Wood Sci Technol (2005) 39: 618–629 DOI 10.1007/s00226-005-0035-8
ORIGINAL
Erlet Kurti Æ D.V. Heyd Æ R. Stephen Wylie
Raman microscopy for the quantitation of propiconazole in white spruce Received: 10 December 2004 / Published online: 1 October 2005 Springer-Verlag 2005
Abstract The Raman signature of propiconazole at 647–693 cm)1 was used to determine the propiconazole distribution in white spruce (Picea glauca). Samples treated with propiconazole were milled at 1.5 mm intervals and analysed by methanol extraction and GC-MS to obtain a depth profile in the longitudinal direction. The concentration of propiconazole in the milled wood layers ranged from 0.5 to 7.3 mg/g (dry mass wood). The average Raman signal from each of the layers was linear (R2=0.933) with the GC-MS-determined concentrations. The effective detection limit (concentration producing a signal three times the standard deviation) was 1.0 mg/g. The standard deviation of the method was approximately 0.3 mg/g.
Background Propiconazole is a fungistatic agent (Ciba-Geigy) that inhibits biosynthesis of ergosterol, an important constituent of fungal cell walls. The agent was first developed in 1979 by Janssen Pharmaceuticals of Belgium. Originally intended for pharmaceutical use, it is currently used for protecting crops and wood. The toxicity of propiconazole is low, making it a good environmental choice; however, determination of the depth of penetration is very time consuming. An in situ technique would require much less sample preparation and analysis time. The objective of this work was to study the applicability of Raman microscopy to the quantitative determination of propiconazole in wood. Raman spectroscopy has been used extensively to study natural components of wood (Agarwal 1999). Studies of lignin (Ibrahim and Oldham 1997; Takayama et al. 1997), carbohydrates (Ona et al. 1998, 1997; Agarwal and Ralph 1997) and pinosylvins (Holmgren et al. 1999) have been reported. Results of such studies have been combined with statistical methods to distinguish soft and E. Kurti Æ D.V. Heyd (&) Æ R. S. Wylie Department of Chemistry and Biology, Ryerson University, 350 Victoria St, M5B 2K3, Toronto, ON Canada E-mail:
[email protected] Tel.: +1-416-9795000 Fax: +1-416-9795044
619
hard woods (Lavine et al. 2001; Yang et al. 1999). Cell types (fibre, ray parenchyma, and axial parenchyma) and softwood tissues can also be distinguished based on their Raman spectra (Ona et al. 1999; Bond et al. 1999). Recently, resonance Raman has been used to study extractives in native woods (Nuopponen et al. 2004a, b). To our knowledge, there are no literature reports of the use of Raman to study propiconazole in wood; in fact, there are very few Raman microscopy studies of treated wood. There has been some interest in Raman to study wood bonded with resins, such as the diphenyl methane di-isocyanate (MDI) binders used in oriented strand board (for example, Yamauchi et al. 1999, 1997). The technique has also been employed to study the absorption of water, formamide and diiodomethane (Shen et al. 2001). The current work demonstrates that propiconazole is quantifiable by Raman microscopy. The Raman signature of propiconazole at 647–693 cm)1 region is free of interference from wood signals. The Raman wood signals about 985– 1175 cm)1 serve as normalisation standards to correct for surface effects. The technique provides chemical information rapidly on a micron scale without the need for grinding, sieving, and extraction. While the technique is not as accurate as gas-chromatography, it is advantageous in terms of speed and spatial resolution.
