Identification of Clay Minerals by Infrared Spectroscopy

and Discriminant Analysis

MICHAL RITZ,* LENKA VACULÍKOVÁ, and EVA PLEVOVÁ
VŠB-Technical University Ostrava, 17. listopadu 15, 708 33 Ostrava-Poruba, Czech Republic (M.R.); and Institute of Geonics of the AS CR,
Studentská 1768, 708 00 Ostrava-Poruba, Czech Republic (L.V., E.P.)

Identification of clay minerals based on chemometric analysis of measured Discriminant analysis allows the quantification of the
infrared (IR) spectra was suggested. IR spectra were collected using the probability with which an object can be assigned to an
diffuse reflection technique. Discriminant analysis and principal compo- individual class. The discriminant function, which gives the
nent analysis were used as chemometric methods. Four statistical models
maximum separability for objects from different classes, is
were created for separation and identification of clay minerals. More than
generated by DA. The major purpose of DA is to assign
50 samples of various clay mineral standards from different localities
were used for the creation of statistical models. The results of this study unknown objects to one of several classes. This method can be
confirm that the discriminant analysis of IR spectra of clay minerals could used either for prediction of analyzed samples into predefined
provide a powerful tool for identification of clay minerals. Differentiation classes or for explanation of the differences between classes.6
of muscovite from illite and identification of mixed structures of illite– The sample is classified as a member of the appropriate class
smectite were achieved. predominately according to the value of the Mahalanobis
Index Headings: Clay minerals; Identification; Infrared spectroscopy; IR distance (MD). The MD between spectra from two samples xi
spectroscopy; Discriminant analysis; DA; Principal component analysis; and xj is defined as:
PCA. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
MDij ¼ ðxi  xj ÞT S1 ðxi  xj Þ ð1Þ

INTRODUCTION where xi and xj are the measured spectra and S is the

Knowledge of mineral composition is essential to charac- covariance matrix.7
terize the geochemical and physico-mechanical properties of There are several methods of discriminant analysis described
rocks. Type and content of clay minerals present in rocks have in the literature: Fisher’s linear discrimination function (LDF),
a significant influence on the behavior and properties of rocks linear discriminant analysis (LDA), quadratic discriminant
as well as on the whole rock massif. analysis (QDA), etc. The major difference between the above-
In general, several conventional analytical methods exist that mentioned methods of discriminant analysis is in the number of
can be used to examine the mineral composition of rocks: predefined classes. Furthermore, the predefined classes are
optical microscopy, electron microscopy, X-ray diffraction dependent on the similarity of the covariance matrices. LDF
(XRD), Fourier transform infrared (FT-IR) spectroscopy, can be applied to two classes only; covariance in the classes
simultaneous thermogravimetry and differential thermal anal- must be similar. LDA and QDA can handle more than two
ysis (TG/DTA), and bulk chemistry analysis.1–5 Unfortunately, classes. The covariance in the classes should be similar for
the exact identification of clay minerals by means of these LDA, while QDA can handle very dissimilar covariance
methods is rather complicated and often inaccurate. The main matrices.
analytical difficulties are related to variable chemical compo- The fundamental problem of DA for spectral data is the large
sition and common structural disorders of clay minerals. In number of variables. With a large number of variables it is
addition, the individual clay minerals occur in the form of impossible to apply DA to the ‘‘raw data’’; rather, it is
mixtures (illite–montmorillonite, chlorite–montmorillonite, necessary to carry out PCA followed by DA.8 PCA is one of
etc.) with various ratios of particular clay minerals. the most common multivariate statistical methods. This method
The current Fourier transform infrared (FT-IR) spectroscopy allows ‘‘extraction’’ of important information from a data set
represents a fast, reliable, and efficient tool for mineral (e.g., from IR spectra). The results of PCA are the so-called
analysis. Very useful for processing of IR spectra are principal components (PCs), which make it possible to
multivariate statistical methods. The primary advantage of dramatically reduce the number of variables in the data set.
multivariate statistical methods is their capability to extract Both of the above-mentioned chemometric methods (DA
required spectral information from IR spectra and explore this and PCA) are described in many statistical textbooks9,10 in full
information for qualitative or quantitative applications. The detail. Likewise, DA and/or PCA of IR spectra are used in
most frequently used multivariate statistical methods (often many studies.11–15
called chemometric methods) are factor analysis (FA), This paper describes a chemometric analysis of the IR
principal component analysis (PCA), discriminant analysis spectra of clay minerals by PCA and DA. The results of DA
(DA), principal component regression (PCR), multiple linear can be applied to the identification of clay minerals. PCA was
regression (MLR), and partial least squares regression (PLS). performed to reduce redundant spectral information and find
DA, FA, and PCA are very useful tools for qualitative analysis. the spectral bands important for successful identification of
clay minerals by DA. All IR spectra were collected using the
diffuse reflection technique (DRIFT). The aim of the study was
Received 19 February 2010; accepted 17 September 2010.
* Author to whom correspondence should be sent. E-mail: michal.ritz@ the creation of statistical models for identification of clay
vsb.cz. minerals by discriminant analysis of measured IR spectra. The

