network-based moving objects.
GeoInformatica , 6(2):153–180, 2002.[9] I. Brodsky. H3: Ubers hexagonal hierarchical spatial index, 2018.[10] F. Chelaru, L. Smith, N. Goldstein, and H. C. Bravo. Epiviz: interactivevisual analytics for functional genomics data.
Nature methods , 11(9):938–940, 2014.[11] L. Chen, M. T. ¨Ozsu, and V. Oria. Robust and fast similarity search formoving object trajectories. In
Proceedings of the 2005 ACM SIGMODinternational conference on Management of data , pp. 491–502, 2005.[12] S. Chen, S. Li, S. Chen, and X. Yuan. R-map: A map metaphor for visual-izing information reposting process in social media.
IEEE Transactionson Visualization and Computer Graphics , 26(1):1204–1214, 2019.[13] S. Chen, X. Yuan, Z. Wang, C. Guo, J. Liang, Z. Wang, X. L. Zhang,and J. Zhang. Interactive visual discovering of movement patterns fromsparsely sampled geo-tagged social media data.
IEEE transactions onvisualization and computer graphics , 22(1):270–279, 2015.[14] W. Chen, F. Guo, and F.-Y. Wang. A survey of traffic data visualization.
IEEE Transactions on Intelligent Transportation Systems , 16(6):2970–2984, 2015.[15] M. Chinazzi, J. T. Davis, M. Ajelli, C. Gioannini, M. Litvinova, S. Merler,A. P. y Piontti, K. Mu, L. Rossi, K. Sun, et al. The effect of travel restric-tions on the spread of the 2019 novel coronavirus (covid-19) outbreak.
Science , 2020.[16] Z. Deng, D. Weng, J. Chen, R. Liu, Z. Wang, J. Bao, Y. Zheng, and Y. Wu.Airvis: Visual analytics of air pollution propagation.
IEEE transactionson visualization and computer graphics , 26(1):800–810, 2019.[17] C. Dunne, M. Muller, N. Perra, and M. Martino. Vorograph: visualizationtools for epidemic analysis. In
Proceedings of the 33rd Annual ACMConference Extended Abstracts on Human Factors in Computing Systems ,pp. 255–258, 2015.[18] N. Ferreira, J. Poco, H. T. Vo, J. Freire, and C. T. Silva. Visual explorationof big spatio-temporal urban data: A study of new york city taxi trips.
IEEEtransactions on visualization and computer graphics , 19(12):2149–2158,2013.[19] W.-j. Guan, Z.-y. Ni, Y. Hu, W.-h. Liang, C.-q. Ou, J.-x. He, L. Liu,H. Shan, C.-l. Lei, D. S. Hui, et al. Clinical characteristics of 2019 novelcoronavirus infection in china.
MedRxiv , 2020.[20] D. Guo. Visual analytics of spatial interaction patterns for pandemicdecision support.
International Journal of Geographical InformationScience , 21(8):859–877, 2007.[21] H. He, R. Li, R. Wang, J. Bao, Y. Zheng, and T. Li. Efficient suspectedinfected crowds detection based on spatio-temporal trajectories. arXivpreprint arXiv:2004.06653 , 2020.[22] A. Horni, K. Nagel, and K. W. Axhausen.
The multi-agent transportsimulation MATSim . Ubiquity Press London, 2016.[23] X. Huang, Y. Zhao, C. Ma, J. Yang, X. Ye, and C. Zhang. Trajgraph: Agraph-based visual analytics approach to studying urban network central-ities using taxi trajectory data.
IEEE transactions on visualization andcomputer graphics , 22(1):160–169, 2015.[24] H. Jeung, M. L. Yiu, X. Zhou, C. S. Jensen, and H. T. Shen. Discovery ofconvoys in trajectory databases.
Proceedings of the VLDB Endowment , 1(1):1068–1080, 2008.[25] R. Jiang, X. Song, Z. Fan, T. Xia, Q. Chen, Q. Chen, and R. Shibasaki.Deep roi-based modeling for urban human mobility prediction.
