The analysis and spectrum exploration were performed on the OH first overtone in the region of 1300–1600 nm, according to the guidelines published in Tsenkova et al. (Fig. 1) [38]. Averaging the spectral data of samples according to either major classes or chemovars was performed to minimize the influence of variations not of primary interest, such as humidity levels [38]. This approach aimed to uncover significant differences between the sample groups concerning the interactions of the water matrix with the secondary metabolites of cannabis inflorescence [38]. The hybrid major class displayed the highest average reflectance values in the entire region of the OH first overtone, while the high CBDA major class exhibited larger average reflectance values than the high THCA major class in the region of 1300–1380 nm and 1550–1600 nm (Fig. 1b). On the other hand, the latter displayed larger average reflectance values in the region of 1380–1550 nm as compared to the high CBDA major class (Fig. 1b). The chemovar analysis revealed that the 621–17 chemovar displayed the largest average reflectance values in the region of 1380–1600 nm, while the 240 and 505 chemovars had the lowest average reflectance values in the entire region of the OH first overtone (Fig. 1e). As revealed by the standard deviation plots, the highest variation in most major classes and chemovars is observed at 1450 nm (C8, Fig. 1c and f) [36].

FT-NIR raw reflectance spectra in the wavelength region of 1300–1600 nm of the complete set of samples labeled according to a major classes and d chemovars. Additionally, b the mean spectra and c standard deviation of the different major classes and e the mean spectra and f standard deviation of the different chemovars are provided

After averaging, the raw data was preprocessed using 2nd derivative, followed by MSC and smoothing, in order to discover activated water bands which are not visible in the original averaged spectrum (Fig. 2) [38,39,40,41]. These spectrum exploration steps revealed several water absorbance bands activated in the different cannabis chemovars, i.e., 1348 nm (C1), 1382 nm (C4), 1404 nm (C5), 1433 nm (C7), 1449 nm (C8), and 1466 nm (C9) [36]. The largest variation in the preprocessed spectrum for both major classes and chemovars was obtained at 1404 nm (C5), 1433 nm (C7), 1449 nm (C8), and 1466 nm (C9) (Fig. 2).

Preprocessed FT-NIR reflectance spectra using 2nd derivative followed by MSC and smoothing in the wavelength region 1300–1600 nm of a the complete set of samples and the mean preprocessed FT-NIR reflectance spectra of b major classes and c chemovars

PLS-DA classification models were developed based on the OH first overtone spectral data to distinguish between major classes and chemovars. These models aimed to identify additional water-activated absorbance bands and evaluate whether classification can be achieved using the OH first overtone spectrum only [38, 39, 42]. To reveal such bands, the loadings and regression vectors of these classification models were analyzed following the guidelines established by Tsenkova et al. [38].

The PLS-DA classification yielded an excellent major class separation and prediction model, utilizing eight LVs (Table 1). The calibration, cross-validation, and prediction groups displayed high sensitivity, specificity and accuracy values (> 0.94, Table 1). Only two misclassifications of the high CBDA major class were observed in the cross-validation data set, whereas no misclassifications were obtained for the independent prediction group. Consequently, sensitivity and specificity values approaching unity were obtained, indicative of highly accurate classification model [53]. The root mean square error of calibration (RMSEC), root mean square error of cross-validation (RMSECV), and root mean square error of prediction (RMSEP) values for the different major classes ranged between 0.190 and 0.241, and the RMSECV/RMSEC and RMSEP/RMSECV ratios were below 1.1, indicative of negligible model over-fitting (Table 1) [19, 54]. Overall, the PLS-DA model provided an excellent classification model of all cannabis major classes.

The chemovar PLS-DA classification model utilizing seven LVs yielded good separation and good chemovar predictive capability (Table 2). The calibration, cross-validation, and prediction groups displayed specificity and accuracy values higher than 0.96, while the sensitivity values were higher than 0.87, except for chemovar 156, which displayed a sensitivity value of 0.75 (Table 2). Furthermore, only eight samples, were misclassified in the cross-validation dataset, three of these sample belong to the 156 chemovar, whereas only one samples was misclassified in the independent prediction group. The RMSEC, RMSECV, and RMSEP values for chemovar classification ranged between 0.184 and 0.273, similar to the corresponding major class classification model values, indicating similar classification performances (Table 2). The RMSECV/RMSEC and RMSEP/RMSECV ratios were below 1.1, indicative of negligible model over-fitting to the data (Table 2). Taken altogether, the classification models provided good to excellent major class and chemovar classifications, and therefore, the loadings and regression vectors of these models were used to further reveal additional water activated absorbance bands.