0003-7028/10/6412-1379$2.00/0
Volume 64, Number 12, 2010 Ó 2010 Society for Applied Spectroscopy APPLIED SPECTROSCOPY 1379
TABLE I. List of samples.

Sample Mineral Locality Description

Bio1 Biotite Ktı́š, Czech Republic fraction  5 lm

Bio2 Biotite Ktı́š, Czech Republic ‘‘raw’’ standard
Bio3 Biotite Dolnı́ Bory, Czech Republic fraction  5 lm
Mus1 Muscovite India ‘‘raw’’ standard
Mus2 Muscovite Měděnec, Czech Republic fraction  5 lm
Mus3 Muscovite Otov, Czech Republic fraction  5 lm
Mus4 Muscovite Měděnec, Czech Republic ‘‘raw’’ standard
Mus5 Muscovite Rožná, Czech Republic fraction  5 lm
Mus6 Muscovite–Chlorite Měděnec, Czech Republic ‘‘raw’’ standard
Ill1 Illite Hungary fraction  5 lm
Ill2 Illite (IMt-1) Montana, USA fraction  5 lm
Ill3 Illite Karlov, Czech Republic ‘‘raw’’ standard
Ill4 Illite (IMt-2) Montana, USA ‘‘raw’’ standard
Ill5 Illite (IMt-1) Montana, USA jet mill
Ill6 Illite (IMt-2) Montana, USA ‘‘raw’’ standard
Ill7 Illite–Smectite (ISCz-1) Czechoslovakia ‘‘raw’’ standard
Ch1 Chlorite (CCa-2) California, USA fraction  5 lm
Ch2 Chlorite Letovice, Czech Republic fraction  5 lm
Ch3 Chlorite Orlické hory, Czech Rep. fraction  5 lm
Ch4 Chlorite (CCa-2) California, USA ‘‘raw’’ standard
Ch5 Chlorite (CCa-2) California, USA jet mill
Kao1 Kaolinite Božı́čany, Czech Republic ‘‘raw’’ standard
Kao2 Kaolinite Hornı́ Břı́za, Czech Republic ‘‘raw’’ standard
Kao3 Kaolinite (KGa-1b) Georgia, USA fraction  5 lm
Kao4 Kaolinite Kadaň, Czech Republic ‘‘raw’’ standard
Kao5 Kaolinite Kaznějov, Czech Republic ‘‘raw’’ standard
Kao6 Kaolinite Sedlec, Czech Republic ‘‘raw’’ standard
Kao7 Kaolinite Únavov, Czech Republic ‘‘raw’’ standard
Kao8 Kaolinite (KGa-1b) Georgia, USA ‘‘raw’’ standard
Kao9 Kaolinite (KGa-2) Georgia, USA ‘‘raw’’ standard
Kao10 Kaolinite Sedlec, Czech Republic ‘‘raw’’ standard
Kao11 Kaolinite Sedlec, Czech Republic fraction  5 lm
Dic1 Dickite Krásno, Czech Republic fraction  5 lm
Hal1 Halloysite Bı́lá Hora, Slovakia ‘‘raw’’ standard
Mon1 Montmorillonite Ivančice, Czech Republic fraction  5 lm
Mon2 Montmorillonite Jelšovy potok, Slovakia fraction  5 lm
Mon3 Montmorillonite (SWy-1) Wyoming, USA ‘‘raw’’ standard
Mon4 Montmorillonite (SWy-1) Wyoming, USA fraction  5 lm
Mon5 Montmorillonite (SWy-2) Wyoming, USA ‘‘raw’’ standard
Mon6 Montmorillonite (SAz-2) Arizona, USA fraction  5 lm
Mon7 Montmorillonite (SAz-2) Arizona, USA ‘‘raw’’ standard
Mon8 Montmorillonite (STx-1b) Texas, USA ‘‘raw’’ standard
Non1 Nontronite (Nau-1) Uley Mine, South Australia ‘‘raw’’ standard
Non2 Nontronite (Nau-2) Uley Mine, South Australia ‘‘raw’’ standard
Hec1 Hectorite (SHCa-1) California, USA ‘‘raw’’ standard
Ver1 Vermiculite China expanded
Ver2 Vermiculite Aldrich standard expanded
Ver3 Vermiculite Aldrich standard expanded, jet mill
Ver4 Vermiculite Letovice, Czech Republic not expanded
Ver5 Vermiculite Brasil not expanded
Ver6 Vermiculite Brasil not expanded, jet mill

