Critical Reviews in Biotechnology

ISSN: 0738-8551 (Print) 1549-7801 (Online) Journal homepage: http://www.tandfonline.com/loi/ibty20

Compositional analysis of lignocellulosic biomass:

conventional methodologies and future outlook

Daniel J. Krasznai, Rachel Champagne Hartley, Hannah M. Roy, Pascale

Champagne & Michael F. Cunningham

To cite this article: Daniel J. Krasznai, Rachel Champagne Hartley, Hannah M. Roy, Pascale
Champagne & Michael F. Cunningham (2017): Compositional analysis of lignocellulosic
biomass: conventional methodologies and future outlook, Critical Reviews in Biotechnology, DOI:
10.1080/07388551.2017.1331336

To link to this article: http://dx.doi.org/10.1080/07388551.2017.1331336

View supplementary material

Published online: 08 Jun 2017.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=ibty20

Download by: [Cornell University Library] Date: 10 June 2017, At: 07:16
CRITICAL REVIEWS IN BIOTECHNOLOGY, 2017
https://doi.org/10.1080/07388551.2017.1331336

REVIEW ARTICLE

Compositional analysis of lignocellulosic biomass: conventional

methodologies and future outlook
Daniel J. Krasznaia, Rachel Champagne Hartleya, Hannah M. Royb, Pascale Champagnea and
Michael F. Cunninghama
a
Department of Chemical Engineering, Queen’s University, Kingston, Ontario, Canada; bDepartment of Civil Engineering & Department
of Chemical Engineering, Queen’s University, Kingston, Ontario, Canada

ABSTRACT ARTICLE HISTORY

The composition and structural properties of lignocellulosic biomass have significant effects on Received 16 September 2016
its downstream conversion to fuels, biomaterials, and building-block chemicals. Specifically, the Revised 8 April 2017
recalcitrance to modification and compositional variability of lignocellulose make it challenging Accepted 8 April 2017
to optimize and control the conditions under which the conversion takes place. Various charac-
terization protocols have been developed over the past 150 years to elucidate the structural KEYWORDS
properties and compositional patterns that affect the processing of lignocellulose. Early character- Biorefinery; characterization;
ization techniques were developed to estimate the relative digestibility and nutritional value of chemometrics; compos-
plant material after ingestion by ruminants and humans alike (e.g. dietary fiber). Over the years, itional analysis;
these empirical techniques have evolved into statistical approaches that give a broader and more lignocellulose; spectroscopy
informative analysis of lignocellulose for conversion processes, to the point where an entire com-
positional and structural analysis of lignocellulosic biomass can be completed in minutes, rather
than weeks. The use of modern spectroscopy and chemometric techniques has shown promise
as a rapid and cost effective alternative to traditional empirical techniques. This review serves as
an overview of the compositional analysis techniques that have been developed for lignocellulo-
sic biomass in an effort to highlight the motivation and migration towards rapid, accurate, and
cost-effective data-driven chemometric methods. These rapid analysis techniques can potentially
be used to optimize future biorefinery unit operations, where large quantities of lignocellulose
are continually processed into products of high value.

Introduction heteropolymer that can be found as a binding agent

Lignocellulose is derived from the cell walls of plants
and is composed of cellulose (30–50%), hemicellulose
(15–35%), and lignin (10–30%) [1]. Cellulose is a linear
polysaccharide comprising b(1 ! 4) linked glucose units
with a high degree of hydrogen bonding that makes a
significant contribution to the structural integrity and
recalcitrance to modification of the lignocellulose com-
posite. Cellulose is found in various macromolecular
conformations that can be generally classified as crystal-
line or amorphous, where crystalline cellulose poses a
greater energy penalty to hydrolyze than amorphous
cellulose. Hemicellulose is a branched heteropolysac-
charide of both pentose and hexose units that interact
with cellulose chains and can sometimes be found as a
crosslinking material between cellulose and lignin,
mainly imparting enhanced structural stability to the
cell wall [2]. Finally, lignin is a complex aromatic
along the outside of the lignocellulose microfibril, pro-
viding structural rigidity and protection to the energy-
rich cellulose fibers [3,4]. Lignin comprises b-aryl ether,
biaryl ether, phenylcoumarane, biphenyl, pinoresinol, or
diaryl propane linked p-coumaryl, coniferyl, and sinapyl
alcohol units [4]. Because of its highly branched and
irregular structure formed via random oxidative radical-
ization of phenols [3,5], lignin poses a formidable barrier
to digestion via enzymatic routes. While enzymes cap-
able of catalyzing cellulose hydrolysis (e.g. cellulase) are
produced by a wide variety of prokaryotes, eukaryotes,
insects, and even crustaceans, the layers of lignin sur-
rounding the cellulose fibers serve to protect the sugar-
rich constituents from rapid degradation. Despite this
protective role of lignin, a number of microorganisms
are capable of lignin degradation; for example, white-
rot fungi produce lignin peroxidase and manganese

