Prediction models for canopy hyperspectral reflectance in wheat breeding data

Vegetation indices (VI) generated by using some bands from hyperspectral cameras are used as predictors of primary traits. This study proposes models that use all available bands as predictors of primary traits. The proposed models were ordinal least square (OLS), Bayes B, principal components with Bayes B, functional B-spline, functional Fourier and functional partial least square (PLS). The results were compared with the OLS performed using as predictors each of the eight VIs individually and combined. The data set comes from CIMMYT’s Global Wheat Program and comprises 1170 genotypes evaluated for grain yield in five environments with the reflectance data measured in 250 discrete narrow bands ranging between 492 and 851 nm. in 9 time-points of the crop cycle. Results show that using all the bands simultaneously produced better predictions than using one VI alone or all the VI together, but when used only the bands with heritabilities > 0.5 in Drought environment, the predictions improved, while in the rest of the environments, using all the bands simultaneously produced slightly better prediction accuracies. The models with highest prediction when using all bands were functional B-spline and Fourier. Time-point 6 gives gave promising prediction accuracies for wheat lines before harvesting.