Changes in mobility patterns in Europe during the COVID-19 pandemic: Novel insights using open source data
Anna Sigridur Islind, María Óskarsdóttir, Harpa Steingrímsdóttir
CChanges in mobility patterns in Europe during theCOVID-19 pandemic: Novel insights using open source data
Anna Sigr´ıdur Islind (cid:89) Mar´ıa ´Oskarsd´ottir (cid:89) * Harpa Steingr´ımsd´ottir (cid:89) Department of Computer Science, Reykjav´ık University, Reykjav´ık, Iceland (cid:89)
These authors contributed equally to this work.* [email protected]
Abstract
The COVID-19 pandemic has changed the way we act, interact and move around in theworld. The pandemic triggered a worldwide health crisis that has been tackled using avariety of strategies across Europe. Whereas some countries have taken strict measures,others have avoided lock-downs altogether. In this paper, we report on findingsobtained by combining data from different publicly available sources in order to shedlight on the changes in mobility patterns in Europe during the pandemic. Using thatdata, we show that mobility patterns have changed in different counties depending onthe strategies they adopted during the pandemic. Our data shows that the majority ofEuropean citizens walked less during the lock-downs, and that, even though flights wereless frequent, driving increased drastically. In this paper, we focus on data for a numberof countries, for which we have also developed a dashboard that can be used by otherresearchers for further analyses. Our work shows the importance of granularity in opensource data and how such data can be used to shed light on the effects of the pandemic.
Introduction
In December 2019, a severe respiratory disease emerged in Whuan, China [1]. Since thefirst identified case of the virus COVID-19 was reported on 8. December 2019 [2, 3], thevirus has spread quickly, causing a worldwide health crisis. Since the crisis broke,different types of crisis management strategies have been actualised, impacting the waypeople in different countries act and interact in ways that have not yet fully beenrealised [4]. The pandemic has affected people living in various countries in differentways, since each country has had its own approach and policy for handling the crisissituation. This goes for testing strategies, requirements around quarantine, imposinglock-downs, social-distancing and closing of schools to name a few [5]. Additionally, asthe virus was spreading around the globe, countries were effected by it to a differentextent, and took longer or shorter time to respond to the imminent threat. In Europealone, the response to the pandemic has varied greatly. For example, although Spainand the United Kingdom both reported their first COVID-19 case on January 31, Spainimposed a lock-down nine days before the United Kingdom (43 and 52 days,respectively, after the first diagnosed case). In contrast, Norway and Denmark reportedtheir first cases on February 26 and 27, and waited only 15 days to significantly restrictthe travels to, from and within their countries by imposing a lock-down. Since thesemeasures were taken, it has become clear that non-pharmaceutical interventionssignificantly and substantially slowed the growth of the pandemic [6].Notice: This is a preprint of a work currently in submission. 1/12 a r X i v : . [ c s . C Y ] A ug revious studies have shown the benefits of developing and researching informationsystems with the aim of pursuing grand challenges such as natural disasters, publichealth issues, dealing with poverty, climate change and the ageing society [7–13].Following that stream of research on grand challenges, and in light of the new grandchallenge that we are now facing with the ongoing pandemic, we suggest that providingknowledge and developing information systems based on open source data can providenew insights on the pandemic and its effects. There is a related call for research on theaggregated flow of people during the pandemic in order to show the impact of differentmeasures taken during the state of crisis [14]. This paper directly contributes to thatcall, and that gap in the literature.In this paper we more specifically report on our findings from using open source datafrom different