CONTACT P. Champagne pascale.champagne@queensu.ca Department of Civil Engineering & Department of Chemical Engineering, Queen’s
University, Kingston, Ontario, K7L 3N6, Canada
Supplemental data for this article can be accessed here.
ß 2017 Informa UK Limited, trading as Taylor & Francis Group
2 D. J. KRASZNAI ET AL.
peroxidase enzymes with lignolytic activity [4].
Interestingly, a number of these microorganisms with
lignolytic activity seem to function in a symbiotic rela-
tionship with the host plant [4]. Human beings, on
the other hand, have discovered that lignin is the pri-
mary hindrance to the facile production of fuels and
chemicals from lignocellulosic materials, including
bioethanol [3].
The characterization of raw materials, whether for
composition, structure, or some other salient property,
is a fundamental first step in most experiments and
industrial conversion processes. As reviewed by
FitzPatrick et al., a promising area of research and indus-
trial interest is biorefineries utilizing lignocellulosic
biomass for the production of value-added chemicals
and fuels from non-food crops [6]. These lignocellulosic
biorefineries seek to capitalize on the benefits of an
inexpensive and renewable feedstock as opposed to
more traditional processes that utilize finite petroleum
sources. However, a thorough characterization protocol
is required prior to the conversion of lignocellulosic bio-
mass because of its highly variable and recalcitrant
composition. These properties ultimately limit the
range of products that can be derived from a particular
lignocellulosic feedstock in a cost-efficient manner.
Furthermore, the composition and structure of the
lignocellulosic feedstock serve a critical role in closing
the mass balance of the process and also have a large
influence on the efficacy of pretreatment and subse-
quent processing steps [7]. Because the most expensive
aspects of lignocellulose conversion depends on the
nature of the raw material, accurate and rapid front-end
biomass characterization – either in the field where the
biomass is grown or on the processing line where the
conversion takes place – is absolutely necessary for an
economically feasible biorefinery.
There are many options to consider with respect to
characterizing lignocellulose, but very few published
papers discuss their relative advantages and disadvan-
tages with respect to industrial applications. For
example, Giger-Reverdin reviewed the various protocols
available at the time for estimating cell wall compos-
ition and digestibility of lignocellulosic materials, but
presented the review from a ruminant nutrition per-
spective [1]. However, one exception is a thorough
review paper recently published by Sluiter et al. [8].
Sluiter et al. reviewed the use of sulfuric acid to affect
various stages of cell wall dissolution during the sum-
mative compositional analysis of lignocellulosic feed-
stocks [8]. The authors also mentioned the limitations of
these methods as well as some suggested improve-
ments, most of which will be discussed in the para-
graphs below [8]. In addition, Sluiter et al. mention that
multivariate calibration methods involving near infrared
(NIR) can expedite the compositional analysis process
[8]. While some univariate analytical techniques are well
suited to the laboratory environment, other multivariate
approaches are more robust and can be implemented
in a real-time process control scenario over a wide array
of feedstock compositions and product formulations.
The following review will serve as a comprehensive
examination of the analytical methods available for the
characterization of lignocellulosic biomass, its deriva-
tives, and their suitability for application in biorefinery
processes. Furthermore, it will be shown that an
increase in research focused on biorefinery-relevant
Process Analytical Technologies (PAT) and data process-
ing techniques could present rapid and inexpensive
alternatives for experimental purposes and quality
assurance in both research and industrial settings.
Historical perspective on the compositional
analysis of lignocellulose
Contemporary methods for the compositional analysis
of lignocellulosic materials have been derived from
those developed to assess the digestibility of plant-
based animal feed. A parallel interest in the assessment
of the chemical composition of lignocellulosic materials
has arisen from the practical needs related to the pro-
duction of renewable fuels, chemicals, and materials.
Both of the aforementioned metrics, nutrient require-
ments and chemical composition, have served as the
basis for empirical studies on lignocellulosic biomass
for over 150 years. The progress and evolution of con-
ventional lignocellulosic characterization methods is
summarized in Table 1 and a chronological timeline is
provided in Figure 1, while a more detailed description
of these methods is provided in the Supplementary
information.
New opportunities: chemometrics and
spectroscopy for lignocellulose
characterization
Introduction
For a complex molecule such as lignocellulose, the pri-
mary hindrance to an expedient and accurate compos-
itional analysis ultimately depends on the expedience
and efficiency of the following criteria:
 The lignocellulose fractionation or separation pro-
cess and
 The analytical methods used to analyze the compo-
nents following fractionation.
 CRITICAL REVIEWS IN BIOTECHNOLOGY 3

