Journal of Molecular Structure 651–653 (2003) 397–404 www.elsevier.com/locate/molstruc Measurement of FT-Raman spectra of Norway spruce needles in stepwise rotating cylindrical cell T. Pekareka,*, P. Matejkaa, F. Skacelb, K. Volkaa a b Department of Analytical Chemistry, Institute of Chemical Technology, Technicka 5, Prague 6, CZ 166 28, Czech Republic Department of Gas, Coal and Air Protection, Institute of Chemical Technology, Technicka 5, Prague 6, CZ 166 28, Czech Republic Received 2 September 2002; accepted 16 September 2002 Abstract FT-Raman spectroscopy of Norway spruce needles was found to be one of the effective methods for environmental characterisation of the forest areas. The idea of this study was to profit from good penetration power of excitation near-infrared radiation and to measure a bigger bundle of needles at once. A glass cylindrical cell and a holder, that allows a stepwise rotation of this cell, were developed. All data were evaluated using cluster analysis and principal component analysis. Firstly, it was proven that repetitive cell filling does not affect the spectra. Secondly, no evident differences among average spectra were observed for different turning steps. Thirdly, comparison of spectra of individual needles and average spectra obtained in the experiments with different turning steps showed statistically significant differences. Nevertheless, these differences are less important than differences among the spectra of needles sampled in different forest areas. Individual forest areas can be distinguished both using cylindrical cell and measuring individual needles. Hence, for characterisation of one tree a 6-h measurement of individual needles can be replaced by the 1-h accumulation in cylindrical cell. q 2003 Elsevier Science B.V. All rights reserved. Keywords: FT-Raman spectroscopy; Rotating cell; Picea abies; Cluster analysis; Principal component analysis 1. Introduction Non-destructive analysis of Norway spruce needles (Picea abies (L.) Karst.) by FT-Raman spectroscopy [1 –3] and ATR technique in mid-infrared range [4] has been already reported. Living cells are possible to measure by FT-Raman spectroscopy with near-infrared (NIR) laser source with only small contribution (or without at all) of fluorescence and * Corresponding author. Tel.: þ420-2-2435-4091; fax: þ 420-22431-0352. E-mail address: tomas.pekarek@vscht.cz (T. Pekarek). with only small risk of damage of living cells [5]. Thus, FT-Raman spectroscopy of Norway spruce needles was found to be a suitable method for environmental characterisation of forest areas [1 – 3]. Nevertheless, analysis of tens of needles from individual areas is needed for such characterisation and consecutive measurement of corresponding amount of needles is very time consuming [1,2]. Therefore, the aim of this study was to develop a new timesaving method, when more than one needle is measured at once, taking into account the high penetration depth of NIR excitation. A bundle of needles was placed in a cylindrical cell that was fixed 0022-2860/03/$ - see front matter q 2003 Elsevier Science B.V. All rights reserved. PII: S 0 0 2 2 - 2 8 6 0 ( 0 2 ) 0 0 6 5 8 - 0 398 T. Pekarek et al. / Journal of Molecular Structure 651–653 (2003) 397–404 in the spectrometer using a stepwise rotating holder. The spectra obtained were evaluated using cluster analysis (CA) and principal component analysis (PCA). Firstly, the effect of multiple fillings of the cell on repeatability of the Raman spectra was examined. Secondly, the differences among the average spectra obtained with shorter and longer turning steps were evaluated. Thirdly, the results of the new timesaving method were compared with results obtained in the experiments based on consecutive measurement of individual needles. Finally, eventual differences among spectra of needles from different areas were analysed. 2. Experimental 2.1. Preparation and treatment of needles for analysis Two-years-old needles of Norway spruce from two areas (Uherske Hradiste (UH)—tree no. 651, and Zdar nad Sazavou (ZR)—tree no. 605) were examined. Needles were carefully torn off branches using tweezers. Needles were immediately packed in marked Al-foils. Prepared packets were put into poly(ethylene) (PE) bags and stored in a freezer (ca. 2 8 8C). Individual needles were placed in a previously developed sample holder [1]. A bundle of needles was arranged using tweezers in a newly developed cylindrical cell with stepwise rotating holder. 2.2. Cylindrical cell and rotating holder Appropriate cell for analysis of a bundle of needles was found to be a hollow cylinder (height ca. 3.0 cm) made of clear optical glass with a poly(tetrafluoroethylene) (PTFE) round basis (diameter ca. 2.5 cm) (Fig. 1). The needles were arranged vertically. Clockwise/anticlockwise rotation of the cell placed into stepwise turning holder was controlled either manually or automatically using control unit with digital indication of cell position. A full turn of 3608 was represented by 200 steps. 