Experimental Sample preparation and extraction Samples of white spruce (Picea glauca) and propiconazole (50% in mineral spirits) were obtained from EverDry Forest Products Ltd. The samples were surveyed by Raman spectroscopy prior to analysis. Raman spectra indicated only small variations in the composition of the wood both within and between samples. The samples were all light in colour, and the Raman spectra had low fluorescence backgrounds. The air-dried spruce samples were cut into blocks with a band saw. The dimensions of the blocks used in this study are shown in Table 1. Each block was submerged in a solution of propiconazole in mineral spirits (CAS# 8052– 41-3) and soaked for 48 h. Soaking times greater than 48 h did not result in significantly different propiconazole concentrations in the treated wood. The concentrations of the soaking solutions used are included in Table 1. Solutions were prepared by diluting the 50% stock solution. (The concentration of the Table 1 Dimensions and soaking solution concentrations for spruce blocks analysed in this study Block
Standard A Standard B Standard C Validation D
Concentration of propiconazole in soaking solution (%) 4.0 2.0 1.0 3.0
Dimensions (mm) Tangential
Radial
Longitudinal
73 71 86 85
27 26 26 24
31 27 16 13
620
stock solution was determined by GC-MS as 50±1%.) The blocks were not evacuated and the ends of the blocks were not coated or treated in any way prior to soaking. The vessels containing the wood blocks and propiconazole were sealed, wrapped in aluminum foil, and stored in a closed cabinet in the dark during soaking. The samples were removed from the solution after 48 h and suspended from a wire with the longitudinal axis vertical. After drip-drying in that position for 15 min, the samples were placed on a radial face, and dried in fumehood at ambient temperature for 36 hours to remove all the mineral spirits (undetectable by GC-MS). A slice of each dried sample was removed for Raman analysis. The slice, taken from the central region, was the full radial thickness, the full longitudinal length, and 3 mm wide in the tangential direction. Samples for gas chromatography analysis were produced by milling layers approximately 1.5 mm thick (longitudinal direction) from the remainder of the sample. The ground wood from each layer was collected in a separate plastic bag. The ground wood was then screened at 40-mesh. The screened wood was weighed and extracted with 25 ml of HPLC-grade methanol for 4 h in a shaker bath. The extracts were filtered (0.45 lm, PTFE) and volume was transferred into a 25 ml flask. A standard amount of azaconazole (an internal standard) was added to each flask, and the volume was made up to the mark with methanol. GC-MS determination of propiconazole The methanol solutions obtained were analysed by GC-MS (Perkin-Elmer AutoSystem XL GC with TurboMass MS and electron impact ionisation) with a 95% methyl polysiloxane and 5% phenyl column. The injector was set at 250C, and the column was ramped from 25C/min to 350C. Chromatograms were first recorded under total ion current mode, scanning from 50 to 350 amu in order to determine the best masses for observing azaconazole and propiconazole. Two isomers of propiconazole were fully resolved, with retention times of 9.15 and 9.20 min, in a ratio of about 1:2. Both peaks were integrated to obtain the total propiconazole signal. The azaconazole peak was observed at 8.69 min. The analytical mass peaks for propiconazole and azaconazole were 69 and 217 amu, respectively. A background-free chromatogram was obtained by monitoring the signal at 217 amu, then switching to 69 amu at 9.00 min. Integration times were set to give at least ten points across each of the three peaks. Raman Spectra Raman measurements were performed on the slice of sample removed prior to grinding. Spectra were collected using a Renishaw System 2000. The instrument was equipped with a Pelletier-cooled CCD array that could capture a 600 cm)1 spectral window (spectrograph) with 2.5 cm)1 spectral resolution and 5 lm spatial resolution. Excitation was achieved with a 785-nm diode laser (25 mW at the head; 1 mW at the sample). Spectra were obtained from the summer wood only. They were collected from four points on a line across the longitudinal/radial face at a longitudinal position corresponding to the centre-line of each milled layer. The four measurements were then averaged. The spectral
621
window from 600 to 1,200 cm)1 (Raman shift) was integrated for 300 s with the cosmic ray elimination feature off.