Table 1. Historical perspectives on lignocellulosic biomass characterization.

Method
Weende method Proposed by Hennerberg and Stohmann for quantifying crude fiber [9–12]. Modified for the analysis of fiber
in food, the Weende method can yield very inaccurate results [13,14]. There are limitations associated
with the complexity of fractionation and analysis of the substrate of interest, as well as the variability of
lignin [10,15]; that being said this method is still routinely used in agriculture [16].
Klason lignin method Klason published two papers regarding the analysis of black liquor from the Kraft pulping process in 1893
followed by development of his techniques over the next 40 years [17]; these techniques are still used
for the quantification of lignin in biomass [18,19]. The Klason technique has been accepted as a TAPPI
standard, but its limitations make analysis complicated [20,21]; however, modifications have been made
for better analysis [23,24]. This method is still used rigorously today [25–27].
USDA FPL methods The USDA Forest Products Laboratory (FPL) applied the Klason lignin methods to wood [28] over the next
20 years, and adopted the methods of Saeman et al. [29] for the saccharification of wood and cellulose
in their research [29–32].
Dietary fiber Hipsley coined the term “dietary fiber” in 1953 to describe plant cell wall materials composed of cellulose,
hemicellulose, and lignin that are indigestible by the human digestive tract [33,34]. Figure 2 shows this
relationship. This term became a basis for future work in the agricultural and food sciences.
Saeman pentosan method Approximately 10 years following these applications, Saeman et al. optimized the chromatograph techniques
that had been adopted as a TAPPI standard for pentosan determination [35], and have recently been
applied in biorefinery for the quantitative determination of carbohydrate and lignin in wheat and corn
stover [31,36].
Monoethanolamine method Nelson and Leming optimized the MEA method for the gravimetric quantification of cellulose after the
removal of lignin and other carbohydrates; they determined the optimal conditions for this process, and
its usefulness with application to three types of agricultural residues [37–41].
Permanganate method Tasman and Berzins [48] subsequently developed a permanganate method for lignin to measure the Kappa
number of wood in the pulp and paper industry [48–50], however, its limitations have resulted in min-
imal use
Detergent fiber methods A detergent fiber method was also developed [52,53,55,56], used for the fractionation and compositional
analysis of plant-based materials, including lignocellulose; these included neutral and acid detergent fiber
analyses (NDF and ADF, respectively). Plant material that may be quantified with such are summarized
elsewhere [14,57–61].
Moore and Johnson Method (modified Moore and Johnson developed a method for the analysis of pulp and wood sugars at the USDA FPL [63,64].
USDA FPL) This includes the hydrolosis and subsequent conversion of pentoses to furfural, followed by quantification
using the orcinol techniques [7]. The Moore and Johnson method was found to underestimate pentose
content [7].
Trifluoroacetic acid method Developed by Fengel and Wegner [65] for the determination of cellulose and hemicellulose by proxy of
their hydrosylate sugars; two methods were described for materials with high and low lignin content
[65]. This method has been applied in a few studies, however has rarely been applied to lignocellulosic
biorefinery research [66,67].
Grohmann method Development of this method involved the modification of the Moore and Johnson methods [63,64] for the
use of a milder acid hydrolysis technique. This method has rarely been used in recent years [68,69].
Prosky dietary fiber method Prosky et al. [70] developed an enzymatic and chemical dietary fiber method that allowed retention of the
cell wall, allowing for the quantification of cell wall components; the technique has been designated as
an AOAC official method [70,71].
Uppsala method This method was developed by Theander [72] to estimate the composition of cell wall components using a
chemical approach and a variety of techniques [14,50,72–75]. It was adopted as an AOAC method, follow-
ing inter-laboratory studies by Milne et al. [76] and Theander et al. [75], for the quantification of soluble
and insoluble polysaccharides. This method is, however, susceptible to contamination and overestimated
Klason lignin values [50].
Acetyl bromide (Derivatization followed by This method stands out as a consistent method for determining the digestibility of lignin, and overcomes
reductive cleavage) method some of the issues associated with some of the previous methods [15,50,51]. In this method lignin is dis-
solved and brominated, followed by reductive cleavage and acetylation [77]. The products are generally
quantified using mass spectrometry. This method is however limited by xylan degradation [77,78].
Nrel methods (laboratory analytical Methods commonly used for the characterization of lignocellulosic biomass for engineering applications
procedures) have been the standards developed by the National Renewable Energy Laboratory (NREL); a number of
laboratory analytical procedures (LAPs) have been developed for the determination of structural carbohy-
drates and lignin in biomass in recent years [79–83]. These techniques, as well as some similar techni-
ques that have been altered slightly, have been used by a number of researchers for the compositional
analysis of lignocellulosic biomass [36,63,64,68,84–93].

