How AI is ploughing a farming revolution

By Ben Payton

As agriculture has evolved over more than 10,000 years, farmers have constantly been searching for ways to increase yields and improve efficiency. Now, they could be set to benefit from one of the biggest innovations in decades, as artificial intelligence is increasingly applied on farms around the world.

A farmer’s success has always depended on their skill at forecasting; for millennia, farmers have made decisions such as which crops to plant, based on their forecasts of growing conditions and of market demand. Meanwhile, AI models, at their most basic level, are based around pattern recognition. Many AI apps are designed to digest much more data than a human, and then analyse this data to make more accurate forecasts.

A precision weed-spraying process using AI developed by Greeneye Technology. Greeneye Technology/Handout via REUTERS

Will Kletter is vice president of operations and strategy at ClimateAI, a company that is using artificial intelligence in climate-risk modelling to help customers improve their resilience to climate change. He says that AI can provide ″more accurate but also more localised″ forecasts of climate trends. This can help agricultural investors to select sites where particular crops will grow well, or decide which crops are best suited to specific sites.

Kletter says that his clients often arrive at the same conclusions as the AI models, but take far longer to collect and analyse data. ″We're helping customers make faster decisions, with less waste, that ultimately helps them bring maybe a new seed variety or new food crop to market more quickly,″ he claims.

But Kletter adds a note of caution, saying farmers should only apply AI selectively, to solve specific problems. Given the economics of farming, ″there isn't a lot of excess profit to go throwing at cool ideas. They have to be really useful ideas.″

The market is producing both winners and losers. Several agtech startups that developed AI and robotics to operate indoor vertical farms have faced financial turmoil over the past year, for example. Kentucky-based AppHarvest became among the highest-profile casualties when it filed for bankruptcy , opens new tabprotection in July.

Several agtech startups developed artificial intelligence for indoor farming. REUTERS/Rula Rouhana

Yet many startups remain convinced that AI can boost productivity and yield better environmental outcomes for farmers.

Several companies have developed technology that combines high-precision cameras with machine learning algorithms that can rapidly analyse and act on the data that ″computer vision″ provides.

Blue River Technology, a company that was acquired by agricultural machinery giant John Deere in 2017, has developed technology that attaches cameras to crop sprayers. The camera is trained by AI to recognise weeds, meaning that the sprayer applies herbicides to a targeted area, rather than an entire field.

Blue River’s chief executive, Willy Pell, notes that farmers that intensively care for every plant on their farm can, in theory, achieve higher yields with lower inputs. The problem, he says, is this kind of intensive plant care is impossible to practise on a large farm without a vast army of labourers.

AI-based targeted spraying technology offers the possibility that farmers can ″scale this micromanagement with machines that can perceive and act in real time″, Pell says. He predicts that targeted spraying technology will become widespread, especially on large-scale farms, within the next decade.

Nadav Bocher, chief executive of Israeli agtech startup Greeneye Technology, which has also developed precision spraying technology, believes the environmental benefits will prompt regulators to incentivise farmers to adopt those technologies.

The Tumaini app allows banana growers to scan plants for signs of diseases and pests. Michael Selvaraj/Alliance of Biodiversity and CIAT/Handout via REUTERS

As technology evolves, there will be many other ways to harness computer vision and artificial intelligence to create much more sustainable farming practices, he says. Crucially, extremely high-resolution data captured through cameras can be used to offer insights and recommendations to farmers.

This supportive role could be particularly important in developing countries. One of the concerns around AI in agriculture is that the benefits will go to major agribusiness corporations, rather than the hundreds of millions of small-scale farmers around the world.

Michael Selvaraj, a senior scientist in digital agriculture at the International Centre for Tropical Agriculture, has developed a smartphone app called Tumaini, which allows banana growers to scan plants for signs of diseases and pests. The app also provides advice to farmers, including on selecting disease-resistant crop varieties that are suited to their regions.

Selvaraj adds that the app can provide an ″early warning system″. Farmers scan plants on their smartphone; if this detects the presence of disease, the app can alert policymakers and support risk-mapping and epidemiological surveys.

Meanwhile, Selvaraj’s colleague, agricultural scientist David Guerena, has helped design the Artemis Project, opens new tab in Tanzania. This uses an app to help with phenotyping, the process of breeding new crop varieties based on observations of plant characteristics.

″Usually, it takes about 10 years to develop a new variety of plant,″ says Guerena. ″Climate change is happening so fast now that 10 years is not fast enough. Breeders need to speed up this process to make sure that the varieties that are coming out are climate resilient.″

The Artemis Project in Tanzania uses an app to help with phenotyping, the process of breeding new crop varieties based on observations of plant characteristics. Eliot Gee/Alliance of Biodiversity and CIAT/Handout via REUTERS

He explains that the basic approach to breeding new crop varieties, which involve humans attempting to cross plants with promising characteristics, have ″remained unchanged for 10,000 years″. This is an inefficient process, involving a great deal of trial and error, he says.

″We're looking at basically replacing what breeders are doing now with computer vision-enabled models,″ says Guerena. Plant growers participating in the Artemis Project collect data by taking photos through the app. This data is then analysed by AI-powered models to help select plant genes that are best-adapted to specific locations and will be more resilient to forecasted climatic changes.

AI can also be used in agriculture to help estimate the level of carbon stored in soils.

The Earth’s great rainforests are often described as the lungs of the planet, but soils are actually much more important carbon sinks. According to the Ecological Society of America, soils contain around 75% of all carbon, opens new tab stored on land.

Soils sequester carbon mainly though organic matter such as decomposing plants and animals. Carbon can potentially remain in soils for thousands of years. However, common agricultural practices, including ploughing, rotating crops and applying fertilisers can release carbon into the atmosphere.

A first step towards managing soil carbon stocks is accurate measurement and monitoring. This, however, is far from easy, says Martha Farella, a data scientist at engineering and environmental consultancy Stantec. ″Quantifying soil carbon over any given area of space is quite difficult,″ she says, with climatic conditions, topographic factors and vegetation types, along with the soil’s porosity and texture, all affecting the amount of carbon that can be stored. ″There's a lot of uncertainty that surrounds soil carbon because of that.″

Seagulls fly around as a farmer ploughs a field. AI is being used to assess the level of carbon in soil. REUTERS/Ints Kalnins

The standard way to measure soil health involves ground sampling. But this is a time-consuming and costly process, that can realistically only be undertaken occasionally at a small number of locations. Farella explains that using remote sensing, combined with machine learning models that provide a picture of how certain factors influence soil health, can be a more effective approach at scale.

″We can look at climatic patterns over space, we can look at vegetation properties over space, we can look at various soil factors to some degree over space,″ she tells us. ″And then we can use (machine learning) to build models based on known soil carbon values and those remote sensing variables to then predict soil carbon across space for all of the areas that we don't have direct measurements for.″

Ben Wark is head of Asia-Pacific at Downforce Technologies, a company that uses remote sensing and machine learning techniques to provide agricultural landowners with a ″digital twin″ of their properties.

Wark says that his company is generating huge interest from landowners who want to ″understand and quantify the current sequestration that they have within their properties″. Downforce’s platform is designed to allow landowners to establish a baseline of soil carbon and then monitor how changes to land-management practices affect the level of carbon sequestration.

Downforce is able to use this approach ″to create carbon credits that can be used for insetting within businesses,″ says Wark.

Having better and more timely data will ultimately help agricultural landowners to manage soil health more effectively, Ward argues. ″We want to demonstrate that understanding what you've done in the past that was good or bad, and understanding what works where, will help you to be able to create better soil health and in turn increase soil carbon.″