On-line characterization of wood chip brightness and chemical composition by means of visible and near-infrared spectroscopy

2.1 Sample preparation

In October 2018, logs were freshly harvested in the vicinity of the town of 100 Mile House, in the province of British Columbia, Canada. Two logs of 2 m long and with an average diameter of about 15 cm in diameter were harvested from the base of two separate trees for four different species common in the area: aspen (Populus tremuloides Michx.), black cottonwood (Populus trichocarpa Torr. & A.Gray ex. Hook.), black spruce (Picea mariana (Mill.) Britton, Sterns & Poggenburg) and Douglas fir (Pseudotsuga menziesii (Mirbel) Franco). In the remainder of this paper, we will refer to these as A, BC, BS and DF, respectively. The logs were sealed in plastics bags immediately after harvesting and shipped to FPInnovations’ facilities in Vancouver. Logs were debarked manually and chipped in a CM&E Model 36 disc chipper. Chips from each species were sealed in separate plastics bags and the chipper was cleaned between species to avoid mixing. In November 2018, four small chip piles of approximately 1 m in diameter and 50 cm in height, were constructed in FPInnovations’ yard and left to age for one year under natural conditions. Each month, a sample of chips of approximately 3 kg was collected from the surface of each pile, sealed in a plastic bag and immediately frozen. In November 2019, all samples collected during the previous 12 months were processed through a Williams classifier for 1.5 min. Chips smaller than 3/8ʺ (0.95 cm) and larger than 1 1/8ʺ (3.49 cm) were discarded.

2.2 Near-infrared spectroscopy

Fresh samples were scanned frozen (−20 °C) and unfrozen (20 °C) with a Polytec PSS-2121 (Polytec GmbH, Waldbronn, Germany) near-infrared (NIR) spectrometer connected to a Polytec PSS-H-A03 sensor head. This spectrometer operates in the wavelength range of 1100–2100 nm. Further details on this system are given in Hans et al. (2017). The sensor head was installed over a turntable, at a distance of 25 cm from the surface of the sample that was placed in a 40 cm diameter pan. The Polytec spectrometer’s integration time was set to 25 ms and 400 spectra were averaged per scan. The samples were then dried over-night at 35 °C in a ventilated drying chamber to a moisture content of 5.5 ± 0.8% (wet basis). In the chamber, the chips were spread on a metallic lattice to ensure even drying. Samples were then re-scanned under the Polytec system and with an ASD LabSpec 4 VIS-NIR spectrometer connected to an ASD turntable. The ASD LabSpec 4 operates in the wavelength range of 350–2500 nm with a resolution of 1 nm. The light source of the ASD turntable was positioned 10 cm above the surface of the sample that was placed in a 14 cm diameter pan. The ASD spectrometer’s integration time was set to 34 ms and 50 spectra were averaged per scan. Finally, following standard T257 (Tappi standard 2014), the chip samples were ground in a Wiley mill (model ED-5; Thomas Scientific, Swedesboro, New Jersey) equipped with a 0.50 mm sieve and then further ground in a second Wiley mill (model 8-338; Thomas Scientific) equipped with a 40 mesh (0.38 mm). Dry ice was added with the chips in the Wiley mills to avoid over-heating (Hu et al. 2011). A subsample of the final wood powder produced was collected from each sample and placed in an aluminum pan with a diameter of 5 cm. This subsample was scanned with the ASD LabSpec 4 connected to an ASD contact probe with an illumination diameter of 1 cm. The contact probe was placed directly in contact with the powder but no pressure was applied (other than that generated by the probe’s own weight).

This succession of spectroscopic measurements generated four distinct spectral datasets acquired from various types of wood material, with different instruments and under diverse conditions of moisture and temperature. The characteristics of these datasets along with their labels are summarized in Table 2.

2.3 Brightness and chemical analysis

For each sample, another subsample of wood powder was collected and ISO brightness was measured as the directional reflectance of light at 457 nm following Hu et al. (2011). A final subsample of approximately 1 g of wood power was extracted with acetone for 20 min on an ASE (Accelerated Solvent Extraction) system (Thermo Fischer Scientific) at 1600 psi and 90 °C. The extractives were evaporated, oven dried for 1 h and weighed to determine the percentage of extractives. The extracted wood powder was weighed in a beaker, hydrolyzed with 72% sulfuric acid and diluted to 3% by adding 84 mL of milli-Q water before autoclaving for 1 h at 121 °C. After cooldown, the resulting samples were filtered in a medium grit glass filter (Corning Pyrex 30 ml Gooch Crucible Medium Fritted 30 mm Disc). The filtered solids were dried and ashed at 525 °C to determine the percentage of Klason or acid insoluble lignin. The filtrate of the hydrolyzed sample was diluted to 100 ml and analyzed by UV spectroscopy (at 205 nm using a quartz cell) to determine the percentage of acid soluble lignin. Total lignin was computed as the sum of acid soluble and insoluble lignin. The filtrate was then further diluted to 100 mL and analyzed by ion chromatography with the ICS-5000+ (Thermo Fischer Scientific) configured for HPAE-PAD (high-performance anion-exchange chromatography with pulsed amperometric detection) and equipped with a CarboPac PA-1 analytical column and a post-column addition. Six standards were used to constitute the calibration curve and an internal standard of deoxy-d-glucose was added to all injections to ensure the stability of the results. Following Acquah et al. (2015), the percentages of arabinan, galactan, glucan, xylan and mannan were determined and summed to obtain holocellulose.