NIR, mid-infrared (MIR) absorption and Raman scat-
tering spectroscopy have been used to characterize a
diverse range of materials, since the required instru-
mentation is inexpensive, expedient, and gives a broad
spectrum of compositional data for macromolecules.
Under most circumstances, scientists are more inter-
ested in comparing lignocellulosic materials rather than
determining absolutes based on thorough compos-
itional analyses [14]. As a result, the proliferation of
calibration techniques involving NIR, MIR, Raman, and
even NMR spectroscopy has enabled large quantities of
compositional and structural data to be acquired in
real-time. Furthermore, technology developments may
lead to an increase in the use of NMR in lignocellulosic
biorefineries (e.g. Benchtop NMR from Nanalysis). The
rapid accumulation of compositional information from
spectroscopic techniques can be further supplemented
through the use of databases and web groups that are
 4
D. J. KRASZNAI ET AL.

Figure 1. Chronology of conventional methods for the compositional analysis of lignocellulosic biomass.

Figure 2. Definitions of dietary fiber fractions (Reproduced with permission from Monro and Burlingame [157]).
congruent to the latest collaborative efforts in the field
of chemometrics [94]. In fact, NIR is a relatively mature
technology and has been used for in situ measurement
of samples since the late 1970 s [95]. The use of Fourier
Transform Infrared (FTIR) and Fourier Transform Raman
(FT-Raman) spectroscopy has also been applied in bio-
medical research to identify disease patterns in human
tissue [96,97]. While NIR, FTIR, and FT-Raman are gener-
ally useful in material science, they may also be used as
a nondestructive tool in the field of archeology to
obtain compositional information of artifacts, which is
required if the item is to be preserved without causing
undue damage or modification [98,99]. In fact, infrared
instrumentation coupled with multivariate statistical
techniques, which will be discussed in the paragraphs
below, has been widely utilized for characterization and
quality control purposes in pulp and paper applications
[100–104]. Reviews have also been published on
the use of infrared and/or Raman spectroscopy for
the analysis of lignocellulosic materials such as plant
materials [105], wood [106], and carbohydrates [107].
Furthermore, Adapa et al. [108] published a review on
the use of infrared and Raman spectroscopy and micro-
spectroscopy for the compositional analysis of agricul-
tural biomass at great spatial resolution. Likewise,
Gierlinger and Schwanninger [109] reviewed the use of
near infrared FT-Raman, resonance Raman, and surface-
enhanced Raman spectroscopy for the chemical and
structural analysis of lignocellulosic materials. As
reviewed by Gierlinger and Schwanninger, Raman imag-
ing can also provide valuable information on the spatial
distribution of various compounds that comprise the
cell-walls of lignocellulosic biomass down to the sub-
micron level [109]. Thus, the quantity and distribution
of cellulose, hemicellulose, and lignin can be elucidated
using infrared absorption or Raman scattering micros-
copy and imaging techniques. The use of quantitative
31
P-NMR in wood chemistry was reviewed by
Argyropoulos [110]; however, the aim of most NMR
approaches is to elucidate minute structural differences
in the lignocellulose polymers. Thus, it is unlikely that
NMR approaches will see widespread adoption as PAT
CRITICAL REVIEWS IN BIOTECHNOLOGY 5
in modern lignocellulosic biorefineries due to their
expense and complexity of interpretation.
It is clear that infrared, Raman, and to some extent
NMR spectroscopy, have shown promise as rapid and
accurate methods for the analysis of numerous materi-
als. However, the main problem associated with these
techniques can be attributed to the convoluted nature
of the spectra when analyzing mixtures, where difficul-
ties arise in accurately choosing peaks associated with
chemicals of interest. However, over the past 20 to
30 years in the published literature, the use of chemo-
metric techniques have increased. Chemometrics is a
term used to describe the science of applying mathem-
atical algorithms – often implemented using computer
