Twenty-three (n = 23) selective timber species including 15 hardwoods and 8 softwoods were collected from different parts of India. The hardwood and softwood species were identified based on morphological characteristics. Each species was collected from three different trees, which resulted in 69 timber samples. The timber samples were obtained by drilling the trunk of the tree at breast height with a Helfinch Cheston 10 mm powerful drill machine (350-W, 2,600 rpm) to obtain a fine powder (2–5 g) that was collected in a zip-lock bag with proper labeling with the name and provenance. The samples were oven–dried at 30°C for 2 weeks to evaporate the moisture.

The selection of timber species was based on their unique characteristics, economic value, and legitimate protection status, which resulted in the collection of eleven timber species that are being protected by various organizations at the global level, and other timber species were selected based on their commercial value (Table 1). The intra-species variability of the samples was assessed by analyzing three different samples of timber species procured from three different trees under a similar set of experimental conditions. Similarly, the repeatability of the samples was checked by analyzing a sample thrice under the same set of parameters. A reproducibility test was conducted by analyzing a single timber sample at distinct points for 3 successive days under the same set of experimental conditions.

All collected timber samples were analyzed using Bruker Alpha ATR-FTIR spectrophotometer containing an ATR accessory with zinc selenide crystal, and DTGS as a detector. The selected spectral range was 4,000–600 cm⁻¹ over 4 cm⁻¹ resolution. The final acquired spectrum was the average of the accumulated 16 scans. All measurement scans were carried out at room temperature (25°C ± 5°C). Homogenous pressure was applied to the selected samples by using an ATR knob and anvil. The ATR zinc selenide crystal was cleaned using spectroscopic-grade acetone before and after the analysis of each sample.

The discriminatory power (DP) was calculated by Smalldon and Moffat’s (Smalldon and Moffat, 1973) formula by dividing the total discriminated sample pairs by the possible number of sample pairs multiplied by a hundred. DP = T o t a l   n u m b e r   o f   d i c r i m i n a t i n g   s a m p l e   p a i r s   T o t a l   n u m b e r   o f   p o s s i b l e   s a m p l e   p a i r s   n × 100

Where “n” can be calculated as, T o t a l   n u m b e r   o f   p o s s i b l e   s a m p l e   p a i r s   n = n n − 1 2 where n = total number of samples.

Total Chemometrics include the application of various statistical tools for the objective interpretation of the dataset based on the similarities in the structural makeup of the samples. In this work, chemometric models such as PLS-DA (Partial least square discriminant analysis), PCA (principal component analysis), and LDA (linear discriminant analysis) have been used for the differentiation of the studied timber species. The spectra obtained were pre-processed before subjecting to chemometric analysis by using different pre-processing methods to enhance the information contained in the spectral data and to condense the noise ratio. Pre-treatment methods such as Savitzky-Golay smoothing, normalization, and baseline correction were used before performing PLS-DA, PCA, and LDA. In addition to this, deresolve transformation (for noise reduction) and orthogonal signal correction transformation were also used for PLS-DA using Unscrambler^®X (64-bit) version 10.5.1, by Camo, Norway.

A blind study was a prerequisite in this study to check the efficacy of the present approach to identify a given unknown timber species. The blind test was performed by randomly selecting seven timber species (three softwoods and four hardwoods) from the list of procured timber species for this study. Additional samples were collected to conduct the blind study and were labeled as X1, X2, X3, X4, X5, X6, and X7. The analyst was unaware of the identity of the timber species. All the samples were analyzed in the same manner as the pristine sample analysis method. The IR spectra obtained from the blind samples were first projected into the already-built PLS-DA model to identify whether the timber samples belonged to softwood or hardwood. After identifying the type of wood, the samples were projected into the respective PCA model to identify the timber species.

All the spectra obtained by ATR-FTIR analysis were visually examined for specific peaks at different wavenumbers. Visual comparison of the spectra was done regarding the presence or absence of specific peaks in the IR spectra at different wavenumbers. However, the comparison did not consider characteristics such as sharpness, broadness, peak height, and intensity. The representative spectra of all the timber samples are shown in Figure 1. It was observed that most of the spectra have specific peaks corresponding to timber components such as cellulose, hemicellulose, and lignin. The details of the peak, IR vibrations, and possible components are given in Table 2. The IR spectra of timber samples showed the presence of cellulose which is indicated by the bands present at 2,848, 1,634, 1,363, 1,317, 1,158, and 1,025 cm⁻¹ which may be attributed to C-H stretching, C-O stretching, or O-H bending, C-H stretching, CH2 bending stretching, C=O and C-O-C stretching, and C-O deformation respectively (Sharma et al., 2020). Similarly, the presence of hemicellulose was indicated by the presence of peaks in the IR spectra at 1730, 1,457, 1,363, and 1,158 cm⁻¹ which may indicate the occurrence of C=O vibration, C-H deformation stretching, C-H stretching, and C=O stretching respectively (Chen et al., 2010; Sharma et al., 2020). In the same manner, the presence of lignin in the timber samples was indicated by the presence of specific bands at 1730, 1,634, 1,588, 1,506, 1,230, and 1,025 cm⁻¹ which may indicate C=O vibration, C-O stretching, C=O stretching, C=C stretching, C-O-C stretching, and C-H deformation respectively (Chen et al., 2010; Sharma et al., 2020).

The IR spectra obtained from the blind samples were projected into the already-built PLS-DA model. The results showed that the softwoods and hardwoods were correctly classified with 100% accuracy (Figure 5). Therefore, with the present approach, it is possible to group hardwoods and softwoods into their respective classes with 100% accuracy.

The softwood blind samples were then projected in the already-built PCA model for softwoods. Similarly, the hardwood blind samples were projected into the already-built PCA model for hardwoods. The results showed that all the softwood blind samples were projected with 100% accuracy (Figure 6). Sample X3 was correctly predicted to be sample S4. Similarly, the samples X4 and X6 were predicted S5 and S7. Similarly, all four blind samples of hardwoods were correctly predicted with 100% accuracy (Figure 7). The sample X1 was predicted as H1. The samples X2 and X5 were predicted as samples H3 and H9 respectively. Similarly, the sample X7 was predicted as H15. A comparison of the results of all the chemometric methods used in this study has been listed in Table 5.

The proposed work supports the work done by Sharma et al. (2020) as the authors differentiated softwood and hardwood using ATR-FTIR spectroscopy and various chemometric tools such as Hierarchical Clustering Analysis (HCA), ANOVA (Analysis of Variance Test), PCA, and LDA. A total of twenty-four timber species were procured from Brno, Czech Republic followed by the identification and the classification of the samples. Similarly, Popescu et al. (2009) differentiated hardwood and softwood using FTIR spectroscopy and X-ray diffraction (XRD). However, no chemometric tools were used for the analysis of timber samples. The present research work involved the identification and differentiation of twenty-three selective timber species from India in which ten timber species are legally protected by various organizations at the global level. The identification and differentiation of these precious timber species is important to combat timber-related crimes. This approach will aid in the identification of the unknown timber species and the linking of an unknown timber sample with a known source. Moreover, the chemometric tools used for the present study included PLS-DA, PCA, and LDA, which delivered significant findings. The application of chemometrics tools to the IR spectra helped in providing objectivity to the results. The results of the blind study demonstrate the applicability of this approach in real-case scenarios in timber forensics.