APPLICATIONS OF UAS IMAGERY IN WHEAT BREEDING

Andrew Walter Herr

doi:10.7273/000006521

Plant breeding is a field of study with goals that have not changed significantly over time: develop cultivars with high yield, disease resistance, and drought tolerance, to name a few. While the goals of a breeding program may not change frequently, the form and technology used with which those goals are achieved are constantly evolving. High throughput phenotyping (HTP) with unoccupied aerial systems (UAS) shows significant promise in improving how crops are bred. Data collected from UAS can provide a breeder with new insights into how cultivars respond to stress and a particular environment, creating potential use cases for improving other areas of breeding, such as genomic selection and how field experiments are designed and analyzed. These new technologies, however, should not be adopted without consideration. The first study, outlined here, utilized three different HTP platforms and collection methodologies, two ground systems and one UAS-based, to determine if there is a difference in the quality of data collected. Across four years, data collected from ground systems only moderately correlated to UAS. It was also shown that data collected with UAS produced more heritable data than that collected with either ground-based system. While manufacturing specifications of the data collected from remote sensors may be similar, it is essential to be aware of the methodology used in the collection. Reflectance data standardization, sensor platform, and environmental conditions can significantly impact the quality of the data obtained and limit utility across platforms and methodologies. In the second study, spectral reflectance indices (SRI) were evaluated for their ability to improve genomic selection (GS). SRIs collected on 11,593 plots across four years were used with genomic data in univariate models as covariates and in multivariate models as secondary response variables for the assessment of prediction accuracy of grain yield. Including SRI data as covariates in univariate genomic prediction models improved prediction accuracy over the control GS model but was unreliable across years. In multivariate models, SRIs improved prediction performance across years, but due to the dataset size, high-performance computational resources were required, which could limit feasibility in an applied setting. The final study highlights the potential for SRI to improve how a breeder deals with field variability in yield trial experiments. Across three years, 47 breeding trials were evaluated under three spatial analysis strategies: linear models incorporating block-effect, row-column effect, and 2D splines. Model fit was improved across all spatial analysis methods when SRIs were incorporated as covariates. Model fitness was most greatly improved in unreplicated early-generation trials. This study highlighted the potential of SRIs to enhance how breeding trials are analyzed despite extreme environmental variables and climate conditions. This collective research highlights the challenges and benefits of utilizing UAS imagery in an applied breeding pipeline. When used strategically, the insights gained from UAS will, like genomic selection, make it an invaluable tool in the plant breeder's toolbelt.

APPLICATIONS OF UAS IMAGERY IN WHEAT BREEDING

Files and links (1)

Abstract

Metrics

Details