key: cord-0530252-5vujmm2i
authors: Suzumura, Toyotaro; Kanezashi, Hiroki; Dholakia, Mishal; Ishii, Euma; Napagao, Sergio Alvarez; P'erez-Arnal, Raquel; Garcia-Gasulla, Dario
title: The Impact of COVID-19 on Flight Networks
date: 2020-06-04
journal: nan
DOI: nan
sha: 1a8fd8c8c40c45d57eeedb55fba0bba2a2da3add
doc_id: 530252
cord_uid: 5vujmm2i

As COVID-19 transmissions spread worldwide, governments have announced and enforced travel restrictions to prevent further infections. Such restrictions have a direct effect on the volume of international flights among these countries, resulting in extensive social and economic costs. To better understand the situation in a quantitative manner, we used the Opensky network data to clarify flight patterns and flight densities around the world and observe relationships between flight numbers with new infections, and with the economy (unemployment rate) in Barcelona. We found that the number of daily flights gradually decreased and suddenly dropped 64% during the second half of March in 2020 after the US and Europe enacted travel restrictions. We also observed a 51% decrease in the global flight network density decreased during this period. Regarding new COVID-19 cases, the world had an unexpected surge regardless of travel restrictions. Finally, the layoffs for temporary workers in the tourism and airplane business increased by 4.3 fold in the weeks following Spain's decision to close its borders.

As COVID-19 transmissions spread worldwide, governments announced and enforced both domestic and international travel restrictions. These restrictions affected the trend of flight networks around the world such as major airlines [1] and tourism-dependent cities [2] . All or a portion of flights connecting these restricted countries and areas were cancelled. The date of travel restriction enforcement varies by country, therefore the period and degree of flight reductions also vary by country and continents.

We quantitatively investigated how flight restrictions affect domestic and international travel in countries and continents through time-series analytics and graph algorithms. We also investigated the negative effect of these restrictions on employment and the number of infected cases in Barcelona, where the majority of revenue depends on the travel industry. *Toyotaro Suzumura and Hiroki Kanezashi contributed equally to this work. The structure of this paper is as follows; In Section II, we introduce the flight dataset which we used for our analysis. We compared the dataset with other flight datasets and discussed the limitations and potential biases. In Section III, we summarize the statistical analytics of airport and flight data by country and continent. In Section IV, we analyze the number of daily flights globally as well as by region (continent and country level), and discuss the relationship between government announcements and the decline of flights. In Section V, we visualize flight networks and apply graph analytic techniques to the network to evaluate the quantitative impacts on travelers. We examine economic effects on travel restrictions and cities depending on tourism industries from the public data from Barcelona in Section VI, and we evaluate the relationship between incoming flights from Europe and the number of new infection cases in the United States in Section VII. Finally, we discuss our findings, limitations and related work in Section VIII, and explain the summary and future perspectives of our research in Section IX.

In order to evaluate the effect of travel restrictions on the flights, we conducted the following analyses step by step.

First, we summarized the total number of airports, flights and potential passengers for each country and continent from an open dataset. Next, we counted up the number of daily flights for countries and continents, and compared the period of the flight reductions and the date of the travel restriction enforcement. In order to evaluate the density of the flight network quantitatively, we also applied graph analytics methods to the time-series flight data. Finally, we analyzed time-series data sets about the number of incoming flights, number of infected cases and unemployment in Barcelona [3] , where tourism industries is a main revenue source, and hypothesized the correlation among these data. [4] from The OpenSky Network [5] . The dataset contains flight records with departure and arrival times, airport codes (origin and destination), and aircraft types. The dataset includes the following flight information during 2020 January 1 to April 30. The dataset for a particular month is made available during the beginning of the following month. Note that the year of all of the following dates is 2020. In total, our dataset is composed by: 2 describes the total number of flights from January 1 to April 30 2020 for each continent and country. More than half of flights in this dataset departed from Europe. 10% of international flights departed from Germany (10%) and the United States (10.7%). In Asia, Hong Kong, the United Arab Emirates, and Singapore had the highest number of flights (2.9%, 2.6%, and 2.2% respectively), yet there were very few flights connecting to China. The number of flights in South America, Africa, and Oceania were too small to analyze, therefore these continents were excluded from the following sections. 

