Real tomato plants in the greenhouse. Photo: Rick van de Zedde
Will vegetable growers soon be able to create the perfect conditions for their crops using computer models? We plunge into the world of the digital twin, a promising new technique in which WUR researchers are doing cutting-edge work. What are digital twins, what can you do with them and what are the benefits? “If this works, we will be able to save a lot on water, energy and raw materials.”
Imagine you are a tomato grower and you want to know whether you could use less water without it affecting your crop yields. Or perhaps you are considering replacing the glazing in your greenhouses to get more light, but you don’t want to risk an investment that turns out to be pointless. Of course, you could start experimenting to see what happens. But it would probably save you a lot of time and money if you could predict beforehand what the consequences of a particular decision might be. That is precisely the power of ‘digital twins’, a new computer technique that WUR scientists are working in three research projects this year. Admittedly, the term ‘digital twin’ is rather abstract. It basically boils down to replicating an existing product or system — which could be a bicycle but equally a farm or a living plant — as precisely as possible on the computer, based on all kinds of data about its properties. You see a 3D image on the screen that looks exactly like its real-life equivalent and behaves the same way too.
Predicting the future of a tomato plant: Wageningen scientists Sjoerd Boersma and Katarina Streit are trying to achieve this by building a digital twin of the plant. It’s a computer model that can show exactly how a plant grows under different conditions. In this video, the scientists explain what goes into building a digital twin.
In theory, you can create twins for almost any product or organism. “But digital twins are still rare in WUR’s domain — the living environment,” says Willem Jan Knibbe, who coordinates the Digital Twins investment programme. “That made us curious to see how this technique would work in our fields of research.” ‘Standard’ plant simulation models have been around for years. What’s new about the digital twin is that the computer model is linked to its physical twin, and gets updated in real time with the twin’s data. In the case of a tomato plant, this encompasses data on how fast the plant is growing, how much sunlight it captures and how much water it consumes. Knibbe: “The idea is that as a farmer, policymaker or researcher, you can see exactly what is happening, for example to the fertility of your land or your plant growth. That gives you much more room to try things out without incurring huge costs or damaging the environment.”
No ordinary greenhouse
What does a digital twin project look like in practice? In the brand-new high-tech experimental greenhouse of the Netherlands Plant Eco-phenotyping Centre (NPEC) on Wageningen campus, a team of researchers are building a digital twin of a tomato crop. You can tell the moment you enter that this is no ordinary horticultural greenhouse. The tomato plants are neatly arranged in rows in a blue rail system, each in its own pot. When it is time for a measurement, the plants shift one by one towards the measuring station along the side of the greenhouse. Dozens of properties of the plant and its surroundings are monitored here, from the plant architecture and moisture content of the fruit to the amount of sunlight in the greenhouse.
Close-up of a young tomato plant in the research greenhouse. Photo: Katarina Streit
Developing digital twins could help growers use energy, water and nutrients more efficiently, explains Jochem Evers, an associate professor at WUR and the leader of the tomato project. “A greenhouse is a major consumer so there is a lot of scope for saving on inputs — or achieving higher yields with the same quantities.” What happens for example if you remove leaves to increase the light? Or if you alter the distance between the plants? “Growers will soon have a much sounder foundation for various decisions, simply because they will have information that they didn’t have in the past.”
Evers shows his computer screen, which displays a lifelike replica in 3D of the greenhouse environment — it almost seems like a live stream. In the real NPEC greenhouse, the tomato plants pass through the measuring station every day. “Our simulation is not just fancy graphics. We are genuinely twinning this specific greenhouse.” But unlike the example of the bicycle, Evers and his colleagues do not have a digital copy of each individual plant. “We look at the crop as a whole. It’s not about having each individual leaf sticking up in exactly the right direction; what matters is whether the overall greenhouse is a good representation of the real-world one.”
All the measurement data is fed into the computer, which automatically refines the model. “If you use that to calculate scenarios, the outcome is more reliable than if you use average data,” explains Evers. “After all, averages don’t say much about your specific conditions. If there is less sunlight in a given year, that will affect the growth of your crop.”
