NC State
Baldin, T., Marchiori, J. N. C., Santos, G. A., Gallo, R., Santos, O., Valente, B. M. R. T., and Hein, P. R. G. (2018). "Evaluation of alternative sample preparation methods for development of NIR models to assess chemical properties of wood," BioRes. 13(3), 5394-5407.


The aim of this work was to investigate the influence of sample preparation including variation in moisture content and particle size on the accuracy of near infrared (NIR) spectroscopy models developed to predict Klason lignin, total lignin, and holocellulose in wood. Seventy-five samples of sawdust obtained from a eucalyptus plantation were divided into aliquots and submitted to three different treatments: traditional (TRAD), large particle dried at room temperature (LPRT), and large particle oven-dried (LPOD). The influence of sample preparation method on models’ accuracy was compared by statistical analysis. Overall, grinding to a larger particle size and drying at room temperature (treatment LPRT) did not decrease the accuracy of the prediction models when compared to the TRAD sample preparation method. These findings were more evident for Klason lignin and holocellulose. This is relevant because resources used for sample preparation (i.e. grinding and drying) can be minimized, which is expected to reduce the costs associated with analysis of wood properties by NIR.

Download PDF

Full Article

Evaluation of Alternative Sample Preparation Methods for Development of NIR Models to Assess Chemical Properties of Wood

Talita Baldin,a,* José Newton Cardoso Marchiori,a Glêison Augusto dos Santos,Ricardo Gallo,b Osmarino dos Santos,c Brígida M. Reis Teixeira Valente,c and Paulo Ricardo Gherardi Hein d

The aim of this work was to investigate the influence of sample preparation including variation in moisture content and particle size on the accuracy of near infrared (NIR) spectroscopy models developed to predict Klason lignin, total lignin, and holocellulose in wood. Seventy-five samples of sawdust obtained from a eucalyptus plantation were divided into aliquots and submitted to three different treatments: traditional (TRAD), large particle dried at room temperature (LPRT), and large particle oven-dried (LPOD). The influence of sample preparation method on models’ accuracy was compared by statistical analysis. Overall, grinding to a larger particle size and drying at room temperature (treatment LPRT) did not decrease the accuracy of the prediction models when compared to the TRAD sample preparation method. These findings were more evident for Klason lignin and holocellulose. This is relevant because resources used for sample preparation (i.e. grinding and drying) can be minimized, which is expected to reduce the costs associated with analysis of wood properties by NIR.

Keywords: Holocellulose; Lignin; Near infrared (NIR) spectroscopy; Eucalyptus wood; Particle size

Contact information: a: Department of Forestry Science, Federal University of Santa Maria, Av. Roraima, 1000, 97105-900, Rio Grande do Sul, Brazil; b: Department of Forest Engineering. Federal University of Viçosa, Av. Peter Henry Rolfs, s/n, Minas Gerais, Brazil; c: CMPC- Celulose Riograndense, São Geraldo, 1680, 92500-000, Guaíba, Brazil; d: Department of Forestry Science, Federal University of Lavras, Av. Dr. Sylvio Menicucci, 1001 – Kennedy, Minas Gerais, Brazil;

* Corresponding author:


Near infrared (NIR) spectroscopy is a successful scientific tool for characterizing many biomaterials, including wood (Tsuchikawa 2007; Pažitný et al. 2011; Tsuchikawa and Schwanninger 2013). The number of publications reporting NIR spectroscopy application to forestry, wood, and wood products is increasing systematically (Tsuchikawa and Kobori 2015; Sandak et al. 2016). These applications include determination of wood quality and pulp characteristics (Birkett and Gambino 1989; Castillo et al. 2012), prediction of lignin syringyl/guaiacyl content (Hein et al. 2010; Lupoi et al. 2014; Ramadevi et al. 2016), and the prediction of ashes, extractives, lignin, holocellulose, and wood cellulose content (Wright et al. 1990; He and Hu 2013; Zhou et al. 2015, 2016). Additional uses include characterization of anatomical, physical, and mechanical properties of wood (Hein 2012; Milagres et al. 2013; Sundaram et al. 2015); wood bioenergy (Fagan et al. 2011; Hou and Li 2011; Ramalho et al. 2017), and discrimination of similar rare woods (Yang et al. 2012; Shou et al. 2014).

NIR is a rapid, nondestructive, and safe method for determining the variability of many wood traits from a single sample. Providing a quick and reliable quantitative estimation of Klason lignin, holocellulose, and α-cellulose content can improve the efficiency of feedstock processing in industrial environments. Additionally, NIR provides great advantages for plant germplasm screening, genetic engineering, and industrial crop cultivation.

