Does AI Hold the Key To A New And Improved “Green Revolution” In Agriculture?
Producing enough healthy food to feed the world—on a changing planet—is going to be a steep challenge. These researchers are giving farmers AI-driven techniques and tools to find solutions.
BY JACKIE SNOW | FEBRUARY 19, 2019 | NOVA NEXT
Automation in agriculture may soon make robots as common in greenhouses as they are on factory floors. Photo credit: Shutterstock
On a stretch of highway in the Netherlands not far outside of Amsterdam, a row of greenhouses at Wageningen University & Research (WUR) poke up like knuckles along the flat landscape. The Dutch university is known for its cutting-edge agricultural research, but some of these greenhouses recently ran an experiment that’s novel even for them: autonomous growing.
Stepping into a humid box from a brisk autumn day, you hear the noises of machines adjusting themselves mixing with the sounds of leaves rustling. The amount of light, water, fertilizers, and carbon dioxide—along with the temperature of the greenhouse—are all set by deep learning algorithms and executed by machines. Humans are still responsible for moving vines up the lattices as they grow, as well as pruning and harvesting.
But it’s pretty clear who—or rather, what—is calling the shots.
The Future of Farming?
By 2050, we’ll need to feed nine billion people with about a third less arable land than we had in the 1970s, experts estimate. Farmers will need all the help they can get, including insights gleaned from artificial intelligence, or AI. Developed carefully—and with the people who will be using it taken into account—AI can be part of the solution to feeding a growing world, according to the Refresh report, a document put together by researchers from Google, university professors, nonprofits, and farmers. And as an added bonus, some of the unsustainable practices developed over the past 70 years could be reversed with more efficient, AI-driven technology.
The Green Revolution was a set of advances that started in the 1950s in areas like high-yield crops, synthetic fertilizers, and irrigation technology that greatly increased food production, especially in developing countries—saving an estimated one billion people from starvation. But it left in its wake a culture of pesticides, reduced agricultural biodiversity, and overuse of chemical fertilizers that deplete the soil and poison waterways.
“It was never meant to be used in the long term,” says Danielle Nierenberg, the president of Food Tank, a non-profit working to build a better food system that also worked on the Refresh document. Farmers were supposed to transition back to organic, Nierenberg adds: It just never happened because increased yields generated by industrial-scale farming put pressure on smaller farms to follow suit.
One of the main ways AI could help agriculture transition out of practices forged in the Green Revolution and into a more sustainable future is with precision farming. Until now, there hasn’t been an easy way for farmers to learn from historical or real-time data. But AI-powered programs can combine data on weather patterns, crop yields, market prices, and more to guide farmers to planting at the right time, adding the appropriate level of fertilizers, and harvesting at peak ripeness.
In a greenhouse at Wageningen University & Research (WUR) in the Netherlands, cucumbers grow with the help of deep-learning algorithms and machines. Photo credit: Dr. Silke Hemming, WUR
WUR is one of the places where big data approaches to growing food are being tested. Last fall, five teams of AI researchers and biologists from around the world competed in growing cucumbers in separate 96-square-meter greenhouses, with a sixth grown manually as a reference. Each team trained its own algorithm, although the teams had the ability to decide how closely to follow the solutions that their AI models came up with. The teams kept an eye on their crops with sensors and cameras, and could feed the algorithms new data and tweak them as needed. To win, teams had to maximize total yield and net profits while minimizing the use of resources.
The winner was a team called Sonoma, made up of Microsoft Research employees and students from Danish and Dutch universities. According to Silke Hemming, head of the scientific research team for greenhouse technology at WUR, Sonoma’s plan used more artificial light earlier and kept carbon dioxide levels higher than a typical gardener might. But other teams also discovered counterintuitive ways to increase yield, such as pruning smaller cucumbers close to harvest or letting bigger ones have a chance to grow a little more.
Like all problems in AI, growing cucumbers and other crops by algorithm demands a food source of its own: data—and lots of it. The cucumber contest was a start at putting information together that other researchers can build on with future projects.
“You have a dataset you would never have,” Hemming says. “You can learn so much from that.”
The researchers organizing the competition chose cucumbers because they are a fast-growing crop cultivated worldwide, and problems like blight show up in them immediately. But this project could transform how other indoor crops are grown. It’s a first step in finding ways to combine humans and AI technology to produce more food, more efficiently.