2.3. Instrumentation a FT-Raman spectra were collected using Fourier transform near-infrared (FT-NIR) Fig. 1. Cylindrical cell used for measurement of a bundle of needles. spectrometer Equinox 55/S with FT-Raman module FRA 106/S (Bruker). The samples were irradiated by the focused laser beam with a laser power 50 mW of Nd:YAG laser (1064 nm, Coherent). The scattered light was collected in backscattering geometry. Quartz beamsplitter and Ge detector (liquid N2 cooled) were used to obtain inteferograms. The number of scans was adapted to the type of experiment. The standard 4 cm21 spectral resolution, ‘zero filling’ 8 and Blackmann– Harris cosine apodisation function were used for all measurements. 2.4. Scheme of experiment In the case of consecutive measurement of needles [1 –3] 1024 scans were used to obtain a spectrum. In preliminary experiments, where the same cylindrical cell (Fig. 1) was filled with a bundle of needles; 1024 scans were accumulated within continuous rotation of the cell with a selected speed. To allow comparison of results of all types of experiments, 1024 scans were accumulated for one filling of the cylindrical cell regardless of the stepwise moving mode. Three different modes of stepwise rotation were tested (Table 1). Firstly, the full turn was divided into 16 segments (mode A). The cell was turned alternately by 22 and/or 238, i.e. 12 and/or 13 steps. One spectrum was measured in each of 16 positions. To obtain 1024 scans per one filling of the cell, 64 scans were accumulated for a particular spectrum. Secondly, the eight positions were examined (mode B); i.e. 128 scans per a spectrum in a particular position were accumulated. After each of eight spectral accumulations the cell was turned by 458, i.e. 25 steps. Finally, the number of segments was reduced to 4 (mode C) that means 256 scans per position were 399 T. Pekarek et al. / Journal of Molecular Structure 651–653 (2003) 397–404 Table 1 Modes of spectral accumulations within stepwise rotation of the cell Mode Number of positions per full turn Angle between two consecutive measurements (8) Steps between two consecutive measurements Number of scans per spectrum measured in one position A B C 16 8 4 22/23 45 90 12/13 25 50 64 128 256 accumulated. The angle between two consecutive measurements was 908 (Table 1). To allow an evaluation of the effect of repetitive cell filing on the Raman spectra obtained, three consecutive fillings were realised for all modes of rotation. 2.5. Treatment and evaluation of spectra All treatments of spectra were undertaken using software OPUS 2.0 (Bruker). Measured spectra were primarily averaged for each of cell fillings of a particular rotation mode. Finally, also averages of all spectra obtained by one mode of rotation and averages of all spectra of one tree obtained either by measurement in cylindrical cell or by consecutive analysis of individual needles were calculated. Every measured and averaged spectrum was cut to the range 3600 –400 cm21 prior chemometric evaluation. Averaged spectra representing individual cell fillings of a particular rotation mode were remitted to CA alone and also together with consecutively measured spectra of individual needles. All cut spectra were separately treated by following operations: (1) correction of baseline, (2) vector normalisation in full limits (3600 – 400 cm21), (3) vector normalisation in region 1625 – 1575 cm21, (4) correction of baseline and vector normalisation in full limits (3600 – 400 cm21), and (5) correction of baseline and vector normalisation in region 1625– 1575 cm21. Finally, all types of cut spectra were exported to JCAMP-DX format. The sets of spectra for a particular evaluation using PCA were submitted to the software The Unscrambler 7.6 (CAMO) to create an appropriate matrix data sheet. The examined category variables (e.g. forest area, tree, and rotation mode) were inserted before running PCA. 3. Results and discussion Preliminary studies with continually rotating cell demonstrated spectral deformation caused by a movement of the cell during individual scans. Thus, the stepwise moving holder was developed to ensure fixed position of the cell in the time of data accumulation. The aspects of number of angular positions of the cell, repetitive filling were studied together with a comparison of the data obtained with the results of conventional analysis of individual needles. The CA and then the PCA were used for such evaluation. Nevertheless, before the chemometric evaluation the measured spectra were visually compared. Spectra of bundles of needles and of appropriate individual needles exhibited usually analogous shape of the baseline, the bands were at the same positions, but their intensity was sometimes apparently different (Fig. 2). 