The bands that exhibited the highest local positive or negative contributions to the classification models, as indicated by the LV loadings and regression vector plots, were identified and recorded as water-activated absorbance bands (Fig. 3) [38]. Although the first two latent variables (LVs) accounted for over 99% of the data variation, it is essential to examine all LV loadings since subtle changes in the water matrix may be captured by higher-order LV loading vectors, which could provide additional significant information [38]. Summarizing the water activated absorbance bands obtained by each LV loadings and regression vector plots revealed that six activated water bands (i.e., C1, C2, C4, C5, C7, and C9) consistently occurred following the aquaphotomics analysis and, thus, may be considered as more informative than the other bands (Table S2) [36, 38]. Five of these bands (i.e., C1, C4, C5, C7, and C9) were also identified in the spectrum exploration step (Figs. 1 and 2). The identification of these bands is crucial for understanding the aqueous system and the water matrix structure of the different cannabis chemovars, as illustrated in aquagrams in the following section [38].

a LV loadings and b regression vectors plots of major class PLS-DA classification model and c LV loadings and d regression vectors plots of chemovar PLS-DA classification model. The dashed grey lines represents the 12 different activated water bands in the first overtone region of the OH stretching ban

Aquagram values according to major classes and chemovars were calculated and plotted according to the guidelines published by Tsenkova et al. [36, 38]. Significant differences in the water spectral patterns were observed for both major classes and chemovars (Fig. 4).

In the major class aquagram, the highest variations were observed at 1412 nm (C5) followed by 1364, 1374, 1384 nm (C2–C4), and 1512 nm (C12, Fig. 4a). These water activated absorbance bands were also occurred in several major class LVs or regression vectors (Table S2).

The aquagram values at 1412 nm (C5) indicate that the high CBDA chemovars exhibit the highest presence of free water in the inflorescence matrix, followed by the high THCA and hybrid major classes (Fig. 4a, Table S3) [36, 38]. Furthermore, the aquagram values at 1364, 1384, 1426, and 1452 nm (C2, C4, C6, and C8, respectively) indicate that the high CBDA chemovars also possess the strongest water solvation shells and the highest water hydration levels (Fig. 4a, Table S3). In contrast, the hybrid chemovar exhibit the weakest water solvation shells and the lowest water hydration levels (Fig. 4a, Table S3) [36, 38].

Conversely, the aquagram values at 1512 nm (C12) reveal that the high CBDA chemovars exhibited the lowest amount of strongly bound water, whereas the hybrid major class displayed a high amount of strongly bound water (Fig. 4a, Table S3) [36, 38]. This observation is further supported by the aquagram values at 1440, 1462, 1476, and 1488 nm (C7, C9, C10, and C11, respectively), which indicate that the high CBDA chemovars displayed the highest number of water molecules with a single hydrogen bond (Fig. 4a) [36, 38]. In contrast, the hybrid major class revealed the highest number of water molecules with two, three, and four hydrogen bonds (Fig. 4a) [36, 38]. However, only 1488 nm aquagram values were found to be statistically significant (Table S3). These results are in agreement with the free water, water solvation, and water hydration results. For example, the high CBDA major class exhibited a significant presence of free water, strong water solvation shells, and high degree of water hydration and also revealed the lowest quantity of strongly bound water but a high number of hydrogen bonds.

Overall, the high THCA chemovars revealed the lowest number of hydrogen bonded water molecules (Fig. 4a). Notwithstanding, due to lower variations in the bands related to hydrogen bonded water molecules, it is reasonable to assume that these differences between the major classes are not that significant as compared to the free water differences (Fig. 4a, Table S3).

Aquagrams of the OH first overtone 12 WAMACS of a major classes and b chemovars

Examination of the chemovar aquagram reveals several significant differences and similarities compared to the major class aquagram. In both cases, the highest variation is observed at 1412 nm (C5), followed by 1364, 1374, 1384 nm (C2–C4), 1488 nm, and 1512 nm (C11–C12, Fig. 4). This indicates that these water-activated bands can serve as effective discriminators and classifiers for both major classes and chemovars, particularly the C2–C5 bands, which were consistently identified throughout the aquaphotomics analysis steps (Fig. 4, Table S2–S4). Thus, the aquagram of each chemovar can act as a unique fingerprint, as it reflects the dominant characteristics and relationships with the chemovar’s water matrix structure [39]. Additionally, the hybrid chemovar Gen12 in the dataset exhibited the lowest aquagram values at 1374, 1384, and 1412 nm (C3–C5), whereas the high CBDA chemovars demonstrated the highest aquagram values at 1374, 1384, and 1476 nm (C3, C4, and C10, respectively), consistent with the major class aquagram values (Fig. 4b, Table S4).