resources, including the Apple Maps Mobility Trend data set and theCOVID-19 Flight data set provided by OpenSky, for the purpose of developing adashboard which shows the changes in mobility patterns in Europe and analyse the datawith the purpose of shedding light on mobility trends following the pandemic. Ourresearch questions are two-fold: i) How has the COVID-19 pandemic changed mobilitypatterns in Europe; and ii) What conclusions can be drawn from open source dataabout the impact of the COVID-19 pandemic on well-being of individuals in Europe inrelation to strategy steps taken by the various countries. Using the data and ourdashboard, we show the impact of the different strategies put into place in Europeduring the pandemic and show how mobility patterns changed in different counties. Thefindings show how the majority of Europeans walked less during the lock-downs andsocial distancing periods, and that even though flights were less frequent, drivingincreased drastically instead. The main contribution of our research is firstly an analysisof mobility patterns in relation to measures taken to mitigate the effects of thepandemic in different countries in Europe and secondly an information system –thedashboard– with data and visualisations showing changes in mobility patterns inEurope. The dashboard will be made available online upon publication of this paper.Additionally, we illustrate how novel use of open source data can elucidate variousaspects of the pandemic which indicates the importance of granularity in open sourcedata, and the significance of updating open source data frequently so that it can beused to tackle grand challenges such as this one.The rest of this paper is organized in the following way. In the next section wediscuss the related literature on mobility, crisis management, predicting the spread ofthe pandemic and the effect of imposed restriction on general population, followed by anexplanation of our proposed methods and materials in Section . In Section we presentthe results of our analyses and discuss their implications in Section . We conclude thepaper with some final remarks and directions for future research in Section . Related Work
Human mobility has been a popular research topic for over a decade. With increasedavailability of granular, geo-located mobility data research has revealed that people tendto follow simple and reproducible patterns; they are likely to return to a few highlyfrequented locations and their trajectories are highly regular both in terms of time andspace [15]. Recently, there is a growing interest in understanding mobility patterns andsocial interactions in healthcare situations through a data-driven approach [16]. Suchdata has also been used to assess population displacement following disasters [17]. Inlight of that and with the COVID-19 disease spreading around the globe, mobilityresearch has gain new heights, whereby researchers, on the one hand, strive tounderstand and predict the spread of the virus using geo-located data, and, on the otherhand, to understand the effects of the pandemic on human mobility through the use ofNotice: This is a preprint of a work currently in submission. 2/12ata.The first line of research is focused on understanding and predicting how the virusspreads, which includes estimating its basic reproduction number R . Over the years,advanced epidemic and compartment models that rely on detailed mobility data havebeen developed to simulate and investigate epidemic spread. These and alternativetechniques have successfully been used for H1N1 as well as seasonal influenza. [18–20].With a new, unknown infectious disease, estimation of the basic reproduction numberoften outlines a first step, and furthermore to model the virus’s spread and effect ofmanagement strategies. Early studies of the spread of COVID-19 in China and itsorigin in Wuhan, indicated that the basic reproduction number could be as high as R = 6 .
47, and that with strict travel restrictions, the disease would peak in about 2weeks time [21]. In an effort to estimate the latent infection ratio in Wuhan before thelock-down, a study showed that a significant number of infected people had left the cityat that time, and gave an estimate of R = 3 .