advantage of using DA for identification of clay minerals is, collected for the preparation of statistical models. Some of the
e.g., the ability to identify mixed structures of clay minerals. mineral standards were treated by jet mill; others were prepared
This method of identification has not been reported so far for by sedimentation to obtain a fraction grain size of less than 5
clay minerals. lm. Some standards were used as unmodified (‘‘raw’’
standards). A list of all samples used as mineral standards is
MATERIALS AND METHODS shown in Table I.
Fourier Transform Infrared Measurements. Approxi-
Samples of Minerals. This research focuses on the basic mately 5–10 mg of a mineral standard sample was ground with
clay mineral specimens (kaolinite, montmorillonite, nontronite, approximately 400 mg dried KBr. IR spectra were collected
illite, chlorite, and vermiculite), including micas (muscovite from this mixture.
and biotite). These mineral standards were obtained from the The IR spectra were collected using a Nexus 470 FT-IR
collection of minerals available at the Institute of Geonics, spectrometer (ThermoScientific, USA) with DTGS detector.
Nanotechnology Centre of VŠB-Technical University Ostrava The diffuse reflection measurement technique (DRIFT) was
and from the Source Clays Repository of the Clay Minerals used. The measurement parameters were as follows: spectral
Society, USA. Fifty-one samples of mineral standards were region 4000–400 cm1, spectral resolution 8 cm1; 128 scans;

1380 Volume 64, Number 12, 2010

 FIG. 1. Scheme of identification of clay minerals by discriminant analysis. Note: Na-MMT: Na–montmorillonite; Ca-MMT: Ca-montmorillonite.

and Happ–Genzel apodization. Freshly dried KBr was used for used in one-member-only classes. The selection of calibration
the background measurement. and validation spectra was carried out using TQ Analyst
Every sample was prepared and consequently measured software. The models in this study were considered correct
several times (three to five times). The mean IR spectrum of when all validation spectra occurred inside their class clusters.
every sample was calculated. All discrimination models mentioned in this paper fulfill this
Discriminant Analysis. The linear discriminant analysis condition.
(LDA) was performed using TQ Analyst software (Thermo-
Scientific, USA). Two spectral regions of each IR spectrum RESULTS AND DISCUSSION
(4000–3000 cm1 and 1300–400 cm1) were used for DA. A Strategy of Model Creation. The creation of a unique
one-point baseline was used in both spectral regions. A statistical model for all used samples was a primary goal of the
multiplicative signal correction of the path length was used for authors. During the process of creation of the unique models,
calculation. The Mahalanobis distance was used to formulate however, serious challenges were encountered. Multiple
the distance between clusters. Before DA, PCA was performed minerals were classified into incorrect classes because of
to reduce redundant spectral information. The number of considerable variability in the spectral bands. The solution to
principal components used for the preparation of all models this problem was the creation of partial models; four partial
was 10. models were created in this study. These partial models allow
Validation of statistical models has been carried out by separation and subsequent identification of clay minerals in a
means of a validation spectra set (TQ Analyst software does few steps. A brief scheme of the identification of clay minerals
not allow cross-validation for discriminant analysis). The IR proposed in this study is shown in Fig. 1.
spectra of minerals were split into two groups in each class General discrimination of clay minerals was performed in
used in DA: a calibration group of spectra and a validation Model 1. The clay minerals were classified into particular
group of spectra. The calibration spectra were used to create a groups of clay minerals in this model: biotite, muscovite–illite,
discrimination model; the validation spectra were used to verify chlorite, kaolin–serpentine, smectite, and vermiculite. The
this model. Most of the IR spectra in each class were used as remaining models were created for identification of individual
calibration spectra; up to three IR spectra in each class were members of selected clay mineral groups. Model 2 allows
used as validation spectra. There were no validation spectra identification of minerals of the kaolin–serpentine group