The success of NIR-based analyses is driven by many factors such as understanding how instrument parameters and sample conditions affect the spectral signatures, as well as how to extract useful information from the spectra. Wood processing before analysis is a key factor because significant variability in prediction may be introduced by the sample preparation method (Hein et al. 2010). For example, moisture content that is not uniform among samples can introduce an undesirable source of variation. Conversely, care must be taken when drying samples in order to avoid potential sample damage or changes in the chemical structure of the wood (Sandak et al. 2016).

The influence of sample moisture content on the NIR spectra has been described previously (Tsuchikawa and Tsutsumi 1998; Thygesen and Lundqvist 2000a, b; Watanabe et al. 2006; Schwanninger et al. 2011). Overall, studies suggest that the samples can be oven-dried to decrease moisture content before being read by the NIR (Lupoi et al. 2014). Because water is a strong absorber in the NIR region, samples with high moisture are strongly dominated by the signature from water. However, it remains unknown whether different temperatures and sample moisture content influence the statistics generated with the prediction models developed from those spectra.

Wood preparation before analysis is normally based on mechanical treatment that produces wood powder. TAPPI T257 CM-02 (2002) recommends grinding wood to a particle size that passes a 40 mesh (0.4 mm) screen, whereas heating the material during grinding or regrinding the material should be avoided. However, the resulting particle size may be too large for NIR spectroscopy analysis, and it then becomes necessary to regrind or fractionate the sample (Schwanninger et al. 2004). Additionally, it is known from TAPPI T264 CM-07 (2007) that different fractions may contain varying quantities of some wood constituents, so that their removal could modify the chemical composition of the sample. Similarly, regrinding can also significantly modify the chemical composition of the wood material.

This study evaluated the effects of sample preparation (i.e., variation in moisture content and particle size) for NIR spectra acquisition on the accuracy of multivariate models for predicting Klason lignin, total lignin, and holocellulose associated with wood quality. While the effects of moisture content and particle size on NIR spectral determination have been previously described, this research attempted to find a less tedious sample preparation method for estimation of chemical components of wood. If costs associated with sample preparation can be minimized (i.e., grinding and drying) without compromising model accuracy, then the costs associated with NIR analysis can be reduced.



The wood material used came from a Eucalyptus benthamii progeny test plot installed in 2011 in the municipality of Encruzilhada do Sul, in the state of Rio Grande do Sul, Brazil (S -30.65268 e W -52.46839). The test plot area was 1.3 ha, and the spacing between trees was 3.50 m by 2.14 m, which allowed a total of 1394 trees in the experiment. Samples for the study were taken 55 months after planting the trees.


Selection of trees and sample collection

An initial sample was collected from all 1394 trees using a non-destructive method. This procedure included using a drill equipped with a special bit to collect core sawdust from four spots at the diameter of breast height, or 130 cm from the ground. The sawdust was processed into smaller particles using a Willye mill equipped with a 1 mm sieve. The portion of the sample that passed through the 1 mm sieve was analyzed using the NIR. Based on the spectra, the variable pulp yield and the syringyl/guaiacyl (S/G) ratio were estimated using calibration models developed for Eucalyptus globulus and Eucalyptus grandis (Alves et al. 2011a, 2012). Subsequently, the estimates of pulp yield and S/G ratio from the 1394 trees were used to select 75 representative Eucalyptus benthamii trees for the development of predictive models based on NIR spectroscopy. The selected subgroup included 20 trees with high pulp yield (55% to 62%), 20 with low pulp yield (36% to 43%), and 25 with intermediate values (44% to 54%). Additionally, 5 trees were selected to represent high S/G value (1.70 to 2.10) and 5 to represent low S/G (1.10 to 1.69).

A log of about 70 cm was collected immediately above the diameter of breast height (approximately 130 cm from the ground) from all 75 trees. This log was then chopped into chips. An aliquot of this material was used for standard wet chemistry tests, and another portion was used for measurements with the NIR as described below.

Wet chemistry analysis

Chips without bark and knots were classified following SCAN CM-40:01 (2001). The accepted fraction was ground in a knife mill and sieved according to TAPPI standard T257 CM-02 (2002), where the fraction that passed through the 40 mesh (40 openings per mm2 or size of the opening equal to 0.47 mm) and retained in the 60 mesh (0.31 mm) was used for analysis. Subsequently, the chemical properties of Klason lignin, total lignin, and holocellulose were determined by wet analysis chemistry and served as the dependent variables for which the calibration models in the NIR were adjusted against. The standards used in the wet analysis chemistry were TAPPI T222 OM-2 (2002), TAPPI UM-250 (1991), and TAPPI T203 CM-99 (1999), for Klason lignin, total lignin, and holocellulose, respectively. All analyses were performed on a dry matter basis.

