Validation of high-throughput phenotyping tools for selection in groundnut (Arachis hypogaea L.) breeding

Abstract

Phenotyping is an important aspect of plant breeding. It is essential in conventional and molecular breeding programs alike. Plant breeding is a game of numbers and breeding programs usually deal with very large numbers of breeding lines and there is a need by breeders to phenotype these populations fast without compromising the quality of the phenotypic data. However, this remains a problem in many breeding programs, including groundnut improvement programs in Sub-Saharan Africa. High-throughput phenotyping (HTP) has been suggested as a solution to bridge the phenotyping gap in the breeding cycle. HTP methods have successfully been integrated into breeding programs of cereals like wheat, barley, maize, and sorghum but have not been tested for application for routine application in groundnut selection. These HTP methods could also be applied to groundnut breeding and therefore, the main objective of this study was to contribute to the development of HTP models for routine selection for LLS, GRD, Drought tolerance, and variety performance in groundnut. Fifty genotypes from NaSARRI and ICRISAT breeding programs were used in this and were planted across two locations; Nakabango and NaSARRI. The genotypes were also planted in the screen house to test for drought tolerance. Data was collected using both traditional and HTP methods. Traditional measurements included visual scores and manual measurements while HTP measurements included the use of sensors like RGB camera, Thermal infrared camera, and GreenSeeker hand-held scanner. Results of the first objective indicate that there is a variation of both agronomic traits and HTP measurements within the population. This variation can be utilized by breeders for selection for various traits of interest. Results of the second objective indicate that HTP measurements were highly related to agronomic parameters. HTP measurements collected closest to traditional measurements had the highest correlation with the agronomic traits. RGB indices hue and GA were highly correlated with plant vigor and biomass, Normalised difference vegetation index (NDVI) and Crop senescence index (CSI) were highly correlated with LLS, CSI was highly correlated with leaf wilting, and greener area (GGA) was highly correlated with both pod and kernel yield. These indices were used to develop regression models under the third objective. The models had 95% accuracy for GRD, 98% accuracy for LLS, 99% accuracy for leaf wilting, and 70% accuracy for plant biomass. These models can be applied to breeding programs for routine use for selection. This study therefore presents novel and alternative methods of phenotyping for LLS and GRD resistance, leaf wilting, and variety performance using simple handheld sensors.