2.4 Data reconciliation

Raw data is subject to errors. Data reconciliation is a technique used to denoise data through the enforcement of equality constraints that typically arise from conservation laws, such as mass and energy balances. It aims at making the minimum adjustments to data to satisfy conservation laws (Crowe et al. 1983). In this study, the measurement of extractives, lignin and holocellulose content was performed on a percent of total mass basis. The sum of these constituents must equal 99.5% for hardwoods and 99.7% for softwoods to meet conservation laws, assuming average ash contents of 0.5 and 0.3, respectively (Table 1). According to this, and following Crowe et al. (1983), the raw data (X) were reconciled ( X cor ) by matrix projection as follows:

where P is a diagonal matrix of the squared standard error of measurements (SEMs) and B is a balance matrix (+1 for inputs and −1 for outputs). For extractives, because a linear regression model could be fit to the original (unreconciled) trends with p-value <0.001 for each species; the root mean square error (RMSE) of this model was used as an estimate of the SEMs. For lignin and holocellulose, the standard deviation of the data for each species was used to estimate the SEMs. The estimated SEMs used for the data reconciliation are given in Table 3 for each species.

3.1 Age and species effects

In order to mitigate variability in feedstock properties, it is common practice in the pulp and paper industry to season the wood. Seasoning is the practice of storing the wood in piles (either of logs or chips) for a certain amount of time. This process typically leads to a reduction of extractives content and to changes in the wood chemical composition (Allen et al. 1991; Gutiérrez et al. 1998, 1999). When chips are stored in piles, only a small layer at the surface is exposed to abiotic factors. This fraction is referred to as the “outer weather belt” in Heinek et al. (2013) and has a depth of 20–40 cm. In large chip piles, the rest of the chips generally offer an aerobic atmosphere protected from most abiotic factors. Under these conditions, several processes and chemical reactions take place including exothermic oxidation reactions, aerobic respiration (from the microorganisms), enzymatic reactions, heat transfer, continuation of life functions (remaining living cells) and evaporation of volatile compounds (Ferrero et al. 2009; Jirjis and Theander 1990; Nugent et al. 1976; Ramnath et al. 2018). A review of the microorganism communities present in large-scale wood chip and log piles is provided in Noll and Jirjis (2012). However, the chip piles created for the purpose of this study, because of the small size, were only representative of the “outer weather belt” and it can be safely assumed that the inner part was not protected. As a result, the chips were mostly affected by abiotic factors such as rain, temperature, and UV radiation (Oberhofnerová et al. 2017) as well as by non-decaying fungi (mould, yeast and staining fungi) that grow at the surface or within cell lumens (Stirling 2005).

Table 3 shows the estimated SEMs after data reconciliation. In comparison with the ones before, it can be seen the estimated SEMs of extractives and lignin both remain essentially unchanged. However, the SEMs of holocellulose were reduced by a factor of 3–8 depending on species. This reflects the relatively higher measurement error for this compound obtained with the reference method (ion chromatography).

Reconciled data illustrating changes in chip brightness and chemical composition with age are shown in Figure 1. It can be seen that brightness and extractives both decreased with age. The statistical significance of these trends was determined by means of an F-test (α = 0.05) over a linear regression model. All trends were significant for both variables and all species with only one exception: the brightness trend of DF had a p-value of 0.056. Loss in brightness and extractives is explained by the abiotic degradation of the wood chips. Brightness loss is attributed mostly to the growth of non-decaying fungi at the surface of the chips and the leaching of water soluble compounds. A 7% drop in ISO brightness of lodgepole pine chips was reported in Stirling (2005) after 115 days of storage and was attributed to the grow of mould. To a lesser extent, brightness loss may also have been caused by the deposition of air dust particles (Oberhofnerová et al. 2017). Loss of extractives was caused by oxidation, leaching and evaporation processes as well as degradation by non-decaying fungi (Gutiérrez et al. 2009; Stirling 2005; Weigel et al. 2002).

Figure 1:

Change in wood chip brightness, extractives, lignin and holocellulose content as well as the sum of the three latter (total) with age for four different species: aspen (A), black cottonwood (BC), black spruce (BS) and Douglas-fir (DF).

