Predicting Grain Protein Content in Wheat Using Hyperspectral Sensing of In-season Crop Canopies and Partial Least Squares Regression

The ability to estimate and map grain protein in cereal crops prior to harvest can benefit Australian grain growers. Segregation of grain by protein content can take advantage of price premiums, as well as to retrospectively assess the effectiveness of nutrient application strategies. This paper explores the relationships between hyperspectral data and grain protein content (GPC) in wheat (Triticum aestivum), with a view of developing predictive regression models. Canopy-scale, ground-based measurements of hyperspectral reflectance were obtained from samples located in the Formatin district of the Darling Downs, Queensland, Australia. Using partial least squares (PLS) regression, we investigated if the raw reflectance spectra, transformed data, and spectral vegetation indices (SVIs) could adequately predict grain protein content. The results showed that there are high correlations (e.g. r=0.86, r=0.81, r=0.80) between reflectance data and grain protein. Cross-validated and tested PLS regression models produced low root mean square error of prediction (RMSEP) values (e.g. 0.5 percent GPC) and high prediction accuracy (e.g. 92%), confirming the usefulness of narrow-band spectral data. Bands in the near infrared (NIR) region were the most significant variables in the prediction. Despite the slightly higher correlation coefficients of SVIs, their predictive power for grain protein estimation was generally comparable with those of the raw spectra when analysed using PLS regression.