Results and discussion Normalisation of Raman Spectra Figure 1 shows Raman spectra of spruce, spruce treated with propiconazole, and standard (99.7%) propiconazole. The propiconazole Raman signal at 647– 693 cm)1 was chosen as the analytical peak for its intensity and because it did not overlap the peaks of the wood spectrum. The absolute Raman signal of a solid sample with a given composition varies with surface characteristics such as roughness, reflectivity, and the direction of the surface normal. For samples such as the wood surfaces used in this study, quantitative determinations require a normalisation procedure. In this work, the analytical peak area was normalised by dividing by the area of the wood signals between 985 and 1175 cm-1. In that way, the wood served as an internal standard for the propiconazole. Spectral features in the region 985–1175 cm)1 are difficult to assign to particular resonances. They are primarily due to C-C and C-O stretching vibrations (Agarwal et al. 1997; Agarwal 1999), which are prevalent in cellulose and hemicellulose. Some variation in the intensity ratios of those signals across a sample would be expected; however, the variation in the overall relative area was low, and so the region was an effective internal standard. Because the wood peaks in the range 985–1175 cm)1 overlapped with some propiconazole peaks, the wood signal had to be corrected before it could be used for normalisation. This was done by subtracting an estimate of the area due to propiconazole signal. The normalised signals were calculated as:
Fig. 1 Spectra of native spruce, spruce treated with 3% propiconazole solution, and standard propiconazole (99.7%). The areas shown (A1 through A4) were used to calculate the normalised Raman signal for propiconazole in treated wood, as described in the text
622
A3 Normalised Intensity ¼ ¼ A4 AA21 A3
A4 A2 1 A3 A1
where A1 is the area under the 647–693 cm)1 region in the standard propiconazole spectrum, A2 is the area under the region 985–1175 cm)1 in the standard propiconazole spectrum, A3 is the area under the analytical region (647–693 cm)1) in the spectrum of the treated wood, and A4 is the area under the region 985–1175 cm)1 in the spectrum of treated wood (see Fig. 1). The subtraction term (A2A3/A1) in the denominator of the normalised intensity expression should be regarded only as an approximation of the contribution of propiconazole in the spectral region 985–1175 cm)1. Implicit in the normalisation procedure is the assumption that the Raman spectrum of the propiconazole is unaffected by uptake into wood. Specifically, the ratio of the peak areas in the propiconazole spectrum under the analytical region (647– 693 cm)1) and the normalisation region (985–1175 cm)1) must be the same for both the propiconazole standard and the propiconazole in treated wood for the approximation to be valid. Because propiconazole is a minor component of the treated wood, it is very difficult to demonstrate that the assumption is true. In principle, the underlying Raman spectrum of propiconazole in the spectrum of the treated wood could be revealed by subtracting a ‘‘wood’’ background; however, the procedure does not work well in practice. Wood is not a single component, but a mixture of many components (cellulose, lignin, extractives, etc.), and the composition varies both within and between samples. Any ‘‘standard’’ wood sample will not have a spectrum identical to the treated wood under investigation because of those small compositional differences. Therefore, subtracting a standard wood spectrum from the treated wood spectrum results in a residual with both propiconazole and wood components. The ‘‘wood’’ residuals are small, but large enough to interfere with the (revealed) propiconazole spectrum. In addition, the propiconazole spectra obtained from subtraction are very noisy because the propiconazole spectral component of the treated wood spectra has lower intensity than that of the wood. The uncertainty in the difference spectrum resulting from the wood residuals and propagation of uncertainty during subtraction severely limits the reliability of peak area ratios in the underlying propiconazole spectrum. While the assumption could not be confirmed directly, the validity of the procedure is corroborated by the fact that it results in a linear relationship between the Raman signal for propiconazole and the concentration of propiconazole in the wood (see Fig. 4 and discussion of the calibration and validation of the method). If the subtracted term overestimated the contribution of propiconazole to the normalisation region, positive curvature in the standard curve would be expected. Likewise, underestimating the contribution of propiconazole would result in negative curvature, as observed if no correction is made. Using the simple procedure described above, a standard curve with negligible curvature was obtained. Fluorescence is arguably the biggest problem for Raman studies of native wood. A broad fluorescence signal overlaps the Raman peaks, and results in a sloping background, as can be seen in Fig. 1. When fluorescence is strong, shot noise can obscure the Raman peaks. The 785-nm laser used in these experiments was chosen because of its low efficiency for electronic excitation. White
623