No noticeable changes in lignin were observed while only two species, BC and DF, showed a significant increase in holocellulose with age. The small pile size did not create the right conditions for wood decaying fungi such as white-rot, brown-rot and soft-rot to develop. This explains the constant percentage of lignin content observed since it is well accepted that losses in lignin and holocellulose are mainly caused by the activity of decaying fungi. The slight increase of holocellulose for BC and DF observed in this study is only relative as the amount of each chemical compound is expressed as a percentage of the total mass, which decreases with time, i.e. the percentage of extractives dropped by an amount similar to the increase in holocellulose. The fact that the relative percentage of holocellulose increased and not lignin is an artifact of measurement error and data reconciliation. Under ideal circumstances one would expect both holocellulose and lignin percentages to increase proportionally to offset the loss of extractives.

Differences in brightness and chemical composition between species are shown in Figure 2. Highly significant differences (p-value <0.0001) between all species were observed for lignin and holocellulose. Differences in brightness were mostly observed between the hardwood and softwood groups, with DF having very low brightness. This difference is visible with the naked eye since DF has a characteristic dark red tint due to its heartwood extractives composition.

Figure 2:

Boxplots showing differences in brightness, extractives, lignin and holocellulose content between species. Differences were determined regardless of age influence by ANOVA followed by Tukey’s test with the following coding for p-value: **** <0.0001, *** <0.001, ** <0.01, * <0.05.

Extractives were significantly higher for A while no noticeable differences were observed between the three other species. Species A is well known for its high extractives content among Canadian species (Allen 1987). The chemical composition of all four species agrees with the literature (Table 1). Similar levels of extractives and differences between A, BS and DF were reported in Allen (1987). For these, the results obtained for total lignin and holocellulose content are in accordance with the data reported in Kaar and Brink (1991) and Rowell (2005). No reported values were found for BC.

3.2 VIS-NIR calibration model development

Calibration models capable of predicting brightness, extractives, lignin and holocellulose (target variables) from VIS-NIR spectra were developed for each dataset presented in Table 2. The significant reduction in the estimated SEMs obtained for holocellulose after data reconciliation (Table 3) directly translated into more accurate calibration models. Calibrations obtained for extractives and lignin before and after data reconciliation showed similar performance. For conciseness, only the results obtained from the reconciled data will be presented. The calibration models were developed and refined in three steps as described below.

3.3 Qualitative validation over historical data

An on-line analyzer (FITNIR MC, FITNIR Analyzers Inc., Vancouver) using a Polytec spectrometer was installed on the feed to the digester in a kraft pulp mill located in Western Canada processing both softwood and hardwood grades. Hardwood chips are a mix of A and balsam poplar (Populus balsamifera L.), a species closely related to BC as the two species have similar chemistry and can both hybridize (Nesom 2002). Softwood chips are a mix of spruce, pine and fir (including BS). The analyzer scans chips reclaimed from the bottom of the chip pile, after outdoor seasoning. The chips are typically subject to high variability in moisture content and temperature. Spectra acquired between October 23rd, 2019, and May 26th, 2020, were retrieved for validation purposes. The models developed based on the Polytec-Fresh-EPO dataset (Table 6) were employed for the prediction of each property using the corresponding pre-processing, desensitizing and variable selection methods, leading to the time series presented in Supplementary Figure S2. To enhance visualization, time series were first-order filtered using a time constant of 100 min. As discussed previously, the spectral range used in the Polytec spectrometer is not the most appropriate for brightness prediction, nevertheless it is shown for completeness.

In comparison with hardwood, softwood runs are characterized by an increase in lignin content and decrease in brightness, extractives and holocellulose. This agrees with the literature (Table 1) and the data reported earlier (Figure 2). In addition, within each grade, variations in wood properties can also be observed, which is consistent with the observation made in Skvaril et al. (2015b) for lignin. The sum of the chemical compounds (total in Supplementary Figure S2) are found within a narrow range (100.01 ± 0.22%) but, considering ash content, are slightly overestimated. When transferring calibration models onto an on-line analyzer, it is common that an offset needs to be adjusted to compensate for height variations between the probe head and the chip surface (which are different between the laboratory and the mill). No attempt at correcting this offset was performed here. This shows that, while this preliminary analysis supports the quality of the measurements, a more formal quantitative validation will be required to fully validate the prediction models.

However, this qualitative validation demonstrates that each property presents variability between runs (Supplementary Figure S2). Between-run variability is quantified here as the difference between the highest and lowest average value for each property of the same grade. Hardwood generally showed less variability than softwood (with the exception of brightness), which was expected as the hardwood chip mixture is composed of only two species while softwood is a poorly defined mixture of several species. Brightness showed the highest between-run variability (2.9% for softwood and 4.7% for hardwood), followed by lignin (1.6% for softwood and 0.8% for hardwood) and holocellulose (1.5% for softwood and 1.0% for hardwood). Finally, extractives presented a smaller between-run variability with 0.3% for softwood and 0.1% for hardwood. However, it is worth pointing out that this measurement does not take into account the extractives content of residual bark that could be present in the chip furnish in small but variable amounts depending on debarking efficiency.