Resilience of Cities Towards Airport Disruptions at Global Scale

Abstract

Improving the efficiency and resilience of air transportation is one of the major challenges in the 21st century. Largecities are often in the catchment area of multiple airports. Accordingly, the services at these airports can be comple-mentary or competitive; an airport-specific view of resilience neglects the effects of such interactions. In this study,we analyze the resilience of cities in the presence of airport disruptions. The major research question we address is:How much would the air accessibility of complete cities lose under the outage of a single airport? We perform a studyat the global level, comparing the resilience of 5,000 largest cities. Two types of passenger-focused measures areproposed to quantify the impact of airport disruption on the air accessibility of cities: The unaffected passengers andthe reroutable passengers. We find that some countries have resilient cities, while others are more vulnerable to air-port disruptions. Our study contributes to the literature by a unique combination of city-view, accurate airport-to-cityaggregation with road driving distance, and presence of real passenger data.

Keywords: Airport failure; City global resilience; Reroutable passengers

1. INTRODUCTION

Given a steadily increasing demand for human mobility, ambitious goals have been set for air transportation, interms of quality and affordability, environment, efficiency, safety, and security (European Commission 2001). With therapid development of economic globalization, resilience of transportation systems to disruptions becomes increasinglyimportant, for improving passenger’s safety and avoiding huge economic losses. The eruption of Eyjafjallajoekullvolcano in 2010 caused airlines lost approximately 1.7 billion US dollars and more than 10 million passengers wereaffected (Wilkinson, Dunn, and Ma 2012). An overnight snowstorm on March 12, 2013 disrupted the transportsystem across northwestern Europe; in particular Frankfurt airport was closed and airlines canceled about 700 flights.In December 2017, an electrical fire in a tunnel beneath Atlanta’s Hartsfield-Jackson airport (the busiest airport in theworld), broke its main and backup power systems, leading to a power outage of nearly 11 hours. Power outage andfailed computing systems can also have large impacts on airlines. For instance, for Delta Airlines lost 150 million USDin August 2016 due to a five-hour power outage, caused by computer crash1. Therefore, it is critical to understand thecomplexity and emergent behavior of air transportation systems.

Network science provides powerful tools for understanding the structures and dynamics of complex socio-economicsystems, including air transportation (Barabasi 2013, Newman 2010, Zanin and Lillo 2013, Vitali et al. 2012). Ac-cordingly, throughout the last decade, many studies have analyzed the delay propagation (Fleurquin, Ramasco, andEguiluz 2013), epidemic spreading (Gomes et al. 2014), temporal evolution (Kotegawa et al. 2011, Wandelt and Sun2015, Wong et al. 2019), port (airport) adaptation to climate-related disasters (Xiao et al. 2015, Wang and Zhang2018, Randrianarisoa and Zhang 2019), and, particularly, the impact of large-scale disruptions (Caschili, Medda, andWilson 2015, Caschili, Reggiani, and Medda 2015, Wandelt, Sun, and Zhang 2019, Wandelt et al. 2018).

The majority of studies on air transportation network resilience focus on the airport network, where nodes repre-sent single airports and links correspond to direct flights between airports. It has been shown that worldwide airportnetwork is rather resilient to random failures (i.e., random malfunctioning airports of any size), while targeted attacks2

(i.e., intentionally disrupt large, critical airports) can break the network quickly (Sun, Wandelt, and Cao 2017, Sun,

1See https://money.cnn.com/2016/09/07/technology/delta-computer-outage-cost/index.html2The term attack is not to be understood literally here. For complex networks, the word attack is used for distinction from random failures,

which do not prefer critical nodes.

1

Figure 1: Multiple airport region for the city of London: Luton Airport (LTN), Stansted Airport (STN), Heathrow Airport (LHR), City Airport(LCY), Southend Airport (SEN), Gatwick Airport (LGW), and Southampton Airport (SOU). From the city center of London (blue circle), a totalof six commercial airports can be reached within a free-flow road travel time less than 100 minutes. Each airport is visualized with its IATA codeand the free-flow road travel time in minutes. The black lines highlight the routes from the city center to the airports.