24 [22]. Strict control measures provedeffective in China. The lock-down in Wuhan delayed the spread of the virus to othercities in China by almost 3 days, and cities that implemented restrictions preemptivelyreported fewer cases in the first weeks than cities that started control later [23]. Anadvanced epidemic simulation model with real-time mobility data from Wuhanfurthermore established that the drastic control measures substantially mitigated thespread of the virus [24]. Moreover, taking global mobility data into account, includingflights, showed that the travel ban in Wuhan only postponed the spread by 3-5 days inChina, but case importation of infections globally was reduced by 80% until midFebruary [25]. Researchers are currently also starting to use geo-located mobility datacoupled with census and demographic data in the USA to predict the effects of a secondwave of the pandemic, concluding that social distancing, high levels of testing,contact-tracing and household quarantine will be most effective in keeping the infectionrate low [26].The second viewpoint of using mobility data to understand the effects of thepandemic on human mobility, has been researched to a less extent. However, eventhough the current literature on the effects of COVID-19 on mobility patterns shows thedetection of dramatic changes in mobility in the wake of COVID-19, most of theresearch is focused on the changes within the USA [27], in Italy [28], and in China [29].Neither analysing more general mobility trends cross continents nor a general levelcomparison between countries have thereby been attempted to date.Bushfires, tornadoes, avalanches, floods and earthquakes are all non-routine eventsthat can be classified as unpredictable extreme situations which call for shifts fromroutine practice [30]. These types of situations, put unfamiliar demands and strains onboth individuals, and on policy-makers [17, 31]. What has characterised the situationsurrounding COVID-19 is that no-one was prepared [32]. As a result, we can see thesituation surrounding the COVID-19 pandemic as a crisis situation due to itsunpredictable non-routine nature where different countries have activated differentstrategies for handling the crisis. The crisis handling of COVID-19 has an impact onmobility patterns since a lock-down is bound to affect how people move about in thatparticular country in their day to day lives. The different countries have enforced andfollowed different rules and policies. When comparing the different strategies, theresearch to date has for instance showed the effects of the different reactions, and crisismanagement strategies. For instance, [33] investigate the effect of major inventions ineleven European countries from February to May. They show that interventions madeat the time were sufficient to drive basic reproduction number ( R ) below one and thusgain control of the pandemic. According to their study, between 3.2% to 4% of theinhabitants in those eleven countries had been infected by May 4. In their study, theyshow that lock-downs in particular had a large impact on reducing transmission.Notice: This is a preprint of a work currently in submission. 3/12he literature on the basic reproduction number and on which strategies haveproven most effective is growing and will continue to grow as more and more is knownabout the virus, its spread as well as short and long term effects. Only when the wholepandemic period is examined will the effects of the pandemic be clear. However, theliterature listed herein gives a view of the first wave of the strategies, spread andmobility during the first responds of crisis mode. Methods & Materials
In our study, we use only open source and publicly available data. As the pandemic wasspreading and to this day, several organisations are sharing aggregated data publicly,often with daily updates. Furthermore, individual counties share additional and specificdata and information for their citizens.To investigate changes in mobility, we considered walking, driving and flying asmeans of travel. For the first two, we use Apple Map Mobility Trends (MT) whichshow the changes in routing requests from day to day, since January 13. These trendsshow a considerable drop in requests in most countries as the virus started spreadingand governments took action, enforcing or requiring social distancing. Similarly, thenumber of requests is gradually increasing since June/July. For flights we use theOpenSky COVID-19 Flight Dataset . The OpenSky Network is a non-profit associationwhich provides open access to real-world air traffic control data for research [34]. Due tohigh demand, they created a data set with detailed information about all flights in 2019and 2020. It was made publicly available in April 2020 and is currently updatedmonthly. The data set is amongst others being used by the Bank of England tonow-cast economic indicators during the crisis [35]. To compare the flights data withthe MT data, we count the number of flights to and from each country per day relativeto a predefined baseline during the same period, see Section . In addition to themobility data, we look at the number of confirmed cases and deaths per day in eachcountry. These data originate from the World Health Organization . Finally, for each ofthe considered countries, we gathered information about the policies taken, such aswhen a lock-down was put in place. These data came from country specific officialinformation websites.The time frame we look at in the data starts on January 13, which is the first dayavailable in the MT data set, and the data presented herein ends on July 31, as theflight data is released in monthly batches. Overall we have over six months of datareported on a daily basis.For the purpose of this paper, we chose to consider only countries in Europe inaddition to New Zealand, where response to the pandemic was fast and policies werevery effective. With this data we created a mobility dashboard using the shiny packagein R. The mobility dashboard will be available online upon publication of this paper . Results
We are interested in knowing more about general mobility trends and well-being inEurope during the pandemic. The following analysis is based on data from Apple (asdescribed above) as well as from the flight data set (also described above). The data iscompiled in our dashboard. https://opensky-network.org/community/blog/item/6-opensky-covid-19-flight-dataset See: https://covid19.who.int/table here we put a link Notice: This is a preprint of a work currently in submission. 4/12 ig 1.