3.1. Evaluation of spectra using cluster analysis The averaged spectra of individual fillings obtained in various rotation modes (Table 1) were evaluated using CA (Ward’s algorithm). Firstly, the effect of repetitive cell filling on the spectra was analysed. Secondly, various rotation modes were mutually compared. Thirdly, the effect of different areas (trees) was examined. Finally, the spectra obtained in cylindrical cell were compared with spectra of individual needles. The main two classes in all results of CA represent the areas (trees) (Table 2). It was proven that both repetitive cell fillings and various rotation modes do not cause any significant clustering (Table 2). While the distance of two main classes is ca. 3.3, the distances of data of various rotation modes and fillings 400 T. Pekarek et al. / Journal of Molecular Structure 651–653 (2003) 397–404 Fig. 2. Example of FT-Raman spectrum of a bundle of Norway spruce needles (A) compared with the spectrum of an individual needle (B). Forest area UH, tree no. 651, laser power 50 mW, focused laser beam. are less than ca. 0.22. The spectra obtained in cylindrical cell were compared with spectra of individual needles. Separate clusters of data from cylindrical cell and from individual measurements Table 2 CA of spectra of bundles of needles measured in various rotation modes Area Tree Filling Mode of rotation 1. Class has nine members: last fusion occurred at 0.125; next nearest class is 2 at 3.294 UH 651 1 A UH 651 3 C UH 651 1 C UH 651 3 A UH 651 2 A UH 651 1 B UH 651 3 B UH 651 2 C UH 651 2 B 2. Class has nine members: last fusion occurred at 0.217; next nearest class is 1 at 3.294 ZR 605 1 A ZR 605 2 A ZR 605 2 B ZR 605 2 C ZR 605 3 A ZR 605 1 B ZR 605 3 C ZR 605 3 B ZR 605 1 C were formed for each of the forest areas. The distances between clusters of data from individual measurements and from cylindrical cell are about 2.5 –3.5 times smaller than the distances between the classes of spectra of needles sampled in different forest areas. But they are at the same time about 10 times bigger than those distances of data of repetitive cell fillings and various rotation modes (Table 3). That means, repetitive cell filings and various rotation modes do not cause any significant differences among spectra obtained, while statistically significant differences among spectra of bundles of needles and data of individual needles are suggested. Nevertheless, the tree (area) of origin of the needles causes the main effect on spectral differences for both types of measurements. Analogous results were obtained when the CA was carried out for spectra with correction of the baseline, and also for normalised both in the full cut range and in the region 1625 – 1575 cm21. The main two classes represented in all Table 3 Ranges of distances of data in results obtained by CA Distances among spectra of repetitive cell filling Distances among spectra of individually measured needles and spectra of bundles of needles Distances among spectra of needles from different forest areas 0.1–0.35 1.0–1.5 3.0–3.5 T. Pekarek et al. / Journal of Molecular Structure 651–653 (2003) 397–404 cases the two different trees (areas) examined, the statistical difference between measurement of individual needles and measurement of bundles of needles were proven and also no significant effect of repetitive cell filing and mode of rotation was confirmed. The ranges of the most important type of data distances are summarised in Table 3. 3.2. Principal component analysis of spectra of needles Both spectra obtained from bundles of needles and spectra of individual needles were evaluated using PCA. The effects of repetitive cell fillings, various rotation modes and origin of needles together with the comparison between measurements of bundles of needles and measurements of individual needles were examined for both untreated spectra and spectra modified by various procedures (Section 2.5). A graph ‘scores’ as a PCA result of evaluation of original cut spectra (Fig. 3) shows the distribution of data along 1st (PC1) and 2nd principal component (PC2). The spectra obtained for the tree 651 are arranged in two well-distinguished and quite compact clusters; the first one located in the range of negative values of both the PC1 and PC2 belongs to spectra measured for individual needles, and the second one spread around zero value of PC1 and in the range of positive values of PC2 is assigned to measurements of bundles of needles. The data of the tree 605 obtained in cylindrical cell forms a third apparent cluster 401 around the zero values of both PC1 and PC2. There is no mutual overlap of these three clusters. The data measured for individual needles of the tree 605 are located in quite wide range along PC1, but in rather narrow range of negative values of PC2. They partially overlap only the cluster of data of the same tree measured in cylindrical cell. That means, that the data of the two different trees are separated. It should be noted, that all data