A digital twin is a computer version of a product that is as close as possible to the original, based on all kinds of data about its properties
Equally, the digital twin can alert you to the threat of deficiencies in the real-world crop. The model then calculates what you need to do to prevent this, for example adjust the temperature in the greenhouse or prune some more leaves. The model also calculates the economic factors, such as the financial consequences of a particular choice.
“We really do want to sketch a comprehensive picture,” says Evers. “One that’s appealing for growers because savings of just half a per cent can result in large sums in profit on an annual basis.”
Model of the virtual tomato crop greenhouse. Photo: Katarina Streit
Transportation of the tomato plants to the measuring station. Photo: Katarina Streit
Various scientific disciplines are involved in the development of digital twins, from artificial intelligence and data science to experts with knowledge of user behaviour. “Each of those research fields is hugely challenging in its own right,” says coordinator Willem Jan Knibbe. “And you need a combination of all of them in these studies.” Furthermore, the researchers in the tomato project have been working closely with Dutch vegetable growers, suppliers and other parties with an interest in the concept from day one. “Their input is incredibly useful for us,” says Evers. “They come up with critical questions that we wouldn’t necessarily think of ourselves. There is no doubt that you would need to measure more than you do at present, but of course growers only want to make such an investment if it really has benefits.”
One thing is clear: it is harder to make a digital copy of a living organism than a model of a bicycle or an aeroplane engine. Knibbe: “You can figure out all aspects of an engine but you have much less of an idea which variables play a role in a plant or a human organ. But if you are successful, the opportunities are enormous. For example, you could see how people would react to medicine, or how the heart functions in certain conditions.”
A digital twin gives you more room to try out things without incurring huge costs or damaging the environment
Apart from the technical development, there are other issues that the researchers have to take into account before users can get to work with their own digital twins. “The models use a lot of privacy-sensitive data,” says Knibbe. “They have to be completely secure. Of course, it’s nice that you can calculate the benefits of new glazing for a tomato grower, but that also means you know the company’s profits and costs. You need to be very careful with that information.”
Huge volume of data
Once the tomato twin has been proven to work, the experts say it will not be difficult to extend the model to include other products. “The complex aspect is developing the technology,” says Knibbe. “Once that’s available, you can just enter different data and create a digital twin for any other organism. In the best-case situation, we may never need to do tests again with lab animals.” That day is still some way off, though; at present, Evers and his colleagues are halfway through their tomato project. “We are measuring as much as possible because we don’t yet know which aspects are relevant,” he says. “That has generated a huge volume of data, as you can imagine. The challenge is to extract the right information.”
Then there is the fact that every study has practical and budgetary limits. Evers: “Our focus is currently on crop yields and the economic aspects. At this stage, we are not looking at diseases and pests, and we are also not considering the fruit quality in detail, although these would be incredibly interesting follow-up studies.”
Will WUR’s digital twins live up to their promise? It’s too early to say. Evers and his colleagues aim to have a proof of concept by the end of next year, a prototype that growers can test in the field. “Then we will know what kind of success we can expect in practice.” However promising the digital twins may be, Knibbe and Evers both stress that the technology is not a holy grail. “It is mainly the potential that is interesting now,” says Knibbe. “We’re aiming to create enough leverage for further development with partners in healthcare, agriculture and the IT sector.” Evers: “I’ve noticed that ‘digital twin’ is a hyped term. The expectations are often huge. It is the solution to everything if some companies are to be believed, but I’m too down to earth for that. In the final analysis, it’s a new tool in addition to the tools we already have. But this is a major advance. You can see it as a high-risk, high-reward approach — if it does work, we will be able to make serious progress.”
OTHER DIGITAL TWIN PROJECTS
In addition to the tomato project of Evers and his colleagues, WUR researchers are working on two other digital twin projects within the WUR investment programme: ‘Me, My Diet and I’, a personalised twin that can predict the fat content in your blood after a meal; and ‘Digital Future Farm’, a model that gives a precise representation of the nitrogen cycle on an arable or dairy farm.