University of Bonn Researchers Develop Software to Simulate Crop Growth

Researchers at the University of Bonn have made significant progress in the field of agricultural digitalization by developing software that can simulate the growth of field crops. Led by Lukas Drees from the Institute of Geodesy and Geoinformation, the team fed thousands of photos from field experiments into a learning algorithm, enabling it to visualize the future development of cultivated plants based on a single initial image. The findings of this study have been published in the journal Plant Methods.

The software, which uses drone photos to predict plant growth, has the potential to revolutionize farming practices. By analyzing images captured during the growth period, the program can accurately estimate parameters such as leaf area and crop yield. This technology aims to assist farmers in making informed decisions, such as assessing the impact of pesticides or fertilizers on crop yield.

To train the learning algorithm, the researchers captured thousands of images over a growth period, documenting the development of cauliflower crops under specific conditions. Based on a single aerial image of an early growth stage, the algorithm generated images depicting the future development of the crop. However, the software’s accuracy is dependent on similar crop conditions to those present during the training photos, and it does not account for sudden weather changes or other external factors.

Additionally, the researchers utilized a second artificial intelligence software that can estimate various parameters from plant photos, including crop yield. This software can also work with the generated images, allowing for precise estimation of cauliflower head size at an early growth stage.

The team is particularly interested in exploring the use of polycultures, where different plant species are sown in one field. Polycultures have shown to enhance crop yield and reduce susceptibility to pests and environmental influences. By feeding results from various mixing experiments into learning algorithms, the researchers aim to determine the most compatible plant species and their optimal mixing ratios.

While process-based models have traditionally been used for plant growth simulations, this software relies solely on the experience gained from training images. The researchers emphasize that both approaches can complement each other, potentially improving the accuracy of forecasts. They are currently investigating how to combine process- and image-based methods to maximize their benefits.