software – to extract useful information from large or
irregular chemical datasets. For example, the phenolic
hydroxyl groups in milled wood lignins were deter-
mined using FTIR and chemometrics (i.e. Partial Least
Squares (PLS) and Principal Components Regression
(PCR)) that were calibrated using aminolysis as the refer-
ence technique [111]. Likewise, Tsuchikawa [112]
reviewed the application of NIR spectroscopy to wood
and paper, which included a discussion on the use of
chemometrics to extract information from infrared spec-
tra. In essence, Tsuchikawa [112] demonstrated that
spectroscopy and chemometrics have applications in
the pulp and paper industry where rapid monitoring
and analysis can improve the process economics.
The application of multivariate statistics in industry
has improved process control and economics, and prod-
uct diversity [113–115]. The proliferation of spectro-
scopic instruments in industry has been aided by the
application of multivariate statistic approaches to large
spectral datasets. For example, Principal Component
Analysis (PCA) can be used to segregate populations of
samples based solely on spectral data, allowing for high
throughput compositional screening of various biomass
populations [116,117]. Likewise, PCR and PLS regression
can be used to rapidly and accurately determine the
composition of lignocellulosic biomass using a well-cali-
brated mass, absorption, or vibrational spectroscopy
model [62,118,119]. The essential motive behind PLS is
6 D. J. KRASZNAI ET AL.
Figure 3. Structure and use of PCA and PLS multivariate datasets applied to lignocellulosic biomass.
to model the hidden structure of two datasets via projec-
tions onto a reduced or latent variable space [120].
Figure 3 provides an illustration of the datasets required
(X spectral data matrix and Y response data matrix) for
the compositional analysis of lignocellulose using PCA
discrimination and PLS regression approaches. A full
mathematical development of the PLS-regression algo-
rithm is available elsewhere [120].
PLS regression produces a predictive model while
PCA simply highlights the differences within a given
dataset. As a result, PCA is often used to simply discrim-
inate between groups of samples based on spectral
information; however, it can be adapted to produce a
regression model. To improve predictability, multivari-
ate calibration models (e.g. PLS) often involve a data-
preprocessing step whereby the data are transformed
via one or more methods. Some of the more common
preprocessing techniques will be discussed below.
Data pre-processing
Mean centering and unit variance scaling
Mean centering and unit variance scaling are ubiquitous
with multivariate calibration methods and are often
referred to as auto-scaling. Mean centering a sample’s
IR, NMR, or mass spectrum can be accomplished by sub-
tracting the mean intensity value from the intensity at
each independent variable. This particular sample-based
mean centering procedure is outlined mathematically in
Equation 1, where k represents the total number of
infrared spectral variables used in the model.
!
MC;row 1 X K
xi;k ¼ xi;k  xi;k (1)
K k¼1
where xi;k  absorbance value of ith sample and kth
spectral variable (wavenumber).
Likewise, the spectra of multiple samples can be
mean centered based on their individual variables (i.e.
wavenumbers) as described in Equation 2.
!
MC;column 1 X i
xi;k ¼ xi;k  xi;k (2)
i i¼1
where xi;k  absorbance value of ith sample and kth
spectral variable (wavenumber).
In summary, Equation 1 operates on X’s rows while
Equation 2 operates on X’s columns.
Graphically, mean centering the dataset will center
the data about the origin, thus eliminating the intercept
term in the regression model [121]. In this way, the
mean centered model becomes easier to interpret since
all of the samples in the dataset are centered at the ori-
gin. Extra caution must be exercised for regression
models involving time-series data since the mean will
drift over time, thereby introducing bias in the dataset.
Therefore, a new mean may need to be calculated at a
set time interval for monitoring processes like batch
polymerizations using multivariate techniques [122].
Unit variance scaling can also be applied to the rows

