Near-infrared spectroscopy used to predict soybean seed germination and vigour

Abstract

Rapid, non-destructive methods for measuring seed germination and vigour are valuable. Standard germination and seed vigour were determined using 81 soybean seed lots. From these data, seed lots were separated into high and low germinating seed lots as well as high, medium and low vigour seed lots. Near-infrared spectra (950-1650 nm) were collected for training and validation samples for each seed category and used to create partial least squares (PLS) prediction models. For both germination and vigour, qualitative models provided better discrimination of high and low performing seed lots compared with quantitative models. The qualitative germination prediction models correctly identified low and high germination seed lots with an accuracy between 85.7 and 89.7%. For seed vigour, qualitative predictions for the 3-category (low, medium and high vigour) models could not adequately separate high and medium vigour seeds. However, the 2-category (low, medium plus high vigour) prediction models could correctly identify low vigour seed lots between 80 and 100% and the medium plus high vigour seed lots between 96.3 and 96.6%. To our knowledge, the current study is the first to provide near-infrared spectroscopy (NIRS)-based predictive models using agronomically meaningful cut-offs for standard germination and vigour on a commercial scale using over 80 seed lots.