Wandelt, and Linke 2017). For instance, PEK (Beijing Capital International Airport) is the second busiest airportin the world. Such hub nodes are critical for the structure of air transportation and they are inherently priorities fortargeted attacks. Takebayashi and Onishi (2018) examine the policies for the reliever airport to regain traffic whenthe main airport is dysfunctional due to a catastrophe, with the two airports linked by high-speed rail. Other studiesinvestigate the resilience of the air transportation city network, where nodes represent cities and links correspond todirect flights between cities. The aggregation of airports into cities relies on the concept of Multiple Airport Regions(MARs). A MAR is defined as a group of two or more major commercial airports in a metropolitan region, serving anoverlapping share of latent passengers (de Neufville 1986). After combining the airports nearby a city into a MAR,which is usually performed by simple spatial distance aggregation (e.g., a distance between city center and airport of50 - 250 km) (Xia et al. 2019), the city network is analyzed with similar techniques as the airport network describedabove. To sum up, existing studies usually analyze the effects of node (or link) failures to the airport network or thecity network.

In this study, we investigate the effect of single airport outages on the air accessibility of global cities. As anexample, consider the area of Greater London, which is being served by seven airports, all of which can be reached in100 minutes by car (Figure 1). Naturally, the roles of these airports for London are transitioning between competitionand cooperation. On one hand, some airports offer similar destinations, essentially competing for passengers. Onthe other hand, airports also offer complementary services. Please note that airports are not exclusively aggregatedto one city, but can be shared by multiple cities. For instance, airport SOU (Southampton Airport) from Figure 1 isaggregated to the city of London and the city of Southampton, among others. While this might be counter-intuitive, webelieve that the methodology underlying our study is rational. Other studies have used the line-by-sight distance (Loo,Ho, and Wong 2005, O’Connor and Fuellhart 2016, Bonnefoy, de Neufville, and Hansman 2010, Hansen and Du1993), which can greatly overestimate or underestimate the actual travel time in presence of specific geographical

2

LCY LGW LHR LTN SEN SOU STN

To airport

LC

YL

GW

LH

RLT

NS

EN

SO

US

TN

Fro

mai

rpor

t

0.043 0.047 0.16 0.62 0.51 0.1

0.81 0.32 0.53 0.91 0.88 0.36

0.82 0.34 0.58 0.91 0.88 0.47

0.71 0.16 0.24 0.84 0.81 0.21

0.22 0.027 0.042 0.057 0.29 0.042

0.22 0.038 0.042 0.12 0.45 0.048

0.8 0.28 0.39 0.5 0.9 0.870.15

0.30

0.45

0.60

0.75

0.90

Figure 2: Potential cooperation and competition between airports for the London example. For each pair of airports, the percentage of destinationoverlap for one airport compared to the destinations of both airports is shown. Red color indicates complementary destinations and light colorcompetitive operations with a large overlap.

conditions and transportation infrastructure. According to the time-threshold in this study, Southampton is accessiblefrom the city center of London within 99 minutes. In fact, there are suggestions to further remove burden from theGreater London area by building new international hubs as far away as Birmingham (and then connect Birminghamto London via a high-speed rail line); Birmingham being 206 km outside of London Center, compared to 112 km forSouthampton.

In Figure 2, this situation is visualized for our London example. For each airport pair (ap1, ap2) around London,we compute |D(ap1)−D(ap2)|

|D(ap1)∪D(ap2)| , where D(ap) denotes the set of destinations for airport ap. A larger value (red cell)indicates that most destinations of ap1 are not covered by ap2; while a smaller value (cell with light color) indicatesthat ap1 does not have many new destinations compared to ap2. For instance, consider the two airports GatwickHeathrow (LGW) and Stansted Airport (STN): |D(LGW)| = 178, |D(S EN)| = 158, |D(LGW) − D(S EN)| = 90, and|D(LGW) ∪ D(S EN)| = 248. Accordingly, |D(LGW)−D(S EN)|

|D(LGW)∪D(S EN)| = 90248 = 0.36. This example highlights the importance of

analyzing the effects of airport outage in a MAR towards the accessibility to the global air transportation system.The remainder of this paper is structured as follows. Section 2 presents the methodology used throughout our

study. We present the results on the resilience of global cities in Section 3. Section 4 summarizes the major contribu-tions and discusses directions for future work.

2. METHODOLOGY AND DATA PREPARATION

The data preparation for the worldwide airports and cities used in this study is presented in Section 2.1 andSection 2.2 separately; Section 2.3 presents the methodology to aggregate the airports to multiple airport region of acity.

