Researchers try to map Africa farms

One square kilometer at a time, a group of University researchers is working to map farmland across sub-Saharan Africa.

The “Mapping Africa” team, led by Wilson School researcher Lyndon Estes and civil and environmental engineering professor Kelly Caylor, has developed a method to crowdsource the project by allowing Internet users to examine 1-kilometer squares of sub-Saharan African terrain from Google satellite imagery. These users can then specify whether they think the land in the square is being actively farmed, based on the image, in return for a small monetary reward: 10 cents a square.

Crowdsourcing refers to any project that distributes a task among a large number of participants. Caylor said he believes this approach will far outperform any existing automated process.

“Right now, we don’t have an effective algorithm to accomplish what the human eye can do,” he said. “We haven’t figured out a better way to do it, so right now it seems that people outmatch computers when it comes to identifying farmland.”

“It’s reassuring to think that all that evolution has prepared us for something,” he joked.

The only existing measurements of active farmland in sub-Saharan Africa consist of government reports — which can be incomplete or outdated — and large-scale, global surface cover surveys heavily reliant on satellite imagery, the research team said. Once the team’s data set is complete, cartographers might be able to use it to better understand defects in existing maps, according to Caylor.

“Satellite imaging is often too fuzzy, but many governments just don’t have the resources to produce high-resolution mapping,” he said. “By being savvy with the maps that we do have, we can improve the maps where they are the worst and supplement them in a strategic way with our own.”

Estes also indicated that the improved agricultural map would provide future insights into changing practices on the continent. Estes explained that the African agricultural sector is growing to meet increased food demand due to population growth. The continent is one of the few places with open land that could sustain increased production, he said.

“Since we’re trying to understand what’s going to happen to agriculture in Africa, it’s really important to have accurate baseline statistics describing where and how much agriculture currently exists in Africa,” he said.

Jonathan Choi ’15, a prospective ecology and evolutionary biology major who is also a member of the research team, said the “Mapping Africa” project would also create a database of information that might be used for future studies in a variety of fields.

“One potential application is global climate change monitoring,” he said. “Having a better understanding of where farms are will give us a better understanding of climate modeling in addition to future agricultural modeling and predictions.”

The “Mapping Africa” project is still in the early stages of development, and the team plans to complete mapping of South Africa before extending to other countries. Eventually, the team hopes to include all of Sub-Saharan Africa in the study.

Get the best of ‘the Prince’ delivered straight to your inbox. Subscribe now »

Both the prototype, to be released this month, and the final version will be hosted on Amazon’s Mechanical Turk service, which allows researchers or companies to offer small sums of money to users in return for completing “Human Intelligence Tasks.” These tasks can range from transcribing an audio file to defining Twitter hashtags or, in the case of “Mapping Africa,” identifying actively farmed land.

“We’ve seen it used to improve classical psychology experiments, which now have access to many more people than were previously available,” Estes commented. “We’re taking advantage of this same idea, especially since our task lends itself to being done on the Internet, given the availability of the Google maps platform.”

But beyond characterization of farmland, Caylor said he believes the model could be used to increase the resolution of a whole host of geographic data.

“What we’re doing here isn’t limited to farmland in Africa. There’s no reason there couldn’t be similar applications on other continents, like in the mapping of rainforest cover in South America,” Caylor said. “While we hope the project doesn’t grow faster than our ability to support it, we’re excited to see how far we can take this.”

In addition to Estes, Caylor and Choi, Office of Information Technology analyst Dennis McRitchie, Wilson School lecturer William Guthe and Reka Zempleni ’16 serve on the “Mapping Africa” team.