APPLIED SPECTROSCOPY 1381

 FIG. 2. Mahalanobis distance plot of classes muscovite–illite and smectite (Model 1).

(kaolinite, dickite, and halloysite), Model 3 allows identifica- and 1010 cm1), the deformation band of inner-surface
tion of minerals of the smectite group (Na–montmorillonite, hydroxyl groups (915 cm1), Si–O stretching bands (795
Ca–montmorillonite, nontronite, and hectorite), and Model 4 cm1, 755 cm1, and 695 cm1), and deformation bands of Al–
allows identification of micas (biotite and muscovite) and illite. O–Si, Si–O–Si, and Si–O (540 cm1, 470 cm1, and 430 cm1,
Model 1: Identification of Clay Mineral Groups. This respectively). Assignment of IR spectral bands of clay minerals
model was created for classification of clay minerals into according to the literature17–19 was used in this paper. PC1
particular groups of clay minerals. Classification of clay described 79.4% of the variance. PC2 described 10.8% of the
minerals according to Weiss and Kužvart16 was used in this variance, and its loadings spectrum included ‘‘negative’’ bands
study. Model 1 contained six classes (biotite, muscovite–illite, of kaolinite and the ‘‘positive’’ broad and very weak band at
chlorite, kaoline–serpentine, smectite, and vermiculite). In this 3400 cm1 (stretching vibration of water). PC3 described 4.0%
model the first two classes represented the mica group of clay of the variance, and its loadings spectrum contained ‘‘negative’’
minerals; the other classes represented the remaining groups of bands of kaolinite and the ‘‘positive’’ band of the stretching
clay minerals. The mica group had to be split into two classes vibration of water (higher intensity than in PC2).
(biotite and muscovite–illite) because of bad creation of The band at 1645 cm1 (in the PC2 and PC3 loadings
clusters. The biotite class contained only samples of biotite; spectra) belongs to the deformation vibration of water. This
the muscovite–illite class contained samples of muscovite, band very probably represents moisture in the samples. The
illite, and the mixed illite–smectite sample. The chlorite class band at 1440 cm1 (‘‘positive’’ in PC2 and ‘‘negative’’ in PC3
contained only chlorite samples. The kaoline–serpentine class loadings spectra) belonged to carbonates present as impurities
contained samples of kaolinite, dickite, and halloysite. The in the samples of biotite (Bio3) and dickite (Dic1). The weak
smectite class contained samples of montmorillonite, non- intensity of the 1440 cm1 band in the IR spectra confirms only
tronite, and hectorite. The vermiculite class contained only traces of carbonates in these standards. In PC loadings spectra
samples of vermiculite. the intensity of this band rises by reason of significant influence
With the exception of sample Ill7, all minerals were clearly of this band on the variability of spectra. The bands at 1645
classified into the appropriate classes after performance of cm1 and 1440 cm1 were not used for the creation of Model 1.
discriminant analysis. The above-mentioned sample (mixed The loadings spectra of the first three PCs (PC1, PC2, and
illite–smectite structure) was classified between the muscovite– PC3) are shown in Fig. 3.
illite class and the smectite class (Fig. 2). Model 2: Identification of Minerals of the Kaoline–
Ten principal components (PCs) were calculated in Model 1. Serpentine Group. This model was created for identification
These ten PCs described 99.9% of the variance of the spectral of individual members of the kaoline–serpentine group of clay
information, with nearly 95% of the variance captured in the minerals. Kaolinite, dickite, and halloysite were used in this
first three PCs. The loadings spectrum of the first principal study as representatives of the kaoline–serpentine group of clay
component (PC1) includes spectral bands of kaolinite only: minerals. Thus, Model 2 included three classes (kaolinite,
stretching bands of inner-surface hydroxyl groups (3695 cm1, dickite, and halloysite).
3668 cm1, 3652 cm1, and 3620 cm1), the Si–O stretching All minerals were clearly classified into the appropriate
band (1100 cm1), in-plane Si–O stretching bands (1030 cm1 classes after performance of discriminant analysis.

