How AI and Drones are Hunting for Hidden Crop Genetics

Researchers at the University of Illinois have developed an innovative artificial intelligence method designed to mine massive amounts of drone-captured field data, revealing hidden, highly heritable genetic traits in crops.

For decades, agriculture has relied on visual cues and simple calculations to gauge plant health. The widely used Normalized Difference Vegetation Index (NDVI), for example, uses just two bands of light reflectance to mathematically determine the greenness of a field. Now, an interdisciplinary team is pushing beyond these traditional boundaries.

Mohammed El-Kebir, an Associate Professor at the University of Illinois, and his research team have created a novel AI tool that processes high-throughput phenotyping data—such as hyperspectral light reflectance—to identify what the researchers call “synthetic traits”.

“We wanted find new indices beyond NDVI, beyond other vegetation indices that have been studied before,” El-Kebir said. “We actually wanted to use heritability as a guide. So we wanted to find new indices, what we call synthetic traits, that have high heritability.”.

To handle the overwhelming scale of high-throughput phenotyping data, the team built a new AI method from the ground up, dubbed H2-opt (Heritability Optimization).

“We needed to come up with new mathematics to do that,” El-Kebir explained. The H2-opt system utilizes differentiable function classes, such as linear models or neural networks, to ensure complex heritability computations can be seamlessly optimized. By selecting synthetic traits based purely on co-heritability estimations, the AI can significantly boost the overall accuracy of genomic prediction models for target traits without needing to immediately define the biological function of the new trait.

The team validated the H2-opt method using a robust dataset of 869 different sorghum lines. The AI successfully discovered completely new, hidden traits that traditional manual phenotyping missed. To confirm the accuracy of these discoveries, researchers cross-referenced the new traits with single nucleotide polymorphism (SNP) genotyping data.

“We actually went back to the SNP genotyping data, and we tried to find associations between SNPs and our new traits,” El-Kebir noted. “We did indeed find some hits. It does show that there is genetic basis, that our traits are indeed genetically driven”.

While early testing has focused on a relatively small number of plants, the research team plans to scale up operations. By flying drones over larger fields, they aim to further decode these synthetic traits and understand their biological functions. Ultimately, this AI-driven approach could serve as a powerful shortcut for crop breeders, rapidly accelerating the development of stronger, higher-yielding, and more resilient crops.

Editor’s note: This article was adapted from a radio broadcast script and formatted for print with the assistance of Google’s generative AI tool, Gemini and has been reviewed by the primary source.