Brazilian Air Traffic and the COVID-19 Pandemic
Analyzing a complex flight network with data visualization techniques
Social distancing is a major strategy used worldwide for containing the spread of corona virus since this approach was indicated by the World Health Organization at early 2020. In Brazil, student classes and face-to-face services, as regular flights, were suspended in march of the same year, delimiting the beginning of the quarantine, which lasts until today in the country.
In this scenario, this study analyze the national flights network through 2019 and 2020 (before and during quarantine) comparing data from these two years in order to verify how this social service was impacted by the pandemic.
Air Traffic
Data about the brazilian national regular flights was collected from a public database available at ANAC (Agência Nacional de Aviação Civil) website. Also, for better spatial visualization, airport geographical coordinates were taken from this public dataset at Datahub.
Analyzing this data, its evident that the COVID-19 pandemic indeed impacted the Brazilian air traffic, drastically decreasing the number of flights (about half a million) as shown on the chart below.
During these years, the two airports that had more flights are from São Paulo, the Brazilian main financial state. However, studying airports alone do not offer information about the whole national air traffic system.
The Network
It is possible to establish a connection between airports to generate a network structure that allows analyzing the air traffic relationship across the country. In this case, airports can represent nodes and their connections can be made by flights. So, given two airports, they will be linked if there were at least one flight between then. Applying this behavior, it is possible to plot the network structure, as it is shown in graphs below. Each black dot is an airport and gray lines are flights.
Eccentricity
In a network, eccentricity is set by the maximum distance (number of links or “steps”) of each node to all other ones. Considering this metric, the minimum value of eccentricity is named radius and the maximum, diameter. If the node’s eccentricity value is equal to the radius, it will be in the center. Nevertheless, if the node’s eccentricity is equal to the diameter, it will be in the network periphery.
For instance, assume the hypothetical network drawn. From node d, the max number of steps to reach any node is 3. Meanwhile, for node a, it takes at most 2 steps to reach each other node. So, d eccentricity is 3 and a is 2. In this case, across all eccentricities, the minimum value is 2 (radius). Then, nodes that have this eccentricity are in network center. And the maximum value is 3 (diameter) which defines the periphery.
In air traffic system, the eccentricity of an airport is the max number of flights that is required to reach all other airports. So, it likely need more flights (steps) to fly through a peripheric airport. While, it usually takes less steps to reach a central airport.
For a spatial view, the following charts shows the airports position by their geographical coordinates. Since there is a lot of airports, labels were omitted.
Notice that, before the pandemic, the central nodes were more scattered all around the country. But in the year after, the number of central nodes decreased and only three airports stayed: Aeroporto Internacional de Viracopos at Campinas (SP), Aeroporto Internacional de Brasília (DF) and Aeroporto Internacional de Manaus (AM). Two Brazilian regions did not show any center node in 2020. This indicates that flight lines that once occurred were suspended during the COVID-19 pandemic, decreasing airports connections.
Centrality
An important feature available by analyzing complex networks is the finding of the most important nodes. In order to find then out, it is possible to evaluate a group of metrics known as Network Centralities. This article only embraces three: Degree, Closeness and Betweenness centrality.
In a short brief, Degree Centrality uses the number of connections (degree) a certain node has and how many it could at most have (considering all nodes in the network). This metric is directly proportional to the amount of connections. In air traffic terms, airports that are linked with many others has a bigger degree centrality.
Meanwhile, Closeness Centrality is set by summing the distance (minimum number of flights required) from a node to all other ones. So, this metric relies in the relation of proximity.
Betweenness Centrality is set by evaluating, for each node, if it is a part of the shortest path between all combinations of two any other nodes in the network. This measurement is a appliance for materializing the concept of attendance in the network flows, determining which nodes are more frequent and required for maintaining the majority of connections.
In the national flights network, considering 2019 and 2020, the Aeroporto Internacional de Viracopos at Campinas (SP) is the airport that has the biggest coefficient. This is certainly an indicator of the importance of this airport to the Brazilian national air traffic even during the pandemic.
Correlation
A way to verify and compare the relation between these centrality metrics is plotting their values for all nodes in the network.
In the charts that contain two measurements, it is noticeable a positive proportion relation between then, since that, as a centrality increases the other one seems to rise as well. This evidence that the importance of the airports at the whole air traffic system does not rely only at one aspect. More important nodes have more links, are less flights away from others and are part of the majority of the national itineraries.
The following graphs labels the airports by their ICAO codes. The intensity of the red color and the size of each node is set by the metrics written in the charts.
In these visualizations (zooming in), the regional clustering becomes clear. The betweenness centrality highlight nodes that act like bridges to airports apart. Degree centrality shows off mostly airports from the Brazilian Southeast region, indicating that there is more flights from/to this region. At long last, the closeness centrality is way more similar across the nodes then other measurements. This evidence that there is a few difference in the amount of flights required to cross the country. Even so, the graphs shows that the Brazilian air traffic system seems to be more concentrated around the Southeast and Midwest regions (economic poles). As the data shows, the COVID-19 did decrease the air traffic. There were less flights, less itineraries and the whole system shrank.
