Near-infrared reflectance (NIR) spectroscopy was used to determine correlations between acoustic velocity and stiffness properties of downed pine trees in the southern coastal plains of the United States. Three acoustic measurement methods (longitudinal, transverse, and offset) were used. From the measurement of the acoustics, the time of flight (TOF) was determined from the downed trees. Increment core samples were obtained from each thirty downed pine trees in the study. NIR spectra were obtained using a fiber probe on the radial surface of each core to rapidly correlate the speed of sound, estimate the strength properties of the downed trees, and the TOF acoustic assessments. The NIR prediction was very good for the transverse and offset methods. The predictability diagnostic was above an R2 of 0.70 for both offset measurements for the transverse methods for the acoustic velocity and dynamic modulus of elasticity (MOE). The longitudinal measurement exhibited the weakest model (R2 < 0.65) for both the acoustic velocity and the MOE with the highest standard error of prediction between 3.0 (ELVLSWV) and 0.31 (VLSWV) for the three measurement types. All the standard errors of calibration were below 1% except in ELVOSWV, which was ∼2%. The dry density measured from the increment cores had a moderate correlation (R2 ∼ 60%), compared with the lower correlation (R2 ∼ 50%) by the green density in the multiple linear regression output. The results of the acoustic model indicated that NIR spectroscopy has the potential to predict the acoustic velocity and corresponding stiffness of downed trees.Abstract
Weather events such as hurricanes and tornadoes cause significant damage to trees, which require a good assessment of their physical and mechanical properties to help extract the best value products. The modulus of elasticity (MOE) is a measure of stiffness and is often used as a surrogate to classify lumber and, more recently, logs. MOE cannot be determined visually for trees; therefore, both destructive and nondestructive approaches have been adopted in tree assessments. However, destructive procedures require extensive and costly sample preparation and testing. Considering the large number of trees that must be assessed during hurricane and tornado events, conventional destructive methods will not be practical and feasible. The industry would thus benefit from rapid, nondestructive, and cost-effective alternatives such as acoustic technologies and near-infrared spectroscopy (NIR) estimations.
To develop accurate models for the prediction of wood mechanical properties, repurposing, and use of downed timber along the supply chain, more rapid and accurate methods of compositional analysis are needed. NIR coupled with multivariate data analysis is proven to be a fast and nondestructive analytical tool in the prediction of internal wood quality traits, including storm-damaged trees (Downes et al. 2000, 2014; Gindl et al. 2001; Thumm and Meder 2001; Via et al. 2003; Kelley et al. 2004; Tsuchikawa 2007; Wang et al. 2010; Cieszewski et al. 2013; Zhao et al. 2013; Tsuchikawa and Kobori 2015).
Reliable and rapid techniques for nondestructive assessment of wood and other forest products can improve financial return and reduce wood waste. Studies by Legg and Bradley (2016) also concluded that having a good understanding of tree stiffness could be used to improve breeding, planting, and silviculture management for better future forests. Field and laboratory-based tools such as NIR (Horvath et al. 2011, Viet et al. 2021), acoustics (Wang et al. 2001, Wang et al. 2007a, Wang et al. 2007b), pilodyn (Cown 1978, Raymond and MacDonald 1998, Wu et al. 2010, Wessels et al. 2011, Chen et al. 2015, Gao et al. 2017), resistograph (Downes et al. 2018, Gendvilas et al. 2021), rigidimeter (Launay et al. 2000, 2002), computed tomography (CT) scanning (Hou et al. 2009, Beaulieu and Dutilleul 2019), the Scion's DiscBot (Schimleck et al. 2019), and SilviScan (Mora et al. 2009), among others, can be used to assess the density and strength properties of standing trees, logs, damaged trees, and samples obtained from an increment core, stem, or disc of the tree. In addition to the NIR, sound wave propagation (acoustic technologies) has also proven a useful and reliable rapid tool for the assessment of tree inner qualities (Wang et al. 2001, Pellerin and Ross 2002, Dickson et al. 2004, Legg and Bradley 2016).