Changes is walking mobility of selected countries
Mobility Trends
Figs. 1, 2 and 3 show, respectively, the changes in walking, driving and flying mobilityin ten European countries as well as in New Zealand (black line). The figures show aseven-day moving average of the daily mobility given in the MT and flight datasets. Thebaseline (0%) is the mobility of the respective country on the first day of the period.A general conclusion that can be drawn from this data is that there is a significantdecrease in all types of mobility in Europe in March 2020, which gradually increasesagain although some countries have to date, not gotten back to their normal level ofphysical activity. Flights are still under the baseline, whereas driving has insteadincreased significantly.Notably, the mobility in Italy (yellow line) dropped a few days before the othercountries. The countries Italy and Spain (grey) drop lower in walking and drivingmobility than the other countries, which can be explained by the strict lock-down whichwas imposed in these countries due to the rapid spread of the virus in March. Clearly,their mobility is also recovering more slowly, since restrictions were only lifted gradually.Sweden (light green) shows a different behaviour from the other countries. Theirwalking and driving mobility dropped less and recovered faster. The other Scandinaviancountries, Denmark (light orange) and Norway (dark orange) have rather similar curvesto begin with since they responded to the pandemic with similar measures at a similartime, but we see that mobility in Norway is recovering faster than in Denmark,especially regarding driving and flying.At the end of February, there is small peak in the flying mobility, which reflects thenumber of people that were rushing to reach their destinations or home countries beforerestrictions were imposed at a global scale.Finally, although the pandemic started residing, at least during the summer monthsin Europe, the imminent second wave and reaction to it, shows that mobility is trendingdown at the end of July, at least in some of the countries.Notice: This is a preprint of a work currently in submission. 5/12 ig 2.
Changes is driving mobility of selected countries
Fig 3.
Changes is flying mobility of selected countriesNotice: This is a preprint of a work currently in submission. 6/12 ction & Reaction
We now attempt to put the mobility trends in perspective with the threat imposed bythe spread of the virus and the actions taken by the various governments to contain it.Table 1 shows important dates for each of the eleven countries we focus on in this paper.These are the date of the first confirmed infection, the date on which a lock-down wasimposed, the date where the number of confirmed cases started decreasing and the dateof the first death caused by the virus. In addition, the number of days from the firstconfirmed infection, is shown in parenthesis. This roughly shows how quicklygovernments responded, how long it took to slow down the spread of the virus and howquickly it escalated.
Table 1. Important dates by country. The numbers in parenthesis () show the number of days from thefirst confirmed case in each country.
Country Date of firstconfirmedcase Date of lock-down( b March 13 (9)Spain January 31 March 14 (43) April 4 (64) March 6 (35)Sweden January 31 – a June 24 (145) March 11 (40)United Kingdom January 31 March 23 (52) May 8 (98) March 10 (39)New Zealand February 28 March 23 (24) March 29 (30) March 28 (29)(a) Sweden never imposed any form of a lock-down.(b) The number of confirmed cases is still rising in Poland.In addition, Table 2 shows the amount of cases at the end of the time period weconsider in this paper. The table shows the number of confirmed infections, the numberper one hundred thousand inhabitants, the confirmed number of deaths and theconfirmed number of deaths per one hundred thousand inhabitants.