Treatments and Fourier transform near-infrared (FT-NIR) acquisition

The aliquot of sawdust designed for NIR analysis was initially processed into smaller particles in a Willye mill equipped with a 1 mm sieve. A fraction of the material that passed through this sieve was collected and split into 3 aliquots for application of the treatments. There were three treatments: traditional (TRAD), large particle dried at room temperature (LPRT), and large particle oven-dried (LPOD). Descriptions of the three treatments, including drying method and particle size, are provided in Table 1. Briefly, TRAD treatment is the protocol commonly used for NIR analysis. The LPRT and LPOD treatments are sample preparation methods that reduce sample grinding and facilitate the drying process.

After the treatments were applied, diffuse reflectance spectra were acquired using a Fourier transform NIR (FT-NIR) spectrometer (Model: MPA, BrukerOptik GmbH, Ettlingen, Germany). The spectra were read between 12000 cm-1 and 4000 cm-1 with a resolution of 8 cm-1 and used to develop prediction models for Klason lignin, total lignin, and holocellulose.

Table 1. Description of the Three Treatments (i.e., Sample Preparation Method) Applied before NIR Spectra Acquisition

Multivariate statistics and data analysis

Calibrations were performed using the Bruker Opus 6.2 software package. In the same software all spectra were pre-processed with four different procedures: 1) first derivative (1stDer); 2) vector normalization (VN); 3) first derivative + vector normalization (1stDer + VN); and 4) first derivative + a multiplicative correction of the signal (1stDer + MSC). Optimization of Savitzky-Golay (SG) smoothing was used for spectroscopy analysis.

Model calibration was determined by regression analysis of the partial least squares (PLS), as recommended by Gierlinger et al. (2002). The models for the treatments were adjusted with the same number of latent variables (LVs) needed to provide the best fit. The number of variables adopted for each model was considered as a function of the decrease of the root mean square error of cross-validation (RMSECV) and the increase of the coefficient of determination (R²) and residual predictive deviation (RPD).

Spectral bands above 10000 cm-1 were excluded because in this region the spectrum presented noise repetitions that did not contain relevant information about the properties of interest. The spectral ranges used for construction of calibration models were based on the recommendations of Schwanninger et al. (2011). The anomalous samples visibly different from the normal distribution were detected as outliers and excluded from the model.

Calibration models were developed for Klason lignin, total lignin, and holocellulose using all 75 observations in each treatment. Subsequently, these models were run in 15 different regions of the spectra to generate replications that allowed statistical testing of model accuracy. A statistical analysis was applied to verify the difference between the mean values of the validation error (RMSECV) of the models using the statistical software Genes (Cruz 2001). When the null hypothesis was rejected, a comparison of means was done by means of a Tukey test (5%) and data compliance was verified through the error normality tests (Shapiro Wilks).


Wet Chemistry

The descriptive statistics of chemical components observed in Eucalyptus benthamii are presented in Table 2. The mean values of Klason lignin, total lignin, and holocellulose found in wood are in agreement with other species of the genus previously studied at similar age (Estopa et al. 2017). Total lignin content was 31.3%, which is lower than what was reported by Trugilho et al. (2015) in Eucalyptus cloeziana (32.3%) and higher than what was determined for 75 Eucalyptus clones (27.3%) commercially used in Brazil (Gomide et al. 2010). Klason lignin averaged 27.7% and was higher than what was previously reported for Ebenthamii (26.2%, Alves et al. 2011b), and for Europhylla × Egrandis (26.9%, Baillères et al. 2002).

Table 2. Descriptive Statistics of Chemical Components of Eucalyptus benthamii

Holocellulose content averaged 64.8%, which is lower than in hardwood trees (79.6%, Zhou et al.2015), but higher than the value reported by Trugilho et al. (2015) for Eucalyptus urophylla x Eucalyptus grandis (59.7%). Of note, Eurograndis is the hybrid that is planted the most in Brazil.

In most Eucalyptus species the lignin content decreases as the tree develops and forms the adult wood. In contrast, the content of carbohydrates tends to increase as time progresses (Trugilho et al. 1996; Estopa et al. 2017).

Although the determination of Klason lignin, total lignin, and holocellulose was relevant for comparison of the dataset against published studies, the main goal of the wet chemistry analysis was to produce reference data for the NIR. The authors observed considerable variation in the dataset and ensured accuracy of the reference methods, which are the two most important factors for calibration and development of accurate NIR models (Alves et al. 2011a).

Fig. 1. a) Score plot obtained by PCA applied to NIR spectra collected from the three different treatments; b) Absorbance versus wavenumber plot for untreated average NIR spectra of the three treatments. Bands assigned to chemical compounds are represented by numbers and listed in Table 3

Effect of Treatments on the Spectra

The PCA applied to the FT-NIR spectra highlighted the discriminating ability of the NIR to separate samples by plotting the averaged spectra of 75 samples for each treatment (Fig. 1a). The PCA scores plot of spectra acquired from the wood specimens shows at least two groups (indicated by dotted circles). The two main components together explain 92% of the variability of the data analyzed, of which 61% is explained by the Principal Component 1 (PC1) and 31% is explained by Principal Component 2 (PC2).

