Researchers used the technique to automatically generate the first nationwide crop map of Thailand, a country of smallholder farmers where small, independent farms are the predominant form of agriculture. The team created a border-to-border map of Thailand’s four main crops—rice, cassava, sugarcane, and maize—and determined which of the four was being grown, every 10 meters and without gaps, across the country. The resulting map achieved 93 percent accuracy, which the researchers say is comparable to on-the-ground mapping efforts in high-income countries with large farms.

The team is applying its mapping technique to other countries such as India, where small farms support most of the population, but historically the types of crops grown from farm to farm have been poorly recorded.

“It’s a huge data gap in our knowledge of what’s being grown around the world,” says Sherrie Wang, the d’Arbeloff assistant professor of professional development in MIT’s Department of Mechanical Engineering and the Institute for Data, Systems, and Society (IDSS). “The ultimate goal is to understand agricultural outcomes, such as yield, and how to farm more sustainably. One of the key preliminary steps is to map what’s even being grown—the more detailed the mapping, the more questions you can answer.”.

Wang, along with MIT graduate student Jordi Laguarta Soler and Thomas Friedel of the agricultural technology company PEAT GmbH, will present a paper detailing their mapping method later this month at the AAAI Conference on Artificial Intelligence.

Small farms are typically run by a single family or farmer who subsists on the crops and livestock they raise. It is estimated that small farms support two-thirds of the world's rural population and produce 80 percent of the world's food. Monitoring what is grown and where is essential for tracking and forecasting the global food supply. But most of these small farms are located in low- and middle-income countries, where few resources are dedicated to tracking crop types and yields on individual farms.

Crop mapping efforts are primarily conducted in high-income regions such as the United States and Europe, where government agricultural agencies oversee crop surveys and send assessors to farms to label crops from field to field. These "on-the-ground truth" labels are then fed into machine learning models that establish connections between the on-the-ground labels of actual crops and satellite signals from the same fields. These models then label and map wider swaths of farmland that assessors don't cover but that satellites automatically do.

“What’s missing in low- and middle-income countries is this terrestrial label that we can associate with satellite signals,” says Laguarta Soler. “First of all, obtaining these basic truths to train a model has been limited in most parts of the world.”.

The team realized that while many developing countries don't have the resources to conduct crop studies, they could use another source of terrestrial data: roadside images, captured by services like Google Street View and Mapillary, which send cars around a region to record continuous 360-degree images with dashcams and roof-mounted cameras.

In recent years, these services have gained access to low- and middle-income countries. While the purpose of these services is not specifically to capture crop images, the MIT team found they could search roadside images to identify crops.

Cropped image

In their new study, the researchers worked with Google Street View (GSV) images taken across Thailand, a country that the service has recently photographed in considerable detail and which is predominantly made up of small farms.

From over 200,000 GSV images randomly taken across Thailand, the team filtered out images depicting buildings, trees, and general vegetation. Around 81,000 images were related to crops. They selected 2,000 of these and sent them to an agronomist, who identified and labeled them