2.1. Airport Data Preparation

The airport data used in this study comes from two distinct sources: The locations of airports come from http:

//ourairports.com, a community-run website for collecting information on airports around the world. The websitereports the location, name, country, runway lengths, navigation types and other information for each airport. Ourair-ports.com is considered as one of the major and complete (open-source) airport datasets, which is updated on a dailybasis by a strong community of pilots and other volunteers. In our own preliminary experiments, we have not foundanything else as complete. We also performed some comparison with the commercial Jeppesen airport database. Thedataset from ourairports.com contains many more airports (52,865) than Jeppesen (13,669). The reason is that notall airports in ourairports.com provide scheduled service; the airports with scheduled service are distinguishable by acolumn in their database; for our study we only analyzed those airports with scheduled service.

Since the airport data does not have any information on the destinations, we merge this dataset with flight leg datafrom the Sabre Airport Data Intelligence (ADI, http://www.airdi.net). ADI reports the number of passengers

3

Figure 3: Worldwide airports in this study. Each green cross corresponds to one airport. Top 20 airports, according to the total number ofpassengers in the year 2015 are highlighted with names. Data sources: OurAirports (http://ourairports.com) and the Sabre Airport Data Intelligence(http://www.airdi.net).

and the number of departures per year for each scheduled airport pair. We extract all airports from ADI for the year2015, keep those listed in http://ourairports.com, filter all airports with less than 50,000 passengers per year.Finally, for each pair of scheduled airports, we convert the number of passengers and the number of departures todaily values. The remaining airports obtained by the filtering step are visualized in Figure 3. In total, we have 2,340scheduled airports for the analysis in this study.

2.2. City Data Preparation

For a collection of cities in the world, we refer to the World Cities Database (https://simplemaps.com/data/world-cities). The data is provided as a CSV file, containing information on cities’ location, name, country,continent, estimated population and others. We sort all cities decreasingly by the estimated population and extract thetop 5,000 cities. According to the source data, we select all cities with a population larger than 25,000, at the time of2015. The total population of all 5,000 selected cities covers roughly two billion people. The cities used throughoutour study are visualized in Figure 4.

2.3. Aggregating Airports to Cities

Traditionally, MARs are defined based on spatial distance threshold. Threshold values range between 50 km to250 km; airports within this distance are aggregated to a MAR. The rationale for such simplified distance-induced ag-gregation criterion is easily being questioned: When analyzing networks at large scale, it becomes intricate to computeall exact distances between airports (and city centers). Perhaps more importantly, for passengers airport accessibilityis a measure of the costliness of reaching an airport using available transportation systems (Reggiani, Nijkamp, andLanzi 2015). The costs depend, in addition to distance, on the availability as well as the quality of land-side trans-portation systems. Recent advances in open Geographical Information Systems (GIS) and increased computationalresources, allow us to overcome this unnatural definition of airport aggregation and compute accurate driving timein this study instead (Sun et al. 2017). Our automatic routing algorithms are based on Openstreetmap (OSM), whichhas become one of the premier resources of worldwide transportation infrastructure. OSM is a community-basedeffort, started in 2004, which is mainly run by volunteer mappers with the goal to create an editable map of the world.

4

Figure 4: Global cities in this study. Each green cross corresponds to one city. Selected capitals are highlighted with their names. Data source:World Cities Database (https://simplemaps.com/data/world-cities).

Over time, OSM has become a highly accurate representation of the transportation infrastructure in the majority ofpopulated areas (Haklay and Weber 2008, Neis and Zielstra 2014, Wandelt, Wang, and Sun 2017, Poppinga 2018)

In this study, we use the road network modeled in OSM for routing between city centers and airports. Our routingimplementation is based on the Open Source Routing Machine (OSRM) (Luxen and Vetter 2011). Within OSRM,scalable techniques have been implemented for routing vehicles with different speed profiles over OSM data. It isreleased as a small standalone server application (see http://project-osrm.org/). Technically, OSRM performsshortest path calculation based on contraction hierarchies, which allows queries to scale up very well to continentscale. OSRM allows for accurate free-flow time estimations, given that it directly exploits road-specific maximumspeed information from OSM. We extracted the driving time for each airport-city pair in our study, by querying forthe fastest driving connection between the latitude/longitude pairs of both places. It should be noted that the resultis an estimation of the minimum free-flow time between the two points, without any information about road blocks,congestion, or other additional (temporary) road conditions.

Figure 5 shows the correlation between spatial distance proxy, as in previous related studies (Loo, Ho, and Wong2005, O’Connor and Fuellhart 2016, Bonnefoy, de Neufville, and Hansman 2010, Hansen and Du 1993), and estimateddriving time (used in our study). It can be seen that, while there is an overall linear correlation between both variables,the deviation of driving times can be rather large for a specific distance. For instance, at a fixed distance of 50 km, theshortest driving time is 40 minutes and the longest driving time is 120 minutes. Treating both cases identical, as donein existing studies, yields misleading results.