On the other hand, the high CBDA and high THCA chemovars exhibit significant differences in their water spectral patterns (Fig. 4b). For example, the 156 chemovar had a much higher aquagram values at 1364, 1374, and 1384 nm (C2–C4) and much lower values at 1488 nm (C11) compared to the 45–3 chemovar, despite both belonging to the high CBDA major class (Fig. 4b, Table S4). Conversely, both high CBDA chemovars showed similar aquagram values at 1342, 1412, 1426, 1440, and 1512 nm (C1, C5, C6, C7, and C12, respectively, Fig. 4b, Table S4). In the case of the high THCA chemovars, the 621–17 chemovar presented much higher aquagram values at 1342, 1364, 1412, 1426, 1452, and 1462 nm (C1, C2, C5, C6, C8, and C9, respectively) and much lower values at 1512 nm (C12) compared to the other three high THCA chemovars (Fig. 4b, Table S4). The aquagram values for the 240, 505, and Erez chemovars were similar, except for the 1364 and 1488 nm (C2 and C11) values in the 240 and Erez chemovars (Fig. 4b, Table S4). It is important to note that adding more chemovars from different major classes to the dataset may reveal additional chemovars with trends differing from their major class aquagram. This is because the classic aquagram is a relative construct that depends on the samples in the dataset and their affiliations [38]. These results suggest that chemovars within the same major class may have completely different water matrix structures.

According to the aquagram values at 1412 nm (C5), the 621–17 chemovar exhibited the highest presence of free water in the inflorescence matrix, followed by both high CBDA chemovars (Fig. 4b, Table S4) [36, 38]. In contrast, the hybrid chemovar, Gen12, had the lowest presence of free water (Fig. 4b, Table S4) [36, 38]. The aquagram values at 1364 and 1384 nm (C2 and C4, respectively) suggested that the 621–17 and 156 chemovars had strong water solvation shells of small molecules, while the Erez and Gen12 chemovars had weak water solvation shells of small molecules (Fig. 4b, Table S4) [36, 38]. Furthermore, the aquagram values at 1426 nm (C6) indicated that the 621–17 chemovar had the highest water hydration, whereas the hybrid chemovar Gen12 and the other three high THCA chemovars—505, 240, and Erez—exhibited the lowest water hydration (Fig. 4b, Table S4) [36, 38].

On the other hand, the Erez chemovar, followed by the 505, 240, and Gen12 chemovars, had the highest amount of strongly bound water, while the 621–17 chemovar had the lowest amount of strongly bound water, as suggested by the 1512 nm aquagram values (Fig. 4b, Table S4) [36, 38]. The Erez chemovar also had the highest number of water molecules with four hydrogen bonds, whereas the 621–17 chemovar had the highest number of water molecules with one and two hydrogen bonds (Fig. 4b, Table S4) [36, 38]. The 1476 nm (C10) aquagram values, which are not statistically significant, suggested that the water matrices of all chemovars contained a similar number of water molecules with three hydrogen bonds (Fig. 4b, Table S4) [36, 38]. The 240 and 505 chemovars had the lowest number of water molecules with one and two hydrogen bonds, while the 156 chemovar had the lowest number of water molecules with four hydrogen bonds (Fig. 4b, Table S4 [36, 38].

These findings revealed that within each major class, chemovars exhibited different water spectral patterns, with the most significant variations related to the presence of free water, solvation shells of small molecules, the amount of strongly bound water, and the number of hydrogen bonds (Fig. 4b) [36, 38]. These results suggest that the most accurate way to explore the water matrix spectral patterns of cannabis inflorescence is by chemovar rather than by major class.

Although the C7 and C9 bands were consistently observed in the previous steps of the aquaphotomics analysis, they exhibited lower variation in their aquagram values (Fig. 4, Table S2–S4). This suggests that the number of water molecules with one and two hydrogen bonds is relatively similar across different chemovar water matrix structures, with the exception of the 621–17 chemovar (Table S4) [36, 38]. In contrast, the aquagram values of the C11 and C12 bands showed high variations, indicating that one of the main differences in the water matrix structures is related to strongly bound water with four hydrogen bonds (Fig. 4, Table S2–S4) [36, 38].

To further evaluate the cannabis inflorescence water matrix spectral pattern, we used PCA analysis and Pearson r correlation coefficients matrix of the WAMACS aquagram values, to reveal major correlations between different WAMACS and chemovars. Positively correlated bands can be identified based on their vectors direction, where smaller angle between two vectors suggests stronger positive correlation and angle approaching 180° suggests stronger negative correlation [22, 48, 55]. The high CBDA major class found to positively correlate to 1342, 1426, and 1440 nm (C1, C6, and C7, respectively) aquagram values, the high THCA major class found to positively correlate to 1512 nm (C12) aquagram values, while the hybrid major class found to positively correlate to 1488 nm (C11) aquagram values (Fig. 5a and b). The 621–17 chemovar was found to positively correlate with the 1412 nm (C5) aquagram values, while the 45–3 chemovar positively correlated with the 1452 and 1462 nm (C8–C9) aquagram values (Fig. 5c and d). The Gen12 and Erez chemovars showed positive correlations with the 1488 nm (C11) aquagram values, and the 505 chemovar positively correlated with the 1512 nm (C12) aquagram values (Fig. 5c and d). The 156 chemovar exhibited positive correlations with the 1364, 1374, and 1384 nm (C2–C4) aquagram values (Fig. 5c and d). These findings are consistent with the original chemovar aquagram values (Fig.