1 23
Wood Science andTechnologyJournal of the InternationalAcademy of Wood Science ISSN 0043-7719Volume 45Number 1 Wood Sci Technol (2010)45:83-102DOI 10.1007/s00226-010-0307-9
Rapid assessment of physical propertiesand chemical composition of thermallymodified wood by mid-infraredspectroscopy
1 23
Your article is protected by copyright and
all rights are held exclusively by Springer-
Verlag. This e-offprint is for personal use only
and shall not be self-archived in electronic
repositories. If you wish to self-archive your
work, please use the accepted author’s
version for posting to your own website or
your institution’s repository. You may further
deposit the accepted author’s version on a
funder’s repository at a funder’s request,
provided it is not made publicly available until
12 months after publication.
ORI GIN AL
Rapid assessment of physical properties and chemicalcomposition of thermally modified woodby mid-infrared spectroscopy
Marcos M. Gonzalez-Pena • Michael D. C. Hale
Received: 2 December 2008 / Published online: 27 February 2010
� Springer-Verlag 2010
Abstract Characterisation, quality assessment and property prediction are several
of the major industrial challenges for widespread acceptance of thermally modified
wood (TMW). This study shows the potential of the multivariate analysis of mid-
infrared (MIR) spectral data for the prediction of impact strength, five mechanical
parameters in bending, moisture content, weight loss, density and chemical com-
position of small specimens of thermally modified beech, Norway spruce and Scots
pine woods. Anti-swelling efficiency was also studied using DRIFT spectroscopy
for spruce wood only. Calibrations were successfully accomplished by partial least-
squares regression, with RY2 and QCUM
2 values [0.96 for 64 out of 67 models.
Predictions were also successful, with relative prediction values[0 and RMSEP:SD
ratios \1 in most cases. Changes in the MIR spectra of TMW show that bands
arising from the lignin environment and new bands appearing due to the degradation
of carbohydrates, giving negative loadings, were related to strength loss, while those
bands arising from the polysaccharides were associated with property retention. It is
concluded that this approach is a powerful tool to characterise a number of prop-
erties of TMW with a single after-treatment measurement.
Introduction
The search for environmentally neutral wood protection systems has been a core
research subject in Europe over the last 10 years (Hill 2006). Thermal modification
M. M. Gonzalez-Pena � M. D. C. Hale
School of the Environment and Natural Resources,
Bangor University, Bangor, Gwynedd LL57 2UW, UK
M. M. Gonzalez-Pena (&)
Centre for Advanced Wood Processing, The University of British Columbia,
2900-2424 Main Mall, Vancouver, BC V6T 1Z4, Canada
e-mail: [email protected]
123
Wood Sci Technol (2011) 45:83–102
DOI 10.1007/s00226-010-0307-9
Author's personal copy
of wood which reduces the water reactivity of the cell wall polymers affords
improved dimensional stability and fungal decay resistance of various timber
species (Gonzalez-Pena et al. 2004; Hale et al. 2005).
In the laboratory, the amount of thermal modification has often been determined
by mass loss of small wood specimens, but for larger planks, this measurement is
not readily achievable on an industrial scale with a hygroscopic material. Colour
changes have also been explored for the prediction of mechanical properties of
thermally modified wood (TMW); results from this approach are scant and
apparently contradictory (cf. Bekhta and Niemz 2003; Johansson and Moren 2006).
Non-destructive predictions of modulus of rupture in bending (MOR) from direct or
indirect measurement of the stiffness of TMW have also been attempted (Repellin
and Guyonnet 2003; Widmann et al. 2007). However, these have the drawback that
the correlation between MOR and modulus of elasticity (MOE) decreases in line
with the treatment severity: MOE remains generally unchanged for weight losses
smaller than 10%, while the MOR is reduced at the early stages of modification
(Bengtsson et al. 2002; Gonzalez-Pena and Hale 2007).
When wood is exposed to heat above 140�C, multiple chemical reactions occur in
the cell wall polymers, which result in composition changes. Correlation between
these modifications and physical changes provides a potential quality control
method for predicting mechanical and other physical properties of TMW from
simple chemical analyses. Vibrational spectroscopy is a fast method for examining
different functional organic groups and changes in their amount. On this basis, wood
has been subject to several studies using near-infrared (NIRS) and mid-infrared
(MIRS) spectroscopies; often wood samples have been examined directly without
further preparation (Kelley et al. 2004; Nuopponen et al. 2006). Multivariate data
analyses such as partial least-squares regression (PLSR) and modern deconvolution
algorithms are now established as powerful techniques to extract the information of
the spectra. The most common spectroscopic method for the assessment of
mechanical properties of untreated solid wood has been NIRS (Tsuchikawa 2007
and ref. therein). Most reports have demonstrated that this can provide physical and
chemical information about wood in a laboratory environment. The extension of this
for the rapid evaluation of physical properties in TMW is appealing, given the losses
in some forms of wood strength that result and the current requirement for quality
control and assurance. Recently, NIRS was applied to model the MOE, MOR and
five other properties of heat-treated pine and eucalyptus woods (Esteves and Pereira
2008) and to predict the cellulose content and crystallinity of heat-treated Hinoki
cypress wood (Inagaki et al. 2007), but the prediction of the chemical composition
or physical properties of TMW has not been attempted in the MIR range.
In this paper, the prediction of the mechanical strength and other physical
properties of three modified woods by multivariate analysis of Fourier-transform
MIR (FTIR) spectra is reported. The determination and prediction of chemical
constituents are also examined to unveil the relationship between physical
properties and chemical changes in TMW. Predictions of anti-swelling efficiency
(ASE) of spruce wood using diffuse reflectance Fourier transform infrared (DRIFT)
spectroscopy are also described.
84 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
Materials and methods
Thermal modification of wood
Matched, oven-dry, small wood specimens (150 mm 9 20 mm 9 10 mm,
l 9 r 9 t) of Scots pine (P. sylvestris L.), Norway spruce (Picea abies (L.) Karst.)
and beech (Fagus sylvatica L.) were thermally modified in N2 atmosphere at room
pressure. Twenty treatments, using five treatment times (0.33, 1, 4, 8 and 16 h) at
four treatment temperatures (190, 210, 230 and 245�C) plus untreated specimens
were studied. Treatment at 245�C for 8 h was ultimately not performed for beech
wood.