For each city in the dataset, we compute the driving time to all airports. If the driving time from the city centerto the airport is less than 120 minutes, we aggregate the airport to the city. For all airport-city pairs, we enforcethat airport and city have to be located in the same country, with the only exception of participating countries in theEuropean Schengen area. For cities in the latter countries, we assume that cities can also aggregate airports in otherSchengen countries. For instance, people in the border region of Western Germany frequently take international flightsfrom AMS (Amsterdam Schiphol), instead of farther away German airports. In reality, other countries could be takeninto consideration, e.g., UK and England, under an airport disruption at either Belfast or Dublin. Future studies couldperform a more fine-grained analysis with an improved distinction of viable cross-border traffic.

As indicated above, both, ourairports and Openstreetmap, are typical open-source data and they have been increas-

5

Figure 5: Correlation between spatial distance proxy (x-axis) and estimated driving time (y-axis) for all city-airport pairs in our study. Each bluepoint is a city-airport pair; the red line is a linear regression curve; the Pearson correlation coefficient is 0.93 and the p-value is 0.

1 2 3 4 5 6 7 8 9 10

Number of airports

101

102

103

Nu

mb

erof

citi

es

Figure 6: Number of cities versus the number of airports aggregated for a city. Most cities in our study have one airport. Two cities (Antwerpenand Oxford) have ten airports within their range. Note that the y-axis is log-scaled for the sake of magnitude comparison.

ingly used in the literature in recent years. These open-source datasets have been built and maintained by volunteersand they are released with open-content licenses. The aim of these open-source projects is exactly to promote any-one to use and share these geospatial data. The free-use and sharing of data sources is just the spirit of open-sourceprojects, which is rather different from commercial data/service providers, such as OAG database. The high dataacquisition cost of OAG prohibits large numbers of researchers to access the data who are interested in solving chal-lenging problems. We believe that research findings based on (accurate) open-source data, can better enable thedissemination of knowledge in a fair and open way across the research community in the future.

3. RESULTS

After aggregating all airports to nearby cities, we obtain 3,511 cities having at least one airport reachable withintwo hours of driving time. Out of these, a total of 1,568 cities have at least two airports as a choice for their population.The two cities with the largest number of reachable airports are Antwerpen and Oxford with ten airports, followed byfourteen cities that can reach nine airports, e.g., New York, Brussels, Cologne, Wuppertal, Liege, Gent, Eindhoven,Newark, Reading, Luton, Paterson, Arnhem, ’s-Hertogenbosch, and Maastricht.

6

Table 1: Number of cities with a given number of aggregated airports (from 2 to 10) at the country level. Top 20 countries according to the totalnumber of cities are shown.

Country 2 3 4 5 6 7 8 9 10United States 110 104 68 33 10 5 2 3 0

China 93 33 15 3 2 0 0 0 0Brazil 59 16 1 0 0 0 0 0 0Japan 10 25 16 3 2 1 0 0 0

Germany 9 9 12 11 6 3 4 2 0France 12 11 13 11 5 0 0 0 0Italy 21 9 1 9 3 1 1 0 0

Great Britain 5 7 7 7 7 4 5 2 1Turkey 27 15 0 1 0 0 0 0 0

Thailand 32 7 1 0 0 0 0 0 0Russia 20 9 2 0 0 0 0 0 0Mexico 23 7 0 0 0 0 0 0 0Spain 9 12 2 3 0 0 0 0 0India 25 1 0 0 0 0 0 0 0

South Korea 1 7 13 2 1 0 0 0 0Canada 10 5 4 3 0 0 0 0 0

Australia 17 5 0 0 0 0 0 0 0Sweden 4 5 8 3 0 1 0 0 0

Indonesia 13 8 0 0 0 0 0 0 0Malaysia 14 6 0 0 0 0 0 0 0

1 2 3 4 5 6 7 8 9 10

Number of airports

105

106

107

Cit

yp

opu

lati

on

Figure 7: Boxplot visualizing the population of cities versus the number of airports a city aggregates. The median population is increasing with thenumber of airports, but there exists no strong correlation over all cities worldwide.

