Researchers at the University of Bonn have developed innovative software that simulates the growth of field crops using artificial intelligence. By feeding thousands of photos from field experiments into a learning algorithm, the software can predict the future development of cultivated plants based on a single initial image. This technology allows for accurate estimation of parameters such as leaf area and yield.

This breakthrough offers significant benefits for farmers, helping them determine optimal plant combinations and fertilizer choices to maximize yield. In the future, farmers will increasingly rely on computer support to answer critical questions about crop management.

“We have developed software that uses drone photos to visualize the future development of plants,” explains Lukas Drees from the Institute of Geodesy and Geoinformation at the University of Bonn. Drees, an early career researcher involved in the PhenoRob Cluster of Excellence, highlights the project’s goal of promoting intelligent digitalization in agriculture to enhance environmental sustainability without compromising harvest yields. The findings are published in the journal Plant Methods.

The software presented by Drees and his team is a key component in this effort. It allows for virtual simulation of agricultural decisions, such as assessing the impact of pesticides or fertilizers on crop yield. The program requires drone photos from field experiments for input. “We took thousands of images over one growth period,” Drees explains. “For example, we documented the development of cauliflower crops under certain conditions.”

The learning algorithm, trained with these images, can generate images showing the future development of the crop based on a single aerial image taken at an early growth stage. While the current model is highly accurate under similar conditions to the training data, it does not yet account for sudden weather changes. However, future iterations aim to incorporate these variables to predict outcomes more accurately.

In addition to growth prediction, a second AI software estimates various parameters, such as crop yield, from plant photos. This tool can also work with generated images, allowing for precise early-stage yield predictions.

One of the research areas includes the use of polycultures, where different species are sown together, such as beans and wheat. Polycultures often yield more because the plants have different requirements and compete less with each other compared to monocultures. Additionally, some species, like beans, can naturally fertilize the soil by binding nitrogen from the air, benefiting other species like wheat.

“Polycultures are also less susceptible to pests and environmental influences,” explains Drees. The success of polycultures depends heavily on the species combination and mixing ratio. By feeding results from various mixing experiments into learning algorithms, researchers can derive recommendations for the most compatible plant combinations and ratios.

Plant growth simulations using learning algorithms are a recent advancement. Traditionally, process-based models, which have an inherent understanding of plant nutrient and environmental needs, have been used. “Our software, however, bases its predictions solely on the experience gained from training images,” stresses Drees.

The two approaches—process-based and image-based—can complement each other. Combining them could significantly enhance forecast quality. “This is also a point we are investigating in our study,” says Drees. “How can we use process- and image-based methods so they benefit from each other in the best possible way?”

By Impact Lab