Studies by Schimleck et al. (2001) estimated several physical and mechanical properties of different pine species using NIR spectroscopy where the density (R2 = 0.81, standard error of cross-validation, SECV = 38.5 kg/m3) and MOE (R2 = 0.81, SECV = 1124 GPa) diagnostics demonstrated that it was possible to obtain general calibrations for estimation of wood properties in several tropical, subtropical, and temperate pine species. Kohan et al. (2013) modeled the ultimate tensile strength, tensile MOE, bending strength, and bending stiffness of strand feedstock mechanical properties and found that the MOE for strands under bending exhibited the strongest correlation (R2 = 0.76). NIR has also been used to estimate density (Alves et al. 2012), and monitor wood heat treatment (Mitsui et al. 2008), and NIR has been applied in many different sectors, including agriculture (Batten 1998, Mitsui et al. 2008, Dale et al. 2013, Vincent and Dardenne 2021), pharmaceuticals (Luypaert et al. 2007, Palou et al. 2012), food industry (Prevolnik et al. 2004), and other fields (Hoffmeyer and Pedersen 1995, Fomina et al. 2011, Maji et al. 2019).
Spectroscopy investigates the interaction between matter and the photons of electromagnetic radiation (EMR) in the form of electric and magnetic waves. The NIR region covers the range from 4,000 to 14,000 cm−1 (2.5 to 0.7 mm), allowing the absorption of electromagnetic wavelength by organic molecules in this spectral range due to vibrations and excitations of overtones and combinations of fundamental elements of C–H, O–H, and N–H (Hoffmeyer and Pedersen 1995, Gäb et al. 2006, Schimleck et al. 2018).
Despite the advancement in the use of NIR in assessing tree properties, which has proven effective for assessing and monitoring biomaterial quality, there is quite a knowledge gap when used with acoustics. While the studies by Nasir et al. (2019) used longitudinal stress waves generated by a pendulum impact from a Metriguard 239A for velocity assessment and estimated dynamic MOE (MOEdy) using the Adaptive neuro-fuzzy inference system (ANFIS), our current approach focused on using the NIR principal component regression (PCR) + quantitative to predict acoustic velocity using the different probe positioning strategies. Hence, this current study focuses on using different probe positioning strategies and NIR as rapid approaches in assessing tree stiffness, which is one of the hindrances of the use of storm-damaged downed timber. Understanding the stiffness will help landowners classify downed timber into product classes that might not have been considered by the landowner previously. Hence, this study used and compared three different acoustic velocity sampling methods and the ability to use NIR spectroscopy as a surrogate prediction tool.
Methodology
Near-infrared spectroscopy
The infrared (IR) region lies between the visible and microwave regions of the electromagnetic spectrum. NIR measures the amount of near-infrared light a sample absorbs, transmits, or reflects based on its chemical composition. A Perkin Elmer Spectrum 400 Fourier-transform infrared (FT-IR)/FT-NIR spectroscope was used to collect the NIR spectra from the radial face of the dried core samples, with the smoothest surface used for spectra collection. To meet the statistical requirements, replicate scans were taken on each of the core samples from the 30 different trees. The 1,860 scans acquired were averaged to 30 spectra, with each spectrum obtained by averaging 62 independent scans. All 30 spectra from the averaged scans were used for prediction, and 15 samples from the crown region were used for calibration. Since the thickness of the samples was 1.0 ± 0.03 mm, it was assumed that the wood chemistry on both faces was equal. Absorbance was recorded from 1,000 to 25,000 nm with a resolution of 0.5 nm with 64 scans. All NIR measurements were made in a controlled environment of 40 percent relative humidity and a temperature of 20°C.