obtained for individual needles are characterised by negative value of PC2, while the data measured in cylindrical cell are located in quite narrow range of values of PC1 (around the zero value) along the PC2 axes. The graph of ‘loadings’ (Fig. 4) is used to specify spectral regions that mostly contribute to the differentiation of data. The general shape of the ‘xloadings’ curve (Fig. 4) is analogous to measured FT-Raman spectra (Fig. 2); the peaks in the ranges around 3000 and 1700– 900 cm21 contribute to the distinguishing of spectra. Nevertheless, the relative intensities of some peaks are very different, especially the quite intense spectral bands at ca. 1525 and 1150 cm21 (Fig. 2) attributed to carotenoids [3] are not pronounced on the x-loadings curve (Fig. 4). Such observation suggests that in this case the carotenoids are not very important in differentiation of data. After the PCA of untreated spectra, the same analysis of baseline-corrected data was proceeded. The range of values of both PC1 and PC2 in the scores graph is smaller in this case (Fig. 5) compared to analogous graph of untreated data (Fig. 3). The data of Fig. 3. Evaluation of untreated FT-Raman spectra of needles using PCA. 605j—spectra of individual needles, tree no. 605; 605v—spectra of bundles of needles, tree no. 605; 651j—spectra of individual needles, tree no. 651; 651v—spectra of bundles of needles, tree no. 651. 402 T. Pekarek et al. / Journal of Molecular Structure 651–653 (2003) 397–404 Fig. 4. Graph loadings for untreated FT-Raman spectra of needles. the two trees are distinguishable, but they are worse separated than in the case of untreated spectra. Nevertheless, many analogies can be found between the results for untreated and baseline-corrected data. The baseline-corrected spectra obtained for the tree 651 are arranged in two well distinguished and quite compact clusters (Fig. 5) like the untreated spectra (Fig. 3). The biggest variance of data is exhibited by the spectra measured for individual needles of the tree 605. The data of the tree 605 obtained in cylindrical cell are partially overlapped by the data of individual needles of the same tree. Spectral regions around 2935 and 1650 – 1120 cm 21 mostly contribute to the distribution of baseline-corrected spectra. These regions are narrower compared to ranges contributing to differentiation of untreated spectra. PCA results of baseline-corrected spectra normalised in region of one of the most intense bands (1625 – 1575 cm21) show higher variance of data for both PC1 and PC2 in comparison with results for only baseline-corrected spectra. The clustering of data based on trees and on methods of measurements of needles is analogous to the previous cases (Figs. 3 and 5), but the data of the tree 605 obtained for bundles of needles and for individual needles are separated in this case. Certain values of PCs can be associated to Fig. 5. PCA results for baseline-corrected FT-Raman spectra of needles. 605j—tree 605, individual needles; 605v—tree 605, bundles of needles; 651j—tree 651, individual needles; 651v—tree 651, bundles of needles. T. Pekarek et al. / Journal of Molecular Structure 651–653 (2003) 397–404 the clusters, e.g. spectra of needles from the tree 605 measured in cylindrical cell are characterised by negative values of PC1 and positive values of PC2. In the case of baseline-corrected spectra normalised in the range 1625 –1575 cm21, the bands of carotenoids (around 1530 and 1160 cm21) contribute apparently to differentiation of spectra. PCA results of spectra normalised in spectral area 1625 – 1575 cm21 (without any correction of the baseline) exhibit higher variance of both PC1 and PC2 than data in all other cases, but the data of both different trees and methods of measurements of spectra are worse distinguished. Only the cluster of spectra of bundles of needles from the tree 651 is well separated. Other clusters partially overlap each other. PCA results of vector normalised spectra in full limits (3600 – 400 cm21) show apparent clustering in according to both the tree of origin of needles (651 or 605) and the type of measurement (individual needles or bundles of needles). Certain ranges of values of both PC1 and PC2 can be associated to individual clusters analogously to previous cases, e.g. the cluster of data obtained for individual needles from the tree 605 has only negative values of PC2; the cluster of spectra of bundles of needles from the tree 651 is characterised by positive values of both PC1 and PC2. The baseline-corrected spectra after normalisation in full limits (3600 – 400 cm21) were also analysed by PCA. The graph scores is given in Fig. 6. The data of 403 the bundles of needles from the tree 651 form very compact cluster in the range of negative values of PC1 and positive values of PC2. These data are well separated from data of the same tree obtained for individual needles that are characterised by