The OpenSky Network Dataset does not contain the final number of passengers for each flight. However, since the data contained each aircraft type used, we used this to estimate the number of passengers for each trip. Since the information regarding the flights were extracted from the capacity data of each aircraft type from their respective Wikipedia pages [6] , we aim to obtain more accurate information on aircraft types and their capabilities for our future analysis.

The proportion of international passengers travelling from Asia was approximately 31.2% of all passengers, and the number of flights were 23.5% of overall flights. On the other hand, the total number of passengers of international flights within Europe is about 46.5% while the proportion of international flights outside of Europe is 55.5%. The gap of proportions between passengers and flights comes from the difference in capacities of the aircraft.

Announcements and enforcement regarding lockdowns (border closures and travel restrictions, etc.) were implemented to slow the rate of transmissions and prevent overburdening of health facilities. The date of implementation, length, and extent vary by country. In the US, the Secretary of Health and Human Services declared a public health emergency on January 31 [7] . The US government suspended incoming flights from Europe (Schengen Area) on March 14 [8] , and closed the border between the US and Canada on March 18 [9] . Germany closed borders with neighboring countries on March 16 [10] . Countries in the European Union agreed to reinforce border closures and restrict non-essential travels from March 17 [11] onwards. In this section, we show the changes in the number of daily flights in each region, and discuss the correlations between travel restrictions and the number of flights in each country. Figure 3 describes the number of daily international flights in the world. Before the end of February, the daily number of international flights was approximately 8,000. However, the number of daily flights gradually decreased to 6,000 by early March and then drastically dropped to less than 1,000 (around 10% of regular seasons) by the end of March. Figure 5 describes the number of continental flights within each continent. In Europe, more than 3,000 flights were operating per day until the beginning of March, but dropped to 10% of the regular number of flights from mid March to the end of March. In Asia, the number of flights has gradually decreased from February. In North America, the number of flights was steady until mid-March, before dropping significantly.

In Europe, the UK, Ireland, and Germany had the highest number of flight passengers [12] . In Germany, international flights decreased from the beginning of March and then dropped after the nation closed its borders on March 16 ( Figure  6 ). This reflected the EU's agreement to ban incoming travel from areas outside the EU, UK, and Switzerland on March 17.

Most of the continental flights had been operating in North America (the US in particular) until mid-March and half number of these flights in the Beginning of January are still in operation today ( Figure 7 ). On the other hand, the number of flights in Asia, Europe, and Oceania have dropped to less than 20% to that of January 1.

In the US, the number of domestic flights suddenly dropped from mid-March. However, roughly 50% of all regular flights were still in operation on March 31 (Figure 8 ), although the US government and airlines planned to ban or reduce flights by that point [13] . In April, the number of flights gradually declined but 30% of regular flights were still in operation.

The number of flights from European countries to the US suddenly dropped after March 14 ( Figure 9 ). The US suspended incoming travel from Europe (Schengen area) from March 13. In the overall number of international flights, the primary destination from the US is Canada (green lines in Figure 10 ). The number of daily flights between the US and Canada gradually decreased after these governments agreed to close their border on March 18. 

We analyzed the number of daily flights for each country and continent, and compared the relationship between the date of travel restrictions and the decline of flights. We found that many international flights suddenly declined during the middle of March. In Europe, the number of flights dropped to about 10% of regular seasons after the EU agreed to restrict incoming travels on March 17. In the US, approximately half of all domestic flights were still in operation even at the end of March. 

With travel restrictions, the global flight network gradually became more sparse, which prevented travelers from: 1) Using direct flights to their destinations 2) Returning to their origin once they depart from the airport 3) Departing from their current location due to absence of flights in nearby airports By constructing daily flight networks, we quantitatively evaluated these effects through graph analytics. Each vertex and edge represent an airport and a flight, respectively. While multiple flights are in operation every day, not all flights are operated every day (e.g., three days a week). To detect long-term trends more precisely, we also aggregated flight edges by week as an optional preprocessing ( Figure 12 ). 