Core samples from downed trees
Thirty downed trees were selected from two different diameter classes at breast height (Chip-N-Saw 8 to 10.9 in. and saw logs 11 in. up) for the development of acoustic velocity measurement. Fresh core samples were taken from each downed tree. The samples were dried in a vacuum freeze drier (The Labconco 6 Liter Console Freeze Dry System with 16-port round drying and Edwards RV8 Vacuum Pump, Hampton, New Hampshire) operating at a condenser temperature of −50°C and a vacuum pressure of 10 Pa 48 hours (Hammami et al. 1999). The moisture-free core samples were cut to a thickness of 1.0 mm and taken to the Perkin Elmer Spectrum 400 FT-IR/FT-NIR spectroscope for analysis.
Downed tree properties and sound waves
The Fakopp 1D Microsecond Timer (Fakopp Enterprise, Agfalva, Hungary) was used to measure stress wave velocity that was transmitted and received via two probes inserted into the sapwood of each tree at a 45-degree angle pointed toward each other as shown in Figure 1 (Wang et al. 2007b). The transverse (VTSWV), longitudinal (VLSWV), and the opposite phase (offset) (VOSWV) methods used in this study are shown in Figure 1 as described in Musah et al. (2022).
Citation: Forest Products Journal 73, 2; 10.13073/FPJ-D-22-00067
The green and dry densities were estimated from the increment core samples of the downed trees (Ross 2010). The stiffness is related to the dynamic modulus of elasticity (MOEdy); EL was estimated by measuring the velocity of the traveling wave and by using the one-dimensional wave equation (Wang et al. 2002, Wang 2013) indicated in Equations 1 and 2.
where MOEdy or EL (Gpa) is the dynamic modulus of elasticity, V (km/s) is the one-dimensional wave velocity, and r (kg/m3) is the bulk density of the wood (estimated from the oven dry methods).
Multivariate analysis
Multivariate calibration is one of the major success stories of chemometrics, owing to a large number of applications in NIR spectroscopy (Martens and Martens 2001). The chemometric approach to quantification provides a choice of multivariate calibration algorithms. This algorithm gives reliable model output obtained using multivariate calibration methods such as partial least squares (PLS) regression and principal component regression (PCR). This study used both PCR and PLS regression to calibrate the NIR spectra and compare the three different methods of acoustic estimation and their corresponding NIR prediction. Studies have indicated that PCR could yield simple and better models for cases where the interpretation of the model was important (Jouan-Rimbaud et al. 1995, Via et al. 2014). PLS is mostly shown to perform better than PCR when dealing with low-detection limit samples (Lipp 1996).
Calibration development and statistics
To analyze the NIR results, PLS regression models with four cross-validation segments were used for calibration development and optimization. The PerkinElmer Spectrum Quant+ software was used for all the analyses with a spectral signature measured between 10,000 and 4,000 cm−1. Cores from all 30 downed timber trees were used for the calibration. Second-derivative spectra with a segment of 10 nm were used to develop the NIR analysis systems. Regression diagnostics were used to compare the predictive power of different models. Each tree model was deduced from the 20 scans taken from the core samples. In total, 15 core samples from the crown region of 15 trees from the site were used for building the validation model. Regression coefficients, root mean square error of validation (RMSEV), R2, and the adjusted R2 were used to compare the model predictability. The equation for the RMSEV used was
where yi is the prediction value of the ith observation, ŷi is the measured value of the observation, and n is the number of observations. The RMSEV for this paper was the root mean square error derived when regressing the calibration model to the validation data. A lower RMSEV and a higher R2 and adjusted R2 suggest a better model. The NIR-based models were developed by correlating the spectra data with basic density, acoustic velocity, and MOEdy by the partial least squares regression. The matrix was composed of estimated values of the stress wave timer and physical analyses (determined by standardized procedures) and by the NIR spectra collected in the respective core samples of the trees. The mathematical models for the spectra included central averaging, normalization, first derivative, and second derivative with a 10-point filter and a second-order polynomial. The leave-one-out method was used for full cross-validations, while independent set validation was performed using spectra from the 15 trees chosen at random.
