Limitations of phenomic prediction for evaluating wheat stem sawfly resistance in wheat breeding programs

Abstract Wheat stem sawfly (WSS, Cephus cinctus Norton) threatens wheat ( Triticum aestivum L.) production in the US Great Plains. Increased stem solidness improves resistance to WSS but developing solid‐stemmed cultivars requires time‐consuming and destructive phenotyping. To expedite development of WSS‐resistant cultivars, a high‐throughput phenotyping method is needed. Therefore, we assessed phenomic prediction using uncrewed aerial systems (UAS) to predict stem solidness, WSS infestation, and yield in wheat. Multispectral and red‐green‐blue UAS data were collected in‐season at two naturally infested locations in western Nebraska from 2023 to 2024. The spectral reflectance indices from the UAS data were compared with agronomic traits (i.e., yield and plant height) and WSS traits (i.e., stem solidness and WSS infestation). Ridge regression, k ‐nearest neighbors (KNN), and random forest (RF) models were then trained to use spectral indices to predict yield, stem solidness, and WSS infestation. Correlations between WSS and spectral traits were temporally and environmentally dependent. The best prediction model depended on the biological trait. RF performed the best for yield ( r = 0.469), KNN for stem solidness ( r = 0.231), and ridge regression for WSS infestation ( r = 0.194). Predicting traits based on spectral data in a new environment was poor for both stem solidness ( r = 0.02–0.38) and WSS infestation ( r = 0.01–0.22). Our ability to predict WSS resistance was low, and UAS‐based phenotyping was not viable with current technology.