Aerial high‐throughput phenotyping enables indirect selection for grain yield at the early generation, seed‐limited stages in breeding programs

Abstract

Breeding programs for wheat (Triticum aestivum L.) and other crops require one or more generations of seed increase before replicated trials can be sown to assess yield. Extensive phenotyping at this stage is challenging because of the small sizes of plots and large numbers of lines under evaluation, and therefore, breeders typically rely on visual selection to promote lines to yield evaluation. Aerial high‐throughput phenotyping (HTP) enables the rapid acquisition of traits that may be useful for selection among early generation lines. With the objective of assessing the potential for aerial measurements recorded on seed increase plots to improve indirect selection for grain yield (GY), two sets of 1,008 early generation bread wheat breeding lines were sown both as replicated yield trials (YTs) and as small, unreplicated plots (SPs) at the International Maize and Wheat Improvement Center during two breeding cycles. Normalized difference vegetation indices (NDVI) collected with an unmanned aerial vehicle (UAV) in the SPs were observed to be heritable and moderately correlated with GY assessed in YTs. Furthermore, NDVI was more predictive of GY than univariate genomic selection (GS), with still higher overall predictive abilities from multitrait approaches. A related experiment showed that selection based on NDVI would have outperformed visual selection, though this approach would have driven a directional response in phenology because of confounding between phenology, NDVI, and GY. A restricted selection index was proposed to address this issue. These results provide a promising outlook for the use of aerial HTP to improve selection at the early generation, seed‐limited stages of breeding programs.

Convolutional Neural Networks to Estimate Dry Matter Yield in a Guineagrass Breeding Program Using UAV Remote Sensing

“…Indirect selection is one of the main applications of HTP in breeding programs [32]. The potential of HTP in improving the efficiency of early generation selection has been studied in sugarcane [30], and wheat [31]. In both studies, the results were encouraging to use HTP traits and indirect selection.…”

“…Natarajan et al [30], using indirect selection based on NDVI as HTP trait, reached 73% of the efficiency of direct selection for yield in sugarcane. Also, using NDVI, Krause et al [31] reported a higher efficiency of indirect selection when compared to visual selection for grain yield in lines of wheat in early generations. Thus, these results in different crops show that HTP is a promising strategy to improve the efficiency of breeding programs.…”

“…Field HTP using remote sensing and CNNs generate new traits throughout the image detection, classification, and retrieval process. These traits can be used in different selection methods in the breeding program, such as indirect selection [30,31], index selection, or in prediction with genomic selection [32]. In the indirect selection, a secondary trait (e.g., HTP trait) can replace the main trait (e.g., DMY) in selection criteria if the secondary trait is highly heritable and correlated to the main trait [33].…”

Genomic Selection for Forest Tree Improvement: Methods, Achievements and Perspectives

“…Juliana et al [98] used aerial HTP platforms in the wheat genomic selection for grain yield under stress conditions (drought, heat) and found that it increased selection intensity due to the large populations used. An unmanned aerial vehicle (UAV) was first used to assess the heritability of vegetation index traits measured on small unreplicated plots and to estimate the extent to which they are predictive of wheat grain yield in replicated yield trials [99]. In that study, aerial HTP provided a substantially better response to selection for grain yield than conventional visual selection.…”

Multi-Species Genomics-Enabled Selection for Improving Agroecosystems Across Space and Time

“…Modern plant and animal breeding is a predictive, data-driven, multi-disciplinary science. Statistical prediction methods that leverage genomic and phenomic data (e.g., drone-based hyperspectral imaging) are greatly accelerating the rate of population genetic improvement ( Jannink et al, 2010 ; Hickey et al, 2017 ; Voss-Fels et al, 2019 ; Krause et al, 2020 ). Decision support tools based on these technologies are now available to large-acreage monoculture systems.…”