Silicon Valley Meets Central Valley

Silicon Valley Meets Central Valley
University of Sydney

In various guises, information technology is taking over agriculture

One way to view farming is as a branch of matrix algebra. A farmer must constantly juggle a set of variables, such as the weather, his soil’s moisture levels and nutrient content, competition to his crops from weeds, threats to their health from pests and diseases, and the costs of taking action to deal with these things. If he does the algebra correctly, or if it is done on his behalf, he will optimise his yield and maximise his profit.

The job of smart farming, then, is twofold. One is to measure the variables going into the matrix as accurately as is cost-effective. The other is to relieve the farmer of as much of the burden of processing the matrix as he is comfortable with ceding to a machine.

An early example of cost-effective precision in farming was the decision made in 2001 by John Deere, the world’s largest manufacturer of agricultural equipment, to fit its tractors and other mobile machines with global-positioning-system (GPS) sensors, so that they could be located to within a few centimetres anywhere on Earth. This made it possible to stop them either covering the same ground twice or missing out patches as they shuttled up and down fields, which had been a frequent problem. Dealing with this both reduced fuel bills and improved the uniformity and effectiveness of things like fertiliser, herbicide and pesticide spraying.

Since then, other techniques have been added. High-density soil sampling, carried out every few years to track properties such as mineral content and porosity, can predict the fertility of different parts of a field. Accurate contour mapping helps indicate how water moves around. And detectors planted in the soil can monitor moisture levels at multiple depths. Some detectors are also able to indicate nutrient content.

All of this permits variable-rate seeding, meaning the density of plants grown can be tailored to local conditions. And that density itself is under precise control. John Deere’s equipment can plant individual seeds to within an accuracy of 3cm. Moreover, when a crop is harvested, the rate at which grains or beans flow into the harvester’s tank can be measured from moment to moment. That information, when combined with GPS data, creates a yield map that shows which bits of land were more or less productive—and thus how accurate the soil and sensor-based predictions were (see chart). This information can then be fed into the following season’s planting pattern.

Farmers also gather information by flying planes over their land. Airborne instruments are able to measure the amount of plant cover and to distinguish between crops and weeds. Using a technique called multispectral analysis, which looks at how strongly plants absorb or reflect different wavelengths of sunlight, they can discover which crops are flourishing and which not.

Sensors attached to moving machinery can even take measurements on the run. For example, multispectral sensors mounted on a tractor’s spraying booms can estimate the nitrogen needs of crops about to be sprayed, and adjust the dose accordingly. A modern farm, then, produces data aplenty. But they need interpreting, and for that, information technology is essential.

Over the past few decades large corporations have grown up to supply the needs of commercial farming, especially in the Americas and Europe. Some are equipment-makers, such as John Deere. Others sell seeds or agricultural chemicals. These look like getting larger still. Dow and DuPont, two American giants, are planning to merge. Monsanto, another big American firm, is the subject of a takeover bid by Bayer, a German one. And Syngenta, a Swiss company, is being bid for by ChemChina, a Chinese one.

Business models are changing, too. These firms, no longer content merely to sell machinery, seed or chemicals, are all trying to develop matrix-crunching software platforms that will act as farm-management systems. These proprietary platforms will collect data from individual farms and process them in the cloud, allowing for the farm’s history, the known behaviour of individual crops strains and the local weather forecast. They will then make recommendations to the farmer, perhaps pointing him towards some of the firm’s other products.

But whereas making machinery, breeding new crops or manufacturing agrochemicals all have high barriers to entry, a data-based farm-management system can be put together by any businessman, even without a track record in agriculture. And many are having a go. For example, Trimble Navigation, based in Sunnyvale, at the southern end of Silicon Valley, reckons that as an established geographical-information company it is well placed to move into the smart-farming market, with a system called Connected Farms. It has bought in outside expertise in the shape of AGRI-TREND, a Canadian agricultural consultancy, which it acquired last year.

By contrast, Farmobile of Overland Park, Kansas, is a startup. It is aimed at those who value privacy, making a feature of not using clients’ data to improve its products, as many farm-management systems do. Farmers Business Network, of Davenport, Iowa, uses almost the opposite model, acting as a co-operative data pool. Data in the pool are anonymised, but everyone who joins is encouraged to add to the pool, and in turn gets to share what is there. The idea is that all participants will benefit from better solutions to the matrix.

