Figure 4.

Predicted versus measured barley leaf P concentrations based on a PLS model with three principal components. A, PLS prediction in the calibration range. Gray circles indicate calibration samples, and red circles indicate validation samples. The blue curve indicates the optimal Y = X line, and the red curve represents best linear fit to data (R² = 0.8). The root-mean-squared error for the 159 calibration samples (root-mean-squared error of calibration [RMSEC] = 369) is seen to be very equivalent to the root-mean-squared error for the 131 independent test samples (root-mean-squared error of prediction [RMSEP] = 379). Together with the low number of principal components used, this shows that the model is not a result of overfitting. B, Predicted versus measured P concentrations for all 448 OJIP transients. The blue curve represents the optimal fit of a Y = X line intersecting a constant line at 2,975 µg g^–1 DW. The two dotted lines indicate 2,000 µg g^–1 DW, and coloring indicates whether the PLS model predicts samples correctly according to this threshold. Two hundred sixty-nine of 291 leaves (92%) from the test set were correctly classified as above or below the 2,000 µg g^–1 DW threshold, and the misclassified leaves position close to the threshold except for one measurement.