In regular times, many flights (edges in the flight network) connect from the airport (vertices in the flight network) to the airport. That helps travelers to get to more destinations with fewer transfers. However, travel restrictions prevent them from using direct flights and getting to their destinations. To measure the effect quantitatively, we computed the following metrics for each airport and each date ( Figure 13 The color and size of each vertex in Figure 14 -19 indicate the number of departed flights from the airport. From mid-March to the beginning of April, many flights were gradually canceled, except for domestic flights in the United States and international flights between the United States and Europe.

The number of flights in Europe suddenly decreased from March 16 (the day before EU countries closed borders) to March 23. In April, only a few flights were operating and many airports became isolated.

In North America, most of the flights in North America were operating between the United States and Canada. In the United States, most of the domestic flights were operating as usual, even at the end of April.

With denser flight networks, travelers can reach more destinations. As the barometer of the flight network density, we computed the following metrics. Density: The ratio of the actual number of edges after aggregation against the number of edges of the complete graph. Clustering Coefficient: The ratio of the total number of triangles against the total number of triplets (connected three vertices). This is mainly used for social network analytics to measure the connectivity of friendships.

The density (red lines in Figure 20 -22) of the global flight network rapidly decreased by half during March 8 to April 1. In Europe, drastic decreases in the flight network density was similar to that of the global network, and fell to approximately 17% of the peak. In the US on the other hand, the density gradually decreased, yet only by 20%. From March 13, the clustering coefficient (blue lines) of the global flight network gradually decreased by half in the next two weeks. In Europe, it rapidly declined to about a quarter in the same period. In the US, the decline of the clustering coefficient is only about 10% at the end of April.

In regular seasons, travelers departing from any airport can typically go to any destination and return to their airport of origin. This is due to flight networks being a strongly connected component (SCC), within which travelers can go from/to any airports in the same SCC. With travel restrictions, however, many flights were canceled and the large SCC (blue circle in Figure 23 ) has decomposed into smaller SCCs (yellow and green circles in Figure 23 ). In the result, travelers cannot: 1) Go to other airports outside local flight networks 2) Return to their origin (a red edge from E to D) 3) Depart from isolated airports (a red vertex)

In the global flight network, the size of the largest SCC (red line) decreased by 30% from March 13 (300) to April 6 (210), and the number of isolated airports (blue line) increased by 30% (from 310 to 400). In Europe, the largest SCC size on April 7 was less than half of March 10 (before border closures), and the number of isolated airports tripled. In the US on the other hand, the largest SCC size and isolated airports are almost stable (80 and 100 respectively). Although the largest SCC size gradually decreased from March 13 to April 6, the rate of the overall decrease is only about 10%.

When many flights are canceled, more transfers are required for travellers to reach the destinations compared to regular seasons. In order to measure the number of required transfers, we applied a graph analysis to compute eccentricity (the maximum distance from a specified vertex to all other connected vertices) of the flight network. An eccentricity of an airport vertex corresponds to the number of required transfers to other destinations in the worst case. With travel restrictions, the diameter (the maximum value of eccentricities) and the radius (the minimum value of eccentricities) of the flight network become larger.

The diameter (red line in Figure 28 -30), radius (blue line), and average eccentricity (green line) of the largest SCC size (black line) in the global flight network increased from April. In the flight network in Europe, diameter and radius rapidly increased from March 16 in two weeks. The overall flight networks trend in the US did not change until the end of April.

Although a number of flights still exist among most of the airports, the flight frequency has gradually decreased. To measure the decline of flight frequencies, we defined the relative weight for each flight edge (pair of vertices as connected airports) and network density with the weighted edges. First, we computed the moving average of the number of daily flights for each pair of vertices, and then computed the ratio of the total number of flights in a week against the baseline of the specified week (January 12 -January 18), before defining the ratio as the relative weight of the edge. Besides the topological density described in V-C, we computed the density of the whole flight network with the weighted The blue and red lines in Figure 32 -34 represent the density of the daily flight network with the unweighted/weighted flight edges. The density of the whole flight network with weighted edges in Europe dropped to almost zero in April.