spruce tends to fluoresce less than most other wood species we have studied. While the current work was restricted to white spruce, it is likely that Raman could be useful for analysing propiconazole in more fluorescent wood species if an even longer wavelength (e.g. 1064 nm) were used. Distribution of propiconazole in springwood and summerwood The spatial resolution of Raman microscopy is very suitable for studies of wood because the sizes of the structures in wood (e.g. pits, cells, and even cell walls) are larger than the resolving power of the microscope. The power of this microscopic technique, as well as one of its potential pitfalls, is apparent in Fig. 2, which shows dramatic differences between springwood and summerwood in the same ring of a sample. Repeated analysis has shown that raw propiconazole signals (i.e. the intensities prior to normalisation) tend to be greater in summerwood. The explanation appears to be that the density of summerwood is much greater than that of springwood. The sampling volume of the Raman microscope contains a greater mass of wood when sampling summerwood; therefore, the wood signal
Fig. 2 Pre-normalised Raman line map across spring and summer wood. Error bars are ±1 SD
624
is also larger when sampling summerwood. The higher density means that summerwood has more adsorption sites for propiconazole per unit volume, resulting in a larger propiconazole signal. Because both the wood and propiconazole signals were larger for summerwood, normalisation (taking the ratio of the propiconazole signal to the wood signal) produced Raman ratios that were indistiguishable for spring and summerwood. However, due to lower raw counts in the springwood spectra, the standard deviations of the springwood Raman ratios were higher. This highlights the need for normalisation even if the surface properties of the wood can be controlled reasonably well. Because the raw (unnormalised) signals from springwood were lower than those from summerwood, all Raman spectra in this study were sampled from summerwood. The practice did not bias the results because the normalised signals were the same for spring and summerwood. The higher raw (unnormalised) signals for summerwood allowed for a reduced sampling time and better signal-to-noise ratios. Depth profiles of propiconazole by GC-MS and Raman Figure 3 shows depth profiles of propiconazole along the longitudinal direction of three wood samples, as determined by both GC-MS and Raman microscopy. The GC-MS results are represented as a step-plot (histogram) because each value is the average concentration of propiconazole in an extracted layer. The Raman data are plotted as points, each representing the average of four Raman spectra measured in a line through the middle of the layer. Error bars represent the standard deviations of the means. Each sample was a different length, as can be seen from the abscissas of the plots and Table 1. The first and last layers of each sample do not appear in Fig. 3. The last layer could not be analysed by GC-MS because the milling machine used to grind the wood could not be operated safely on such a thin piece. The concentration in the first layer was much higher than in the subsequent layers. Light microscopy and Raman microscopy indicated that most of the propiconazole in that layer was located in a waxy deposit on top of the surface. The deposit might have formed while the sample was drip-drying; the transverse faces were oriented on the top and bottom of the sample during the 15min drip-drying stage. (During the 36-h air-drying stage, the sample was oriented with the longitudinal axis horizontal, so that the two radial faces were exposed.) The propiconazole that was deposited as a waxy film on the first layer was not associated with wood fibre, as it was in subsequent layers. Therefore, the normalised Raman intensity (the ratio of the Raman signals for propiconazole and wood) of the first layer did not follow the same concentration relationship as the second, third, and deeper layers, for which there was no possibility of a waxy layer forming. For that reason, the first layer of each sample was eliminated from Fig. 3. The depth profiles obtained by GC-MS and Raman are qualitatively similar, although the Raman data are more scattered than those of the GC-MS curves. The standard deviations for the GC-MS curves were small (relative standard deviations were 1%) and were omitted from the figure to improve readability. The depth profiles of the three samples all show a characteristic ‘‘bowl’’ shape. The concentration decreased from the edges to a flat minimum in the middle of the sample. The shape is consistent with a diffusion mechanism.
625
Fig. 3 Depth profiles of propiconazole in the longitudinal direction by Raman and GC-MS. Samples were soaked in a 4% b 2% c 1% propiconazole in mineral spirits. Surface layers are excluded, as described in the text. Raman values are mean of four measurements, and error bars represent ±1 SD of the mean
The apparently asymmetrical distribution (higher on the left-hand side) probably arose from the way the samples were drip-dried after soaking; however, because the last layers were not analysed, the distribution may be more symmetrical than the data show. In any case, the symmetry of the distribution does not affect the results of this study; the concentration of each layer was determined by GC-MS independently. There are two reasons for the larger standard deviations of the Raman profiles compared to the GC-MS, and both arise from the small sampling volume of the Raman microscope. First, the volume sampled by the Raman
626
