Quantifying Physical Distancing

Physical distancing is key to avoid or slow down the spread of viruses. Each country has taken different policies and actions to restrict human mobility. In this project, we investigate how policies and actions affect human mobility in certain cities and countries. By referencing our analysis of policies and secondary impacts, we hope that decision makers can make effective and appropriate actions at suitable timings. Furthermore, by analyzing human mobility, we also aim to develop a physical distancing risk index to monitor the risk on areas with high population density and probability of contraction.

Research Topics

1. Understanding Global Mobility Using Flight Data More details

Each country has implemented different policies to restrict travel. In order to understand global mobility, we provide a real-time dashboard that visualizes travel restrictions in each country.

2. Changes in Mobility (Tokyo) More details

We explored human mobility via mobile phone data as an early indicator of confirmed cases. We analyzed NTT DoCoMo data and accessed high-resolution hourly population data (Mobile Kukan Toukei™) within Tokyo.

3. Changes in Mobility (New York City) More details

We analyze mobility changes in New York City - one of the most densely populated cities in the world. After the announcement of the lockdown on March 13th, our analysis of the NYC open data set clearly shows traffic dramatically decreased, and how there was a surge in citizens who relied on the bike sharing system after the city council recommend people to refrain from using public transportation. 

4. Mobility Changes in Barcelona More details

We analyze mobility changes such as bike sharing systems and road traffic in Barcelona - the capital of the autonomous region of Catalonia. Changes in mobility are noticeable from March 13th when schools were suspended, indicating that the local population did not alter their mobility patterns in response to milder governmental actions. After the national lockdown from March 14th, the change was drastic, as the population displayed a strong adherence to the established restrictions. 

5. Physical Distancing Posts on Instagram More details

As countries and cities restricted activities outside, social networking service users in affected areas began utilizing their SNS channels to disperse mass amounts of information regarding their socially distanced conditions. We analyze the increase in hashtags related to COVID-19.

Topic 1. Understanding Global Mobility using Flight Data

Contributors: Hiroki Naganuma and Euma Ishii

To avoid or slow the spread of viruses, each country has taken different policies in restricting travel. In order to understand global mobility, we provide a real-time dashboard that visualizes travel restrictions in each country, notice to airmen (NOTAM) messages that are issued from airports, and confirmed infection cases in each country.  

We aim to develop a country-level and airport-level physical distancing metric. . We plan to build a temporal network where a country or an airport is represented as a vertex and a connection between 2 countries or 2 airports is represented as an edge.  By building such a temporal network and compute shortest paths and their length between 2 countries or 2 airports, we can measure how travels are restricted in a quantitative manner by using graph analytics.


Topic 2. Mobility Changes in Tokyo

Contributors: Kunihiko Miyoshi and Hiroshi Maruyama

One of the difficulties in dealing with COVID-19 is the approximate 12-day delay between an infection and its registration in the official statistics of confirmed cases. We explored human mobility via mobile phone data as an early indicator of confirmed cases. We analyzed NTT DoCoMo data and accessed high-resolution hourly population data (Mobile Kukan Toukei™) within Tokyo. Our preliminary analysis suggests that the  mobility level correlates with the daily growth rate of reported confirmed cases within the following 12 days. We are studying the detailed structure of this correlation by looking at demography, time, and area.

The data set divides Tokyo into 8,500 grid cells of 0.5km x 0.5km. The provided data is a collection of population vectors Pt where Pt[i] is the population of the grid cell i at time t (t is an hourly time point between 0:00 on January 1st and 23:00 on March 31st). We defined the overall mobility within Tokyo at time t as L1(Pt - Pt+1) where L1 is the L1 norm. Intuitively, this metric counts the sum of the number of people who came into or left each cell during the given hour. Note that this metric underestimates actual mobility since incoming and outgoing people within an hour cancel each other out.
Fig. 1 shows the daily mobility index and the growth rate of reported confirmed cases. The colored arrows show the correspondence between them, which suggests that the mobility index may be an early indicator of confirmed cases.

Fig. 1. (a) Daily mobility index of Tokyo in February and March 2020. (b) The daily growth rate of reported confirmed cases in Tokyo Metropolis. The blue arrows show the drop of mobility around March 2nd may be correlated to the drop of growth rate on March 14th. The red arrows show the pick up of mobility on March 18th may be correlated to the peak of the growth rate on March 29th.  