Mechanical testing and measuring physical parameters
Test specimens were conditioned at 65% RH at 20�C for at least 6 months, and then
subjected to a three-point bending test according to BS 373 (BSI 1957) (n = 10 per
treatment). From this test, five strength parameters were computed, namely: fibre
stress at limit of proportionality (RLP), resilience or work to the limit of
proportionality (R), work to maximum load (WML), modulus of elasticity (MOE)
and modulus of rupture (MOR). The load at limit of proportionality to calculate
RLP was obtained graphically from the stress–strain plot while the R and WML
were derived from the respective areas of the same plot. Additionally, impact tests
were conducted following the Charpy method according to BS EN ISO 179 (BSI
1997). The flat-wise impact strength (IS) of un-notched type 3 specimens was
investigated (n = 10 per treatment). Oven-dry heat-induced weight loss relative to
the initial wood weight (WL), nominal density (ND = oven-dry weight/volume at
65% RH, 20�C), oven-dry-specific gravity [SGOD = (oven-dry weight/oven-dry
volume)/1 g cm-1] and equilibrium moisture content at the conditioning atmo-
sphere (EMC) were also computed. The dimensional stability was appraised in
terms of the anti-swelling efficiency, ASE (%) = [(Su - Sm)/Su] 9 100, where Su
and Sm are the swelling coefficients of unmodified wood and modified wood,
respectively (Hill 2006).
Wet chemical analysis
Following bending tests, three specimens were randomly chosen for each
treatment and milled. The milled 30–60 mesh fraction was Soxhlet-extracted
using toluene, methanol and acetone (4:1:1, v/v). Acid insoluble lignin (Klason
lignin) and acid soluble lignin were obtained according to Effland (1977) and
TAPPI um-250 (TAPPI 1991), respectively. Total lignin content is reported as the
sum of Klason and acid soluble lignins. Quantification of monosaccharides in the
hydrolysate was performed by HPAE-PAD chromatography (Worrall and
Anderson 1993). Based on monosaccharide data, the carbohydrate content of
TMW was calculated as for the T249 cm-85 standard (TAPPI 1994). Total
carbohydrate content was computed from lignin by difference to give 100% of the
extracted oven-dry weight of the specimen. From monosaccharide composition,
Wood Sci Technol (2011) 45:83–102 85
123
Author's personal copy
cellulose content and the two main hemicellulose types (glucomannan, GluMan,
and glucuronoxylan, GluXylan) were calculated based on several assumptions on
the structure of the hemicelluloses according to Janson (1970). One determination
of chemical composition was carried out for each treatment per wood species.
Further details on the chemical determinations are given elsewhere (Gonzalez-
Pena et al. 2010).
Mid-infrared spectroscopy
FTIR spectra were obtained with the KBr pellet technique in the wavenumber range
4,000–400 cm-1, at a resolution of 4 cm-1, on a Bruker spectrophotometer (Tensor
27, Bruker Optik GmbH, Germany). The interferograms were Blackman Harris
apodized, and the resulting spectra stored in transmission mode. Spectra were
baseline corrected, next converted to absorbance units (using A = Log 1/T), vector
normalised and finally stored in text format for data processing in the SIMCA-P
software (version 11.0.0.0, Umetrics AB, Sweden). Only one spectrum was taken
from each treatment-species combination. This is taken as a representative of the
wood sample because of the original pooled sampling method included three
specimens, and the ball-milling of the KBr preparation technique randomises the
material within the disc.
Another set of treated matched specimens was used to investigate the ASE of
spruce wood only, for treatments at 210 and 230�C (10 schedules ? control), using
DRIFT spectroscopy. Five specimens were used for each schedule, and only one
spectrum was taken from each specimen; the surface of the wood specimens was
scanned without further preparation. Spectra were acquired in a Bomem FTIR
spectrometer MB100, equipped with a Spectratech diffuse reflectance unit; 200
scans were collected at a 4 cm-1 resolution. The background spectra were obtained
against a silver plate supplied by Spectratech, and the spectral intensities were
calculated in Kubelka–Munk units. No baseline correction or normalisation was
performed. Another set of measurements was taken with a smaller number of scans
(24) at reduced resolution (8 cm-1); only Norway spruce specimens treated at
210�C plus untreated controls were analysed. Three specimens were used for each
treatment; only one spectrum was acquired for each specimen.
For the qualitative evaluation of changes in the IR spectra of TMW, analysis of
the differences of relative bands intensities was performed. The absorbance spectra
were baseline corrected at 3800, 1875 and 800 cm-1, and normalised using the
closest band to 1,374 cm-1 as a maximum (CH bending deformation in cellulose,
symmetric) and 1,394.25 cm-1 as a minimum. To obtain the difference spectra, the
spectra of treated specimens were subtracted from the spectrum of the untreated
control.
Multivariate analyses and calibration statistics
Calibrations were performed using PLSR (Martens and Næs 1989; Eriksson et al.
2001). Spectral ranges employed were 3,500–2,800 and 1,800–550 cm-1 (Nuop-
ponen et al. 2006), giving 505 data points for a resolution of 4 cm-1. For the
86 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
calibration and subsequent prediction modelling using FTIR spectra, the following
values were used to build the Y matrix: for MOR, MOE, R, RLP, WML, WL, ND,
SGOD and EMC, the average values of the three specimens milled; for IS, the
average values of three specimens per treatment with similar levels of WL to those
of the bending test; for chemical constituents, the results of the single wet chemical
determinations for each treatment. For building each calibration PLS model for
mechanical properties, WL, ND, SGOD, EMC and chemical constituents, data from
13 or 14 treatments, chosen systematically in each species (Table 1), were used; full
cross-validation was used for building all models. The data for the remaining seven
to eight treatments were not included in the calibration modelling and were used as
the prediction set. For the modelling of ASE and WL using the DRIFT spectra in
spruce wood, individual specimen values were used for building the Y matrix; 36
specimens were used for the calibration set, while the remaining 19 specimens were
left apart for the prediction set. In all models, spectral data (X matrix) were mean-
centred, whereas physical and chemical data (Y matrix) were mean-centred and
scaled to unit variance. Two-component orthogonal signal correction (OSC) was
used for all spectra (Wold et al. 1998); prior to predictions, spectra for samples in
the prediction set were also two-component OSC-filtered. The relative prediction
error, RPE, was used to compare the efficiency of the predictions of different data
sets. This was calculated as: RPE = (SD2 - MSEP):SD2, where SD2 is the
variance of the Y data in the calibration model and MSEP = RMSEP2. RMSEP,
root-mean-square error of prediction is defined as:
Table 1 Treatments included in the calibration and prediction sets in all PLS models using FTIR spectra
Beech setsa Scots pine sets Norway spruce sets
Calibrationb Prediction Calibration Prediction Calibration Prediction
Untreated 190/1 Untreated 190/4 Untreated 190/1
190/0.33 190/8 190/0.33 210/1 190/0.33 190/8
190/4 210/1 190/1 210/4 190/4 210/1
190/16 210/8 190/8 230/4 190/16 210/8
210/0.33 230/1 190/16 230/8 210/0.33 230/1
210/4 230/8 210/0.33 230/16 210/4 230/8
210/16 245/4 210/8 245/4 210/16 245/1
230/0.33 210/16 230/0.33 245/8
230/4 230/0.33 230/4
230/16 230/1 230/16
245/0.33 245/0.33 245/0.33
245/1 245/1 245/4
245/16 245/8 245/16
245/16
a No treatment at 245�C for 8 h exists for beech woodb First value for treatment temperature (�C), second value for treatment time (h)
Wood Sci Technol (2011) 45:83–102 87
123
Author's personal copy
RMSEP ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
NP
i¼1
yi^ �yi
� �2
NP
v
u
u
u
t
ð1Þ
where yi^
is the value of the property of interest for specimen i estimated using the
calibration model, yi is the experimental value of the property of interest of
specimen i, and NP is the number of specimens in the prediction set. An RPE value
close to one indicates an excellent prediction for specimens not included in the
calibration model, while a zero or negative value suggests a poor prediction for the
specimens in the prediction set (Thumm and Meder 2001).