The differences in the spectra due to treatment effect were assessed by plotting and overlapping the average of the 75 samples for each treatment (Fig. 1b). The reference for the NIR region interpretation is based on Watanabe et al. (2006), Schawninger et al. (2011), and Sandak et al. (2011) (Table 3).

Overall, the three spectra showed strong absorption bands. This was due to the overtone of the OH stretching vibration at 7200 cm-1 to 6000 cm-1 and the combination of OH stretching and deformation vibrations at 5400 cm-1 to 4800 cm-1 and 4900 cm-1 to 4600 cm-1. The spectra of the three bands increased progressively, with LPRT having the highest peak, followed by TRAD and LPOD (Fig. 1b). The lowest absorbance peaks were seen in the spectra recorded from samples with the lowest moisture content (LPOD treatment). This observation agrees with results published by Inagaki et al. (2008). Variation in the NIR spectra due to sample moisture content was observed throughout the NIR range, which indicated that the absorption peaks due to OH bonding of water were substantially overlapped in the original spectra. This variation in absorbance due to moisture content is governed by the adsorption/desorption mechanism, where the wood–water interaction or the physicochemical properties of water play an important role. Because the NIR light absorption by wood overlapped with that of water, the spectroscopic information was compromised. Importantly, the mechanism of which moisture content in wood influences the acquisition and behavior of NIR spectra has been extensively discussed previously (Tsuchikawa and Tsutsumi 1998; Thygesen and Lundqvist 2000a, b; Watanabe et al. 2006; Schwanninger et al. 2011; Pecoraro et al. 2015).

Table 3. NIR Absorption Bands Associated with Components Present in Wood Specimens. Index Numbers Related to the Specifics Bands Indicated in Fig. 1.

The principal component analysis grouped LPOD treatment in a separate cluster and combined LPRT and TRAD treatments in another (Fig. 1a). In LPOD treatment, the dried wood showed only hygroscopic water, and the water band was expressed with less intensity at the range of approximately 5200 cm-1 (combination of stretching and deformation vibrations for OH). However, the location of the band can be modified with increased moisture content in the wood, with NIR absorption also apparent at 5000 cm-1, 5080 cm-1, and 4920 cm-1 (Schwanninger et al. 2011). From a different prospective, analyzing the NIR spectra could be a tool to screen the content and the state of water in unknown samples (Tsuchikawa and Tsutsumi 1998; Terazawa et al. 2003; Inagaki et al. 2008). It should be noted that the shapes of the bands were similar among treatments (Fig. 1b), and the most conspicuous changes in the spectra of treated wood were only observed when analyzing the wavenumbers (Table 3).

Overall, the spectral absorption was more influenced by sample moisture content than particle size (Fig. 1b). The greatest absorbance values were evidenced in the TRAD and LPRT treatments, which had greater moisture content compared with LPOD (12.0% vs. 7.0%). This result is somewhat expected because water has great capacity to absorb electromagnetic radiation (Achata et al. 2015). Therefore, the results suggested that when preparing wood samples for NIR analysis, controlling sample moisture content might be more important than trying to grind samples to a particle size smaller than 1 mm. Thus, LPOD may be preferred particularly for preparation of a large number of samples. It should be noted that alternative methods that minimize the drying time should be studied.

Differences between NIR spectra recorded from 0.5 mm or 4.0 mm particle size appear to be less significant with absorbance variation continuously similar along the NIR spectral range (Hein et al. 2010). These same authors mentioned that the ball milling process resulted in more homogeneous spectra when compared to solid wood particles (e.g. wood chips).

This was not surprising, since the band at 1400 nm of the fine powder was stronger than that of the raw wood meal. The more intensive the ball milling, the higher the percentage of fine material in the sample, and this explains the increase in absorbance values in some spectra regions (Schwanninger et al. 2004). However, ball milling is a time-consuming laboratorial procedure that can preclude processing of a large number of samples. Variations in particle size between 30 mesh and 60 mesh of Eucalyptus wood did not have a significant effect on the NIR spectra (Baillères et al. 2002). Nevertheless, in Eucalyptus globulus the NIR-estimated lignin content was influenced by particle size (Poke and Raymond 2006).

Calibration Models

The partial least square regression models for chemical components predicted by NIR spectroscopy for each treatment are presented in Table 4. The Klason lignin models presented promising statistics and showed potential for providing a method of accessing quality control for individual samples (Figs. 2a to 2c). Predicted and observed values for Klason lignin contents exhibited good correlations with R² of 0.87, 0.90, and 0.86 for LPRT, LPOD, and TRAD treatments, respectively. The RPD values of validations were lower than 3, indicating that such models may be adequate for screenings or for genetic studies (Schimleck et al. 2003).

