October 28, 2016
The view from 2 million feet
How ultra-detailed satellite data are opening up new frontiers for economists
Deforestation in the remote Brazilian state of Rondônia. This series of images was captured by NASA's Moderate Resolution Imaging Spectroradiometer (MODIS), 2001-2012.
Rebecca Lindsey, Robert Simmon, and Reto Stöckli/NASA
So much of modern economic research relies on accurate data. But what happens when you are surveying a Nairobi slum where your research assistants are viewed with suspicion and must contend with pickpockets and speeding buses? Or trying to study illicit activity like deforestation in Indonesia where corrupt local officials may have been persuaded to cook the books?
Increasingly, economists are turning to the skies – and reams of newly accessible satellite data that can reveal economic patterns that aren’t always visible from ground level.
In an article appearing in the Fall issue of the Journal of Economic Perspectives, authors Dave Donaldson and Adam Storeygard discuss new applications of remote sensing technologies like satellites, which are poised to play a bigger role in social science research in the decade to come.
Engineering advances in remote sensing technology mean that more and more powerful satellites are launched into orbit each year, just as new cloud computing methods and platforms like Google Earth Engine provide better means to process the gargantuan data streams that satellites generate.
Meanwhile, new platforms for U.S. government satellite data are allowing easier access and regulatory changes are allowing American companies to sell fine-grained satellite imagery to non-governmental customers for the first time.
Private companies are already using these new data sources to gain a competitive edge, for instance by counting cars in parking lots to estimate demand in real time, or scouting far-flung locations in inner Mongolia for business opportunities.
Researchers are getting in on the game too. Even though it would have been pretty novel a decade ago, now “you are no longer surprised in a seminar if someone says ‘I got a variable I am using from satellite data,’” says Donaldson, one of the article’s authors who has used satellite data to evaluate the potential impact of climate change on agricultural productivity around the world. He credits an attention-getting 2012 paper from his co-author Storeygard on the connection between nighttime illumination and economic growth with helping economists to understand the vast potential of these new data sources.
Remote sensing can help economists get around some thorny data problems, like questionable statistics from government sources or data from remote, inaccessible areas that would be too expensive to collect in traditional ways. Satellite data isn’t interrupted when civil servants go on strike or when a region descends into conflict. It can also help bridge what is being called the “data divide” – a growing disparity between the quality of government-collected data in the developed and developing worlds.
Some of the more recent applications have used satellite data in conjunction with more traditional economic data, like surveys or government statistics. One pioneering 2003 study explored the re-forestation of India, which like some countries has actually experienced increasing forest cover in recent decades. Judging by where the growth was happening, the authors concluded that this was the result of local economies demanding more timber.
Satellite readings can also be used to infer properties that wouldn’t necessarily be obvious to the naked eye, like the temperature of the ocean or the density of pollutants like nitrogen dioxide. One recent study in the American Economic Journal: Applied Economics used satellite data from NASA’s MODIS program to analyze fishing conditions in the waters off the Indonesian coast and found that poor fishing conditions in an area tended to lead to an uptick in pirate attacks.
This research agenda for economists and other social scientists based on remote sensing data has made great strides in the last decade, but it seems safe to say that the real excitement lies ahead.
Donaldson and Storeygard (2016)
Higher-resolution data from private satellites can be used to analyze objects much smaller than cities, forests, or open oceans. In a study of rental arrangements in a Nairobi slum, the authors use satellite data to assess the quality of individual housing units, specifically looking at the reflectivity of each roof to estimate how recently it had been replaced.
The ability to analyze individual buildings, structures, and farmsteads in daytime satellite imagery holds a lot of promise for answering a whole new set of economic questions. But for a study ranging beyond one neighborhood or one city, the task of making economic sense of the raw data (does this cluster of pixels represent a prosperous town or an impoverished village?) needs to be automated.
The authors argue that economists will need to start borrowing techniques from the fields of machine learning and computer vision to take full advantage of the wealth of information that is becoming available.
One recent study used a technique called “transfer learning” that takes advantage of the well-established connection between nighttime luminosity and economic activity. A model was trained to identify recognizable features in high-resolution satellite images of Africa, like roads and farmland, that were correlated with nighttime luminosity.
The model then showed an ability to predict the relative wealth of various neighborhoods in Uganda simply based on what the daytime satellite images of those neighborhoods revealed. In principle, such a model could be used to estimate economic well-being across a large area at a very high resolution – without the need for human input. As more and better satellite data becomes available every year, these approaches will be an increasingly important part of the economist’s toolkit.
The View from Above: Applications of Satellite Data in Economics appears in the Fall 2016 issue of the Journal of Economic Perspectives.