(samples) or columns (variables) of a dataset. The tech-
nique is analogous to mean centering: each data point
is divided by either the standard deviation of the
sample’s spectrum or the standard deviation of a single
variable for all samples.
Standard normal variate (SNV)
Row-operating unit variance scaling is often employed
in conjunction with mean scaling in what is referred to
as standard normal variate (SNV) scaling. Applying SNV
scaling to spectra reduces particle size (multiplicative)
effects that are especially troublesome for solid mix-
tures [123]. During SNV scaling, each sample spectrum
is first mean centered as shown in Equation 1. The
standard deviation of the sample spectrum is then cal-
culated and the mean centered data is then divided by
this value. Equation 3 describes the SNV procedure in
mathematical terms.
X
K
xi;k  K1 xi;k
k¼1
SNV
xi;k ¼ (3)
s
where xi;k  ith sample comprising kth spectral variable
(wavenumber), s  calculated standard deviation of all
k variables of the ith spectrum.
Spectra treated using SNV have a mean of zero and a
variance equal to unity. Detrending is sometimes used
in conjunction with SNV-treated data since the latter
are still susceptible to baseline curvature issues, which
can be corrected by the former.
Smoothing
Savitzky–Golay smoothing is an averaging algorithm
that fits a polynomial to the dataset and then predicts
the point of interest from the resulting polynomial
equation. Savitzky–Golay smoothing is often applied in
conjunction with first and second derivatives [124].
First derivative
Taking the first derivative of a spectrum gives the slope
of the curve at each data point. Taking the first deriva-
tive of a spectrum can reduce baseline offset and linear
baseline issues; however, it has been noted to cause a
shift in the characteristic peaks, thus making the result-
ing spectrum difficult to interpret using peak-picking
techniques. Nkansah et al. preprocessed their raw NIR
spectra using Savitzky–Golay first derivatives (i.e. 15
smoothing points, second order polynomial fit) in order
to improve the quality, interpretation, and predictive
ability of the PLS model [116].
CRITICAL REVIEWS IN BIOTECHNOLOGY 7
Second derivative
Taking the second derivative of a spectrum gives the
change in the slope of the curve at each data point.
Similarly, to first derivative, second derivative spectra
remove baseline offset and linear baseline issues. In
addition, second derivative spectra retain the position
of peaks so interpretation is generally easier than first
derivative spectra. It should be noted that both first and
second derivatives might intensify noise in the model.
In addition to first derivative preprocessing, Nkansah
et al. also preprocessed their raw NIR spectra using
second derivatives although no significant improve-
ment to the predictive ability of the PLS regression
model was observed [116]. Krongtaew et al. applied
Savitzky–Golay second derivatives (i.e. 17 smoothing
points, second order polynomial fit) to their raw NIR
spectra, thus improving the predictive ability of their
PLS regression models [125].
Multiplicative signal correction
Multiplicative Signal/Scatter Correction (MSC) is a
method that can be used to reduce multiplicative and
additive scatter effects in a dataset, such as those
caused by path length variations, offset shifts, interfer-
ences, and particle size effects [123]. Mathematically,
MSC is a transformation that first involves regressing