Table 4. Cross-validation Statistics of Partial Least Square Regression Models for Calibration of Chemical Components in Eucalyptus benthamii

Overall, values of R2 for total lignin and holocellulose can be considered acceptable (i.e. 0.82 in LPOD and 0.84 in LPRT), and the RMSECV values were below 0.61% and 0.69%, respectively (Table 4, Figs. 2f and 2h). Validation of these calibrations produced good correlations between wet chemistry predicted values in the laboratory and NIR-predicted values. Other researchers have published calibration models for total lignin in Eucalyptus trees. However, these models showed lower R² (0.71) and higher RMSECV (0.90%) (Milagres et al. 2013). A better calibration (R² = 0.78) was developed using NIR analysis for predicting Klason lignin content from solid wood in Eucalyptus globulus (Poke and Raymond 2006). For holocellulose, Hou and Li (2011) developed a model containing 8 PLS vectors, which resulted in high R² (0.98) and low RMSECV (0.34%).

With regard to the lignin content in Eucalyptus benthamii wood, calibration models with coefficient of determination of 0.64 and RMSECV of 0.68% have been published (Estopa et al. 2017). The present study yielded excellent cross-validation statistics for lignin content, with higher coefficient of determination and lower RMSECV for all treatments (Table 4).

Comparison of PLS-R Models

An important goal of this study was to evaluate differences between the calibration models related to different sample preparation protocols prior to collecting the spectra. It was anticipated that if no difference between treatments was evident, then the sample preparation work can be minimized, which reduces costs and increases the speed of NIR analysis.

When a clear decision cannot be made regarding the choice of the best model, more rigorous examination of the results is required. Examining the cross-validation of the models resulting from the combination of multiple data sets could help in making such a decision (Alves et al. 2011a). Alternatively, statistics such as comparison of means can be applied to differentiate the values of RMSECV.

Fig. 2. Cross-validation predictions of Klason lignin, total lignin, and holocellulose in Eucalyptus benthamii

The mean comparison test applied on the 15 RMSECV values of the calibration models for each treatment is presented in Fig. 3. Regarding Klason lignin, LPOD treatment yielded satisfactory prediction and was not statistically different than TRAD treatment (Fig. 3a). Additionally, no difference was observed between the LPOD and LPRT treatments. This finding indicated that for measurement of Klason lignin, samples do not need to be ground to a particle size smaller than 1 mm or reduced to a moisture content smaller than 12%. This is relevant as it substantially reduces sample preparation time and labor cost.

Fig. 3. Treatment effect on RMSECV values for calibration models of Klason lignin (a); total lignin (b); and holocellulose (c)

* Means followed by the same letter in the column do not differ by Tukey test at 5% probability

With regard to total lignin, the model error in LPOD and TRAD treatments was smaller than that in LPRT treatment (Fig. 3b). This indicated that, for analyses of total lignin, samples should be ground to particles sizes less than 1 mm or, if grinding to 1 mm, sample moisture content should be reduced to less than 12%. A noteworthy observation was that LPOD treatment, which had samples with lower moisture content, presented the overall best statistics for total lignin model fit. This was mainly because the bands related to water adsorption were excluded from the calibration models. Moisture content was important for selecting the best models for total lignin, and was also an indicator of dryness of wood particles after sample preparation. For holocellulose, TRAD treatment produced better predictions than LPOD treatment, but did not differ from LPRT (Fig. 3c). This indicated that in addition to measurements of Klason lignin, LPRT treatment can be applied on samples when analyzing holocellulose.

NIR analysis has commonly been performed using the fraction of the wood milled through the 40 mesh to 60 mesh, particularly because this is the fraction used for gold standard wet chemistry analyses. However, this study demonstrated that accurate NIR models could be developed using samples of larger particle size (i.e. 1 mm, LPRT and LPOD treatments). Furthermore, except for the determination of total lignin that required either smaller particle size (TRAD treatment) or lower sample moisture content (LPOD treatment), this study demonstrated that samples of larger particle size (1 mm) can be dried at room temperature. The authors expected that the alternative sample preparation methods evaluated in this study could be applied to wood samples other than Eucalyptus. However, special attention should be given to sample final moisture content because of the disturbing effects of water on the quality of the NIR spectra. Additionally, further research is warranted using similar sample preparation methods, particularly on wood samples that expand the range of Klason lignin, total lignin, and holocellulose outside the boundaries evaluated in this study.


  1. This study demonstrated that sample preparation work can be minimized (i.e., grinding to 1 mm and drying at room temperature), which can considerably reduce the costs associated with NIR analysis.
  2. Compared with the traditional sample preparation method, grinding to a larger particle size and drying at room temperature (LPRT treatment) did not affect the accuracy of the prediction models, particularly for Klason lignin and holocellulose. For analysis of total lignin, samples can be ground to a larger particle size, but the quality of the calibration model was only equivalent to the traditional method when moisture content was reduced to 7% (LPOD).
  3. Promising NIR models were developed for predicting chemical components (i.e., total lignin, Klason lignin, and holocellulose) associated with wood quality. Statistical analysis of the error associated with the developed NIR models demonstrated minor differences between the treatments.