The PLS algorithm used the data decomposition technique to extract the systematic variation that exists in the data set by using the raw untreated NIR spectra (i.e., absorption of NIR light at different wavelengths) as the independent (X) variables and regressed them on the results acquired from the conventional testing methods (that is dependent − Y variables). The regression of the PLS involves simultaneous and interdependent principal components analysis (PCA) decomposition of both the X and Y matrix. The measured absorbance of NIR of the cores of the downed trees at the wavelength is related to the acoustic and strength properties or other concentrations of the analyte of interest based on the Bouguer-Lambert-Beer or, more commonly, Beer's law, indicated in Equation 3 (Burns and Ciurczak 2007). Several statistics, including standard error of calibration (SEC), standard error of prediction (SEP), coefficient of determination (R2), and residual predictive deviation/ratio of performance to deviation (RPD), were used to evaluate the performance of our models. Models that had the least error values were selected and used to predict the acoustic velocity and MOEdy of the downed timber. The measure of how well a calibration fits the data is the SEC. The measure of how well the calibration predicts the constituent of interest for a set of unknown samples that are different from the calibration test set is given by the SEP. Formulas used to estimate the SEC and SEP given in Schimleck and Evans (2002) are indicated in Equations 4 and 5, respectively.
where A is the absorbance; I0 is the intensity of radiation entering the sample; I is the intensity of radiation transmitted by the sample; 
is the value of the constituent of interest for the validation sample in SEC and the value of the constituent of interest for the sample in the SEP; i is estimated using the calibration in SEC and predicted by the calibration in SEP; yi is the known value of the constituent of interest of sample i; NC is the number of samples used to develop the calibration; k is the number of factors used to develop the calibration; yi is the known value of the constituent of interest for sample i in both SEC and SEP; NP is the number of samples in the prediction set; NC is the number of samples used to develop the calibration; and k is the number of factors used to develop the calibration.
The PCR+ algorithm
This study used the PCR+ quantitative technique to establish the relationship between the property of interest. One for each component between the spectra of a set of calibration standards, and the corresponding property values determined by independent means (density, MOEdy, and velocity). These relationships can be used for the subsequent prediction of unknown samples.
Results
NIR model calibration and evaluation
Organic compounds consist of various functional groups whose molecular vibration is observed in the IR range. The absorptivity of NIR light in organic materials is very weak compared to that of IR light, which could measure high-density and concentration materials directly using NIR light (Tsuchikawa and Kobori 2015). Chemometrics, which involves the aid of computers and mathematics, is then applied to the spectra to derive meaningful interpretations of chemical information from samples of varying complexity. The NIR spectra for downed loblolly pine trees (Fig. 2) and property calibrations developed for the downed trees are given in Table 1. To avoid overfitting, the minimum SEP was used for choosing the best number of principal components. There were positive and strong associations with the acoustic velocity (VTSWV, VOSWV) and modulus of elasticity (ELVTSWV, ELVOSWV) with coefficients of determination (R2) above 70 percent. The longitudinal measuring method for both acoustics velocity (VLSWV) and the modulus of elasticity (ELVLSWV) demonstrated a moderate association (R2 ∼ 50%). The R2 for the dry density was above 60 percent, which was higher than the green density (R2 = 51.5%). The standard errors (SEC) were not far from each other and below 1.0, except in ELVLSWV (SEC =2.37). However, similar SEC across categories indicates that calibrations based on the NIR spectra of downed loblolly pine core samples can be used to predict wood properties that are shown in Raymond et al. (2001) and Raymond and Schimleck (2002), where NIR spectra from milled increment cores were used to estimate properties of wood in Tasmanian bluegum (Eucalyptus globulus Labill.) and to assess the quality of its wood pulp.
