Indirect selection based on multispectral aerial images in the early stages of a wheat breeding program

Abstract To optimize a breeding program's phenotyping process, vegetation indices (VIs) obtained from multispectral images can serve as potential tools for the indirect selection of wheat ( Triticum aestivum L.) families. This study evaluates the use of VI and multispectral bands for the indirect selection of superior families in the early stages of a wheat breeding program. We analyzed 852 wheat families in the F 2:4 generation, grown under an augmented block design. Images of the experimental field were collected using a multispectral camera attached to a drone at three timepoints: heading, and 15 and 30 days later. The selection and selection gain (SG) were performed via the multitrait genotype‐ideotype distance index (MGIDI). The receiver operating characteristic (ROC) curve was used to evaluate the classification performance of the VI/bands. ROC analysis revealed that, at the heading stage, the red, green, blue, and near‐infrared (NIR) bands and the modified chlorophyll absorption in the reflectance index provided the best discrimination, with area under curve values up to 0.70, indicating satisfactory classification performance. However, indirect selection resulted in limited SG: the best performing classifier, NIR band, achieved only 18% of the gain obtained using the MGIDI index. Despite this limitation, early spectral data, particularly from the heading stage, proved useful for identifying superior families. These findings highlight the potential and constraints of using multispectral data for indirect selection in early breeding stages and emphasize the importance of growth stage and variable choice in improving classification performance.