The changing legal status of Cannabis sativa and derived products has drawn global research interest. In the U.S., the 2018 Farm Bill removed hemp from Schedule I of the Controlled Substances Act, defining it as cannabis containing ≤0.3 % total delta-9-tetrahydrocannabinol (Δ9-THC) on a dry weight basis. This study explores near-infrared (NIR) spectroscopy combined with statistical data analysis as a rapid method for differentiating cannabis plant materials. Two main objectives were pursued: (i) classification of cannabis samples using a 2 % total Δ9-THC threshold and (ii) quantitative estimation of total Δ9-THC. Partial least squares discriminant analysis (PLS-DA) yielded highly accurate classification models, with 98.9 % accuracy in cross-validation and 96.7 % accuracy in the test set (60/40 split). Only one high-cannabigerol (CBG) sample was misclassified. Partial least squares (PLS) regression produced two quantitative models: a full-range model and a low-range (<1 % Δ9-THC) model. The full-range model showed RMSECV and RMSEP of 0.673 % and 0.741 %, respectively; the low-range model showed improved performance with RMSECV of 0.079 % and RMSEP of 0.073 %. Two high-cannabidiol (CBD) samples (1–2 % Δ9-THC) were well predicted by the full-range model but poorly by the low-range model due to atypical THC/CBD ratios. Comparison of PLS loadings and spectra of pure cannabinoids indicated the low-THC model relied heavily on CBD features due to strong THC–CBD correlation. While further development is needed, these results demonstrate that NIR spectroscopy with multivariate analysis holds strong potential for rapid classification and quantification of Δ9-THC in cannabis plant materials.
(Publisher abstract provided.)