positive values of PC1 and with one exception also by positive values of PC2. There is no overlap of the data of tree 651 with data of the tree 605. The data of the tree 605 for both individual needles and bundles of needles are overlapping and they are located relatively close to the zero value of PC1. All the data of individual needles of the tree 605 are characterised by negative values of PC2, while the data of bundles of needles are placed around zero value of PC2 both in positive and negative range. Thus, the basic scheme of clustering is analogous with most of the previous cases (Figs. 3, 5, and 6), i.e. the data of the two trees are distinguished and the data of the tree 651 form two quite compact well-separated clusters with respect to the methods of measurement of needles, while the data of the tree 605 obtained by the two methods are mutually overlapped. The effects of spectral treatment on the results of PCA are usually quite minor, although some specific effects for individual methods of treatment can be demonstrated. For example, the enormous separation of clusters along PC1 is observed for the data of bundles of needles (tree 651) and data of individual needles (tree 651) in the case of baseline-corrected spectra normalised in full range (Fig. 6). Fig. 6. PCA results for baseline-corrected and in full limits vector-normalised FT-Raman spectra of needles. 605j—tree 605, individual needles; 605v—tree 605, bundles of needles; 651j—tree 651, individual needles; 651v—tree 651, bundles of needles. 404 T. Pekarek et al. / Journal of Molecular Structure 651–653 (2003) 397–404 4. Conclusions Concluding, the newly developed technique of measurement of FT-Raman spectra for bundles of needles in a cylindrical cell enables to distinguish the forest areas (trees), which the needles were taken from. Any type of chemometric evaluation of both original and treated spectra does not show any significant effect of repetitive cell filing, thus the repeatability of measured spectra for a particular type of needles was proven. Even the mode of stepwise rotation of the cell does not significantly affect the data evaluated, when the total number of scans per averaged spectrum of one cell filling is preserved. The FT-Raman spectra obtained for a bundle of needles are rather different compared with spectra of individual needles taken from this bundle. This effect can be explained by distinctions of optical arrangement in the sample compartment of the spectrometer using the two mentioned methods. While an individual needle is directly irradiated by excitation laser beam and on the rear side of the needle is in the distance of ca. 5 mm non-reflecting black metal body, the bundle of needles is irradiated through a glass wall of the cell and a needle is surrounded by other needles. Nevertheless, the results of CA demonstrate that the spectral differences between the two methods of measurements of needles are less important than the differences caused by the origin of needles. Also, the results of PCA show differentiation of data of different forest areas (trees) both for spectra obtained in the cylindrical cell and using individual needles. In conclusion, individual forest areas can be distinguished both using the cylindrical cell and measuring individual needles. Both results of CA and PCA demonstrate that any used treatment of spectra does not fundamentally affect the general scheme of differentiation of data with respect to the origin of needles and to the method used for their measurement. Only some minor effects can be pronounced by the data treatment. Summarising all results mentioned earlier, we can consider that the method of measurement of a bundle of needles in stepwise rotating cell can be used to characterise individual trees and/or areas. Furthermore, for such characterisation of one tree a 6-h measurement of individual needles can be replaced by the 1-h accumulation in cylindrical cell to save instrumental time, to shorten the time of an individual analysis and/or to enable more comprehensive analysis of wide range of various types of needle samples, e.g. to enlarge the number of trees and areas examined. Acknowledgements Financial support of the Ministry of Environment of the Czech Republic (grant VaV340/1/01) and of the Ministry of Education, Youth and Sports of the Czech Republic (grant MSM 223400008) is gratefully acknowledged. References [1] J. Krizova, P. Matejka, G. Budinova, K. Volka, J. Mol. Struct. 480/481 (1999) 547. [2] P. Matejka, L. Pleserova, G. Budinova, K. Havirova, J. Nahlik, F. Skacel, K. Volka, Proc. SPIE-Int. Soc. Opt. Eng. 4199 (2001) 130. [3] P. Matejka, L. Pleserova, G. Budinova, K. Havirova, X. Mulet, F. Skacel, K. Volka, J. Mol. Struct. 565/566 (2001) 305. [4] L. Pleserova, G. Budinova, K. Havirova, P. Matejka, F. Skacel, K. Volka, J. Mol. Struct. 565/566 (2001) 311. [5] B. Schrader, H.H. Klump, K. Schenzel, H. Schulz, J. Mol. Struct. 509 (1999) 201.