Table 2. Number of confirmed cases and deaths
Country Total number of infections(per 100k) Total number of deaths(per 100k)Belgium 69756 (609) 9845 (86)Denmark 13789 (237) 615 (11)Hungary 4535 (46) 597 (6)Iceland 1907 (534) 10 (3)Italy 248070 (411) 35154 (58)Norway 9208 (173) 255 (5)Poland 46894 (123) 1731 (5)Spain 288522 (615) 28445 (61)Sweden 80422 (786) 5743 (56)United Kingdom 304699 (457) 46201 (69)New Zealand 1217 (25) 22 (0.5)Comparing countries that impose a lock-down before and after they experience thefirst death due to the virus, we have on one hand Denmark, Iceland, Norway, PolandNotice: This is a preprint of a work currently in submission. 7/12nd New Zealand, and on the other hand, Belgium, Hungary, Italy, Spain, Sweden andthe United Kingdom. In the first group, there have been fewer deaths per 100kinhabitants as well as fewer number of confirmed infections per 100k inhabitants(excluding Hungary and Iceland). For the countries in the first group, less than 50 dayspassed before the number of daily new infections started to decrease, where as thecountries that took longer to react also had to wait longer for the virus to start residing.It is worth mentioning that we are only looking at lock-downs as a form ofgovernment response. However, alongside this, governments insisted on socialdistancing, hand washing, ban on social gatherings over a certain size, closing of schoolsetc. that also had an effect on the spread. What is clear from these observations,however, is that timely response was crucial. This is inline with [6].
Discussion
Time plays a pivotal role in a crisis situation, and that is also true about this particularpandemic [8]. Using open source data, and pulling together data from different sourceswhich shows the importance of when actions were taken and to what extent, and we seethat as an important contribution in the wake of the pandemic. Technology and thedevelopment of information systems such as the dashboard presented herein which relieson open source data and shows actions in a transparent way, is, as we see it, at theessence of our role as researchers, alongside the ability to analyse vast amounts of dataaround the unprecedented upheaval that COVID-19 has posed. Those two aspectsthereby outline the main contribution of this paper. There are several insights we canobtain from the data presented in the previous section and here we will discuss thefollowing aspects: i) changes in mobility trends, ii) flying versus driving, iii) specificcountries in relation to the level of harshness in policy taken across Europe iv) longterm effects on well-being and the role of habit and rules on everyday activity, v) theuse of open source data and the importance of granularity and accuracy of open data.A general conclusion that can be drawn from our analyses of the open source data isthat there is a significant decrease in all types of mobility in Europe. The drop starts inMarch 2020, and gradually increases again although some countries have to date, notgotten back to their normal level of physical activity. Regarding flights, they are to datestill under the baseline. The sudden drop in mobility is not surprising given the heavyrestrictions in most countries. What is more interesting is how differently these trendsincrease again.Even though flying has decreased significantly, driving has in turn increased,especially after the ease on travel restrictions. An assumption that can be made isthereby that even though we are flying less, we are not really travelling less, or at leastwe are not catering to the sustainable aspect of flying less in general. Instead wecompensate flying less with driving more. Behavioural changes like travelling less, andacting in an increasingly sustainable way in everyday activities are rooted in our humanbehaviour and a pandemic can staple that type of behaviour short term. In contrast,acting in an increasingly sustainable way long term, is something that needs to happenin smaller steps over a longer period of time. Travel restrictions were eased in Europewith summer and summer holidays around the corner, so it is logical to assume thatpeople drove to their holiday destination instead of flying. To shed further light on this,it would be interesting to know where people went on holiday, that is, how far awayfrom their homes.Notably, the mobility in Italy dropped a few days before the other countries. Thecountries Italy and Spain drop lower in walking and riving mobility than the othercountries over the period examined, which can be explained by the strict lock-downwhich was imposed in these countries due to the rapid spread of the virus in March.Notice: This is a preprint of a work currently in submission. 