Around the World in 2.2 Million Stops

Chris Shughrue - December 9, 2019

There may be 150 million places in the world. What if you wanted to visit all of them? How would you plan this epic trip—and make sure it didn’t take a lifetime?

To answer this question, we consulted the models we’ve built. These models predict the location of all the places in the world, including places that haven’t yet been mapped.

Figure 1. Share of earth’s surface coverage. Areas occupied by places accounts for less than 1% of the planet.

Let’s tackle this problem by starting at a far zoom Powers of Ten style. Earth’s surface area is 196 million mi2, of which 29% is land. Humans directly modify and control 55% of this vast area. Most human-controlled land is used for farming and livestock—not especially inviting stops for a world tour.

Human settlements—cities, towns, and villages—occupy only a small fraction of the world, approximately 1.7% of Earth’s surface (Figure 1). For now, we’ll focus on the land actually inhabited by humans.

Figure 2. Global place density predictions (top). Insets show detailed predictions for San Francisco, USA (lower left) and Kuala Lumpur, Malaysia (lower right).

Our predictive modeling takes over from here. We can break the globe down into 691 million small hexagonal tiles and predict how many places are located in each of them (see more in Methods below). Next, we can predict which of these tiles are occupied by places, and how many places we expect to show up in each (Figure 2).

It turns out that of these 691 million tiles, only 2.2 million are occupied by places. This means the planetary place-footprint accounts for only 0.3% of the earth’s surface. This seemingly tiny footprint makes sense when we consider that places tend to be clustered, even within cities and towns. Residential areas, highways, and parking lots, for example, cover large areas but contain very few places to visit.

So where exactly are the places?

Most tiles with places are in cities -- not surprising given that now more than half of the world’s population lives in urban areas. Some regions of the world are more densely urbanized than others, giving rise to uneven patterns in the distribution of places. In regions such as Southern and South-Eastern Asia, about three-quarters of places are in urban areas, whereas Southern and Northern Africa are less urbanized and have the majority of places outside of cities (Figure 3).

Figure 3. Fraction of places located in urban areas by region. In a majority-urban world, places are mostly situated in cities, except in Northern and Southern Africa.

How does this help us plan our trip? It turns out places in urban areas are more densely clustered together. Globally, our models show, tiles in urban areas have nearly twice as many places as those outside urban areas. So before we worry about scouring the countryside for places, starting in cities is a quick way to cover a lot of ground. And in particular, parts of Asia and Oceania might be great places to start.

But, of course, not all cities are the same. Social and economic aspects of cities appear to be highly skewed, a pattern known as Zipf’s Law. For example, the most populous city in a country generally has about twice as many people as the second-most-populous city. The second-most-populous city will have significantly more inhabitants than the third city, and so on. Most cities have relatively few people compared to the biggest cities (think New York versus a typical American city such as Lynchburg, VA).

We observe the same kind of pattern in the distribution of places (Figure 4). The 25 most place-dense cities account for 14% of all the places in the world. That means you could visit a seventh of all the places in the world in just 25 trips between particular cities. We could use this information to focus our tour around hotspots of activity, starting in the cities with the dominant share of places and working our way down the list.

Figure 4. Share of places in each city versus the ranking of the city from most to fewest places. The top 10 most place-dense cities account for 16% of all the places in cities in the world.

We can extend this example to think about specific tiles, rather than grouping by city or region. Let’s imagine we could teleport between the densest city block in Shanghai to the densest street in Mumbai. If we could line the tiles up in this way, from most to fewest places, we could shorten our trip significantly (Figure 5). After visiting only 18% of these carefully ordered tiles, we would have seen more than half of all the places in the world. By the time we made it to half of the tiles, we’d have seen more than 75% of the places in the world. After that, things would start to slow down on our journey as we visited less densely packed tiles.

Figure 5. Fraction of places encountered versus the fraction of all tiles visited in our hypothetical optimally efficient journey. By arranging our stops in order of most to least dense, our return on trips starts off high.

Though this thought experiment may not provide a concrete itinerary to navigate the world, it does highlight an application of StreetCred technology. The vast majority of the potential places we would want to visit haven’t been mapped, but we can use this kind of analysis to help figure out how to most efficiently map the world. After that, it’s smooth sailing (or just 2.2 million point-to-point flights) to see it all.

Methods

We developed a global machine learning model to predict the density of POIs for each of 691 million hexagonal H3 tiles. Our models make predictions based on VIIRS nightime light data, satellite-derived historic land-cover and land-use data, street network statistics, and administrative geography such as city, world region, and development level. Models are trained on a combination of StreetCred data and OSM point of interest data which have been adjusted for undersampling where appropriate using a Leslie Depletion method.