New technologies like satellite imagery offer a promising remedy to this blind spot. The resolutions of satellite images vary, but in general they have increased over the years. In the 1990s, a single pixel of a typical image might have represented about 30 square meters of land area. By 2000, 10 or fewer meters a pixel was commonplace, and now 0.5 meters (1.5 feet) of resolution is possible. Today’s images also contain more data, making it possible to discern building heights, plant species, and other environmental details. Most importantly, the satellites are constantly circling the Earth, gathering consistent, reliable, and low-cost data. Collecting information remotely from machines can allow for wider coverage of urban regions, and consistent collection allows for better benchmarking and detection of changes. We can see changes in the larger urban region better than ever.
For urbanists, the greatest bottleneck is how to process this data. While we are making advances in our ability to manipulate and interpret the data, our models risk becoming disconnected from the situation on the ground. Researchers primarily conduct this computational research sitting in an office, ultimately making assumptions about what they are seeing in the data.
One common assumption is that shiny, dry things are urban. That is, urban spaces are those that use construction materials that are not found in rural areas, such as concrete. From the satellite view, urban spaces are also assumed to exhibit geometric spatial patterns, such as straight roads and contiguous, rectilinear buildings. These assumptions might hold for planned, formal urban development, as in the center of a city. But informal urban structures might use less shiny, wetter materials, such as plant materials. These self-built buildings might also might not line up in a straight row, spaced at regular intervals.
To overcome these issues, my research group trained our maps on a computational model derived from on-the-ground fieldwork, commonly known as ground-truthing. I tromped out into the swampy south side of Saigon confirming that, yep, this blurry blip in the picture is definitely a house. We adapted an algorithm to reflect this reality, by allowing things that were slightly less shiny to be classified as urban if the texture of the area was too variegated to be rural vegetation. Our results recovered 12 percent more newly urbanized land area defined by the parameters determined through field survey than would have been discerned using conventional methods.
To find out why our results differed, we convened a panel with the other two research groups at a World Bank conference in 2017. We wanted to understand how our different results could impact urban management, and how technologies for urban research might adjust accordingly. We discovered that differences between the three groups’ studies emanated from the researchers’ particular interests in using satellite imagery. That is to say, the results they achieved were fundamentally shaped by the goals with which they began their urban data-analysis effort.