Some firms focus on market niches. iTK, based in Montpellier, France, for example, specialises in grapes and has built mathematical models that describe the behaviour of all the main varieties. It is now expanding into California.

Thanks to this proliferation of farm-management software, it is possible to put more and more data to good use if the sensors are available to provide them. And better, cheaper sensors, too, are on their way. Moisture sensors, for example, usually work by measuring either the conductivity or the capacitance of soil, but a firm called WaterBit, based in Santa Clara, California, is using a different technology which it says can do the job at a tenth of the price of the existing products. And a sensor sold by John Deere can spectroscopically measure the nitrogen, phosphorous and potassium composition of liquid manure as it is being sprayed, permitting the spray rate to be adjusted in real time. This gets round the problem that liquid manure, though a good fertiliser, is not standardised, so is more difficult than commercial fertiliser to apply in the right quantities.

Things are changing in the air, too. In a recapitulation of the early days of manned flight, the makers of unmanned agricultural drones are testing a wide range of designs to find out which is best suited to the task of flying multispectral cameras over farms. Some firms, such as Agribotix in Boulder, Colorado, prefer quadcopters, a four-rotored modern design that has become the industry standard for small drones, though it has limited range and endurance. A popular alternative, the AgDrone, built by HoneyComb of Wilsonville, Oregon, is a single-engine flying wing that looks as if it has escaped from a 1950s air show. Another, the Lancaster 5, from PrecisionHawk of Raleigh, North Carolina, vaguely resembles a scale model of the eponymous second-world-war bomber. And the offering by Delair-Tech, based in Toulouse, France, sports the long, narrow wings of a glider to keep it aloft for long periods.

Even an endurance drone, though, may be pushed to survey a large estate in one go. For a synoptic view of their holding, therefore, some farmers turn to satellites. Planet Labs, a firm in San Francisco, provides such a service using devices called CubeSats, measuring a few centimetres across. It keeps a fleet of about 30 of these in orbit, which it refreshes as old ones die by putting new ones into space, piggybacking on commercial launches. Thanks to modern optics, even a satellite this small can be fitted with a multispectral camera, though it has a resolution per pixel of only 3.5 metres (about ten feet). That is not bad from outer space, but not nearly as good as a drone’s camera can manage.

Satellite coverage, though, has the advantage of being both broad and frequent, whereas a drone can offer only one or the other of these qualities. Planet Lab’s constellation will be able to take a picture of a given bit of the Earth’s surface at least once a week, so that areas in trouble can be identified quickly and a more detailed examination made.

The best solution is to integrate aerial and satellite coverage. That is what Mavrx, also based in San Francisco, is trying to do. Instead of drones, it has an Uber-like arrangement with about 100 light-aircraft pilots around America. Each of the firm’s contracted planes has been fitted with a multispectral camera and stands ready to make specific sorties at Mavrx’s request. Mavrx’s cameras have a resolution of 20cm a pixel, meaning they can pretty much take in individual plants.

The firm has also outsourced its satellite photography. Its raw material is drawn from Landsat and other public satellite programmes. It also has access to these programmes’ libraries, some of which go back 30 years. It can thus check the performance of a particular field over decades, calculate how much biomass that field has supported from year to year and correlate this with records of the field’s yields in those years, showing how productive the plants there have been. Then, knowing the field’s biomass in the current season, it can predict what the yield will be. Mavrx’s method can be scaled up to cover entire regions and even countries, forecasting the size of the harvests before they are gathered. That is powerful financial and political information.

A truly automated, factory-like farm, however, would have to cut people out of the loop altogether. That means introducing robots on the ground as well as in the air, and there are plenty of hopeful agricultural-robot makers trying to do so.

At the University of Sydney, the Australian Centre for Field Robotics has developed RIPPA (Robot for Intelligent Perception and Precision Application), a four-wheeled, solar-powered device that identifies weeds in fields of vegetables and zaps them individually. At the moment it does this with precise, and precisely aimed, doses of herbicide. But it, or something similar, could instead use a beam of microwaves, or even a laser. That would allow the crops concerned to be recognised as “organic” by customers who disapprove of chemical treatments.

