Reconstructing geography from airline connections
What does the world look like if you draw it based on airline connections instead of geography? Take 1,087 major airports and the 14,570 direct routes between them. Forget latitude and longitude. Compute the shortest path (in number of flights) between every pair of airports, then use multidimensional scaling to reconstruct a 2D map from these distances alone.
The result is a distorted mirror of the real world. Recognizable, but warped — stretched and compressed by the invisible force of connectivity. Hub cities pull inward, poorly connected regions drift to the edges, and political barriers carve gaps where geography says there should be continuity.
Drag the slider below to warp the world. On the left: real geography — airports at their true latitude and longitude. On the right: the flight network's view — positions reconstructed purely from connection distances. Watch how Europe inflates (it's disproportionately connected), the Pacific compresses (few routes span it directly), and hub cities migrate toward the center.
The flight network is strikingly small. Every airport in this dataset can reach every other in at most 6 flights. But the distribution is far from uniform: over half of all airport pairs are connected by exactly 3 hops, and only 2.5% have direct flights. The average path length is 2.814 hops.
This is the small-world property in action. The handful of mega-hubs — Frankfurt, Amsterdam, Paris — collapse the diameter of the network. Remove them and the world gets much bigger.
The flight network is not a democracy. A handful of airports control access to the world. Frankfurt, with 241 direct connections, can reach 94.5% of the network within just 2 hops. It would take only 3 flights to get from Frankfurt to essentially anywhere on Earth with commercial service.
This extreme hub dominance has a geometric consequence: in the reconstructed map, hub airports are pulled toward the center because they're equidistant (in network terms) from everything. Istanbul has the least displacement of any airport in the dataset — it sits close to its true geographic position because its network position happens to match its geographic one. It is, in a very real sense, the center of the connected world.
| Airport | City | Direct | 2-hop reach |
|---|
Geography says Armenia and Azerbaijan are neighbors. Aviation says they don't exist to each other. Despite being only 305 km apart and collectively having 7 airports, there are zero direct flights between them — a direct consequence of the frozen Nagorno-Karabakh conflict.
The flight network reveals geopolitical fault lines with striking clarity. These are country pairs that are geographically close (within 600 km) but have no direct air connections at all:
| Country 1 | Country 2 | Distance | Airports |
|---|
The pattern is clear: post-Soviet frozen conflicts (Armenia–Azerbaijan, Georgia–Armenia, Uzbekistan–Tajikistan), Balkan fragmentation (Montenegro, Bosnia, Croatia, Serbia against Slovakia), and Caribbean island isolation. The map doesn't just show where planes fly — it shows where political walls stand.
Not all parts of the world are equally "shrunken" by aviation. Europe, with its dense network of low-cost carriers, is the most internally connected region: 359 airports with an average degree of 38.1 connections each. Compare this to Africa, with 58 airports averaging only 13.4 connections.
The inter-region picture is equally revealing. Despite being on opposite sides of the planet, North America and Europe average only 2.90 hops apart. But Asia and South America — a comparable geographic distance — average 3.47 hops. The flight network reflects not just distance, but economic ties, colonial history, and tourism patterns.
The single most "overconnected" pair in the dataset: Sydney to Dallas-Fort Worth — 13,808 km in a single direct flight. That's one hop spanning more than a third of Earth's circumference.
The most "underconnected": Ciudad Juarez to El Paso — cities that are 20 km apart (you can walk between them) but require 3 flights to connect by air. The US-Mexico border is thin on the ground but thick in the sky.
The correlation between flight-network distance and geographic distance is r = 0.64. That means the network captures about 41% of the variance in real-world geography — enough to be recognizable, but far from complete. The remaining 59% is where it gets interesting.
The distortions in the reconstructed map are not noise. They are signal — a precise encoding of where the world is more connected than geography would predict, and where it is less. Europe inflates because low-cost carriers have made it a dense mesh. The Pacific compresses because only a handful of ultra-long-haul routes cross it. Africa drifts because its internal connectivity is sparse, and most of its connections route through European or Middle Eastern hubs.
The flight network is, in a sense, a map of human priorities. We connect places we trade with, vacation in, emigrate to. The routes are drawn by economics, politics, and desire — not by Euclidean distance. The map that results is the world as we use it, rather than the world as it is.