The authors gratefully acknowledge the financial support from the CMPC – Celulose Riograndense, Grant No. 100476 and the CNPq – Conselho Nacional de Desenvolvimento Científico e Tecnológico, Brazil, Grant No. 1498930 for this study.


Achata, E., Esquerre, C., O’Donnell, C., and Gowen, A. (2015). “A study on the application of near infrared hyperspectral chemical imaging for monitoring moisture content and water activity in low moisture systems,” Molecules 20(2), 2611-2621. DOI: 10.3390/molecules20022611

Alves, A., Simões, R., Santos, C., Potts, B. M., Rodrigues, J. C., and Schwanninger, M. (2012). “Determination of Eucalyptus globulus wood extractives content by NIR-based PLS-R models: Comparison between extraction procedures,” J. Near Infrared Spectrosc. 20(2), 275-285. DOI: 10.1255/jnirs.987

Alves, A., Simões, R., Stackpole, D. J., Vaillancourt, R. E., Potts, B. M., Schwanninger, M., and Rodrigues, J. C. (2011a). “Determination of the syringyl/guaiacyl ratio of Eucalyptus globulus wood lignin by near infrared-based partial least squares regression models using analytical pyrolysis as the reference method,” J. Near Infrared Spectrosc. 19(5), 343-348. DOI: 10.1255/jnirs.946

Alves, I. C. N., Gomide, J. L., Colodette, J. L., and da Silva, H. D. (2011b). “Caracterização tecnológica da madeira de Eucalyptus benthamii para produção de celulose kraft,” Ciência Florestal21(1), 167-174. DOI: 10.5902/198050982759

Baillères, H., Davrieux, F., and Ham-Pichavant, F. (2002). “Near infrared analysis as a tool for rapid screening of some major wood characteristics in a eucalyptus breeding program,” Ann. For. Sci. 59(5-6), 479-490. DOI: 10.1051/forest:2002032

Birkett, J. A., and Gambino, M. J. T. (1989). “Estimation of pulp kappa number with near-infrared spectroscopy,” Tappi J. 72(9), 193-197.

Castillo, R. P., Baeza, J., Rubilar, J., Rivera, A., and Freer, J. (2012). “Infrared spectroscopy as alternative to wet chemical analysis to characterize Eucalyptus globulus pulps and predict their ethanol yield for a simultaneous saccharification and fermentation process,” Appl. Biochem. Biotechnol.168(7), 2028-2042. DOI: 10.1007/s12010-012-9915-1

Cruz, C. D. (2001). “Programa Genes: Aplicativo Computacional em Genética e Estatística,” Editora UFV, Viçosa, MG.

Estopa, R. A., Milagres, F. R., Gomes, F. J. B., and Amaral, C. A. S. (2017). “Caracterização química da madeira de Eucalyptus benthamii por meio de espectroscopia nir,” O Papel 78(2), 75-81.

Fagan, C. C., Everard, C. D., and McDonnell, K. (2011). “Prediction of moisture, calorific value, ash and carbon content of two dedicated bioenergy crops using near-infrared spectroscopy,” Bioresource Technol. 102(8), 5200-5206. DOI: 10.1016/j.biortech.2011.01.087

Gierlinger, N., Schwanninger, M., Hinterstoisser, B., and Wimmer, R. (2002). “Rapid determination of heartwood extractives in Larix sp. by means of Fourier transform near infrared spectroscopy,” J. Near Infrared Spectrosc. 10(3), 203-214. DOI: 10.1255/jnirs.336

Gomide, J. L., Fantuzzi Neto, H., and Regazzi, A. J. (2010). “Análise de critérios de qualidade da madeira de eucalipto para produção de celulose kraft,” Rev. Árvore 34(2), 339-344. DOI: 10.1590/S0100-67622010000200017

Hein, P. R. G., Lima, J. T., and Chaix, G. (2010). “Effects of sample preparation on NIR spectroscopic estimation of chemical properties of Eucalyptus urophylla S.T. Blake wood,” Holzforschung 64(1), 45-54. DOI: 10.1515/HF.2010.011

Hein, P. R. G. (2012). “Estimating shrinkage, microfibril angle and density of Eucalyptus wood using near infrared spectroscopy.” J. Near Infrared Spectrosc. 20(4), 427-436. DOI: 10.1255/jnirs.1005

He, W., and Hu, H. (2013). “Rapid prediction of different wood species extractives and lignin content using near infrared spectroscopy,” J. Wood Chem. Technol. 33(1), 52-64. DOI: 10.1080/02773813.2012.731463

