Estimation of soybean seed protein accumulation by measuring canopy hyperspectral reflectance

Within the H2020 project ECOBREED, a set of soybean genotypes widely differing in seed protein content was tested in three environments in Austria in replicated single-row plots. Between the soybean developmental stages of full flowering (R2) and full seed (R6), i.e. the seed filling period, canopy hyperspectral reflectance data were collected at about weekly time intervals using a hand-held spectroradiometer. Reflectance data at particular wavelength points were then utilized for calculating indices or regression models for predicting seed protein content. From over 40 spectral reflectance indices calculated, nitrogen reflectance index (NRI), greenness index (GI) and different ratio vegetation indices (RVI) with optimized wavelengths for soybean canopies revealed highest correlations to seed protein content within particular environments. These indices were also highly correlated with leaf chlorophyll content (SPAD-meter-values). The ratio vegetation index term MA1_R showed the highest correlation to seed protein content across all three environments and all genotypes. In a second approach utilizing the whole range of spectral data available and partial-least-square regression modelling (PLSR), correlations of up to r = 0.91 were achieved for seed protein content data across all three environments. In addition to seed protein content which is directly related to nitrogen fixation, hyperspectral reflectance data appeared to be useful for prediction of additional traits such as time to maturity, oil content or 1000-seed weight as well. Moreover, on the level of individual genotypes, particular indices could be utilized for better characterization of genotypes in terms of water use efficiency, biomass production, and nitrogen metabolism.