AGROS, the research program in which Wageningen University & Research (WUR) and 26 private partners worked on an 'Evolution towards sustainable AGRicultural Business Systems' recently concluded with a final meeting. It was also announced there that the program will be continued with AGROS-II.
'Modelling is not objective'
Modeling and artificial intelligence (AI) have become indispensable within the program. About 75 project participants and other attendees heard keynote speaker Data Scientist Dr. Marc Jacobs says that researchers should never view a model as objective. “A model helps to clarify (large amounts of) data and for discussions, but a model itself is never objective. It merely expresses the modeler's opinions (unintentionally) in a mathematical form. A model is therefore an “extension of the modeller,” Jacobs indicated. “With modeling you, as it were, integrate technology with people. That is why a model, such as the AERIUS model, should never become part of legislation and/or regulations. A model simplifies reality and can 'only' provide decision support.”
In his keynote, Jacobs was also surprised about the so-called Gartner hype cycle of AI and in particular ChatGPT. “ChatGPT's hype cycle is the fastest hype cycle I've ever seen!” At the same time, he warned that researchers should not be misled by AI. “Always ask yourself whether you understand exactly what an AI program does and whether you can explain it to someone who has nothing to do with your research or subject matter. Don't use it because others use it too. That's a non-argument. The use of AI must be able to add something.”
Technology for ecology as a guideline
Dr.ir. Anja Dieleman, senior researcher in plant physiology at the WUR Greenhouse Horticulture business unit, briefly explained the principles of the AGROS program. “It is explicitly about the use of technology for the benefit of ecology. And this in the field of arable farming, dairy farming and horticulture. Consider optimal use of resources, available labor, animal and crop health, greenhouse gas emissions and biodiversity.”
Autonomous greenhouse cultivation
Within horticulture, the autonomous cultivation of crops such as cucumbers has been examined in order to autonomously and automatically control the greenhouse climate and crop growth based on sensor data. For example, sensors/vision are used to control the speed at which new leaves are formed because that speed determines how many fruits can develop. One crop/greenhouse was controlled by growers, a second by a digital copy of the greenhouse (digital twin) and a third based on AI. It can be concluded from the research that autonomous greenhouse cultivation is feasible. The follow-up project AGROS II investigates to what extent a (cucumber) crop has boundaries, what they are and how this should be incorporated into the greenhouse arrangement.
Cow monitoring
DLO researcher, veterinarian and data scientist Dr. István Fodor discussed research aimed at developing methods for individual monitoring of dairy cows. In order to use resources more efficiently and monitor greenhouse gas emissions for more sustainable milk production. To also take animal health to a higher level. “For example, how can we lose less energy from feed and reduce cow emissions through optimized feed conversion? Breath analysis proves to be a great tool. We can even use the composition of exhaled air to say something about the composition of the blood of dairy cows.” Fodor also explained how 2D and 3D computer vision can provide insight into abnormal animal behavior (lameness, disease) and feed intake. In the coming months, the research results and perspectives will be further explored and optimized.
Crop cultivation of the future
Ir. Jan Kamp, team and project leader at WUR Open Teelten in Lelystad, outlined developments in arable farming by pointing out the significant increase in scale and productivity in recent decades. The downside is that the power of agroecology has faded into the background, leading to soil compaction and loss of biodiversity. He showed three future scenarios for food production in the Netherlands with technology for ecology (T4E) as a starting point. “We see three imaginary scenarios that take into account the social and climatic impact, the impact on an arable farmer and his company and on the value chain or the food production chain. Scenario 1 concerns 'data for your convenience', scenario 2 'ecology at the helm' and scenario 3 'crisis for action'.” He also explained the work packages in the project.
Smart detection of potato diseases
In one of the workshops, Jan Kamp discussed the use of vision and AI in the (automated) detection of the potato Y virus and Erwinia in seed potatoes in Dutch arable farming. “Sick potato plants can cause significant yield loss. With vision and AI (deep learning) we have shown that we can make a big leap forward in (early) recognition of infected plants. Especially at Erwinia. The results are at the same level as those of the official inspection body (NAK). Moreover, the automated application is economically feasible with a reasonable operating speed.” Work has been done on the Agria, Esmee and Fontane varieties, among others. Training the model with recent annotated images from the current season increased the R-values (improved correlations/scores). In the AGROS-II project, the developed model will be further improved and one of the private partners will implement the model in a robotized application.
Automated spot spraying of weeds
In the same workshop, Bram Veldhuisen MSc, WUR researcher in precision agriculture and robotics, discussed the possibilities of automated spot spraying of weeds. This has been investigated in a sugar beet crop (control of potato storage) and in an onion crop with low and high weed pressure respectively. The research initially focused on predicting savings on crop protection product use, predicting crop damage and the user-friendliness of spot spraying.
In conclusion, it can be said that modeling the deposition of crop protection products works. The selection of the right spray nozzles and determining the correct dosage of agents is challenging. AGROS has developed a spot spray fact sheet for arable farmers with tips and examples of available technologies.
Model-based disease detection in horticultural crops
Dr. Kirsten Leiss, senior crop health researcher at WUR, talked about the results of early detection and diagnosis of (yet) invisible pests and diseases in horticultural greenhouses (decorative tent and vegetables) using sensors and vision. We looked for thrips (insect), mildew (fungal infection) and the cucumber mosaic virus (virus infection).
The model was able to detect thrips in tomato plants within two days after infection, while a tomato grower normally only notices a thrips infection six to seven days after infection. Leiss also looked at the possibility of using volatile organic compounds (VOCs) as biological markers to detect mildew in a timely manner by collecting and analyzing the air around tomato plants. After five to six days the first infection was measurable in 'hydrocarbon'.
Breath analysis in dairy cattle
In his part of the workshop, István Fodor told more about breath analysis in dairy cattle to not only measure emissions but also detect (metabolic) diseases. “Exhaled air appears to contain more than 1,000 different compounds. It is a complex mix of molecules and it is striking that conclusions can also be drawn from the composition of exhaled air about compounds in the blood of dairy cows.”
Research has included detecting ketosis based on the amount of acetone in exhaled air. This showed that the level of acetone in the breath strongly correlates with the BHB value in the blood of the mother cow. The research results offer potential for non-invasive (and therefore animal-friendly) and almost real-time disease detection.
Three future crop cultivation scenarios
In one of the workshops, research manager Herman Schoorlemmer further explained the future crop cultivation scenarios for 2040 previously mentioned by Jan Kamp. Technology also serves ecology in these scenarios.
In the 'data for your convenience and technology driven' scenario, the balance of power in crop cultivation and food production has shifted compared to today. Manufacturers are no longer suppliers of technology but of services. The arable farmer has become an operator who produces personalized food based on data.
Growers are the heroes in the 'ecology at the wheel' scenario. They provide ecosystem services for which they are also (well) paid. What is striking in this scenario is the great involvement of a particularly active government, driven by the objective of making agriculture more sustainable.
In the third and final scenario called 'crisis force action', lack of trust and agreement in society is leading. They only trust science when it comes to sustainable food production. Crop protection products are only available on prescription and the number of products is severely limited. The storage of fresh water leads to new revenue models. The accompanying English report can be found here: https://edepot.wur.nl/630284
Presentations of the day
Do you want to take a look at the presentations that were given that day? View them here.