Hou, S., and Li, L. (2011). “Rapid characterization of woody biomass digestibility and chemical composition using near-infrared spectroscopy,” J. Integr. Plant Biol. 53(2), 166-175. DOI: 10.1111/j.1744-7909.2010.01003.x

Inagaki, T., Yonenobu, H., and Tsuchikawa, S. (2008). “Near-infrared spectroscopic monitoring of the water adsorption/desorption process in modern and archaeological wood,” Appl. Spectrosc. 62(8), 860-865. DOI: 10.1366/000370208785284312

Lupoi, J. S., Singh, S., Davis, M., Lee, D. J., Shepherd, M., Simmons, B. A., and Henry, R. J. (2014). “High-throughput prediction of eucalypt lignin syringyl/guaiacyl content using multivariate analysis: A comparison between mid-infrared, near-infrared, and Raman spectroscopies for model development,” Biotechnol. Biofuels 7(93), 1-14. DOI: 10.1186/1754-6834-7-93

Milagres, F. R., Gomide, J. L., Magaton, A., and Fantuzzi Neto, H. (2013). “Influência da idade na geração de modelos de espectroscopia NIR, para predição de propriedades da madeira de Eucalyptusspp,” Rev. Árvore 37(6), 1165-1173. DOI: 10.1590/S0100-67622013000600018

Pažitný, A., Boháček, Š., and Russ, A. (2011). “Application of distillery refuse in papermaking: novel methods of treated distillery refuse spectral analysis,” Wood Res. 56(4), 533-544.

Pecoraro, E., Pizzo, B., Alves, A., Macchioni, N., and Rodrigues, J. C. (2015). “Measuring the chemical composition of waterlogged decayed wood by near infrared spectroscopy,” Microchem. J.122, 176-188. DOI: 10.1016/j.microc.2015.05.005

Poke, F. S., and Raymond, C. A. (2006). “Predicting extractives, lignin, and cellulose contents using near infrared spectroscopy on solid wood in Eucalyptus globulus,” J. Wood Chem. Technol. 26(2), 187-199. DOI: 10.1080/02773810600732708

Ramadevi, P., Hegde, D. V., Varghese, M., Kamalakannan, R., Ganapathy, S. P., and Gurumurthy, D. S. (2016). “Evaluation of lignin syringyl/guaiacyl ratio in Eucalyptus camaldulensis across three diverse sites based on near infrared spectroscopic calibration modelling with five Eucalyptus species and its impact on kraft pulp yield,” J. Near Infrared Spectrosc. 24(6), 529-536. DOI: 10.1255/jnirs.1251

Ramalho, F. M. G., Hein, P. R. G., Andrade, J. M., and Napoli, A. (2017). “Potential of near-infrared spectroscopy for distinguishing charcoal produced from planted and native wood for energy purpose,” Energy Fuels 31(2), 1593-1599. DOI: 10.1021/acs.energyfuels.6b02446

Sandak, A., Sandak, J., Böhm, K., Zitek, A., and Hintestoisser, B. (2016). “Near infrared spectroscopy as a tool for in-field determination of log/biomass quality index in mountain forests,” J. Near Infrared Spec. 24(6), 587-594. DOI: 10.1255/jnirs.1231

Sandak, A., Sandak, J., and Negri, M. (2011). “Relationship between near-infrared (NIR) spectra and the geographical provenance of timber,” Wood Sci. Technol. 45(1), 35-48. DOI: 10.1007/s00226-010-0313-y

SCAN CM 40:01 (2001). “Wood chips for pulp production, Size distribution,” Scandinavian Pulp, Paper and Board Testing Committee, Stockholm, Sweden.

Schimleck, L. R., Roran, J. C., and Rimbawanto, A. (2003). “Near infrared spectroscopy for cost-effective screening of foliar oil characteristics in a Melaleuca cajuputi breeding population,” Journal of Agricultural and Food Chemistry (Easton) 51(9), 2433-2437. DOI: 10.1021/jf020981u

Schwanninger, M., Rodrigues, J. C., and Fackler, K. (2011). “A review of bands assignments in near infrared spectra of wood and wood components,” J. Near Infrared Spectrosc. 19(5), 287-308. DOI: 10.1255/jnirs.955

Schwanninger, M., Rodrigues, J. C., Pereira, H., and Hinterstoisser, B. (2004). “Effects of short-time vibratory ball milling on the shape of FT-IR spectra of wood and cellulose,” Vib. Spectrosc. 36(1), 23-40. DOI: 10.1016/j.vibspec.2004.02.003

Shou, G., Zhang, W., Gu, Y., and Zhao, D. (2014). “Application of near infrared spectroscopy for discrimination of similar rare woods in the Chinese market,” J. Near Infrared Spectrosc. 22(6), 423-432. DOI: 10.1255/jnirs.1136

Sundaram, J., Mani, S., Kandala, C. V. K., and Holser, R. A. (2015). “Application of NIR reflectance spectroscopy on rapid determination of moisture content of wood pellets,” Am. J. Analyt. Chem. 6(12), 923-932. DOI: 10.4236/ajac.2015.612088

TAPPI T257 CM-02. (2002). “Sampling and preparing wood for analysis,” TAPPI Press, Atlanta, GA.