1382 Volume 64, Number 12, 2010

 FIG. 3. Loadings spectra of the first three principal components of Model 1.

Ten principal components (PCs) were calculated in Model 2. minerals. Model 3 included five classes (Na–montmorillonite,
These ten PCs described 99.9% of the variance of the spectral Ca–montmorillonite, nontronite, hectorite, and mixed illite–
information, with nearly 97.5% of the variance captured in the smectite structure).
first three PCs. PC1 described 82.5% of the variance, PC2 All minerals were classified into the appropriate classes after
described 11.3% of the variance, and PC3 described 3.6% of performance of discriminant analysis.
the variance. All three PC loadings spectra included ‘‘negative’’ Ten principal components (PCs) were calculated in Model 3.
bands of kaolinite. The particular loadings spectra differed only These ten PCs described 99.9% of the variance of the spectral
in the intensities of their bands. Bands of the highest intensities information, with nearly 97.5% of the variance captured in the
were in the PC1 loadings spectrum. The PC1 loadings first three PCs. The loadings spectrum of the first principal
spectrum includes all bands of kaolinite as well as the PC1 component (PC1) included spectral bands of montmorillonite
loadings spectrum in Model 1 (but ‘‘negative’’). The bands of only: the stretching band of structural hydroxyl groups (3640
kaolinite visible in the PC2 loadings spectrum were the ones at cm1), the broad stretching band of water (3440 cm1), the
795 cm1, 755 cm1, 695 cm1, 540 cm1, 470 cm1, and 430 deformation band of water (1650 cm1), the band of Si–O
cm1. The bands of kaolinite visible in the PC3 loadings stretching (1040 cm1), the deformation band of the Al–Al–
spectrum were the ones at 1100 cm1, 1030 cm1, 1010 cm1, OH vibration (915 cm1), the deformation band of the Al–Mg–
915 cm1, 695 cm1, 540 cm1, 470 cm1, and 430 cm1. OH vibration (840 cm1), the Si–O stretching band of silica
The band at 1440 cm1 (‘‘negative’’ intensity in PC2 (800 cm1), the coupled out-of-plane vibration band of Al–O
loadings spectrum and ‘‘positive’’ intensity in PC3 loadings and Si–O (625 cm1), and the deformation bands of Al–O–Si
spectrum) belongs to the stretching vibration of carbonates and Si–O–Si (525 cm1 and 470 cm1, respectively). PC1
present as impurities in the sample of dickite (see description of described 86.0% of the variance. PC2 described 7.2% of the
Model 1). This band was not used for the creation of Model 2. variance and its loadings spectrum included ‘‘negative’’ bands
The loadings spectra of the above-mentioned PCs (PC1, at 3640 cm1, 3440 cm1, and 1650 cm1 and ‘‘positive’’ bands
PC2, and PC3) of Model 2 are shown in Fig. 4. at 800 cm1, 625 cm1, 525 cm1, and 470 cm1. PC3
Model 3: Identification of Minerals of the Smectite described 4.3% of the variance and the loadings spectrum
Group. Model 3 was created for identification of individual contained ‘‘negative’’ bands at 3640 cm1, 3440 cm1, 1650
members of the smectite group of clay minerals. Na– cm1, 1040 cm1, 915 cm1, 800 cm1, 625 cm1, 525 cm1,
montmorillonite, Ca–montmorillonite, nontronite, and hectorite and 470 cm1. All bands in the loadings spectra of PC2 and
were used in this study as representatives of this group of clay PC3 belonged to montmorillonite as well.