Results and discussion
Model calibration and prediction of mechanical properties
The statistics in Table 2 reveal that excellent1 calibrations can be obtained for each
mechanical strength parameter based on the analysis of MIR spectra of TMW. The
calibrations developed for all five strength parameters in bending and for impact
strength have goodness of fit RY2 coefficients [0.99 and goodness of prediction
QCUM2 values = 0.97, irrespective of the species studied. For the predicted values,
the modelling is also very good, with RPE values[0 in most cases. In terms of the
RPE, the best prediction in bending for the three species was for MOR, although the
best strength prediction in pine was for IS (QCUM2 = 0.998, RPE = 0.85). The
lowest prediction ability in beech was for the models of MOE, WML and R,
although these still were within acceptable ranges (RPE [ 0). Except for MOE in
beech, all the models for mechanical strength parameters gave only one factor after
extraction of two OSC components (A = 1), making it easier to interpret the model
compared to the raw data. For the latter, model dimensionalities were generally
higher (between 2 and 4 dimensions, data not shown). Although the RMSEC values
for the calibration of each strength property were up to one order of magnitude
smaller than the RMSEP results (Table 2), the former values were very small, in
some cases almost zero. RMSEP values were well below one SD for each property
in question, indicating that calibrations based on FTIR spectroscopy can potentially
be effectively used to predict most of the strength properties from specimens not
included in the calibration model. Calibration models for energy-associated
properties (R and WML) gave smaller prediction abilities for specimens not
included in the calibration set. This was in part caused by the inherently larger
variation of these properties, and in part probably by a smaller correlation of these
properties with the spectral information of TMW.
To illustrate graphically the estimation and prediction abilities of the models,
plots of the experimentally obtained versus FTIR-fitted individual strength
properties are shown in Fig. 1 for MOR, MOE, RLP and IS. Predicted values for
1 The term excellent refers to a QCUM2 value[0.9,[0.75 is very good and[0.5 is good (SIMCA-P User’s
Manual, version 11.0.0.0, Umetrics AB, Sweden, 2006).
88 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
each property are also plotted therein. For the predictions (filled symbols), the
model for MOR in beech and spruce was excellent, while the predicted values for
pine have a larger scatter. MOE was confidently predicted with the calibration
model up to around 12,000 MPa. At higher values, one predicted value was under-
estimated for pine, while one value was over-estimated for beech. RLP showed
some degree of under-and over-estimation in the three species, but the general
model prediction ability was acceptable. The prediction of IS was also good for
pine, but in most predicted values were to some extent under-estimated. To the
knowledge of the authors, there is no previous modelling of mechanical properties
of TMW using MIR data. A recent study has dealt with the estimation of MOR and
MOE in eucalyptus and pine woods using NIRS though, attaining RY2 values from 47
to 89% for MOE, and from 75 to 77% for MOR, but the capability of using these
models to predict MOR and MOE in samples not included in the calibrations was
not determined (Esteves and Pereira 2008).
Table 2 Summary of statistics for calibrations and predictions for mechanical properties of thermally
modified beech, Scots pine and Norway spruce woods from FTIR spectra
Species Property Mean SD A RY2 QCUM
2 RMSEC RMSEP RPE
Beech MOE 11,840 1,122 2 0.996 0.985 88.40 935.32 0.31
MOR 96.64 33.76 1 0.999 0.993 0.86 14.45 0.82
RLP 67.30 16.66 1 0.997 0.976 0.91 12.13 0.47
R 0.0223 0.0080 1 0.997 0.987 0.0004 0.0064 0.37
WML 0.0687 0.0560 1 0.997 0.971 0.0032 0.0451 0.35
Scots pine MOE 12,812 1,180 1 0.996 0.994 73.71 901.27 0.42
MOR 94.11 27.51 1 0.999 0.999 0.68 13.22 0.77
RLP 64.92 12.99 1 0.994 0.990 1.06 12.55 0.07
R 0.0195 0.0057 1 0.996 0.991 0.0004 0.0058 -0.04
WML 0.0565 0.0320 1 0.998 0.996 0.0013 0.0192 0.64
IS 17.63 8.87 1 0.999 0.998 0.33 3.45 0.85
Norway spruce MOE 9,254 1,311 1 0.999 0.970 46.98 786.07 0.64
MOR 64.80 22.02 1 1.000 0.988 0.38 6.04 0.92
RLP 36.75 11.58 1 0.997 0.993 0.71 7.43 0.59
R 0.0087 0.0041 1 0.994 0.989 0.0003 0.0024 0.66
WML 0.0416 0.0260 1 0.998 0.993 0.0011 0.0071 0.93
IS 12.27 7.57 1 0.999 0.995 0.31 3.62 0.77
Number of samples in calibration and prediction sets: 13 and 7 in beech, 14 and 7 in pine, and 13 and 8 in
spruce, respectively (see Table 1)
Units: MOE, MOR and RLP = MPa; R and WML = mm N mm-3; IS = kJ m-2
Statistics: Mean, mean value of the calibration set; SD, standard deviation of the calibration set; A, factors
in the calibration model; RY2 : rCalibration
2 , goodness of fit (explained variation of the Y matrix);
QCUM2 : rCross-validation
2 , cumulative goodness of prediction; RMSEC, root-mean-square error of cali-
bration; RMSEP, root-mean-square error of prediction; RPE, relative prediction error
For property abbreviations, see ‘‘Materials and methods’’. IS in beech not studied
Wood Sci Technol (2011) 45:83–102 89
123
Author's personal copy
Model calibration and prediction of WL, ND, EMC and SGOD
Excellent calibrations were obtained for each physical property studied based on the
analysis of MIR spectra of TMW (Table 3). The calibrations developed for WL,
ND, MC and SGOD have in all cases coefficients of the goodness of fit RY2 [ 0.98
and goodness of prediction QCUM2 [ 0.96, irrespective of the species studied. For the
predicted values, the fit was also good, with RPE values[0.2 in most cases. Based
on the RPE statistic, the best predictions using FTIR spectra were for WL and EMC
in the three species. Apparently, the prediction of ND in pine was not completely
acceptable, although from the scatter in the plot of measured versus fitted values,
this is not evident (Fig. 2). Figure 2 also includes the plots for other individual
physical properties for the experimentally obtained versus FTIR-fitted, namely for
WL, SGOD and EMC. Predicted values for each property are plotted in the same
charts as well. From plots in Fig. 2, it is noticeable that excellent fits for the
calibration model were obtained for each property (open symbols). Also from the