The mobility index above was for overall Tokyo Metropolis area as defined by the local government, which include large rural areas that may less contribute to COVID-19 transmission. To study the structure of the mobility, we clustered ~8,500 grid cells into 6 groups based on hour-of-day population patterns (each grid cell is represented as a 24-variable vector).  The results are shown in Fig. 2.

Fig. 2. (a) Hourly population of each cluster. Cluster 3 (magenta) represents night districts (e.g, Shinjuku and Shibuya) while Cluster 2 (purple) are business districts (e.g., Otemachi). (b) The clusters shown on a map.
Different clusters show different responses in mobility to COVID-19. Fig. 3 shows how mobility changes over time depending on the cluster.

Fig. 3. (a) Mobility timeseries for each cluster. We took three reference time points, February 4th (baseline), March 3rd (school closure begins), and March 31st (lockdown is imminent) for comparison. (b) Mobility reduction rate for each cluster. The first set of bars shows reduction from February 4th to March 3rd, and the second set shows reduction from February 4th to March 31st. The night districts show the largest drop of mobility, up to 35%.

Latest Mobility Update

We will further investigate demographic and time subcomponents of mobility to refine our model as an early indicator of reported cases. The following image is developed by Mitsuha Miyake.

Data Provider

Mobile Kukan Toukei™,  NTT DOCOMO, INC.
“Mobile Kukan Toukei” is a trademark of NTT DOCOMO, INC.
NTT DOCOMO’s “Mobile Kukan Toukei” services are only available to subscribers in Japan.

Topic 3. Changes in Mobility (New York City)  

Contributors: Hiroki Kanezashi

Here, we analyzed mobility changes in New York City - one of the most densely populated cities in the world. After the announcement of the lockdown on March 13th, our analysis of the NYC open data set clearly shows traffic dramatically decreased, and how there was a surge in citizens who relied on the bike sharing system after the city council recommend people to refrain from using public transportation.

Road Traffic in New York City

After the first confirmed case of COVID-19 in New York City on March 1st, freeway traffic drastically decreased. The overall daily average traffic speed in NYC in 2020 is higher than that of 2019, and the average travel time on the freeway became significantly shorter after March 13th in comparison to previous years. Similar trends are found in Manhattan, the Bronx, and Queens. In Brooklyn, the roads became vacant after March 5th. In Staten Island, the traffic speed is higher and the time spent on the freeway has become overall longer than that of 2019.
We extracted the daily average travel time and speed in 20 day ranges. The x-axis of the graph shows the number of days from the first Sunday of March (Blue lines: 2019 3/3 - 3/21, Orange lines: 2020 3/1 - 3/19). We observed 3 representative locations within New York City: Lincoln Tunnel, Robert F. Kennedy Bridge, and Brooklyn Bridge. There overall was a significant decrease in traffic volumes especially from March 11st.

Figure 1: Examples of traffic volume changes in NYC freeways

The overall trend of NYC freeways has changed after 14th. Figure 2 shows the overall average of the daily average speed in NYC freeways. The average traffic speed before March 13th was relatively low (40 mph or less) and frequently changes by date, but it increased to around 45 mph after March 14th and became more steady.

Average Traffic Speed of NYC Freeways

Figure 2: Overall average traffic speed (mph) in NYC freeways

Bike Sharing System in New York City

After the NYC municipality recommended citizens to ride bikes instead of using public transportation, there was a surge in the usage of Citi Bike, a privately owned public bicycle sharing system in New York City. By measuring how many bikes are used on an hourly and daily basis, we estimate the number of people are outdoors during the lockdown.
For this analysis, we track the number of people using bikes at each bike station every minute. Since we can only track the number of available bikes in each station, we estimate the number of departed bikes by computing the difference in the number of available bikes between two timestamps. We developed an interactive visualization dashboard that illustrates how bikes are used over time since March 23th. We plan to incorporate additional information such as changes in the number of infection cases, weather, and policy changes.

Figure 3: Interactive visualization dashboard of the number of used bikes.

The usage of Citi Bike varies by day and time of day (Figure 4). The hourly usage of overall Citi Bike usages at night is around 100, but thousands of people used them around 3 PM. Each daily peak of the bike usage is completely different by day as well. On April 4th, about 4,000 people used bikes in the peak time while only about 1,000 people used on March 28th. The main factor of the difference is probably climate and weather rather than day of week.