Citation: Forest Products Journal 73, 2; 10.13073/FPJ-D-22-00067
The data for the absorbance of downed timber spectra grouped by age and separated by their mean density are shown in Figure 3. The spectra were collected over the whole range of 4,000 cm−1 (2,500 nm) wave number to about 10,000 cm−1. The data for the transmittance, based on the mean density of the trees, showed a clear distinction in the transmittance percentage for each age group. The denser group (39 yr) recorded a low transmittance percentage compared to the less dense age group (15 yr). The older trees absorbed more incident radiation compared to the younger trees, probably due to the increased density in the visible light section of the electromagnetic spectrum (4 × 10−7 m to 7 × 10−7 m), which resulted in low transmittance. The relatively lower transmittance at the blue-light spectral region is most likely due to absorption by the lignin. However, the wavelength dependence does not have a strong effect and does not strongly influence the total transmittance. Fujimoto et al. (2012) accurately predicted wood density using NIR spectra obtained from 100 Japanese larch (Larix kaempferi) wood samples containing various amounts of moisture, where characteristic absorbance bands in the vicinity of 7,320 cm−1 (1,366 nm), 7,160 cm−1 (1,400 nm), and 7,000 cm−1 (1,428 nm) were key in predicting wood density. The air-dried wood showed the highest absorption values at peaks 7,320 cm−1 and 7,000 cm−1, indicating that wood density could be estimated independently of moisture content.
Citation: Forest Products Journal 73, 2; 10.13073/FPJ-D-22-00067
The transmittance data for the downed pine trees and the second-derivative spectra, which measured the rate of the absorbance change to the wavelength at the first derivative with the different age groups, are presented in Figure 4. Comparing the labeled spectra to the literature on band assignment in near-infrared spectra of wood and wood components (Schwanninger et al. 2011), the peak at 4,411 cm−1 (2,267 nm) can be assigned to lignin, 4,808 cm−1 (2,080 nm) is assigned to semicrystalline or crystalline regions in cellulose, 5,245 cm−1 (1,907 nm) is assigned to hemicellulose, and 6,003 cm−1 (1,666 nm) is assigned to hemicellulose at a bond vibration of first overtone C–H str. (CH3 groups). The peak at 7,092 cm−1 is assigned to lignin/extractives that are phenolic hydroxyl groups or from lignin (Baillères et al. 2002) as reported by Wüst and Rudzik (1996).
Citation: Forest Products Journal 73, 2; 10.13073/FPJ-D-22-00067
The number of factors used in the model was comparable with those reported by Schimleck et al. (2018), who modeled the NIR spectroscopy for the estimation of density (at 12% moisture content), MOEdy, and modulus of rupture using clear wood samples obtained from several pine species with transmitted NIR spectra from increment cores. However, the required number of factors in the present study was considerably more than those reported by most of the other studies. This is probably due to the large spectral variation caused by the differences in the stand age and diameter at breast height (dbh) classes.
Estimated acoustic velocity and NIR predicted acoustic velocity
The primary objectives of this study were to access the predictive accuracies of three different types of acoustic measurement positions and corresponding predictive acoustic stiffness with NIR as a nondestructive evaluation approach. NIR has been used to model nonchemistry secondary traits such as density (Fujimoto et al. 2012), acoustic velocity, and strength properties of wood with the aid of chemometrics where cellulosic features have an underlying effect. Several studies have shown that acoustics have been used for both juvenile wood (Lindstrom et al. 2002, 2004), mature wood (Chauhan and Walker 2004, Lindstrom et al. 2005), and standing or felled logs (Legg and Bradley 2016). Even though the relationship between the reduced model (PC ≤ 7) of the estimated acoustics velocity and the NIR predicted measurements for the three measurements ranged from 50 to 70 percent, the calculated data from the multiple linear regression for the full model (7 ≤ PC ≤ 10) of the properties ranged from about 60 to ∼80 percent (Fig. 5). Caution needs to be established in the use of the full model due to overfitting. The full model indicates the association of the NIR predicted and established measured properties. The regression analysis indicated a linear correlation between the time of flight (TOF) of downed trees for each method tested and the core sample predicted by the NIR. The range of acoustic velocity values predicted for the downed loblolly pine was from 2.40 to 3.33 km/s for the transverse acoustic velocity (VTSW), 5.16 to 6.05 km/s for the offset/opposite face method (VOSW), and 3.24 to 4.10 km/s for the longitudinal acoustic velocity. The relationship between estimated acoustics and the NIR predicted acoustic velocity based on the coefficient of determination for the offset (VOSW) and the transverse (VTSW) were similar (>70%), while the longitudinal (VLSW) was a little on the low side (63% at PC 8). This difference between the three different methods was attributed to the differences in the path of travel by the stress wave and the amount of wood through which the stress wave must pass (Schimleck et al. 2018). The 30 downed loblolly pines used in this study ranged from 15 to 40 years of age. The NIR predicted acoustic velocity of the downed pines was comparable to several other studies' TOF acoustic measurements: An average of 1.88 to 2.88 km/s was estimated for stands with ages within 8 to 25 years for studies conducted by Chauhan and Walker (2006). The larger and older trees recorded higher acoustic velocities both in the TOF estimation and in the NIR predicted compared to younger trees.