Pro-ceedings of the ACM on Interactive, Mobile, Wearable and UbiquitousTechnologies , 2(1):1–29, 2018.[26] R. Jiang, X. Song, Z. Fan, T. Xia, Q. Chen, S. Miyazawa, and R. Shibasaki.Deepurbanmomentum: An online deep-learning system for short-termurban mobility prediction. In
AAAI , pp. 784–791, 2018.[27] X. Kang, L. Liu, D. Zhao, and H. Ma. Trag: A trajectory generationtechnique for simulating urban crowd mobility.
IEEE Transactions onIndustrial Informatics , 2020.[28] Z. Li, B. Ding, J. Han, and R. Kays. Swarm: Mining relaxed temporalmoving object clusters.
Proceedings of the VLDB Endowment , 3(1-2):723–734, 2010.[29] C. Liu, C. Wu, H. Shao, and X. Yuan. Smartcube: An adaptive datamanagement architecture for the real-time visualization of spatiotemporaldatasets.
IEEE transactions on visualization and computer graphics ,26(1):790–799, 2019.[30] D. Liu, D. Weng, Y. Li, J. Bao, Y. Zheng, H. Qu, and Y. Wu. Smartadp:Visual analytics of large-scale taxi trajectories for selecting billboardlocations.
IEEE transactions on visualization and computer graphics ,23(1):1–10, 2016.[31] H. Liu, Y. Gao, L. Lu, S. Liu, H. Qu, and L. M. Ni. Visual analysis ofroute diversity. In , pp. 171–180. IEEE, 2011.[32] S. Liu, W. Cui, Y. Wu, and M. Liu. A survey on information visualization:recent advances and challenges.
The Visual Computer , 30(12):1373–1393,2014.[33] M. F. Mokbel, L. Alarabi, J. Bao, A. Eldawy, A. Magdy, M. Sarwat,E. Waytas, and S. Yackel. Mntg: An extensible web-based traffic generator.In
International Symposium on Spatial and Temporal Databases , pp. 38–55. Springer, 2013.[34] K. Ouyang, R. Shokri, D. S. Rosenblum, and W. Yang. A non-parametricgenerative model for human trajectories. In
IJCAI , pp. 3812–3817, 2018.[35] L. Pappalardo, F. Simini, G. Barlacchi, and R. Pellungrini. scikit-mobility:a python library for the analysis, generation and risk assessment of mobilitydata, 2019.[36] W. Pei, Y. Wu, S. Wang, L. Xiao, H. Jiang, and A. Qayoom. Bvis:urban traffic visual analysis based on bus sparse trajectories.
Journal ofVisualization , 21(5):873–883, 2018.[37] H. Qu, W.-Y. Chan, A. Xu, K.-L. Chung, K.-H. Lau, and P. Guo. Visualanalysis of the air pollution problem in hong kong.
IEEE Transactions onvisualization and Computer Graphics , 13(6):1408–1415, 2007.[38] J. M. Read, J. R. Bridgen, D. A. Cummings, A. Ho, and C. P. Jewell. Novelcoronavirus 2019-ncov: early estimation of epidemiological parametersand epidemic predictions.
MedRxiv , 2020.[39] R. Scheepens, C. Hurter, H. Van De Wetering, and J. J. Van Wijk. Visu-alization, selection, and analysis of traffic flows.
IEEE transactions onvisualization and computer graphics , 22(1):379–388, 2015.[40] X. Song, Q. Zhang, Y. Sekimoto, R. Shibasaki, N. J. Yuan, and X. Xie. Asimulator of human emergency mobility following disasters: Knowledgetransfer from big disaster data. In
Twenty-Ninth AAAI Conference onArtificial Intelligence , 2015.[41] G. Sun, R. Liang, H. Qu, and Y. Wu. Embedding spatio-temporal informa-tion into maps by route-zooming.
IEEE transactions on visualization andcomputer graphics , 23(5):1506–1519, 2016.[42] H. Tian, Y. Liu, Y. Li, C.-H. Wu, B. Chen, M. U. Kraemer, B. Li, J. Cai,B. Xu, Q. Yang, et al. An investigation of transmission control measuresduring the first 50 days of the covid-19 epidemic in china.