8/12learly, their mobility is also recovering slower, since restrictions were only lifted slowly.In the data, Sweden shows a different behaviour in comparison to the other countries.Sweden did not use lock-down strategies and as a result, the data shows a differenttrend. Their walking and driving mobility dropped less and recovered faster. The otherScandinavian countries show rather similar curves to begin with, as their respond to thepandemic and their strategy included similar measures at a similar time whereas themobility in Norway recovering faster than in Denmark, especially when looking towardsdriving and flying. This particular aspect –which strategy is the best– will likely remainan open research question for a long time after the pandemic has passed. Even thoughlock-downs and social distancing are accepted as effective measures, we can not knowwhat the long term economic and public health consequences will be [6].As mentioned above, the long term effects of the pandemic remain unknown.Millions of people were restricted to their homes for weeks, which caused drasticchanges in daily routines as we have demonstrated with the mobility data. In addition,according to open source data from Withings , the average person in France has gained.19 lbs (.084 kg), in Spain .26 lbs (.117 kg), in United Kingdom .35 lbs (.16 kg), inGermany .41 lbs (.189 kg), in Italy .42 lbs (.195 kg) and in China .55 lbs (.25 kg). Dataon steps taken on a daily basis from China shows a decrease by over half (56%) inHubei during the lock-down period. According to the same data set from Withings,daily step count decreased by 8% in the United Kingdom, by 28% in Italy, 27% inFrance and an increase of 1% during self-isolation in Germany. What this data shows, isthat the impact of the measures taken, differs greatly in the countries listed here aboveand bear in mind that this data is only from Withings. This data from Withings, incombination with the mobility data shows that during the lock-down period, there is ashort-term effect on weight, and a connection between reduced walking. The data fromWithings is however only from a short period of time and in additon to that, thenumbers are based on data from a limited part of the population (only those that own aWithings wearable) whereas the long-term effects on well-being in general need to bestudied further, using larger data sets over an extended period of time.The analysis herein, which is based on open and publicly available data, shows theimportance of opening up the data that is already being gathered across differentplatforms. What we would therefore like to emphasise is the importance of granularityand accuracy of open source data. There is much added value in combining data fromdifferent sources, and presenting it in an intuitive and clear manner, as we have donehere. Links and correlations that were unknown before can easily be seen andinterpreted. As data privacy and ethics of data usage are currently widely beingdiscussed and disputed, it is important to emphasise that data in its most aggregateform –as we used here– can come to great use, when presented correctly. With the vastand widespread data collection that is taking place at every front, for example by socialmedia platforms, there is a call for stronger incentives for these organizations to shareinsights and preserve historically significant data [36]. Conclusion
The COVID-19 pandemic has triggered a worldwide health crisis and the strategiestaken across Europe have differed where some countries have taken strict measures,while others have avoided lock-downs. In this paper we report on findings from puttingtogether open data from different sources in order to shed light on the changes inmobility patterns in Europe during the pandemic. Using that data we show the impactof the different strategies put into place in Europe during the pandemic and how Withings produces activity tracking devices such as watches and Bluetooth scales and compiledweight data from the pandemic time: https://blog.withings.com/tag/health-data/
Notice: This is a preprint of a work currently in submission. 9/12obility patterns changed in different counties. Our data shows how the majority ofEurope walked less during the lock-downs, which is bound to have a long-term effect onwell-being. Also, while flights were less frequent, driving increased drastically. Throughour analysis we show the importance of data granularity and accuracy through ournovel study of using open source data to shed light on mobility trends in the pandemicof COVID-19.Our research offers several opportunities for future work. As the summary statisticsabout weight gain during lock-down indicate, analysing other sources of data alongsidewhat we do in this paper would be an interesting way of continuing this line of work.Especially in regards to data describing health and well-being in order to examine thelong-term effects of the pandemic. The OpenSky data set only provided informationabout flights, not the quantity of people on each flight. These numbers could beinteresting to take into account, since during the travel restrictions, relatively fewpeople were on each flight. In our results, we described the variation in the impact ofthe pandemics with respect to temporal factors, such as the delay between first infectionand imposing of lock-downs in Europe. In our future work, we will investigate suchcorrelations to a greater extent at a global level in order to contribute to anunderstanding what measures were the most effective when designing strategies duringa fight against a pandemic.