MOR
Fitted (MPa)0 20 40 60 80 100 120 140 160
Mea
sure
d (M
Pa)
0
20
40
60
80
100
120
140
160MOE
Fitted (MPa x 1000)
4 6 8 10 12 14 16
Mea
sure
d (M
Pa
x 10
00)
4
6
8
10
12
14
16
RLP
Fitted (MPa)
0 20 40 60 80 100
Mea
sure
d (M
Pa)
0
20
40
60
80
100IS
Fitted (kJ m-2)
0 10 20 30 40
Mea
sure
d (k
J m
-2)
0
10
20
30
40
Fig. 1 Calibration plots for MOR, MOE, RLP and IS for thermally modified woods. Predicted values foreach property are also included. Open symbols, calibration set; filled symbols, prediction set; circles,beech; triangles, Scots pine; squares, Norway spruce. Calibration and prediction models were calculatedfor each species separately
90 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
plots, good predictions were found for WL, ND, SGOD and EMC, except for one
specimen in beech for the latter, which was over-estimated by about 2%. No
previous reference exists for the estimation of ND, SGOD or EMC in TMW in the
MIR range. In the case of the prediction of WL, normalised DRIFT spectra (in
Kubelka–Munk units) of heat-treated Norway spruce and Scots pine powdered
woods were used for the prediction of WL by PLSR (Kotilainen et al. 2000).
Spectral variables with correlation coefficients between the measured WL and the
individual spectral intensities [0.145 were included in the model. These authors
found a correlation coefficient (R2) for the calibration set of 0.960, and a R2 of 0.894
for the validation set, with six factors in their model.
Prediction of physical properties using selected IR bands
Calibrations containing from 8 to 10 wavenumbers instead of spectral ranges 3,500–
2,800 and 1,800–550 cm-1 were attempted for the evaluation of MOE, MOR, WL
and ND in spruce wood only. The reduction in predictors from 505 to the 8 to 10
more significant ones led to similar RPE values in both models (selected bands vs.
complete range) for MOE and WL, with only small reductions in the RPE for MOR
and ND (Table 4, FTIRfew). This highlights the possibility of using a simple,
portable device for the prediction of mechanical and other physical properties of
TMW at a larger scale. Portable MIR instruments reading a small number of
wavelengths have been used for many years in other industries, primarily for
kinetics studies in processes concerning either gaseous or liquid substances
(Rouessac and Rouessac 2000); several hand-held prototypes have also been
Table 3 Summary of statistics for calibrations and predictions for physical characteristics of thermally
modified beech, Scots pine and Norway spruce woods using FTIR spectra
Species Property Mean SD A RY2 QCUM
2 RMSEC RMSEP RPE
Beech WL 11.06 9.87 1 1.000 0.995 0.18 1.27 0.98
EMC 6.53 1.60 1 1.000 0.992 0.03 1.07 0.55
ND 618.72 39.00 1 0.992 0.961 3.55 27.48 0.50
SGOD 0.6425 0.0433 1 0.997 0.983 0.0024 0.0336 0.40
Scots pine WL 7.01 6.88 1 1.000 0.999 0.09 0.99 0.98
EMC 7.48 1.63 1 0.984 0.981 0.21 0.56 0.88
ND 514.40 19.05 1 0.987 0.980 2.24 19.82 -0.08
SGOD 0.5373 0.0233 1 0.995 0.991 0.0018 0.0201 0.26
Norway spruce WL 7.75 7.94 1 1.000 0.984 0.10 1.09 0.98
EMC 7.63 1.99 1 0.999 0.997 0.07 0.31 0.98
ND 357.98 17.99 1 0.998 0.986 0.76 11.19 0.61
SGOD 0.3751 0.0224 1 1.000 0.994 0.0004 0.0122 0.70
Number of samples in calibration and prediction sets: 13 and 7 in beech, 14 and 7 in pine and 13 and 8 in
spruce, respectively (see Table 1)
Units: WL = %; EMC = %; ND = kg m-3; SGOD = unitless. For statistics abbreviations see Table 2.
For property abbreviations see ‘‘Materials and methods’’
Wood Sci Technol (2011) 45:83–102 91
123
Author's personal copy
developed in modern times for measuring selected MIR bands in solid materials
(Workman 1999).
Model calibration and prediction of chemical constituents
In general, the modelling of chemical constituents was superior in both the
calibration and prediction ability to all other models, except for the models for
galactan in beech and pine, which gave unacceptable predictions (Table 5).
Excellent calibrations were obtained for the remaining monosaccharides and
polymers based on the analysis of MIRS (Table 5). The calibrations developed for
wood polymers (lignin, glucomannan, glucuronoxylan, hemicelluloses and cellu-
lose) have all goodness of fit RY2 and goodness of prediction QCUM
2 statistics[0.975,
irrespective of the species studied. For the predicted values, the fit was also good,
with RPE values[0.54 in most cases. Model complexity for all polymers was also
low (A = 1). Based on the RPE, the best predictions using FTIR spectra differed
between softwoods and beech. For the latter, the best prediction was for
WL
Fitted (%)-5 0 5 10 15 20 25 30
Mea
sure
d (%
)
-5
0
5
10
15
20
25
30EMC
Fitted (%)4 6 8 10 12
Mea
sure
d (%
)
4
6
8
10
12
ND
Fitted (kg m-3)
200 300 400 500 600 700
Mea
sure
d (k
g m
-3)
200
300
400
500
600
700 SGOD
Fitted (unitless)0.2 0.3 0.4 0.5 0.6 0.7 0.8
Mea
sure
d (u
nitle
ss)
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Fig. 2 Calibration plots for WL, EMC, ND and SGOD. Open symbols, calibration set; filled symbols,prediction set; circles, beech; triangles, Scots pine; squares, Norway spruce. Calibration and predictionmodels were calculated for each species separately
92 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
hemicelluloses, followed by GluXylan, while for gymnosperms the best prediction
was for lignin, followed by GluMan and hemicelluloses. In all three species, the
least efficient prediction was for cellulose. This may be because significant changes
in cellulose were not reflected by mass change only, but also by structural
modifications (e.g. increased crystallinity), which are not expressed in the
gravimetric determinations. For the calibrations for wood monosaccharides
(converted to individual oligosaccharides), arabinan, galactan, glucan (including
glucose in cellulose), xylan and mannan, the RY2 and QCUM
2 statistics were[0.99 and
[0.98, respectively, regardless of the species concerned, except for galactan in
beech and pine. Predictions were also good, with RPE values[0.58 in 12 out of 15
models. Based on the RPE, the best predictions were for xylan in beech and for
mannan in softwoods. The poor fit for galactan, the constituent with the second
lowest contribution in the chemical composition, may be influenced by the close
elution of this with glucose as seen in the chromatogram, making the analytical
quantification less exact for this sugar.