“It’s not all about winning.” Hemming says. “It’s also about learning.”
FARMWAVE Founder and CEO Craig Ganssle uses its smartphone app with an automated kernel count feature to assess corn yield. Photo credit: FARMWAVE
AI on the Farm
“Farming is a lot more complicated than other industries,” says Joshua Woodard, an agricultural business and finance professor at Cornell and founder of the farming data company Ag-Analytics. “It’s a really complex system of environment and management practices."
Ag-Analytics’s wants to bridge that gap with easy-to-use data analysis tools to help farmers plan and monitor their fields. Their farm management platform takes data from sensors in John Deere farm equipment and combines it with other datasets, like satellite imagery and weather forecasts, to develop predictions for individual farms.
Algorithms working from afar could make a huge impact for less tech-heavy farms, too. Farmers in the developing world are working with minimal data and stand to make leaps in productivity with algorithms in the cloud instead of expensive machinery in their fields. According to the United Nations, 20 to 40 percent of crop yields are lost each year due to pests and diseases. AI tools like Plant Village and FARMWAVE allow farmers to take photos with their phones of sickly plants, bugs, and weeds, and then have computer vision-powered algorithms diagnose the problem from afar in seconds. FARMWAVE is already working with farmers in countries across the world, who, despite their distance, are all dealing with similar problems that AI can spot.
"Army worm in corn looks the same in Africa versus the U.S.,” says Craig Ganssle, the founder and CEO of FARMWAVE.
In India, a team at the International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) is working on providing real-time pest predictionsto help Indian farmers take specific actions to protect their crops. ICRISAT uses cloud computing, machine learning, and data from IoT (short for the “Internet of Things”) sensors to come up with personalized predictions about pest risks.
Dr Avijit Tarafdar of ICRISAT converses with chickpea farmer Mr Srinivasa Boreddy in Adilabad District, Telangana. Photo credit: D Chobe, ICRISAT
“Whenever [farmers] see the pests in the field, they simply go for pesticides,” says Dr. Mamta Sharma, a principal scientist at ICRISAT. “It will help them reduce the amount of sprays that farmers are applying."
ISCRISAT has offices in Africa that could eventually use the tool, with interest coming from South America as well. As these offices collect more data, Sharma says, it could be used to spot new risks due to climate change.
“It helps us recognize emerging threats,” she says.
Robot Green Thumbs
Indoor farming currently occupies around 2.3 million square feet worldwide. But based on information from growers, the analysis firm Agrilyst predictsthis number will balloon to 22 million square feet over the next five years. Despite the expense of setting up these spaces and the limited types of produce that can currently be profitably grown, much of AI research is being done in greenhouses and other indoor spaces because, with the reduction of arable land, these production methods will become more critical. Indoor farming can also produce up to 20 times as much fruit and vegetables per square foot as outdoor farming, while using up to 92 percent less water, according to one study, with one company claiming it needs 99 percent less water.
In San Carlos, California, two robots cruise within a hydroponic farm developed by the start-up Iron Ox. These robots, which plan, care for, and harvest produce, are overseen by a computer program affectionately nicknamed “the Brain.” Even before the advent of AI, hydroponic systems were known to use less water, need fewer pesticides, grow faster, and produce more plants in less space. However, hydroponics are notoriously labor-intensive, requiring plants to be moved to different vats throughout the growing phase. Training robots for this monotonous task could make razor-thin profit margins a little less tight.
“A lot of things that weren’t feasible outside of a lab five years ago are possible now,” says Brandon Alexander, the CEO of Iron OX.
In the end, improved agricultural processes lead to better food options. And making small indoor farms more efficient could open up the possibilities of food grown closer to city centers. Most produce travels an average of 2,000 miles from farm to shelf in the U.S., which forces farmers to plant fruits and vegetables that can handle being transported—not necessarily those that taste good.
“Fresh produce isn’t that fresh,” Alexander says.
After improving its robotic systems, Alexander says, Iron OX’s long-term plans include breeding plants using data currently being gathered on its farm. Algorithms crunching this data and other local information, like what sells best, could replace tasteless, homogenized tomatoes and lettuce with more varieties suited to different communities’ tastes.
“We could make delicious, extra healthy things that people want to eat,” Alexander says.