Remote sensing data and machine learning models estimate sorghum grain yield in a plant breeding program

Authors: Pugh, Nicholas; Young, Andrew; Nesbitt, Matthew; Hayes, Chad

Submitted to: Plant Phenomics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/30/2026
Publication Date: 2/5/2026
Citation: Pugh, N.A., Young, A.W., Nesbitt, M.E., Hayes, C.M. 2026. Remote sensing data and machine learning models estimate sorghum grain yield in a plant breeding program. Plant Phenomics. 8(2). https://doi.org/10.1016/j.plaphe.2026.100179.
DOI: https://doi.org/10.1016/j.plaphe.2026.100179

Interpretive Summary: Drought-resistant, high-yielding sorghum is vital for feeding a growing population, but breeding new varieties requires a great deal of labor and money. This “phenotyping” step—recording how much each plant produces—is the biggest roadblock in speeding up sorghum improvement. To tackle this, we flew small drones over experimental fields and took overhead images at key growth stages. From those photos, we pulled out simple traits—how green the leaves were, how tall the plants grew, and details of the grain heads—using an automated object-detection program. We then tested six different computer-learning methods to see how well they could predict the actual grain weight harvested from each plot. Our best models, which used leaf greenness and plant height, accurately predicted 74–76% of the differences in yield. Grain-head measurements alone were much less reliable (28–42%). We also found that straightforward regression formulas gave more consistent and repeatable results than complex “tree” methods. By showing that drones plus simple image analysis can predict yield almost as well as hands-on measurements, we’ve taken a big step toward faster, non-destructive phenotyping. Breeders can use this approach to screen more plant lines in less time and at lower cost. Future work will add images taken at multiple time points and refine how we pick out plant features, making predictions even more accurate.

Technical Abstract: Phenotyping remains a critical bottleneck in sorghum (Sorghum bicolor L. Moench) breeding programs, limiting genetic gain due to labor-intensive yield estimation methods. To address this concern, this study investigates the potential of integrating remote sensing data with machine learning (ML) and deep learning (DL) models to improve sorghum grain yield predictions. UAS-based imagery was collected across multiple field trials, extracting standard vegetation indices, canopy height features, and panicle traits using a YOLOv11-based object detection model, YOLO-SORG. Six ML models—including ridge regression (RR), elastic net (EN), LASSO regression (LR), support vector regression (SVR), Random Forest (RF), and XGBoost (XGB)—were trained to predict plot-level yield using three distinct feature sets: panicle traits, canopy traits, and a combination of both. Results indicate that models relying solely or partially on canopy-derived features provided the most consistent and accurate yield estimates (R² ˜ 0.74–0.76), whereas models relying solely on panicle traits performed poorly (R² ˜ 0.28–0.42), indicating nadir-derived panicle metrics were potentially being indirectly captured with the canopy traits. Traditional regression models outperformed tree-based ensemble methods in variance partitioning and repeatability (R ˜ 0.59-0.60), making them more suitable for breeding applications. The findings highlight the promise of UAS-driven ML pipelines for non-destructive yield prediction but underscore limitations of nadir imagery for capturing panicle morphology and use in a robust yield prediction model. Future research should explore the inclusion of multi-temporal imaging and refined feature extraction approaches to enhance predictive accuracy in sorghum breeding programs.