Model calibration and prediction of the ASE
The plot for the experimentally measured versus DRIFT-fitted ASE in spruce wood is
given in Fig. 3. Predicted values are also included in this plot. An excellent fit can be
seen for the calibration model. Additionally, rather good predictions were obtained
for ASE for the specimens in the prediction set, given the high inherent variability of
this property, and considering that the prediction set consisted of ASE values of
individual specimens (compared to the values in the prediction set for mechanical
properties, which were the mean values of three specimens). Calibrations and
predictions of WL were also successfully accomplished with DRIFT-spectra data
Table 4 Summary of statistics for calibrations and predictions for physical properties of thermally
modified Norway spruce wood using selected bands of FTIR spectra, DRIFT spectra and low-resolution
DRIFT spectra
Spectra Property Mean SD A RY2 QCUM
2 RMSEC RMSEP SD n
Cal Pred
FTIRfew MOEa 9,254 1,311 1 0.96 0.96 263.6 751.8 0.67 13 8
MORb 64.80 22.02 1 0.99 0.99 2.07 7.89 0.87 13 8
WLb 7.75 7.94 1 0.99 0.99 0.75 0.58 0.99 13 8
NDa 358.0 17.99 1 0.92 0.92 5.22 14.07 0.39 13 8
DRIFT WL 4.59 3.47 1 1.00 1.00 0.00 1.66 0.77 36 19
ASE 21.30 16.32 1 1.00 1.00 0.00 5.96 0.87 36 19
DRIFTlow WL 3.13 3.13 1 1.00 1.00 0.00 1.29 0.83 12 6
ASE 15.16 16.20 1 1.00 1.00 0.02 4.86 0.91 12 6
FTIRfew, FTIR selected wavenumbers (a =3500, 1712, 1600, 1290, 1200, 1057, 1031 and 620 cm-1;b =3220, 1712, 1600, 1508, 1290, 1200, 1159, 1107, 1057 and 1031 cm-1). DRIFTlow, DRIFT low
resolution (24 scans at 8 cm-1 resolution). For property abbreviations, see ‘‘Materials and methods’’. For
property units and statistics abbreviations, see Table 2
n is the number of cases in the calibration (Cal) and prediction (Pred) sets
Wood Sci Technol (2011) 45:83–102 93
123
Author's personal copy
Table 5 Summary of statistics for calibrations and predictions for chemical composition of thermally
modified beech, Scots pine and Norway spruce woods using FTIR spectra
Constituent (%)a Mean SD A RY2 QCUM
2 RMSEC RMSEP RPE
Beech
Lignin 28.62 4.97 1 1.000 0.995 0.098 2.322 0.78
GluXylan 12.65 8.26 1 1.000 0.995 0.133 2.185 0.93
GluMan 1.42 1.58 1 1.000 0.998 0.019 0.653 0.83
Hemicellulose 14.07 9.46 1 1.000 0.997 0.120 2.033 0.95
Cellulose 42.90 5.06 1 0.995 0.975 0.377 2.998 0.65
Arabinan 0.20 0.23 1 1.000 0.993 0.002 0.135 0.65
Galactan 0.51 0.51 1 0.994 0.938 0.041 0.515 -0.03
Glucan 44.99 5.70 1 0.999 0.983 0.194 3.075 0.71
Xylan 12.40 8.23 1 1.000 0.996 0.120 1.951 0.94
Mannan 0.79 0.89 1 1.000 0.998 0.011 0.367 0.83
Scots pine
Lignin 31.87 3.32 1 1.000 0.998 0.071 0.956 0.92
GluXylan 5.58 2.97 1 0.996 0.989 0.206 1.413 0.77
GluMan 10.37 4.37 1 0.999 0.996 0.143 1.208 0.92
Hemicellulose 15.95 7.24 1 0.997 0.991 0.420 2.460 0.88
Cellulose 43.01 3.62 1 0.989 0.983 0.388 2.440 0.55
Arabinan 0.59 0.49 2 0.995 0.988 0.038 0.399 0.33
Galactan 1.23 0.64 1 0.892 0.873 0.220 0.850 -0.74
Glucan 46.47 4.21 1 0.992 0.988 0.381 2.543 0.64
Xylan 4.03 2.15 1 0.995 0.989 0.150 1.024 0.77
Mannan 8.00 3.37 1 0.999 0.996 0.107 0.906 0.93
Norway spruce
Lignin 32.06 4.92 1 1.000 0.975 0.067 1.332 0.93
GluXylan 4.74 2.63 1 0.999 0.991 0.061 0.971 0.86
GluMan 10.15 4.81 1 1.000 0.986 0.096 1.749 0.87
Hemicellulose 14.89 7.40 1 1.000 0.988 0.172 2.554 0.88
Cellulose 42.96 5.51 1 1.000 0.998 0.105 4.321 0.39
Arabinan 0.46 0.37 1 1.000 0.995 0.008 0.143 0.85
Galactan 0.91 0.46 1 0.997 0.993 0.025 0.166 0.87
Glucan 45.73 6.34 1 1.000 0.993 0.076 4.062 0.59
Xylan 3.47 1.95 1 0.999 0.992 0.055 0.691 0.87
Mannan 7.95 3.78 1 1.000 0.987 0.076 1.367 0.87
For the modelling of galactan in Norway spruce wood, values for treatment at 245�C for 8 h and at 230�C
for 1 h are not included in the calibration and prediction sets, respectively, due to inconsistency in the
gravimetric determinations
Number of samples in calibration and prediction sets: 13 and 7 in beech, 14 and 7 in pine and 13 and 8 in
spruce, respectively (see Table 1)
For statistics abbreviations see Table 2a Content of the before-treatment oven-dry weight, in %
94 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
(Table 4). The results show that the calibration model derived from the OSC-
corrected spectra was excellent in terms of model complexity (A = 1 for WL and
ASE) and predictive power (QCUM2 = 1.00 in WL and ASE) compared with raw
spectra (A = 2 and QCUM2 = 0.986 and 0.925 for WL and ASE, respectively). During
the collection of the DRIFT spectra, the possible sources of systematic variation may
have been light scattering due to the rough surface of the wood specimens, and path
length differences arising from the positioning of the individual specimens during
scanning. These interferences tend to dominate over the chemical signals in the raw
spectra, but are handled efficiently by the OSC filtering. This algorithm is typically
used in NIRS, although there is no theoretical reason for not using this tool in MIRS
(Fearn 2000); as it has been shown herein, the OSC pre-treatment improved the
models using data from the MIR range. The DRIFT spectra were collected with no
further specimen preparation, a condition probably required if this approach would
have any practical application. However, the collection of each spectrum took
*10 min, so another set of measurements was taken with fewer scans (24) at
reduced resolution (8 cm-1), DRIFTlow. In this way, the collection of spectrum took
39 s. In the low-resolution spectra, the absorbance intensity is reduced, and the
spectra also appear noisy (Fig. 4). Nevertheless, results confirm that predictions of
ASE and WL from the low-resolution DRIFT-spectra data are still good (Table 4,
DRIFTlow), with values of RPE in fact slightly larger than those of the calibration
using the higher-resolution spectra. This result is encouraging, since it substantiates
the possibility of using an IR system for the prediction of the ASE and WL of
modified wood in a continuous process environment.