Figure 6 visualizes the number of cities aggregating a given number of airports. It is interesting to note that theperfect powerlaw-fit underlying the results in Figure 6, with the exception of cities with 8–9 airports. Table 1 reportsthe number of cities with 2–10 airports for selected 20 countries. The countries were chosen regarding the totalnumber of cities in this study. It can be seen that cities with many airports are particularly prevalent in United Stated(USA), Germany (DEU), and Great Britain (GBR), followed by China (CHN), France (FRA), Italy (ITA), and SouthKorea (KOR).

In order to analyze the dependency between the number of airports and population, we performed additionalexperiments and the results are shown in Figure 7. It can be seen that the median of the population slightly increaseswith the number of airports aggregated into a city. Nevertheless, it should be noted that the variation of populationis much larger than this trend: For cities with one airport, we have outliers with population ten times as high asthe cities with ten airports. So there exists no strong correlation for all cities. In a similar way, we assess thedependency between the number of airports and the number of destinations and the results are shown in Figure 8.We find that the number of destinations grows significantly with the number of airports; indicating a complementaryoperation. Another explanation is that cities with many airports are located around areas with larger populations and

7

1 2 3 4 5 6 7 8 9 10

Number of airports

0

50

100

150

200

250

300

350

Nu

mb

erof

des

tin

atio

ns

Figure 8: Boxplot visualizing the number of destinations versus the number of airports a city aggregates. The median value of destinations isincreasing significantly with the number of airports, which indicates a complementary operation among different airports.

with manifold travel destinations, which are served by the airports around these cities.In the following, we assess the resilience of cities towards disruptions at their major airports. Here, we choose

the major airport as the airport with the largest number of destinations. During the disruption, we simulate thatall passengers traveling from a city via the disrupted airport have to find an alternative airport offering the samedestination. We distinguish two types of passengers here.

1. The percentage of so-called unaffected passengers.: These passengers are not affected by the airport closure. Alarge fraction of unaffected passengers indicates that the airport disruption has a small effect on the travelers. Asmall fraction, on the other hand, shows that the airport disruption is potentially hazardous.

2. The percentage of so-called reroutable passengers: These passengers can be rerouted through other airportsand still make their journeys under disruptions. A large fraction of reroutable passengers indicates that theairport disruption has a smaller effect on the travelers; while a small fraction of reroutable passengers indicatessignificant operational problems.

Figure 9 visualizes the results of these two types of passengers under disruptions. We can see that the resilience ofcities under disruptions varies considerably, from rather resilient cases, e.g., Yokohama and Moscow, to very fragilecities, e.g., Barcelona, Taipei, and Nairobi. This highlights that the choice of airports around a city, and also theassignment of destinations (overlap) is crucial for obtaining cities resilient to air transportation disruptions.

Figure 10 further aggregates the results at the country level. It can be seen that some countries have rather resilientcities, e.g., Japan (JPN), Russia (RUS), and Italy (ITA). Cities from other countries are much more vulnerable todisruptions at their main airports, e.g., Mexico (MEX), Indonesia (IDN), and India (IND). The country-level insightsare interesting, given that they very likely represent structural properties of the government’s city planning.

4. CONCLUSIONS

In this study, we explored the resilience of cities in the global air transportation network to single airport disrup-tions. The major contributions of our study are summarized as follows.

1. We aggregate airports to cities based on an accurate free-flow driving time estimation, compared to traditionalspatial constraints. As shown in Figure 5, spatial constraints are only an inaccurate proxy for the accessibilityfrom a city center.

8

−0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Unaffected passengers

0.0

0.2

0.4

0.6

0.8

1.0

Rer

outa

ble

pas

sen

gers

Tokyo

Mexico City

Shanghai

Sao Paulo

New YorkBuenos AiresMoscow

Istanbul

Seoul

Beijing

Jakarta

Los Angeles

London

Tehran

Manila

Osaka

Rio de Janeiro

Lahore

Guangzhou

Chicago

Bangkok

Chennai

Tianjin

Paris

Taipei

Lagos

Toronto

Ahmedabad

DongguanShenzhen

Shenyeng

HoustonChengdu

Belo Horizonte

Pune

YokohamaPhiladelphia

Busan

Nanjing

Barcelona

Berlin

Casablanca

Dallas

Miami

Hanoi

Nairobi

New Taipei

Quezon City

Johannesburg

Durban

Figure 9: Effects of disruptions in the largest airport of a city (according to the number of destinations).

2. We report statistical properties about the derived city-airport regions, including the country-level distributions,correlation between the number of airports and city population (Figure 7) versus the number of destinations(Figure 8). The latter is more positively correlated with the number of airports than the former. Furthermore,we find that the cities follow a powerlaw-distribution, regarding the number of aggregated airports (Figure 6).