APPLIED SPECTROSCOPY 1383

 FIG. 4. Loadings spectra of the first three principal components of Model 2.

The band at 1650 cm1 was not used for creation of Model 3 following spectral bands applied to the creation of Model 4: the
because of potential interference intensity of this band by stretching band of structural hydroxyl groups (3600 cm1), the
moisture in the samples. band of Si–O stretching (1010 cm1), the deformation band of
The loadings spectra of the above-mentioned PCs (PC1, the Al–Al–OH vibration (935 cm1), the deformation band of
PC2, and PC3) are shown in Fig. 5. the Al–O–Si in-plane vibration (755 cm1), the Si–O
Model 4: Identification of Minerals in the Mica Group. deformation band (700 cm1), and the deformation bands of
This model was created for identification of individual Al–O–Si and Si–O–Si (535 cm1 and 480 cm1, respectively).
members of the mica group. Biotite, muscovite, and illite were The bands at 3600 cm1, 1010 cm1, 700 cm1, and 480 cm1
used in this study as representatives of this group of clay belonged to biotite; the bands at 3600 cm1, 1010 cm1, 935
minerals. Model 4 included four classes (biotite, muscovite, cm1, 755 cm1, 700 cm1, 535 cm1, and 480 cm1 belonged
illite, and mixed illite–smectite structure). to muscovite and illite. PC1 described 90.6% of the variance.
All minerals were classified into the appropriate classes after PC2 described 6.8% of the variance and its loadings spectrum
performance of discriminant analysis. The Mahalanobis included ‘‘negative’’ bands at 3600 cm1, 935 cm1, 755 cm1,
distance plot for the muscovite and illite classes is shown in 535 cm1, and 480 cm1.
Fig. 6. The band at 1440 cm1 in both PC loadings spectra belonged
The clusters of muscovite and illite are relatively close to to carbonates present as impurities in the sample of biotite (see
each other but they are clearly separated in this plot. The IR description of Model 1). The band at 1440 cm1 was not used
spectra of muscovite and illite can be very similar (on account for the creation of Model 4.
of their similar chemical composition and structure) and The loadings spectra of the above-mentioned PCs (PC1 and
PC2) are shown in Fig. 7.
common identification of these minerals by their IR spectra
is practically impossible.5
Ten principal components (PCs) were calculated in Model 4. CONCLUSION
These ten PCs described 99.9% of the variance of the spectral The discriminant analysis of IR spectra of 51 samples of clay
information, with nearly 97.5% of the variance captured in the mineral standards is presented in this paper. Identification of
first two PCs. The loadings spectrum of PC1 included the individual clay minerals is possible using the four created

1384 Volume 64, Number 12, 2010

FIG. 5. Loadings spectra of the first three principal components of Model 3.

FIG. 6. Mahalanobis distance plot of muscovite and illite classes (Model 4).

APPLIED SPECTROSCOPY 1385

 FIG. 7. Loadings spectra of the first two principal components of Model 4.

FIG. 8. Score plot of the first two principal components of Model 4.