Changes in the FTIR spectra and its relationship with property change and
loadings of the PLSR modelling
Within the MIR spectral regions chosen for this study, the plot of the PLS loadings
for the MOR in Norway spruce wood showed an analogous profile with the
difference spectrum obtained by subtracting the spectrum of the treated specimens
Fitted (%)0 20 40 60
Mea
sure
d (%
)
0
20
40
60 Prediction Calibration
Fig. 3 Calibration plot for ASEof thermally modified Norwayspruce wood using DRIFTspectroscopy. Predicted valuesare also included
Wood Sci Technol (2011) 45:83–102 95
123
Author's personal copy
from that of untreated wood (Fig. 5). Wood specimens were separated by PLS
regression according to their strength in the direction of the single factor extracted.
The largest loadings in the loading plot match somehow the bands with the largest
changes in the difference spectra (Fig. 5). Evidently, the MIR region was effective
in detecting gradual changes in wood polymers, because steady changes in several
bands became increasingly evident among spectra representing successively more
severe treatment conditions. Several bands are identified where changes occur due
to the thermal conversion as the WL increases (e.g. 1712, 1290 and 1057 cm-1).
The band with the largest negative loading in the loading plot is the one with
maxima at 1,712 cm-1 in the difference spectra and therefore has the largest
importance for the calibration. Other major loadings that accounted negatively for
the estimation of MOR, also appearing in the difference spectra as absorption
maxima, were at 1600, 1290, 1200, 1057 and 1031 cm-1, while absorption minima
at 3500, 3220, 2920, 2850, 1508, 700, 670 and 625 cm-1 had the largest positive
influence on the single factor extracted for MOR (Table 6). Most of the negative
loadings are related to lignin increase and carboxylation of polysaccharides. An
exception to this was the band at 1,508 cm-1 (aromatic skeletal vibrations), which
showed a clear decrease with the treatment. This has been previously reported
(Grandmaison et al. 1987) and could be linked to the decreasing native lignin
remaining in the residue (Windeisen et al. 2007). On the other hand, bands with
decreasing absorbance (giving positive loadings for the MOR) were mostly related
K -
M u
nits
0
20
40
60
80
100
Wavenumber (cm-1)
100015002000250030003500
K -
M u
nits
0
20
40
60
80
100
Fig. 4 Diffuse reflectance spectra for heat-treated Norway spruce wood. Above: 200 spectra acquired at4 cm-1 resolution. Below: 24 spectra acquired at 8 cm-1 resolution
96 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
1712
1600
15083220
1290
12001057
1031
625670700
752806
845922
883
11391380142128502920
3000
1768 15731540 1517
1487
1454
11091159
3500a
-5
-4
-3
-2
-1
0
1
2
5507509501150135015501750
Wavenumber (cm-1 )
Wavenumber (cm-1 )
Arb
itrar
y un
its
1.0
1.41.82.7
4.96.010.5
15.017.726.7
0
Weightloss (%)
280029503350 31503500
b
1650170017501800
Arb
itrar
y un
its
0
c
WL = 26.7%
WL=1.0%
Fig. 5 a Loading plot for the PLS model of MOR in thermally modified Norway spruce wood;b difference spectra for control treatment minus treated wood at ten levels of WL; c enlarged view for thespectral region marked in b; spectra are for increasing levels of WL in the same order as in the legend in b
Wood Sci Technol (2011) 45:83–102 97
123
Author's personal copy
to the polysaccharide portion of wood, which in turn are the components most
degraded during the modification (Table 6). Changes in bands associated with
carbohydrates also indicate the conversion from amorphous to a more crystalline
form of cellulose, and polymer dehydration in TMW. However, a few bands arising
from the frequencies normally assigned to the cellulose environment and from the
hydrogen bonding system do not appear to agree well with a decrease in
Table 6 Band assignments in the FTIR spectra of thermally modified Norway spruce wood
Wave-number
(cm-1)
Assignment and features Occurrenceb Referencec
Increasinga
845 1,3,4—substituted benzene ring in syringyl units L 1, 2
883 1,3,4—substituted benzene ring in guaiacyl units L 1, 2
922 CH out of plane aromatic vibration L 3
1,031 Caryl–O ester vibrations, both methoxyl and b–O–4 in
guaiacyl units
L 4
1,057 C–O stretching vibration, condensation reactions C, H 2, 5
1,109 O–H association band C, H 1
1,159 C–O-C asymmetric bridge stretching vibration C, H 1,5
1,200 O–H in plane bending vibration C, H 1
1,517 Aromatic skeletal vibration L 6
1,600 Aromatic skeletal vibration (smaller than untreated wood at
WL \ 10%)
L 6
1,600 Ring stretch vibration related to quinone formation L 7
1,712 C=O groups in unconjugated ketone, carbonyl and esters L 8
1,720 C=O groups due to condensation reactions, at the expense of
carbonyl groups at 1,654 cm-1L 9, 10
1,730 C=O groups due to carboxylation of polysaccharides C, H 11, 12
3,000 CH stretch in CH3 and CH2 groups L 3
Decreasing
670 C–O–H out of plane bending mode C 2
700 C–O–H out of plane bending mode C 2
806 Ring breathing (b) C 3
1,508 Aromatic skeletal vibrations L 13
2,850 CH2 symmetric stretching C 3
2,920 CH2 anti-symmetric stretching C 3
3,220 OH stretching vibration due to intermolecular hydrogen
bonding (shifting from 3,176 cm-1)
C 3
3,500 OH stretching vibration due to intermolecular H bonding,
becoming a doublet with 3,220 cm-1C 3
a Increasing refers to a higher absorbance compared to untreated wood at increasingly higher WLb L, lignin; C, cellulose; H, hemicellulosesc 1, Harrington et al. 1964; 2, Kuo et al. 1988; 3, Baeza and Freer 2001; 4, Collier et al. 1992; 5, Faix and
Bottcher 1992; 6, Grandmaison et al. 1987; 7, Kimura et al. 1994; 8, Faix 1992; 9, Sudo et al. 1985; 10,
Funaoka et al. 1990; 11, Teratani and Miyazaki 1968; 12, Chow 1971; 13, Hergert 1971
98 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
polysaccharides (Table 6). As noted elsewhere, the hydrogen bonding array appears
to be disrupted upon heating; the increase in absorption in some wavenumbers may
also indicate the formation of aliphatic alcohols during heating or condensation
reactions (Faix 1992; Weiland and Guyonnet 2003).