In the US, the density with weighted edges has dropped from 0.025 to 0.005 in March. Although the topology of the flight network remains unchanged, the frequency of domestic flights in the US significantly decreased as seen in other 

We applied some graph analytics to the flight data in order to evaluate the density of flight networks. Many graph metrics such as density, maximum SCC size, and diameter remarkably 

According to the unemployment dataset [14] , the number of affected temporary workers (red line in Figure 35 ) gradually declined in April after it suddenly increased on Match 24. The number of flights (blue line) arriving in Barcelona gradually decreased until the beginning of April. These numbers seem to have some correlations, but more investigation is required to understand the causality and relationship between these numbers.

The number of flights to Barcelona (blue line in Figure  36 ) has been decreasing since the beginning of March, and suddenly halved on March 16 just before Spain closed its borders [15] . Then, the number of affected temporary workers 

Approximately half of all flights to Barcelona (BCN) come from France, Germany, and the UK. The government of Spain closed its borders on March 16, and the number of flights from these three countries decreased by half during the following day. However, the number of new infection cases for these countries has been gradually increasing even before border closures. Considering the latent period of COVID-19 is commonly five to six days [16] , the reason the trend of increase of new infection cases continued for a week after the border closure may be due to the immigration of infected individuals before the border closure. 

We compared the number of incoming flights, affected temporary workers, and the number of new infection cases in Barcelona to estimate how the travel restriction affected the tourism industry. We found the number of flights halved after Spain closed borders on March 16, and the number of affected workers jumped just after the government opened the application of unemployment on March 23. The number of new infection cases increased until March 25 after the border closures, potentially due to infections spreading during its latent period.

A sequence analysis of COVID-19 virus found out that the virus type of the most infection cases in New York City is from Europe [17] . The first case was confirmed in the State of New York on February 29, and the number of infection cases gradually increased from March 15 ( Figure 38 ). To investigate the effects of incoming flights from Europe, we compared the number of daily flights from Europe and new infection cases in the US. 

The trend of increase in newly infected cases in Europe began a week before the US (Figure 39 ). In particular, confirmed new infections in Italy started to increase in the beginning of March. Spain and Germany also confirmed cases before the US (Figure 40 ). The US announced to enforce travel restrictions from Europe on March 13 (EU mainland) and March 14 (the UK and Ireland). After these restrictions, infection cases gradually increased in the US from March 15 and exceeded 30,000 on April 2. 

With the travel restrictions from European countries to the United States, the number of incoming flights (blue line in Figure 41 ) suddenly decreased from March 13 to March 24. However, the number of infection cases (red line in Figure  41 ) began to increase just after these restrictions. However, the latent period of COVID-19 is commonly five to six days as we mentioned in Section VI-C, therefore we cannot rule Figure 42 shows the number of daily international flights from these countries to the US and the number of daily new infections in the US (red lines). In Italy (blue line), more infection cases were reported than other European countries from March 1 (Figure 41 ), but the number of flights to the US became almost zero just after the travel restriction was enforced. On the other hand, some flights from Spain (orange line) and Germany (green line) to the US were still in operation, even after the restrictions were announced and infection cases in the US began to increase. These flights may be one of the reasons for the gradual increase of infections. In Section 4, we found out when and how international flights in each continent and countries declined due to travel restrictions. The number of flights on April 1 had dropped to 10% of that of regular seasons. The number of flights from Europe sharply decreased in the second half of March after the US announced the travel restriction from most of European countries on March 11 and the EU agreed to close borders on March 17. On the other hand, half of regular domestic flights in the US were still in operation at the end of March.

Besides counting the number of daily flights, we also applied several graph analytics to daily flight networks in order to explain how these networks became sparse in Section 5. With the result of analytics, we found out the density of the global flight network halved in March, and it has been divided into small strongly connected components from the mid of March to the end of April.