1386 Volume 64, Number 12, 2010

statistical models. Model 1 was created for the splitting of clay ACKNOWLEDGMENTS
minerals into individual groups of clay minerals. Model 2, This work was supported by the Czech Science Foundation Project No. 105/
Model 3, and Model 4 were used for the identification of clay 08/1398, Project No. 105/07/P416, and by Research plan No. AVOZ
30860518.
minerals from the groups of kaoline–serpentine, smectite, and
mica minerals, respectively.
Principle component analysis was not alone sufficient to 1. J. Srodon, Mineral. Mag. 66, 677 (2002).
2. H. Kodama, L. S. Kotlyar, and J. A. Ripmeester, Clays Clay Minerals 37,
make clear separation of samples to appropriate classes. The 364 (1989).
score plot of the first two principle components (PC1 and PC2) 3. C. Vogt, J. Lauterjung, and R. X. Fischer, Clays Clay Minerals 50, 388
of Model 4 is shown as an example (see Fig. 8). Samples of (2002).
biotite and mixed illite–smectite structure are clearly separated 4. S. J. Chipera and D. L. Bish, Clays Clay Minerals 49, 398 (2001).
5. L. Vaculı́ková, New possibilities of identification of clay minerals and
in this score plot. Samples of muscovite and illite do not create micas in sedimentary rocks using infrared spectroscopy with Fourier
separated clusters and samples of these minerals have been transformation (in Czech) (Final Report of Post-Doc Project of Czech
mutually mixed in the score plot. Contrary to PCA, Science Foundation, Ostrava, 2006).
discriminant analysis was effective in the classification of all 6. T. Isaksson and A. H. Aastveit, ‘‘Classification Methods’’, in Handbook of
Vibrational Spectroscopy, J. M. Chalmers and P. R. Griffiths, Eds. (John
the minerals to the appropriate classes. The task of PCA was Willey and Sons, Chichester, 2002), vol. 3, pp. 2113–2121.
rather reduction of redundant spectral information and finding 7. R. D. Maesschalck, D. Jouan-Rimbaud, and D. L. Massart, Chemom.
of significant spectral bands for subsequent discriminant Intell. Lab. Syst. 50, 1 (2000).
8. T. Fearn, ‘‘Discriminant Analysis’’, in Handbook of Vibrational Spectros-
analysis. The most important spectral bands of clay minerals copy, J. M. Chalmers and P. R. Griffiths, Eds. (John Willey and Sons,
found by PCA were vibration bands of hydroxyl groups and Chichester, 2002), vol. 3, pp. 2087–2090.
vibration bands of Si–O and Al–O bonds. Also considerably 9. M. Meloun and J. Militky, Statistical Analysis of Experimental Data (in
significant were bands of kaolinite in Model 1 and Model 2. Czech) (Academia, Praha, 2004), 2nd ed., pp. 301–405.
10. J. Hendl, Review of statistical methods of data processing (in Czech)
Identification of clay minerals by discriminant analysis of IR (Portál, Praha, 2006), 2nd ed., pp. 486–484.
spectra was successful in this study. Differentiation of 11. S. Tesch and M. Otto, Fuel 74, 978 (1995).
muscovite from illite was even achieved using this method. 12. D. B. Dahlberg, S. M. Lee, S. J. Wenger, and J. A. Vargo, Appl. Spectrosc.
51, 1118 (1997).
Muscovite and illite had too similar IR spectra and these two 13. A. Kher, M. Mulholland, B. Reedy, and P. Maynard, Appl. Spectrosc. 55,
clay minerals are almost indistinguishable when common 1192 (2001).
methods of identification by IR spectroscopy are used. The 14. S. Tsuchikawa, K. Inoue, J. Noma, and K. Hayashi, J. Wood Sci. 49, 29
probable reason for a successful differentiation of muscovite (2003).
15. M. Ritz, Vib. Spectrosc. 43, 319 (2007).
from illite was the use of PCA as a part of the discriminant 16. Z. Wiess and M. Kužvart, Clay Minerals–their nanostructure and
analysis. Principle component analysis is able to find the very utilization (in Czech), (Karolinum, Praha, 2005), p. 29.
slight differences in the IR spectra of the muscovite and illite 17. J. Madejová and P. Komandel, Clays Clay Minerals 49, 410 (2001).
standards. Another advantage of discriminant analysis as an 18. J. D. Russel and A. R. Fraser, ‘‘Infrared Methods’’, in Clay Mineralogy:
Spectroscopic and Chemical Determinative Methods, M. J. Wilson, Ed.
identification tool is the ability to identify mixed structures of (Chapman and Hall, London, 1994), Chap. 2, pp. 19–39.
clay minerals (e.g., the illite–smectite mixed structure). 19. L. Vaculı́ková and E. Plevová, Acta Geodyn. Geomater. 2, 163 (2005).

APPLIED SPECTROSCOPY 1387