Number of bikes departed from CitiBike stations per hour

Figure 4: Total hourly number of departed bikes from all Citi Bike stations in NYC

Topic 4. Changes in Mobility (Barcelona)

Contributors: Sergio Alvarez Napagao, Dario Garcia-Gasulla, Raquel Pérez-Arnal, Dmitry Gnatyshak, Anna Arias Duart

The first detected case of COVID-19 within Spain was on January 31st in the Canary Islands, located more than 1,000 km from peninsular Spain. By late February, imported cases were detected in the mainland, and on February 26th the first endemic case was diagnosed. On March 9th, 999 cases were diagnosed and certain regions in Spain started implementing local restriction policies. By March 13th, cases had been detected across all 50 provinces. The following day, March 14th, the Spanish Government announced a state of emergency, and implemented a lockdown for the whole population. Citizens were only permitted to travel for work, and all social events were prohibited. This lockdown was reinforced on March 29th with total mobility restrictions, and only essential services were an exception.
Awareness and interest in COVID-19 is also noticeable by the increased coverage of the disease on the frontpage of “El Pais” newspaper, the most read national newspaper in Spain.

The first restriction against COVID-19 by the regional government of Barcelona was on March 11th that informed citizens to avoid gatherings of more than 1,000 people. Two days later, on March 13th, with 508 confirmed cases in the region, all classes were suspended. By March 14th, a national lockdown was declared by the Spanish government. The effects on mobility are only visible from March 13th, indicating that the local population did not alter their mobility patterns in response to earlier and milder governmental restrictions.

Public Bike Service in Barcelona

Similar to New York City, public bike sharing is available in Barcelona. By analyzing the availability of docking stations throughout the city, we measure the changes in population mobility. The first measures affecting Barcelona were implemented on March 13th, after which the availability of stations was 89% on average when compared to the previous week. On March 14th, availability was down to 62% in comparison to the previous week. March 15th saw minimal activity, and bike services were indefinitely suspended on March 16th.

Road Traffic in Barcelona

We analyzed road traffic data from March 2019, and found that traffic mobility (fluid) increased on the weekends, especially on Sundays.

The same data source showed minor changes in road congestion until March 13th, and then a dramatic change after the national lockdown was announced on March 14th.


Topic 5. Posts Regarding Physical Distancing on Instagram

Contributors: Keita Suzuki, Hiroki Kanezashi, and Naomi Nakagawa

As more and more countries and cities restrict outdoor activities, users in affected areas have begun utilizing their social networking channels to disperse mass amounts of information regarding situations regarding physical distancing.

Users have utilized them as an effective way to increase their interactions on the platform and raise the level of engagement and interactions that occur on their account pages. A post on Instagram would be summarized and categorized by hashtags on the post. Since March 2020, specific hashtags that represent physical distancing have increased significantly; #stayhome, #stayathome, #socialdistancing, #workfromhome, #zoom (online meeting software).

As of April 11, 2020, more than 16 million posts with #stayhome have uploaded on Instagram, including 6 million posts in March 2020. During mid March 2020, the number of posts with #stayhome gradually rose up, with states across the U.S. announcing a stay-at-home order. On March 24, 2020, Instagram announced that it launched a “Stay Home” sticker to help those practicing physical distancing connect with others. This might have also boosted the number in March 2020.

#Hourly Posts with #stayhome (2020/03/01 - 2020/04/12)

Hourly Instagram posts with hashtag "stayhome"

We have analyzed #socialdistancing posted with positional information among 292,163 #socialdistancing from April 3, 3 p.m. to April 5, 3 p.m. (UTC). Top countries tagged with the hashtag in this period are the US, the UK, Canada, Indonesia, India, Australia, and Germany. Posts with the hashtag were posted during local daytime, mainly in late afternoon in each country.

Since some states and counties introduced work from home measures in response to the COVID 19 pandemic in mid March 2020, the number of posts with #zoom, online meeting software, has been increasing as the animated map video below. The hashtag tends to increase on weekdays rather than weekends.

The stock price of Zoom Video Communications, Inc. (ZM) at the Nasdaq rose up along with the number of posts on Instagram until the end of March. These facts might suggest that people started to work on Zoom on weekdays, which enhanced the value of Zoom in the stock market.