Evidently, the success in predicting bending strength in the MIR region could be
credited to divergence in lignin and polysaccharide contents, whereby the lignin
content had a negative influence on wood strength, while the signals arising from
the carbohydrate component are linked to strength retention.
The observations made for the difference spectra and loading plot of Norway
spruce still hold for Scots pine; peaks in the difference spectra of pine wood and the
loading plot for MOR are similar to that of spruce wood, with only small variations
(plot not shown). Moreover, the findings made for softwoods for the difference
spectra and loading plot are still valid for beech, but several bands appeared at
slightly lower or higher wavenumbers compared to the difference spectra in
softwoods. Additionally, considerable changes in the magnitude of the loadings and
band heights in four bands were noted for beech at 670, 788 cm-1, the broad peak at
3,270–3,100 cm-1 and at 1,747 cm-1 (plots not shown). The largest difference in
the loading plot of beech compared to the one in softwoods is the decrease at
1,747 cm-1 (C=O stretch in carboxyl groups from xylan, Owen and Thomas 1989).
Although in the plot for the difference spectra in softwoods, a decrease for this
group at 1,736 cm-1 was observed at WL \ 6.0%, in the loading plot no decreasing
peak appeared at this band. In softwoods, the OSC algorithm probably superimposes
the large increase in other bands in the spectral range between 1,740 and
1,700 cm-1 over the initial decrease at 1,736 cm-1 due to the initial loss of esters
and carboxyl groups particularly in xylan and assigns a negative loading to this
band. In contrast, a sharp peak develops in the difference spectra of beech at
1,747 cm-1 and a large positive loading also appears in its loading plot. This is
probably due to the larger proportion of xylan in untreated beech, and also to the
large, fast degradation of xylan upon heating. In the remaining strength properties in
bending and in impact strength, the loading plot showed similar profile for the MOR
loading plot but for small changes in the loadings (plots not shown). For chemical
constituents increasing in line with the thermal modification (e.g. lignin content),
the profile in the loading plot was also analogous but the loadings had opposite sign
to the ones in the plot for MOR. For most other chemical polymers or monomers
and physical properties (WL, ND, SGOD), the profile of the loading plot was also
alike to the one of MOR. The only noticeable difference was the loading plot for the
prediction of EMC, where most of the loadings dominated by the carbohydrate
component in the region 1,290–550 cm-1 had negative values.
Conclusion
The results reported here demonstrate for the first time the capability of the MIRS
combined with multivariate analysis, as a prospective powerful tool for the rapid
prediction of several properties of thermally modified wood. By characterising the
material with training sets including untreated specimens, it is possible to predict
Wood Sci Technol (2011) 45:83–102 99
123
Author's personal copy
five mechanical parameters in bending, impact strength, WL, density, EMC, ASE
and chemical composition of TMW by collecting uncomplicated spectra measure-
ments. All properties are estimated with only one after-treatment measurement of
the spectra to arrive to predictions with low margin of error, at the laboratory scale.
This method itself is inexpensive and has the potential to be incorporated into online
process control.
While the properties of small-sized specimens of TMW appear to be confidently
detected by FTIR and DRIFT spectroscopies, the viability of this technique for the
evaluation of properties of large-sized specimens remains to be assessed. Common
knowledge indicates that the ASE, WL, ND, SGOD and EMC of long-sized
specimens would be little influenced by the presence of knots or other wood defects,
rendering this approach feasible for the study of larger specimens. Calibrations
using larger members are probably required for the prediction of some strength
parameters, but these could in first principle be derived from calibrations of small
specimens applying safety factors.
Acknowledgments The National Council for Science and Technology, Mexico (CONACYT) is
thanked for financial support to undertake this research (Grant 178663 to MMGP). Dr. Simon F. Curling
is acknowledged for conducting the HPLC-PA determinations.
References
Baeza J, Freer J (2001) Chemical characterization of wood and its components. In: Hon DN-S, Shiraishi
N (eds) Wood and cellulose chemistry. Marcel Dekker, New York, pp 275–384
Bekhta P, Niemz P (2003) Effect of high temperature on the change in color, dimensional stability and
mechanical properties of spruce wood. Holzforschung 57:539–546
Bengtsson C, Jermer J, Brem F (2002) Bending strength of heat-treated spruce and pine timber. 33rd
Annual meeting of the international research group on wood protection, Cardiff, Wales. Document
IRG/WP 02-40242, 9 p
BSI (1957) BS 373: Methods of testing small clear specimens of timber. HMSO, London, 31 p
BSI (1997) BS EN ISO 179. Plastics—determination of charpy impact strength. European Committee for
Standardization, Brussels, 15 p
Chow SZ (1971) Infrared spectral characteristics and surface inactivation of wood at high temperatures.
Wood Sci Technol 5:27–39
Collier WE, Schultz TP, Kalasinsky VE (1992) Infrared study of lignin: reexamination of aryl-alkyl ether
C-O stretching peak assignments. Holzforschung 46:523–528
Effland MJ (1977) Modified procedure to determine acid-insoluble lignin in wood and pulp. Tappi
60:143–144
Eriksson L, Johansson E, Kettaneh-Wold N, Wold S (2001) Multi- and Megavariate data analysis.