3. We identify the general phase transition for cities when their largest airports are disrupted (Figure 9), distin-guishing the unaffected passengers and the reroutable passengers. The majority of cities can be found along alog-curve.

In addition to demonstrating that (open) big data can empower researchers to perform transport network-resilienceanalysis at an unprecedented level of scale, our study yields useful implications for the industry and governments.Given that resilience is an important performance indicator for an airport or a city, our analysis will help airportmanagers and policy makers monitor the network performance against that of other airports and cities. Furthermore,presumably management and policy practices will have an impact on the spatial transformation of air transport systemsin the form of, e.g., connectivity and resilience changes. To see such an impact, measuring resilience by quantifyingand visualizing the overwhelming data is a first step. This in turn will assist managers and policy makers for i)better monitoring and planning of their initiatives, and ii) optimal management and control of the transformation. Theimplications include facilitation of networking and routing decisions, identification of business opportunities (e.g.if a city is ranked low, there may be opportunities for improvements and investments), and designing strategies toimprove the competitive positons of airports/cities. Another important implication of our analysis is for assessing thecompetitive airport environment that each city faces. Effective competition of an airport may come not only fromairports in the same city, but also from airports located in neighbour cities. Moreover, multiple airports in the samecity can help each other in case of major disruptions. Management and policy practices should be designed with thiscomplementary in mind.

For our study, we made a set of simplifying assumptions; some of which are caused by lack of data and otherssimply by the planet-wide coverage of this study. Future work could address these limitations as follows:

1. This study measured the accessibility of airports based on driving time; the literature, usually only computesline-by-sight distance-based accessibility for large-scale studies. Future work could take into account the actualtravel costs, degree of convenience, travel frequency and other factors. However, it should be noted that obtain-ing such data at planet-scale is very challenging, given the lack of a commonly used, global database for suchpurposes.

9

USA CHN BRA JPN DEU FRA ITA GBR TUR THA RUS MEX ESP IND KOR CAN AUS SWE IDN MYS

Country

0.0

0.2

0.4

0.6

0.8

1.0

Fra

ctio

nof

pass

enge

rs

unaffected

reroutable

Figure 10: Effects of disruptions in the largest airport of a city, aggregated at the country level, showing the top 20 countries as in Table 1.

2. Similarly, any disruptions will have different effects on different passenger types. For instance, business/firstclass passengers will probably prefer other transportation options than the majority of economy class passen-gers. With additional ticket/price data, future work could analyze this effect further. Moreover, even the typeof airline could be used for making decisions regarding the viable alternatives, e.g., low-cost carrier versustraditional airlines.

3. We did not consider a time constraint for the connectivity during disruptions, since from the data availableto us we can only tell whether two airports are connected and how many passengers took that connection.Nevertheless, a review of the well-established literature on airport connectivity (e.g., (Burghouwt and Veldhuis2006, Zhu, Zhang, and Zhang 2018, Burghouwt and de Wit 2005)) indicates that a viable connection mayinvolve some realistic assumptions such as the maximum transit time (e.g., within 4 hours) and maximumnumber of stops allowed; for a literature survey see (Burghouwt and Redondi 2013). The incorporation offurther connection criteria is an important area of future research.

4. Throughout this study we assume that cross-border traffic takes place inside Schengen area only. In reality,other countries could be taken into consideration, e.g., UK and England, under an airport disruption at eitherBelfast or Dublin. Future studies could perform a more fine-grained analysis with an improved distinction ofviable cross-border traffic.

5. All airports in our study have a sufficient infrastructure to cope with typical international services. The onlyexception are wide-body aircraft such as A380. Given that we do not have information about aircraft types andrunway configurations for all airports at planet scale, we did not make decisions regarding specific fleets andthe availability of adequate runways on each airport. It is an interesting and important extension, which couldbe considered in future work, given available data.

6. We analyze the robustness of air transportation on its own. This is motivated by the lack of infrastruc-ture/schedule data for alternative transportation modes around the world. Future work can aim for a com-prehensive analysis of multi-modal robustness at larger scales. But the existing, available datasets do not allowfor such studies yet. Only giant companies like Google could perform such studies at planet scale, and still facelarge computational challenges. Our study should be seen as a first step towards planet-scale city resilience ofair transportation.

Acknowledgement

This study is supported by the National Natural Science Foundation of China (Grants No. 61861136005, No.61851110763, No. 61601013, No. 71731001).