microscope is on the order of the cube of the focus of the laser, which was about 5 lm at the waist in these experiments. The GC data, on the other hand, was derived from sample approximately 1010 times larger by volume, and so should be expected to produce better signal-to-noise. Second, and perhaps more importantly, the distribution of propiconazole was inhomogeneous. The scatter in the data, primarily due to sampling error, could be reduced by collecting more spectra or by using a sampling technique with a larger spot size (a fibreoptic probe, for example). On the other hand, the micro-distribution of propiconazole (and other preservatives) in wood is important information for understanding wood treatments. Raman microscopy could be an important tool for the direct determination of the localisation of preservatives in the microstructure of wood. The ultimate resolution of Raman microscopy is about 1–2 lm, smaller than the width of a single cell wall. Calibration and validation of the Raman method In Fig. 4 the averaged, normalised Raman signals from Fig. 3 are plotted against the concentrations determined by GC-MS. The fit to the data is linear, with R2 = 0.933. The dashed lines indicate the error on the determination of concentration (± 1 S.D.), from the average of four measurements of the normalised Raman intensity. The dashed lines show that the standard deviation of the technique is approximately 0.3 mg/g at the mid-range concentrations. The standard deviation of the intercept (sb) was calculated as 6.2·10)4 (unitless because the analytical signal is the Raman ratio) and the standard deviation of the regression (sr) was 2.4·10)3. The limit of detection (LOD) was 1.0 mg/g, using the commonly accepted definition, LOD = 3sr/m, where m is the slope of the line of best fit. Points above 7.3 mg/g were omitted from Fig. 4. The inclusion of the two points at concentrations higher than 7.3 mg/g resulted in a regression line that did not follow the low-concentration values as well; the intercept was higher and the slope lower. The change in the slope and intercept due to inclusion of
Fig. 4 Calibration curve for Raman determination of propiconazole in spruce. Plot was constructed from same data (45 points) as Fig. 3. Dashed lines represent ±1 SD of the determination (x-direction)
627
Fig. 5 Validation of Raman method with spruce soaked in 3% propiconazole (block D in Table 1). Surface layers excluded. Error bars represent ±1 SD on the Raman-determined values
higher concentrations might indicate the presence of curvature in the calibration curve at higher concentrations; however, the effect was more likely due to random scatter. There were only two points above 7.3 mg/g, but their positions, so far from the origin, resulted in a disproportionate influence on the calculation of the fitting parameters. The focus of this study was on lower concentrations, as they are more relevant to the wood preservation industry. Because of that the calibration curve was limited to concentrations <7.3 mg/g, where a linear relationship was observed. The low R2 value and high detection limit put the method in the ‘‘semiquantitative’’ range; however, there is good reason to believe that the method can be improved. As was previously mentioned, the scatter should be reduced by increasing the number of Raman spectra per layer (for this study four spectra were averaged per calibration point), or by using a probe with a wider spot size. Higher raw signals could be obtained by using a more powerful laser, thus reducing the influence of shot noise. Ordinary Raman signals increase linearly with laser power. In this study, the laser power at the sample was approximately 1 mW, which is modest. In Fig. 5, the method was validated by analysing a sample soaked in 3% propiconazole. The sample was prepared for GC-MS and Raman analyses as previously described. Eight layers were obtained from the sample. Normalised Raman intensities, averages of four measurements per layer, were used in conjunction with the calibration curve in Fig. 4 to predict the propiconazole concentration (points in Fig. 5). These values are compared in Fig. 5 to the GCMS results on the same layers (histogram bars). The error bars on the points represent ±1 SD of the determination. Raman determinations agreed with GC-MS results within 1 SD in six of the eight layers. That accords with the prediction of Gaussian statistics (±1 SD represents a 68% confidence interval). Percent deviations from the GC-MS values ranged from )12% (layer 6; 1.3 mg/g by GC-MS) to +38% (layer 2; 3.0 mg/g by GC-MS). The absolute standard deviations of the Raman predictions were approximately constant over the range of concentrations represented in Fig. 5.
628
The coefficients of variation (standard deviation as a percentage of the average value) of the Raman-predicted results varied from 5% (layer 1; 6.9 mg/g by GC-MS) to 31% (layer 6; 1.3 mg/g by GC-MS).
Conclusions Raman spectroscopy shows potential as a means of analysis of propiconazole in wood. The results of the current work lack the precision of GC-MS; however, the advantage of speed makes the technique promising, at least for semiquantitative analysis. The inhomogeneous distribution of propiconazole in the sample leads to high scatter in the Raman data. The precision of the technique should be improved by using a probe, such as a fibre-optic, that encompasses a larger area. A detection limit of 1.0 mg/g and an uncertainty of 0.3 mg/g are readily attainable using the technique described, even with a low laser power and only four spectra averaged. Higher laser power and more signal averaging should improve that. Weaker Raman signals were obtained from springwood than from summerwood, so only summerwood was used in this study; however, the Raman ratios were similar for both spring and summerwood, so the use of springwood (with longer signal averaging) is not precluded. A major drawback of Raman is that the spectra of some species of wood may be obscured by fluorescence. That problem could be overcome by using a longer wavelength laser (e.g. 1064 nm). Acknowledgments The authors would like to thank Materials and Manufacturing Ontario (MMO), the Natural Sciences and Engineering Research Council (NSERC), and Janssen Pharmaceuticals for funding this project. We are also indebted to Ron Bobker at EverDry Forrest Products for his expert advice and materials (spruce and propiconazole solutions), to Paul Cooper and Tony Ung (University of Toronto) for helpful discussions and use of equipment, and to Janssen Pharmaceuticals for supplying the standard propiconazole.