TAPPI T264 CM-07. (2007). “Preparation of wood for chemical analysis,” TAPPI Press, Atlanta, GA.

TAPPI T222 OM-02. (2002). “Acid-insoluble lignin in wood and pulp,” TAPPI Press, Atlanta, GA.

TAPPI UM 250. (1991). “Acid-soluble lignin in wood and pulp,” TAPPI Useful Method 250, TAPPI Press, Atlanta, GA.

TAPPI T203 CM-99. (1999). “Alpha-, beta- and gamma-cellulose in pulp,” TAPPI Press, Atlanta, GA.

Terazawa, Y., Miyazawa, M., Kawano, S., and Maekawa, T. (2003). “Infrared spectra analyses of gelatinized and retrograded wheat starch,” J. Japanese Soc. Food Sci. 50(4), 162-166. DOI: 10.3136/nskkk.50.162

Thygesen, L., and Lundqvist, S. (2000a). “NIR measurement of moisture content in wood under unstable temperature conditions. Part 1. Thermal effects in near infrared spectra of wood,” J. Near Infrared Spectrosc. 8(3), 183-189. DOI: 10.1255/jnirs.277

Thygesen, L., and Lundqvist, S. (2000b). “NIR measurement of moisture content in wood under unstable temperature conditions. Part 2. Handling temperature fluctuations,” J. Near Infrared Spectrosc. 8(3), 191-199. DOI: 10.1255/jnirs.278

Trugilho, P. F., Goulart, S. L., Assis, C. O., Couto, F. B. S., Alves, I. C. N., Protásio, T. P., and Napoli, A. (2015). “Características de crescimento, composição química, física e estimativa de massa seca de madeira em clones e espécies de Eucalyptus jovens,” Ciênc. Rural 45(4), 661-666. DOI: 10.1590/0103-8478cr20130625

Trugilho, P. F., Lima, J. T., and Mendes, L. M. (1996). “Influência da idade nas características físico-químicas e anatômicas da madeira de Eucalyptus saligna,” Revista Cerne 2(1), 94-111.

Tsuchikawa, S. (2007). “A review of recent near infrared research for wood and paper,” Appl. Spectrosc. Rev. 42(1), 43-71. DOI: 10.1080/05704920601036707

Tsuchikawa, S., and Kobori, H. (2015). “A review of recent application of near infrared spectroscopy to wood science and technology,” J. Wood Sci. 61(3), 213-220. DOI: 10.1007/s10086-015-1467-x

Tsuchikawa, S., and Schwanninger, M. (2013). “A review of recent near-infrared research for wood and paper (part 2),” Appl. Spectrosc. Rev. 48(7), 560-587. DOI: 10.1080/05704928.2011.621079

Tsuchikawa, S., and Tsutsumi, S. (1998). “Adsorptive and capillary condensed water in biological material,” J. Mater. Sci. Lett. 17(8), 661-663. DOI: 10.1023/A:1006672324163

Watanabe, A., Morita, S., and Ozaki, Y. (2006). “A study on water adsorption onto microcrystalline cellulose by near-infrared spectroscopy with two-dimensional correlation spectroscopy and principal component analysis,” Appl. Spectrosc. 60(9), 1054-1061. DOI: 10.1366/000370206778397452

Wright, J. A., Birkett, M. D., and Gambino, M. J. T. (1990). “Prediction of pulp yield and cellulose content from wood samples using near infrared reflectance spectroscopy,” Tappi J. 73(8), 164-166.

Yang, Z., Lv, B., and Fu, Y. J. (2012). “The relationship between near infrared spectroscopy and surface color of eight rosewoods,” Adv. Mater. Res. 479-481, 1772-1776. DOI: 10.4028/

Zhou, C., Jiang, W., Via, B. K., Chettyd, P. M., and Swain, T. (2016). “Monitoring the chemistry and monosaccharide ratio of Eucalyptus dunnii wood by near infrared spectroscopy,” J. Near Infrared Spectrosc. 24(6), 537-548. DOI: 10.1255/jnirs.1254

Zhou, C., Jiang, W., Via, B. K., Fasinac, O., and Han, G. (2015). “Prediction of mixed hardwood lignin and carbohydrate content using ATR-FTIR and FT-NIR,” Carbohydr. Polym. 121, 336-341. DOI: 10.1016/j.carbpol.2014.11.062

Article submitted: February 2, 2018; Peer review completed: March 18, 2018; Revised version received: May 15, 2018; Accepted: May 16, 2018; Published: May 25, 2018.

DOI: 10.15376/biores.13.3.5394-5407