10

References

Barabasi AL, 2013 Network science. Philosophical Transactions of the Royal Society A: Mathematical,Physical and EngineeringSciences 371(1987), URL http://dx.doi.org/10.1098/rsta.2012.0375.

Bonnefoy P, de Neufville R, Hansman R, 2010 Evolution and development of multiairport systems: Worldwide perspective. Journalof Transportation Engineering 136(11):1021–1029.

Burghouwt G, de Wit J, 2005 Temporal configurations of european airline networks. Journal of Air Transport Management11(3):185 – 198, URL http://dx.doi.org/https://doi.org/10.1016/j.jairtraman.2004.08.003, eighth An-nual Air Transport Society Conference, Toulouse, July 2003.

Burghouwt G, Redondi R, 2013 Connectivity in air transport networks: An assessment of models and applications. Journal ofTransport Economics and Policy 47:35–53.

Burghouwt G, Veldhuis J, 2006 The competitive position of hub airports in the transatlantic market. European Constitutional LawReview - EUR CONST LAW REV 11:106–130.

Caschili S, Medda FR, Wilson A, 2015 An interdependent multi-layer model: Resilience of international networks. Networks andSpatial Economics URL http://dx.doi.org/10.1007/s11067-014-9274-2.

Caschili S, Reggiani A, Medda F, 2015 Resilience and vulnerability of spatial economic networks. Networks and Spatial Economics1–6, URL http://dx.doi.org/10.1007/s11067-015-9283-9.

de Neufville R, 1986 Multi-airport systems in metropolitan regions. Technical report, Massachusetts Institute of Technology.

European Commission, 2001 European Aeronautics: a Vision for 2020. Meeting Society’s Needs and Winning Global Leadership(2001). Luxembourg: Office for Official Publications of the European Communities.

Fleurquin P, Ramasco J, Eguiluz V, 2013 Data-driven modeling of systemic delay propagation under severe meteorological con-ditions. Tenth USA/Europe Air Traffic Management Research and Development Seminar, URL http://arxiv.org/ftp/

arxiv/papers/1308/1308.0438.pdf.

Gomes MFC, Pastore, Rossi L, Chao D, Longini I, Halloran ME, Vespignani A, 2014 Assessing the International Spreading RiskAssociated with the 2014 West African Ebola Outbreak. PLoS Currents URL http://dx.doi.org/10.1371/currents.

outbreaks.cd818f63d40e24aef769dda7df9e0da5.

Haklay M, Weber P, 2008 Openstreetmap: User-generated street maps. Pervasive Computing, IEEE 7(4):12–18.

Hansen M, Du Q, 1993 Modeling multiple airport system: A positive feedback approach. Technical report, Institute of Transporta-tion Studies, University of California.

Kotegawa T, DeLaurentis D, Noonan K, Post J, 2011 Impact of commercial airline network evolution on the u.s. air transportationsystem. Ninth USA/Europe Air Traffic Management Research and Development Seminar, Berlin, Germany.

Loo BP, Ho H, Wong S, 2005 An application of the continuous equilibrium modelling approach in understanding the geographyof air passenger flows in a multi-airport region. Applied Geography 25(2):169 – 199, URL http://dx.doi.org/https:

//doi.org/10.1016/j.apgeog.2005.03.008.

Luxen D, Vetter C, 2011 Real-time routing with openstreetmap data. Proceedings of the 19th ACM SIGSPATIAL InternationalConference on Advances in Geographic Information Systems, 513–516, GIS ’11 (New York, NY, USA: ACM), ISBN 978-1-4503-1031-4, URL http://dx.doi.org/10.1145/2093973.2094062.

Neis P, Zielstra D, 2014 Recent developments and future trends in volunteered geographic information research: The case ofopenstreetmap. Future Internet 6(1):76–106.

Newman M, 2010 Networks-An Introduction (Oxford University Press).

O’Connor K, Fuellhart K, 2016 Airports and regional air transport markets: A new perspective. Journal of Transport Geography53:78 – 82, URL http://dx.doi.org/http://dx.doi.org/10.1016/j.jtrangeo.2015.10.010.

Poppinga M, 2018 Large Scale Accessibility Analysis Using OpenStreetMap Data. Ph.D. thesis, URL http://dx.doi.org/10.

13140/RG.2.2.12272.51204.