References Agarwal UP (1999) An overview of Raman spectroscopy as applied to lignocellulosic materials. In: Argyropoulos DS (ed) Advances in lignocelluolosics characterisation, U.S. Dept. of Agriculture, pp 201–225 Agarwal UP, Ralph SA (1997) FT-Raman spectroscopy of wood: identifying contributions of lignin and carbohydrate polymers in the spectrum of black spruce (Picea mariana). Appl Spectrosc 51(11):1648–1655 Bond JS, Atalla RH (1999) A Raman microprobe investigation of the molecular architecture of loblolly pine tracheids. In: Proceedings of the 10th international symposium on wood and pulping chemistry, main symposium, June 7–10, 1999, Yokohama, Japan. TAPPI Press, Atlanta vol 1, pp 96–101 Ciba-Geigy Technical Information Bulletin for Propiconazole Fungicide, Greensboro, NC.15pp Holmgren A, Bergstrom B, Gref R, Ericsson A (1999) Detection of pinosylvins in solid wood of Scotch pine using Fourier-transform Raman and infrared spectroscopy. J Wood Chem Technol 19(1–2):139–150 Ibrahim A, Oldham PB, Conners TE, Schultz TP (1997) Rapid characterization of wood pulp lignin by Fourier transform Raman spectroscopy. Michrochem J 56:393–402 Lavine BK, Davidson CE, Moores AJ, Griffiths RP (2001) Raman spectroscopy and genetic algorithms for the classification of wood types. Appl Spectrosc 55(8):960–966 Nuopponen M, Willfor S, Jaaskelainen A-S, Sundberg A, Vuorinen T (2004a) A UV resonance Raman (UVRR) spectroscopic study on the extractable compounds of Scots pine (Pinus
629 sylvestris) wood; Part I: lipophilic compounds. Spectrochimica Acta. Mol Biomol Spectrosc 60(13):2953–2961 Nuopponen M, Willfor S, Jaaskelainen A-S, Vuorinen T (2004B) A UV resonance Raman (UVRR) spectroscopic study on the extractable compounds in Scots pine (Pinus sylvestris) wood; Part II. Hydrophilic compounds. Spectrochimica Acta. Mol Biomol Spectrosc 60(13):2963–2968 Ona T, Sonoda T, Ito K, Shibata M, Kato T, Ootake Y (1998) Non-destructive determination of lignin syringyl/guaiacyl monomeric composition in native wood by Fourier transform Raman spectroscopy. J Wood Chem Technol 18(1):27–41 Ona T, Sonoda T, Ito K, Shibata M, Kato T, Ootake Y (1997) Non-destructive determination of wood constituents by Fourier transform Raman spectroscopy. J Wood Chem Technol 17(4):399–417 Ona T, Sonoda T, Ito K, Shibata M, Ootake Y, Ohshima J, Yokota S, Yoshizawa N (1999) In situ determination of proportion of cell types in wood by Fourier transform Raman spectroscopy. Analyt Biochem 268(1):43–48 Shen, Sr Q, Ciovica L, Rosenholm JB (2001) Quantitative comparison of the capillary and non-capillary absorption ability of wood by FT-Raman spectroscopy and spontaneous liquid penetration. Abstr Pap Am Chem Soc 221st Nat. Mtg. CELL)054 Takayama M, Johjima T, Yamanaka T, Wariishi H, Tanaka H (1997) Fourier-transform Raman assignment of guaiacyl and syringyl marker bands for lignin determination. Spectrochimica Acta Mol Biomol Spectrosc 53A(10):1621–1628 Yang H, Lewis IR, Griffiths PR (1999) Raman spectrometry and neural networks for the classification of wood types. 2. Kohonen self-organizing maps. Spectrochimica Acta Mol Biomol Spectrosc 55A(14):2783–2791 Yamauchi S, Tamura Y, Kurimoto Y, Koizumi A (1999) Vibrational spectroscopic studies on wood and wood-based materials (IV) solvent effects of diluent on MDI penetration behavior in wood. Nippon Setchaku Gakkaishi 35(7):300–308 Yamauchi S, Tamura Y, Kurimoto Y, Koizumi A (1997) Vibrational spectroscopic studies on wood and wood-based materials. III. Identification of adhesives in wood by using FTRaman spectroscopy as a nondestructive analytical method. Nippon Setchaku Gakkaishi 33(10):380–388