At the Research Station for Plant Sciences in Lindau, postdoctoral researcher Helge Aasen is piloting a large, black drone with six rotor blades. The drone is helping him to determine the phenotype – that is, quantify the characteristics – of 350 or so different wheat varieties under trial cultivation. Using the drone, he can do this with a speed and level of precision that, until recently, would have been inconceivable. The drone carries a 6-kilogram payload comprising a range of high-tech equipment. A thermal imaging camera measures the temperature of the wheat when the drone is airborne. This enables Aasen to calculate the degree of water evaporation. Two hyperspectral cameras serve to determine chlorophyll levels and therefore plant productivity. An RGB camera generates a 3D model of the crop, accurate to a matter of centimetres. This is used to calculate stalk height and leaf area. The latter is a crucial indicator of the level of resistance to pests displayed by any wheat variety. What Aasen is showcasing here is the future of farming. Indeed, drones and all the various high-tech cameras could one day form part of basic farmyard inventory.
The fourth agricultural revolution
The research station in Lindau is the summer residence of the Crop Science group, to which Aasen belongs. It is here that Achim Walter, ETH Professor of Crop Science, and his team are working to bring about the fourth agricultural revolution – a revolution that he, along with fellow ETH academics, first proclaimed in a widely read opinion piece published by the journal PNAS back in 2017. There the authors predicted that agriculture in the 21st century would be increasingly determined by the smart use of data. Technology is by no means a new arrival to the farmyard. The 1990s saw the arrival of “precision farming”, when equipment manufacturers introduced machinery fitted with GPS and infrared sensors as a means of boosting productivity. Yet “smart farming” – and all the systems this entails – promises to go one step further. With the help of sophisticated imaging technology, efficient machine learning and huge volumes of data, it won’t be long before “intelligent” agricultural machinery is taking decisions autonomously.
Experts see big potential in smart farming. According to the FAO, between 20 and 40 percent of the world’s grain harvest is lost to pests and disease, notwithstanding the use of some 2 million tonnes of pesticides a year. It is hoped that the smart use of data and new technology will help boost yields and, at the same time, substantially reduce farming’s carbon footprint. On the basis of data gathered with drones, as in Lindau, it will be possible to target the use of fertilizer, pesticides, herbicides and fungicides much more precisely, with spraying restricted to only those areas that actually need it. Based on the studies now available, Walter calculates that savings of up to 90 percent could be achieved.
News of the work being done at the Lindau research station has spread within the international smart farming community. It is in Lindau that Walter has set up a unique field trial for determining plant phenotypes (FIP). The key piece of equipment is a sensor unit equipped with a laser scanner, multispectral cameras, an infrared camera and two spectrometers. Suspended from Kevlar cables strung between four high masts, this glides quietly above a test field measuring 100 metres by 130 metres. Growing beneath are various varieties of wheat, soybean, maize, buckwheat and forage grasses. Back in a control room equipped with monitors, the researchers use electric cable drives to manoeuvre the sensor unit to any position above the field. With the help of high-resolution colour, thermal and multispectral images provided by the sensor unit, they can monitor plant growth, photosynthesis and the ripening behaviour of the different varieties under cultivation. What’s more – and this distinguishes Walter’s set-up from others – plant characteristics can be monitored continuously without disturbing the crop below and in almost any weather conditions. And, unlike a drone with rotor blades, the suspended sensor unit doesn’t create any wind turbulence that would make the plants move and thereby falsify the readings. For this reason, the FIP is also very useful for calibrating drone-based scanning methods.
The future of farm work
Flourish, a three-and-a-half-year EU project, has now looked at how drones might be used in future farming. An international team coordinated by Roland Siegwart, ETH Professor of Autonomous Systems, developed an autonomous tractor-drone system for the cultivation of sugar beet, a major crop in Europe. The drone identifies weeds in the ground under cultivation and directs the tractor to the exact location. The tractor then crushes the weeds with hydraulically powered steel rams. This kind of robot would not only lighten the load for farmers but also make it once again economically attractive to destroy weeds mechanically rather than with a chemical herbicide. Flourish was completed in 2018. The tractor-drone system is now undergoing further development with the Bosch company Deepfield Robotics, which was also involved in the EU project.
“Drones are ideal for rapidly collecting data on crop conditions across a large area,” Siegwart says. However, as he explains, processing sensor data in order to develop practical applications for agriculture is a lot more complicated: “Biological systems are diverse and complex. That makes our work more difficult.” In Switzerland alone, for example, there are over 30 different wheat varieties under cultivation. Similarly, shadows from clouds and crop movement caused by wind can make it difficult to obtain accurate readings. In other words, it takes large volumes of data, sophisticated machine learning and lots of patience to develop autonomous systems that will function reliably. In the long term, Siegwart sees drones being used not only for data capture but also for crop spraying and other field work. Equipped with a spray and tank, autonomous drones could be used to apply fungicides and pesticides to only those plants that are actually infested. In order to advance the development of this and similar technologies, Siegwart and Walter are planning to set up a laboratory in Lindau, where agricultural scientists, biologists, robotics experts and computer scientists can all work together.
Who owns the data?
Smart farming offers great opportunities, but there are risks involved. For Nina Buchmann, ETH Professor of Grassland Sciences and co-author of the 2017 article, one key question concerns ownership of the data and what happens to it. “There’s certainly a danger of becoming reliant on a small number of global players. As many countries as possible, including Switzerland, should therefore join forces to prevent this happening.” Moreover, there is also the question of liability in the event of accidents involving drones or environmental damage stemming from wrong decisions made by autonomous systems. Who is then responsible? The farmer, the programmer or the manufacturer of the sensor system? And, finally, this all inevitably raises the question of whether, one day, there will be fully automated farms without the need for farmers. “That’s certainly not my vision of smart farming,” says Walter. “But here in Switzerland, in particular, digitalisation could help farming as we know it – small-scale, diverse, geared to quality – to hold its own on the international market, despite having reduced the use of fertilizers and pesticides.”
The reaction in the farming community is mixed. As part of the Sustainable Economy national research programme (NRP 73) funded by the Swiss National Science Foundation, farmers were asked which applications would make practical sense and how willing they would be to accept the help of robots and drones out in the fields. “Some said they would shoot down any drone they saw flying over their land,” Walter recounts. “But then others said, ‘Cool! I fly a drone in my spare time, anyway. So why not use the images to improve my yields?’ ”