Insights on multi-spectral vegetation indices derived from UAV-based high-throughput phenotyping for indirect selection in tropical wheat breeding

High-throughput phenotyping (HTP) approaches are potentially useful for designing indirect selection strategies, which consists in the selection of a primary target trait X based on secondary trait Y. Usually, the secondary trait Y is correlated to the target trait X and is easier, faster and cheaper to measure. In this context, the use of secondary traits (such as vegetation indices derived from HTP platforms) could potentially lead to a fast, non-invasive, accurate and efficient selection of superior genotypes. Considering the lack of information in the literature regarding high-throughput phenotyping approaches in tropical wheat breeding, this study aimed to (i) determine the best stages to carry out image acquisition for applying multi-spectral vegetation indices for genotype evaluation and selection; (ii) evaluate the heritability and accuracy of multi-spectral vegetation indices; (iii) understand the relationships between vegetation indices and target agronomic traits; and (iv) evaluate the efficiency of indirect selection via UAV-based high-throughput phenotyping. A diversity panel of 49 tropical wheat cultivars was evaluated during the 2022 winter season. Weekly flight campaigns were performed to further build the dataset with multi-spectral vegetation indices, which were then analyzed together with four target agronomic traits. Statistical analysis based on Mixed Effect Model was performed to estimate genetic parameters and predict genetic values, which were subjected to correlation analysis. Additionally, factor analysis was applied, and the factorial scores were used in an indirect selection strategy (indirect via HTP). This strategy was compared to three alternative strategies: direct via grain yield, direct via days to heading, and the multi-trait genotype-ideotype distance index. The results indicate that vegetation indices are suitable for indirect selection strategies and highly efficient for the indirect selection of grain yield and cycle. These findings will help in the decision making regarding the adoption of remote or proximal sensing-based approaches in Brazilian public wheat breeding programs.