Citation: Forest Products Journal 73, 2; 10.13073/FPJ-D-22-00067
Laboratory estimated density and NIR predicted density
The ranges of the NIR predicted green density (0.65 to 0.99 g/cm3) and dry density (0.40 to 0.58 g/cm3), compared to the acoustic measurements (0.69 to 0.95 g/cm3 for green density and 0.40 to 0.59 g/cm3 for the dry density) is shown in Figure 6. The R2 for the dry density was higher than the green density, possibly due to the near equal moisture and weight in the cores leading to less variability in the green density as seen in both the acoustically measured and the NIR predicted outcomes for the downed trees in this study. The NIR results output for this study is similar to the ranges of results obtained by (Belonger et al. 1997) (0.44 to 0.51 g/cm3) for a 10-year-old loblolly pine trees using X-ray densitometry and also the ranges reported by Jones et al. (2008) ranging from 0.39 to 0.56 g/cm3 determined for 12-year-old loblolly pine trees. The overall low density of the predicted NIR could be attributed to the spectra that were obtained from the radial section of the core. Several studies have indicated that the spectra collected on the transverse sections were better than models based on spectra acquired from the tangential or radial section, where the difference between transversal and radial measurements was lower in the case of density estimation (Schimleck et al. 2005, Defo et al. 2007, Fujimoto et al. 2007). These differences are related to the direct interference of incident radiation by the cell wall as opposed to the free water in the cell lumen when measured transversely (Tsuchikawa et al. 1996), or they could also be the result of radiation influence on the scattering surface due to roughness differences of the transverse section when compared to the radial section, which is related to particle size differences (Schwanninger et al. 2011).
Citation: Forest Products Journal 73, 2; 10.13073/FPJ-D-22-00067
The effect of wood density on structural wood quality and wood strength properties makes it an important trait for assessing wood quality (Megraw 1985). Several studies have indicated that an increase in absorbance is a characteristic result of increased density (Gindl et al. 2001, Schimleck et al. 2002). Via et al. (2003) showed that an increase in density brought about a 0.27 mean increase in absorbance, which was attributed to baseline shift explained based on having more cell wall material and less lumen diameter, implying that more material is available for absorbance across all wavelengths in the NIR region. The relatively weak dynamic MOE (MOEdy) for both the acoustic estimated and NIR predicted could be attributed to the green density of the outer wood portion of the trunks, and not to a weighted measure of air-dry density at breast height (Isik et al. 2011).
Estimated MOEdy from the acoustic velocity and NIR predicted MOEdy
The MOEdy was estimated from the acoustic velocity and the density of each downed tree. The NIR predicted MOEdy was used to compare with the TOF-estimated MOEdy (Fig. 7). The MOEdy from the offset or opposite face acoustic velocity and the transverse methods were similar and moderately high at estimated predictability of R2 ∼ 0.70 at PC 8, compared with the lowest results obtained from the longitudinal method (R2 = 0.53 at PC 7). Similarly, the low MOEdy from the longitudinal measurement may be attributed to the length of travel of the wave (Schimleck et al. 2019). The model for the downed pine dynamic MOE was satisfactory because the prediction capacity was quite good, although the R2 was not as high as for all the different measuring methods.
