Characterization of soybean seed yield using optimized phenotyping

Abstract

Crops research moving forward faces many challenges to improve crop performance. In breeding programs, phenotyping has time and economic constraints requiring new phenotyping techniques to be developed to improve selection efficiency and increase germplasm entering the pipeline. The objectives of these studies were to examine the changes in spectral reflectance with soybean breeding from 1923 to 2010, evaluate band regions most significantly contributing to yield estimation, evaluate spectral reflectance data for yield estimation modeling across environments and growth stages and to evaluate the usefulness of spectral data as an optimized phenotyping technique in breeding programs. Twenty maturity group III (MGIII) and twenty maturity group IV (MGIV) soybeans, arranged in a randomized complete block design, were grown in Manhattan, KS in 2011 and 2012. Spectral reflectance data were collected over the growing season in a total of six irrigated and water- stressed environments. Partial least squares and multiple linear regression were used for spectral variable selection and yield estimation model building. Significant differences were found between genotypes for yield and spectral reflectance data, with the visible (VI) having greater differences between genotypes than the near-infrared (NIR). This study found significant correlations with year of release (YOR) in the VI and NIR portions of the spectra, with newer released cultivars tending to have lower reflectance in the VI and high reflectance in the NIR. Spectral reflectance data accounted for a large portion of variability for seed yield between genotypes using the red edge and NIR portions of the spectra. Irrigated environments tended to explain a larger portion of seed yield variability than water-stressed environments. Growth stages most useful for yield estimation was highly dependent upon the environment as well as maturity group. This study found that spectral reflectance data is a good candidate for exploration into optimized phenotyping techniques and with further research and validation datasets, may be a suitable indirect selection technique for breeding programs.