Artificial intelligence (AI) is gaining attention in various scientific fields as researchers from agricultural, biological, and technological backgrounds collaborate to utilize algorithms and models to analyze datasets and better understand and predict a world impacted by climate change. A recent study published in Frontiers in Plant Science by Purdue University geomatics PhD candidate Claudia Aviles Toledo, along with her faculty advisors Melba Crawford and Mitch Tuinstra, demonstrated the use of a recurrent neural network to predict maize yield from remote sensing technologies and environmental and genetic data. Remote sensing, including the use of UAVs and satellites, is making plant phenotyping, the process of examining and characterizing plant characteristics, more efficient and cost-effective.
The advancement in UAV-based data acquisition and processing, coupled with deep-learning networks, is shown to contribute to predicting complex traits in food crops like maize. Hyperspectral cameras and LiDAR instruments on robots and UAVs can now provide detailed reflectance measurements and generate maps of the geometric structure of plants. This technology enables the collection of novel information on plants that human senses alone cannot discern, allowing researchers to analyze patterns within massive datasets to predict outcomes such as crop yield and crop stress.
Plant breeders like Tuinstra are incorporating genetic data into AI models to understand how different traits react to varying conditions and select traits for future resilient crop varieties. The neural network developed in the study combines remote-sensing data, genetic markers of corn varieties, and environmental data to make predictions about future outcomes. By training the network in one location or time period, it can be updated with limited training data in another geographic location or time, reducing the need for reference data and providing valuable insights to growers about which varieties might perform best in their region.
The neural network model used in the study includes long short-term memory, which allows past data to be constantly considered alongside present data as it predicts future outcomes. Attention mechanisms in the model highlight physiologically important times in the plant growth cycle, such as flowering. While remote sensing and weather data are already incorporated, researchers aim to further process genetic data and incorporate complex traits into their dataset to provide growers with more meaningful information to make informed decisions about their crops and land. This approach has the potential to reduce labor costs and improve crop management practices.
Overall, the collaboration between researchers from different scientific backgrounds highlights the potential of AI in revolutionizing the way plant phenotyping and crop prediction are conducted. By combining advanced technologies such as remote sensing and deep-learning networks with genetic data, researchers are paving the way for more efficient and precise methods of understanding and predicting plant traits and crop outcomes. This multidisciplinary approach is crucial in addressing the challenges posed by climate change and ensuring food security for future generations.