In Section 6, we evaluated the effect of reduction of incoming flights to Barcelona on the tourism industry (e.g. the number of affected temporary workers). The number of incoming flights dropped just after the travel restriction by the government of Spain on March 16, and the number of affected workers drastically increased after the government of Catalonia published the number of these workers from March 23. However, the number of infection cases still increased even after travel restrictions were announced.

We also investigated the relationship between incoming flights from Europe and the number of infections in the US in Section 7. We found that the trend of daily infections in Europe began a week ahead of the US, and that some European countries confirmed cases from March 1 while the number of infection cases in the US began to increase from March 15. We also found that flights from Europe were still accepted in the US even after travel restrictions were announced on March 14. Hien et al. [19] evaluated the correlation of domestic COVID-19 cases and flight traffic volumes with the OpenSky Network dataset, and concluded the number of international flight routes and passengers have a correlation with the risk of COVID-19 exposures.

Since the flight dataset from the OpenSky Network is aggregated voluntarily, it has some limitations. First, some flight entries do not contain origin or destination airport codes.

We excluded such entries before analysis to construct valid flight networks. Moreover, the dataset also has a significant regional bias. For example, many international and domestic flights connecting China, Oceania, South America, and some countries in Africa are missing.

After the declaration of a pandemic by WHO on March 11 and government announcements of travel restrictions during the following week, the number of international flights around the world drastically declined, particularly in Europe. Although many domestic flights were still in operation in the US, the frequency gradually declined.

These travel restrictions affected tourism-dependent countries. In Barcelona in particular, many temporary workers made applications regarding their unemployment soon after the government opened applications on March 23. Moreover, the number of newly infected cases increased after travel restrictions were declared on March 16, most likely due to COVID-19s longer latency period.

Due to the limitation of publicly available data, we could not conduct the same analysis for many countries as Barcelona in Section 6 for this paper. Therefore, we aim to analyze relationships among the number of flights, infection cases, and the impact on the economy for more countries and continents by combining various other data sets with the flight data in our next analysis. We also aim to focus on counties where the number of infections has soared, and investigate how the increase of cases and reduction of flights affect the economy.

We also plan to trace and predict viral mutations with the flight network data. The type of COVID-19 virus varies from region to region [20] , [21] , and this may provide insight to analyzing the route of infections with viral types. With the daily flight network data, we hope to estimate the date and route of infections more precisely.

In order to predict the economic impacts such as propagations of revenue declines in travel industries for each region, we will apply advanced machine learning techniques for time-series graph data such as Graph Convolutional Network (GCN). We will also evaluate the importance of flight network features as well as demographic features of regions (countries and continents) in our next analysis.

Remarks by president trump and vice president pence at a coronavirus briefing with airline ceos

Travel and tourism direct contribution to gdp

Barcelona in numbers: Tourism, between wealth and residents complaints barcelona metrpolis

Bringing up opensky: A large-scale ads-b sensor network for research

The opensky network -free ads-b and mode s data for research

List of aircraft type designators -wikipedia

Proclamation on declaring a national emergency concerning the novel coronavirus disease (covid-19) outbreak

Proclamation -suspension of entry as immigrants and nonimmigrants of certain additional persons who pose a risk of transmitting 2019 novel coronavirus

President donald j. trump is taking necessary safety measures at the border to prevent further spread of the coronavirus

Coronavirus: Germany to impose border controls over coronavirus

Video conference of the members of the european council

Air transport, passengers carried

Coronavirus forces airlines to consider a once unthinkable possibility halting us flights

Temporary employment regulation files (erte) presented daily and number of affected workers

Government restricts access to travellers at spain's external borders

coronavirus disease 2019 -world health organization

Introductions and early spread of sars-cov-2 in the new york city area

Estimating and projecting air passenger traffic during the covid-19 coronavirus outbreak and its socio-economic impact

The association between international and domestic air traffic and the coronavirus (covid-19) outbreak

TreeTime: Maximumlikelihood phylodynamic analysis

Nextstrain: real-time tracking of pathogen evolution