Randrianarisoa LM, Zhang A, 2019 Adaptation to climate change effects and competition between ports: Invest now or later?Transportation Research Part B: Methodological 123:279 – 322, URL http://dx.doi.org/https://doi.org/10.

1016/j.trb.2019.03.016.

Reggiani A, Nijkamp P, Lanzi D, 2015 Transport resilience and vulnerability: The role of connectivity. Transportation ResearchPart A: Policy and Practice 81:4 – 15, URL http://dx.doi.org/https://doi.org/10.1016/j.tra.2014.12.012,resilience of Networks.

11

Sun X, Wandelt S, Cao X, 2017 On node criticality in air transportation networks. Networks and Spatial Economics 17(3):737–761.

Sun X, Wandelt S, Hansen M, Li A, 2017 Multiple airport regions based on inter-airport temporal distances. TransportationResearch Part E: Logistics and Transportation Review 101:84–98.

Sun X, Wandelt S, Linke F, 2017 On the topology of air navigation route systems. Proceedings of the Institution of Civil Engineers-Transport, volume 170, 46–59.

Takebayashi M, Onishi M, 2018 Managing reliever gateway airports with high-speed rail network. Transportation Research PartA: Policy and Practice 118:133 – 146, URL http://dx.doi.org/https://doi.org/10.1016/j.tra.2018.08.036.

Vitali S, Cipolla M, Micciche S, Mantegna R, Gurtner G, Lillo F, Beato V, Pozzi S, 2012 Statistical regularities in ATM: Networkproperties, trajectory deviations and delays. Second SESAR Innovation Days.

Wandelt S, Sun X, 2015 Evolution of the international air transportation country network from 2002 to 2013. TransportationResearch Part E: Logistics and Transportation Review 82:55 – 78, URL http://dx.doi.org/https://doi.org/10.

1016/j.tre.2015.08.002.

Wandelt S, Sun X, Zanin M, Havlin S, 2018 QRE: Quick robustness estimation for large complex networks. Future GenerationComputer Systems 83:413 – 424, URL http://dx.doi.org/https://doi.org/10.1016/j.future.2017.02.018.

Wandelt S, Sun X, Zhang J, 2019 Evolution of domestic airport networks: a review and comparative analysis. TransportmetricaB: Transport Dynamics 7(1):1–17, URL http://dx.doi.org/10.1080/21680566.2017.1301274.

Wandelt S, Wang Z, Sun X, 2017 Worldwide railway skeleton network: Extraction methodology and preliminary analysis. IEEETransactions on Intelligent Transportation Systems 18(8):2206–2216, URL http://dx.doi.org/10.1109/TITS.2016.

2632998.

Wang K, Zhang A, 2018 Climate change, natural disasters and adaptation investments: Inter- and intra-port competition andcooperation. Transportation Research Part B: Methodological 117:158 – 189, URL http://dx.doi.org/https://doi.

org/10.1016/j.trb.2018.08.003.

Wilkinson S, Dunn S, Ma S, 2012 The vulnerability of the european air traffic network to spatial hazards. Natural Hazards60(3):1027–1036, URL http://dx.doi.org/10.1007/s11069-011-9885-6.

Wong W, Cheung T, Zhang A, Wang Y, 2019 Is spatial dispersal the dominant trend in air transport development? a globalanalysis for 2006–2015. Journal of Air Transport Management 74:1 – 12, URL http://dx.doi.org/https://doi.

org/10.1016/j.jairtraman.2018.09.011.

Xia W, Jiang C, Wang K, Zhang A, 2019 Air-rail revenue sharing in a multi-airport system: Effects on traffic and social wel-fare. Transportation Research Part B: Methodological 121:304 – 319, URL http://dx.doi.org/https://doi.org/

10.1016/j.trb.2018.10.002.

Xiao Y, Fu X, Ng AK, Zhang A, 2015 Port investments on coastal and marine disasters prevention: Economic modeling andimplications. Transportation Research Part B: Methodological 78:202 – 221, URL http://dx.doi.org/https://doi.

org/10.1016/j.trb.2015.04.009.

Zanin M, Lillo F, 2013 Modelling the air transport with complex networks: A short review. The European Physical Journal SpecialTopics 215(1):5–21, URL http://dx.doi.org/10.1140/epjst/e2013-01711-9.

Zhu Z, Zhang A, Zhang Y, 2018 Connectivity of intercity passenger transportation in china: A multi-modal and network approach.Journal of Transport Geography 71:263 – 276, URL http://dx.doi.org/https://doi.org/10.1016/j.jtrangeo.

2017.05.009.

12