4.2.1. Results for the PLS-Full Model

At the plant level (Table 3, top), the PLS-Full models varied considerably in performance between the two sites with cross-validated calibration coefficients of determination (*R*<sup>2</sup> *CV*) ranging from 0.03 (Mg US) to 0.88 (Mg ET) at the plant level. The validation coefficients of determinations (Mean *R*2) closely matched the calibration values, except in the case of ET (Ca and Protein (SG)) where the validation *R*<sup>2</sup> was approximately double that of the calibration. Interestingly, when comparing the US and ET, one generally outperformed the other, but the relationship changed depending on the nutrient. Differences in filtering (SG vs. FD) were largely negligible, with the exception of protein in ET, where the model fit for the FD was about twice that for the SG data (0.41 vs. 0.18). The bootstrapped *R*<sup>2</sup> values were normally or semi-normally distributed in all cases except for Mg in the US and Ca in ET (Figure S1, Supplementary Material). Importantly, *t*-test comparisons for replicability indicate

the PLS-Full regression model did not replicate well across the two environments, with significant differences in fit for every nutrient at the plant level (Table 3, top).



\*\*\* *t*-test significant at 0.001, ET served as first group.

At the grain level (Table 3, bottom), results were more consistent across the study sites and nutrients. This is believed to be the case in-part to grain spectral values measured in a controlled setting, while plants were measured in a field setting. Calibration *R*<sup>2</sup> ranged [0.47, 0.93], and validation *R*<sup>2</sup> ranged [0.49, 0.90] (Table 3). The results for ET protein using the SG data were a slight outlier though, and when considering only the FD values, the minimums increase to 0.64 and 0.62, respectively. The differences between the SG and FD filters were again negligible, except in the Mg ET case. The *t*-test comparisons for grain indicate the PLS-Full regression method did not replicate well across the environments.
