Visible and Near-Infrared Spectroscopy Analysis of a Polycyclic Aromatic Hydrocarbon in Soils

Table 3

Sample statistics and results of partial least-squares (PLS) models for the prediction of phenanthrene in cross-validation and prediction datasets for diesel-contaminated soil samples by visible and near-infrared (VisNIR) spectroscopy.

Variable statistics						Model quality
No. of samples	Min. (mg kg^−1)	Max. (mg kg^−1)	Mean (mg kg^−1)	SD	No. of outliers removed	Reflectance spectra				Category^a	First derivative spectra				Category^a
						𝑟^2	RMSE (mg kg^−1)	LV	RPD		𝑟^2	RMSE (mg kg^−1)	LV	RPD
Cross-validation set (76%)
114	0.58	2.49	1.18	0.48	3	0.65	0.28	10	1.71	B	0.62	0.30	6	1.63	B
Prediction set (24%)
36	0.63	2.20	1.40	0.50	N/A	0.83	0.21	10	2.32	A	0.75	0.25	6	2.00	A

aCategory of prediction is the ability of PLS regression analysis for parameter validation and prediction. A if RPD > 2.0, B if 1.4 ≤ RPD < 2.0, and C if RPD < 1.4 [27].
LV: latent variable; N/A: not applicable; RMSE: root-mean-square error; RPD: residual prediction deviation; SD: standard deviation.