Science , 2020.[43] C. Tominski, H. Schumann, G. Andrienko, and N. Andrienko. Stacking-based visualization of trajectory attribute data.
IEEE Transactions onvisualization and Computer Graphics , 18(12):2565–2574, 2012.[44] T. Von Landesberger, F. Brodkorb, P. Roskosch, N. Andrienko, G. An-drienko, and A. Kerren. Mobilitygraphs: Visual analysis of mass mobilitydynamics via spatio-temporal graphs and clustering.
IEEE transactionson visualization and computer graphics , 22(1):11–20, 2015.[45] Z. Wang, M. Lu, X. Yuan, J. Zhang, and H. Van De Wetering. Visual trafficjam analysis based on trajectory data.
IEEE transactions on visualizationand computer graphics , 19(12):2159–2168, 2013.[46] D. Weng, R. Chen, Z. Deng, F. Wu, J. Chen, and Y. Wu. Srvis: Towardsbetter spatial integration in ranking visualization.
IEEE transactions onvisualization and computer graphics , 25(1):459–469, 2018.[47] D. Weng, H. Zhu, J. Bao, Y. Zheng, and Y. Wu. Homefinder revisited:inding ideal homes with reachability-centric multi-criteria decision mak-ing. In
Proceedings of the 2018 CHI Conference on Human Factors inComputing Systems , pp. 1–12, 2018.[48] A. Witayangkurn, T. Horanont, M. Nagai, and R. Shibasaki. Large scalemobility analysis: Extracting significant places using hadoop/hive and spa-tial processing. In
International Conference on Knowledge, Information,and Creativity Support Systems , pp. 205–219. Springer, 2015.[49] J. T. Wu, K. Leung, and G. M. Leung. Nowcasting and forecasting thepotential domestic and international spread of the 2019-ncov outbreak orig-inating in wuhan, china: a modelling study.
The Lancet , 395(10225):689–697, 2020.[50] W. Wu, J. Xu, H. Zeng, Y. Zheng, H. Qu, B. Ni, M. Yuan, and L. M. Ni.Telcovis: Visual exploration of co-occurrence in urban human mobilitybased on telco data.
IEEE transactions on visualization and computergraphics , 22(1):935–944, 2015.[51] Y. Yang, T. Dwyer, S. Goodwin, and K. Marriott. Many-to-manygeographically-embedded flow visualisation: An evaluation.
IEEE trans-actions on visualization and computer graphics , 23(1):411–420, 2016.[52] Z. Yang, Z. Zeng, K. Wang, S.-S. Wong, W. Liang, M. Zanin, P. Liu,X. Cao, Z. Gao, Z. Mai, et al. Modified seir and ai prediction of theepidemics trend of covid-19 in china under public health interventions.
Journal of Thoracic Disease , 12(3):165, 2020.[53] M. Yin, M. Sheehan, S. Feygin, J.-F. Paiement, and A. Pozdnoukhov. Agenerative model of urban activities from cellular data.
IEEE Transactionson Intelligent Transportation Systems , 19(6):1682–1696, 2017.[54] J. Zhang, Y. Zheng, and D. Qi. Deep spatio-temporal residual networksfor citywide crowd flows prediction. In
AAAI , pp. 1655–1661, 2017.[55] K. Zheng, Y. Zheng, N. J. Yuan, S. Shang, and X. Zhou. Online discoveryof gathering patterns over trajectories.
IEEE Transactions on Knowledgeand Data Engineering , 26(8):1974–1988, 2013.[56] Y. Zheng, W. Wu, Y. Chen, H. Qu, and L. M. Ni. Visual analytics in urbancomputing: An overview.
IEEE Transactions on Big Data , 2(3):276–296,2016.[57] Z. Zhou, L. Meng, C. Tang, Y. Zhao, Z. Guo, M. Hu, and W. Chen. Visualabstraction of large scale geospatial origin-destination movement data.
IEEE transactions on visualization and computer graphics , 25(1):43–53,2018.[58] M. Zhu, W. Chen, J. Xia, Y. Ma, Y. Zhang, Y. Luo, Z. Huang, and L. Liu.Location2vec: a situation-aware representation for visual exploration ofurban locations.