Citation: Forest Products Journal 73, 2; 10.13073/FPJ-D-22-00067
The ability of wood to resist loads is characterized by its strength. The strength of the material is usually characterized by the values of its modulus of elasticity (MOE) and its modulus of rupture, MOR. The MOE determines the propagation of the speed of longitudinal waves in the material for a small amplitude of vibration. As wood is considered an elastic material, the MOE is defined as the slope of the stress-strain curve below the linear proportionality limit (Roylance 2001). Acoustic tools have been used to assess standing trees using the TOF acoustic wave, which is measured between the transmitter and receiver probe, in which the sound wave is generated by tapping one end with a light hammer (Wang et al. 2002, Wessels et al. 2011). This can be done even before harvest to enable management, planning, harvesting, and wood processing to be carried out in a way that maximizes extracted value from the resource (Schimleck et al. 2019). It can also be used for downed timber to determine which trees are of the best quality for use in certain applications. The ability to predict the stiffness with these rapid methods could help maximize the values and determine usage, particularly the lumber with dbh around 8 to 11 inches classified as Chip-N-Saw (TMS 2019), the group that is hard to consider in sorting and determination of use and prices due to relative size (bigger than pulp wood but not big enough to be considered saw logs).
Conclusion
This study investigated the applicability of NIR spectroscopy to predict acoustic velocity and estimated MOEdy from acoustic velocity and density. It was demonstrated that NIR spectroscopy from an incremental core sample of downed trees can be successfully used as a high throughput tool to nondestructively screen downed trees for important wood traits such as acoustic velocity, density, and TOF-estimated dynamic MOE. Our results showed that the NIR prediction model for the oven dry density was robust and applicable with a relatively high coefficient of determination when compared to green density. The offset method predicted a relatively high acoustic velocity when compared to the longitudinal and transverse methods.
From the operationality of acoustics and NIR at various points from the stump to the mill point of view, these results are promising, since they demonstrate that acoustics and NIR spectroscopy are efficient tools for the nondestructive measurement/prediction of various wood quality traits. Despite the challenges of determining the properties of interest for fitting the calibration regressions and the relatively time-consuming methods required in taking and collecting cores for the NIR models, when compared to the acoustics, it thus predicts values with detailed profiles which are closer to the acoustics. Therefore, such indirect methods should be considered in the prospects to optimize such technologies in the wood supply chain. The combined features of acoustic and NIR strategy could also be adopted to get a better result. This preliminary study with a smaller sample size is aimed at laying a foundation for further verification to assess the ability of acoustics and NIR in the assessment of damaged trees; more samples are needed in building a more robust model.
The three different patterns of acoustic measurement (a) longitudinal, (b) transverse, and (c) offset method.
Typical absorbance spectra of downed pine trees.
The transmittance of downed pine trees, classified by age: 15-yr-old trees, 30-yr-old trees, and 39-yr-old downed pine trees.
Second derivative of the transmittance of the downed pine trees.
Relationship between measured acoustic velocity and NIR-fitted velocity.
NIR predicted and estimated density of downed trees.
Relationship between measured and NIR predicted MOE.
Contributor Notes
The authors are, respectively, Assistant Professor, Dept. of Environ. Conservation, Univ. Massachusetts, Amherst; Research Scientist, Research Assistant, and Research Scientist, School of Forestry and Wildlife Sci., Auburn Univ., Auburn, Alabama, and Forest Products Development Center, Auburn Univ., Auburn, Alabama; Professor, School of Forestry and Wildlife Sci., Auburn Univ., Auburn, Alabama; Professor and Professor, School of Forestry and Wildlife Sci., Auburn Univ., Auburn, Alabama, and Forest Products Development Center, Auburn Univ., Auburn, Alabama; Technology Manager, U.S. Dept. of Energy, Bioenergy Technol. Office, Washington, DC; Research Forester, USDA Forest Serv., Southern Research Sta., Auburn, Alabama; and Professor, School of Forestry and Wildlife Sci., Auburn Univ., Auburn, Alabama, and Forest Products Development Center, Auburn Univ., Auburn, Alabama. This paper was received for publication in November 2022. Article no. 22-00067.