The whole world is now fighting the coronavirus (COVID-19). Social/physical distancing and limiting travel are effective approaches to contain the virus. Everyone’s effort counts. By analyzing world population flows through the lens of geotagged Twitter data during the COVID-19 Pandemic, this article (story map) showcases how our collective efforts of fighting the virus are reflected on maps and how big social media data can be used for such analyses.
Figure 1 illustrates the contiguous US population flows from March 12, 2020 to March 24, 2020. It’s clear that the intensity of population movement declines dramatically from March 12 to March 24. Note that only tweets geotagged at the place level (e.g., Columbia, SC) are used in the maps for privacy concerns.
Figure 2 shows the world population flows from March 12, 2020 to March 24, 2020. Similar to the US, the intensity of population movement declines over the past days, though the decline is not as clear as in the U.S. due to the small scale maps.
Note that the users on each map only include those who posted tweets on the two consecutive days of that map. We observed a clear increase of the twitter users for both US and world, which is likely due to the fact that more people tweet during the crisis.
Let’s look at the movement from another perspective by calculating the average cross-day travel distance (in mile) of all user samples for each day (details about the cross-day distance calculation can be found here). Figure 3 reveals that average travel distance for both US (right) and the world (left) show similar patterns with two exceptions: 1) the dramatic decline in the U.S. occurred around March 12 which is several days later than the whole world; and 2) there is no clear increase in travel distance is observed in July in the U.S.
By comparing the population flows of a region during the pandemic to a baseline flow (e.g., the same week of the year before the pandemic), the impact can be revealed. Figure 4 and Figure 5 show the comparison of the population flows from 03/23 to 03/24 between 2019 and 2020 for Europe, Japan, and South Korea.
An interactive spatial web portal is being developed to allow users to generate dynamic mobility maps at varying spatial and temporal resolutions. Figure 6, screenshots of the portal, shows the county level average travel distance on March 7, 2020 (left map) and April 11, 2020 (right map). Both days are Saturdays and the color ramps on each map have the same value groups so they are comparable. This again shows a significant decline in travel distance in a majority of counties.
Also derived from the web portal, Figure 9, 10, and 11 show the temporal trend of average travel distance for selected geographic areas at different geographic levels including county (Figure 7), state (Figure 8), and country (Figure 9) from January 1, 2020 to April 29, 2020. Travel distance drops are observed for all selected counties and states in the U.S. starting from around March 10.
To analyze the decline in a more quantitative way, Figure 10 shows the daily mobility change at the state level by comparing to the baseline mobility in January and February of 2020. The y-axis is the ratio (R) between each day’s average movement distance and the baseline distance. ( R > 1.0 indicates increase in average movement distance; R < 1.0 indicates decrease in average movement distance; R =1.0 indicates no change;).
This study is significant not only because it shows the efforts of limiting our travels to contain the virus, but also because monitoring population flows within a region or between different places could help us gain better insights into the current and future infectious risk, as human movement is an important driver of geographic spread of COVID-19. Figure 11 shows the outgoing population flows from Italy to other parts of the world between 03/01/2020 and 03/11/2020.
Our next step is to conduct further quantitative analysis and modeling to examine, for example, the aggregated county or community level population movement analysis, the compliance of the social distancing measures, and the effectiveness of the control measures in containing the spread of the virus. The analysis will be conducted at different geographic scales including country, state, and county.
All the population flow maps are captured from a web portal providing near real-time movement using twitter data (Figure 13). Try the interactive app here.
The aggregated human mobility maps and time series are generated from another interactive spatial web portal (Figure 15). This web portal is currently being developed and tested.
Updated on June 10 2020