For the less fussy, Rowbot Systems of Minneapolis is developing a bot that can travel between rows of partly grown maize plants, allowing it to apply supplementary side dressings of fertiliser to the plants without crushing them. Indeed, it might be possible in future to match the dose to the plant in farms where individual plants’ needs have been assessed by airborne multispectral cameras.

Robots are also of interest to growers of fruit and vegetables that are currently picked by hand. Fruit-picking is a time-consuming business which, even though the pickers are not well rewarded, would be a lot faster and cheaper if it were automated. And robot pickers are starting to appear.

The SW6010, made by AGROBOT, a Spanish firm, uses a camera to recognise strawberries and work out which are ripe for the plucking. Those that are have their stems severed by blades and are caught in baskets before being passed on by a conveyor belt for packing by a human operator sitting on the robot. In the Netherlands, researchers at Wageningen University are working on a robot harvester for larger produce such as peppers.

All these devices, and others like them, still exude a whiff of the Heath Robinson. But robotics is developing rapidly, and the control systems needed to run such machines are getting better and cheaper by the day. Some think that in a decade or so many farms in rich countries will be largely robot-operated.

Yet others wonder just how far farmers will let their farms be robotised. Self-guiding agricultural machinery such as that sold by John Deere is all but robotic already. It is like an airliner, in which the pilot usually has little to do between landing and take-off because computers do the work for him. Yet Deere has no plans to hand over complete control to the cloud, because that is not what its customers want.

Tunnel vision

If total control still seems some way off in outdoor farming, it is already close for crops grown in an entirely artificial environment. In a warren of tunnels beneath Clapham, in south London, Growing Underground is doing exactly what its name suggests. It is rearing around 20 types of salad plants, intended for sale to the chefs and sandwich shops of the city, in subterranean voids that began life as second-world-war bomb shelters.

In many ways, Growing Underground’s farm resembles any other indoor hydroponic operation. But there is one big difference. A conventional greenhouse, with its glass or polycarbonate walls, is designed to admit as much sunlight as possible. Growing Underground specifically excludes it. Instead, illumination is provided by light-emitting diodes (LEDs). These, in the minimalist spirit of hydroponics, have had their spectra precisely tuned so that the light they emit is optimal for the plants’ photosynthesis.

As you would expect, sensors watch everything—temperature, humidity, illumination—and send the data directly to Cambridge University’s engineering department where they are crunched, along with information on the plants’ growth, to work out the best regimes for future crops.

For now Steven Dring, Growing Underground’s boss, is confining output to herbs and vegetables such as small lettuces and samphire that can be brought to harvestable size quickly. He has reduced the cycle for coriander from 21 to 14 days. But tests suggest that the system also works for other, chunkier crops. Carrots and radishes have already been successfully grown this way, though they may not command a sufficient premium to make their underground cultivation worthwhile. But pak choi, a Chinese vegetable popular with trendy urbanites who live in inner-London suburbs like Clapham, is also amenable. At the moment growing it takes five weeks from start to finish. Get that down to three, which Mr Dring thinks he can, and it would be profitable.

The firms that make the LEDs could also be on to a good thing. Mr Dring’s come from Valoya, a Finnish firm. In Sweden, Heliospectra is in the same business. Philips, a Dutch electrical giant, has also joined in. In conventional greenhouses such lights are used to supplement the sun, but increasingly they do duty in windowless operations like Mr Dring’s. Though unlike sunlight they do not come free, they are so efficient and long-lasting that their spectral advantages seem clinching (see chart).

This kind of farming does not have to take place underground. Operations like Mr Dring’s are cropping up in buildings on the surface as well. Old meatpacking plants, factories and warehouses the world over are being turned into “vertical farms”. Though they are never going to fill the whole world’s bellies, they are more than a fad. Rather, they are a modern version of the market gardens that once flourished on the edge of cities before the land they occupied was swallowed by urban sprawl. And with their precise control of inputs, and thus outputs (see article), they also represent the ultimate in what farming could become.