Principles and applications. Umetrics Academy, Umea 533 p
Esteves B, Pereira H (2008) Quality assessment of heat-treated wood by NIR spectroscopy. Holz Roh
Werkst 66:323–332
Faix O (1992) Characterisation in solid state: Fourier transform infrared spectroscopy. In: Lin SY, Dence
CW (eds) Methods in lignin chemistry. Springer, Berlin, pp 83–109
Faix O, Bottcher JH (1992) The influence of particle size and concentration in transmission and diffuse
reflectance spectroscopy of wood. Holz Roh Werkst 50:221–226
Fearn T (2000) On orthogonal signal correction. Chemom Int Lab Syst 50:47–52
Funaoka M, Kako T, Abe I (1990) Condensation of lignin during heating of wood. Wood Sci Technol
24:277–288
Gonzalez-Pena MM, Hale MDC (2007) The relationship between mechanical performance and chemical
changes in thermally modified wood. Third European conference on wood modification, 15–16
October, Cardiff, UK, The Angel Hotel, pp 169–172
100 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy
Gonzalez-Pena MM, Breese MC, Hill CAS (2004) Hygroscopicity in heat treated wood: effect of
extractives. 1st International conference on environmentally-compatible forest products, 22–24
September, Porto, Portugal, Fernando Pessoa University, pp 105–119
Gonzalez-Pena MM, Curling SF, Hale MDC (2009) On the effect of heat on the chemical composition
and dimensions of thermally-modified wood. Polymer Degrad Stab 94:2184–2193
Grandmaison JL, Thibault J, Kaliaguine S (1987) Fourier transform infrared spectrometry and
thermogravimetry of partially converted lignocellulosic materials. Anal Chem 59:2153–2157
Hale MD, Ghosh SC, Spear MJ (2005) Effects of artificial UV weathering and soft rot decay on heat-
treated wood. 36th Annual meeting of the international research group on wood protection,
Bangalore, India. Document IRG/WP 05-40302, 13 p
Harrington KJ, Higgins HG, Michell AJ (1964) Infrared spectra of Eucalyptus regnans F. Muell. and
Pinus radiata D. Don. Holzforschung 18:108–113
Hergert HL (1971) Infrared spectra. In: Sarkanen KV, Ludwig CH (eds) Lignin: occurrence, formation,
structure, reactions. Wiley, Toronto, pp 267–297
Hill CAS (2006) Wood modification. Chemical, thermal and other processes. Wiley, Chichester 239 p
Inagaki T, Yonenobu H, Mitsui K, Yamamoto H, Sasaki Y, Tsuchikawa S (2007) Investigation of thermal
degradation mechanism of softwood by NIR spectroscopy. The 13th international conference on
near infrared spectroscopy, 15–21 June, Umea, Sweden, Umea Folkets Hus, 1 p
Janson J (1970) Calculations of the polysaccharide composition of wood and pulp. Pap Puu 52:323–329
Johansson D, Moren T (2006) The potential of colour measurement for strength prediction of thermally
treated wood. Holz Roh Werkst 64:104–110
Kelley SS, Rials TG, Snell R, Groom L, Sluiter A (2004) Use of near infrared spectroscopy to measure
the chemical and mechanical properties of solid wood. Wood Sci Technol 38:257–276
Kimura F, Kimura T, Gray DG (1994) FT-IR study of the effect of irradiation wavelength on the colour
reversion of thermomechanical pulps. Holzforschung 48:343–348
Kotilainen RA, Toivanen TJ, Alen RJ (2000) FTIR monitoring of chemical changes in softwood during
heating. J Wood Chem Technol 20:307–320
Kuo M, McClelland JF, Luo S, Chien P-L, Walker RD, Hse C-Y (1988) Applications of infrared
photoacoustic spectroscopy for wood samples. Wood Fiber Sci 20:132–145
Martens H, Næs T (1989) Multivariate calibration. Wiley, Guildford 419 p
Nuopponen MH, Birch GM, Sykes RJ, Lee SJ, Stewart D (2006) Estimation of wood density and
chemical composition by means of diffuse reflectance mid-infrared Fourier transform (DRIFT-MIR)
spectroscopy. J Agric Food Chem 54:34–40
Owen NL, Thomas DW (1989) Infrared studies of hard and softwoods. Appl Spectrosc 43:451–455
Repellin V, Guyonnet R (2003) Evaluation of heat-treated beech by non-destructive testing. First
European conference on wood modification, April 3–4, Ghent, Belgium, Ghent University, pp 73–82
Rouessac F, Rouessac A (2000) Chemical analysis. Modern instrumentation methods and techniques.
Wiley, Chichester, pp 161–188
Sudo K, Shimizu K, Sakurai K (1985) Characterization of steamed wood lignin from beech wood.
Holzforschung 39:281–288
TAPPI (1991) Method um-250: acid-soluble lignin in wood and pulp. In: TAPPI (ed) TAPPI useful
methods. TAPPI, Atlanta, pp 47–48
TAPPI (1994) Method T249 cm-85: carbohydrate composition of extractive-free wood and wood pulp by
gas-liquid chromatography. In: TAPPI (ed) TAPPI test methods. TAPPI, Atlanta 5 p
Teratani F, Miyazaki K (1968) Effect of thermal treatment in wood hemicelluloses. I. The change of
arabinogalactan by heating. Mokuzai Gakkaishi 14:91–97
Thumm A, Meder R (2001) Stiffness prediction of radiata pine clearwood test pieces using near infrared
spectroscopy. J Near Infrared Spectrosc 9:117–122
Tsuchikawa S (2007) A review of recent near infrared research for wood and paper. Appl Spectrosc Rev
42:43–71
Weiland JJ, Guyonnet R (2003) Study of chemical modifications and fungi degradation of thermally
modified wood using DRIFT spectroscopy. Holz Roh Werkst 61:216–220
Widmann R, Beikircher W, Fischer A (2007) Mechanical properties of thermal treated hardwood (beech):
bending and tension strength and stiffness of boards. Third European conference on wood
modification, 15–16 October, Cardiff, UK, The Angel Hotel, pp 187–190
Windeisen E, Strobel C, Wegener G (2007) Chemical changes during the production of thermo-treated
beech wood. Wood Sci Technol 41:523–536
Wood Sci Technol (2011) 45:83–102 101
123
Author's personal copy
Wold S, Anttia H, Lindgrenb F, Ohman J (1998) Orthogonal signal correction of near-infrared spectra.
Chemom Int Lab Syst 44:175–185
Workman JJ (1999) Review of process and non-invasive near-infrared and infrared spectroscopy: 1993–
1999. Appl Spectrosc Rev 34:1–89
Worrall JJ, Anderson KM (1993) Sample preparation for analysis of wood sugars by anion
chromatography. J Wood Chem Technol 13:429–437
102 Wood Sci Technol (2011) 45:83–102
123
Author's personal copy