Fenotipagem digital integrada a modelos mistos para seleção de genótipos em trigo

Abstract

High Throughput Phenotyping (HTP), based on images acquired by unmanned aerial vehicles (UAVs), has emerged as a promising tool to complement conventional phenotyping by enabling rapid, precise, and non-destructive data collection on plant development. In wheat (Triticum aestivum L.) breeding programs, the integration of digital phenotypes with robust statistical methods may increase the efficiency of selecting superior genotypes. In this context, the objective of this study was to evaluate the potential use of digital phenotypes derived from multispectral drone imagery as auxiliary variables in selection methods for grain yield in wheat under different statistical modeling approaches. Public data from a wheat breeding experiment conducted by the North Dakota State University (NDSU) breeding program were used. The dataset comprised 106 wheat lines evaluated in a randomized complete block design during the 2021 growing season. Throughout the crop cycle, 15 UAV flights equipped with a multispectral sensor were performed, generating RGB orthomosaics, multispectral images, and digital surface models. Statistical analyses were conducted in the R environment, including the estimation of heritability, correlations between digital phenotypes and agronomic traits, and the comparison of different selection methods based on observed means, Best Linear Unbiased Estimators (BLUE), and Best Linear Unbiased Predictions (BLUP), with and without the inclusion of secondary variables derived from digital phenotyping. The results revealed high heritability for some digital phenotypes, particularly traits associated with estimated plant height and individual bands. Correlations among the different selection models ranged from 0.47 to 0.99, indicating agreement in genotype ranking. Overall, the evaluated models showed high consistency in identifying superior lines, suggesting that simpler statistical approaches may be applied without substantial loss of information in genotype selection.