Josse De Baerdemaeker graduated from the KU Leuven, Belgium and has an MSc and PhD in Agricultural Engineering from Michigan State University. He is recognised as the founding father of the concept of precision agriculture, and his focus is on improving technology in crop cultivation, harvesting and handling to minimize losses and optimize yield and income for farmers. He is Emeritus Professor at the KU Leuven and was a visiting Professor at Kyoto University, Japan, and is currently chairman of VILT, the Flanders Infocenter for Agriculture and Horticulture.
What is the most exciting opportunity for bringing AI into agriculture?
I think that in the future AI (Artificial Intelligence) will help improve our crop production and our crop protection. We could better understand the biology and the ongoing processes at the smaller scale to better manage the whole process of growth and production, and eventually bring it to the market.
Many publications are appearing about soil processes, because what happens in the soil essentially determines what happens in the air. The best sensor about what is happening in the soil is the plant, so if it does not grow then you know something is wrong in the soil. But what? The crop does not tell us the details. I see this coming together with AI to bring great potential.
AI will also be useful in the supply chain as we can use it to optimise production according to the demand of the consumers. It will, for example, direct farmers to grow particular products because there is a market for it in one place. This integration in the food chain, linking genetics, farm management and consumers, can also be called ‘Artificial Intelligence for Intelligent Agriculture’ (AI for IA). It relies on communication (processes as well as technology), sensors and data analytics for an efficient supply chain.
The European Green Deal stipulates reduction of land use, and less use of chemicals in agriculture towards a common goal of sustainability. Here as well we rely on communication between all actors as well as on data analytics and sensors. We should produce more with less land, but in a way that does not downgrade quality. I think that better understanding biology – and building that into the way we produce food – would help farmers to implement that goal. Also, give them the machines that can do it!
What is the difference between image-based pest-detection types of AI and farm management AI?
In the end, they have to come together. Image processing to detect pests is a different implementation in the sense that a sprayer on the tractor stays there. But the fact that it detects a pest and has to spray affects farm management over a long period.
These automated machines for spraying are the easy (obvious!) part. But an observation you do today cannot stand on its own. You have to bring together all the data, and this requires AI. Let’s say we have different test plots in a field, and we have bugs in some of them. Now we do a satellite area observation, and we see where the disease is and send the tractor there to spray pesticide. That is not enough. The spread of the disease changes over time, so we have omitted the temporal dynamics. If we have observations at different locations and different times, we can link these together to see the evolution.
What I would like to know as a farmer is: when can I expect that pest to arrive in my field? How fast does it move? And how widely is it spread around my area? Why do I have that disease here? Instead of having just precision farming, which is often thought of as a local application for a local problem, we will have to look at spatiotemporal variability not just in one location, but over a wide area.
This seems like it will require a lot of data sharing over a long period of time. What factors play into this?
AI in agriculture will have to rely on a very good telecommunication system, which is a limiting factor as it will put farmers at a disadvantage if they do not have this system in place. Similarly, in remote areas we do not have good communications everywhere in Europe.
But let’s say you are harvesting grapes, and you collect data on the grapes as you are harvesting them. You send this data to the winery so that by the time you arrive at the winery they know the grapes will be for the normal wine, or for the premium wine. If you do this completely online, this entails high data traffic from the field to the winery. If there is more local intelligence on the grape harvester, on the machines, or in the local distributor, then you only have to send a little bit of data once in a while, so that traffic and communication systems are not overloaded.
AI gives us strength, but it needs to recognise and overcome weak points in the food supply chain. If the chain is broken somewhere between the farmer and the data analysis operator or the market, what do you do? There needs to be redundancy in the system so that it still works.
So is there a cybersecurity element to agricultural AI?
You seldom hear about cybersecurity pertaining to agriculture, but it often involves big data that someone is storing and handling. Farmers are relying on the transmissibility or the use of those data, so that supply chain can be disrupted. What I am concerned about is if someone slips in a virus, the whole thing could go haywire and there goes our AI.
This is one of the challenges to implementation: safe supply chains that are robust to intrusions. It could be that suddenly we lose access to information, but for food security, systems have to keep going. If we put intelligence into edge computing, even if the system breaks down we still have a minimum of local intelligence.