the sample spectrum against an “ideal” spectrum, often
estimated as the set-mean-spectra, ½strnsx, to fit
parameters that describe additive and multiplicative
effects [126]. The additive and multiplicative effects are
estimated from the ordinary least squares solution to
Equation 4.
xi ¼ ai 1 þ bi ½strnsx þ ei (4)
where
xi  vector of spectral absorbances of the ith sample,
ai  regression coefficient describing additive effects,
1  vector of 1’s introduced for formality,
bi  regression slope describing multiplicative effects,
½strnsx
Xi  set-mean-spectra for all i samples,
1
i x ,e  regression model error.
i¼1 i i
Once the intercept (ai) and slope (bi) parameters
have been determined, each sample spectrum is cor-
rected as shown in Equation 5.
ðxi  a^iÞ
xMSC ¼ (5)
i ^i
b
where
xMSC
i  MSC – corrected vector of spectral absorbances
of the ith sample,
^ i  estimated regression coefficient describing addi-
a
tive effects,
8 D. J. KRASZNAI ET AL.
^ i  estimated regression slope describing multiplica-
b xi;k  absorbance at wavenumber k of the ith sample,
tive effects.
The similarities between MSC and SNV can be noted
by considering Equations 3 and 5 as well as the discus-
sion presented by Dhanoa et al. [123]. A more detailed
mathematical description of the MSC and Extended
MSC (EMSC) techniques are described elsewhere
[126–128]. Wallb€acks et al. (1991) found that MSC pre-
processing of their raw NIR spectra slightly improved
the predictability of their PLS regression models [129].
However, Nkansah et al. also preprocessed their raw NIR
spectra using MSC although no significant improvement
to the predictive ability of the PLS regression model
was observed [116]. Liu et al. applied EMSC to their raw
NIR spectra in order to remove physical interferences
and to subsequently improve the predictive ability of
their PLS regression model [130].
Orthogonal signal correction
Orthogonal Signal Correction (OSC) is a data prepro-
cessing technique that removes information from X
that is unrelated (orthogonal) to Y. OSC is analogous to
the PLS algorithm (NIPALS) except that OSC attempts to
minimize instead of maximize the covariance between
X and Y [131]. Thus, each OSC component will remove
information from X that shares minimal covariance
(maximal orthogonality) with Y. Caution should be exer-
cised when choosing the number of OSC components
followed by PLS: if too many OSC components are used,
the result will approach the multiple linear regression
(MLR) solution and overfitting may take place [131].
A full mathematical description of the OSC process is
outlined elsewhere [131].
Baseline correction
Baseline correction seeks to minimize linear offset in
the spectral data by subtracting either a minimum or
user-defined value from the dataset. Similarly, con-
structing a line between two user-defined values can
impose a linear baseline. For spectral data, a linear base-
line is generally within a fingerprint region and each
subsequent wavenumber in the spectrum is normalized
accordingly. For example, the general form of baseline
correction is to subtract a baseline absorbance value,
which is a function of the wavenumber, from the corre-
sponding absorbance value at the same wavenumber
for the spectrum to be corrected, as described by
Equation 6.

xi;k ¼ xi;k  f ðkÞ
where
xi;k  baseline corrected absorbance at wavenumber k
of the ith sample,
f ðkÞ  absorbance value corresponding to some func-
tion of the wavenumber, k, for sample i (e.g. min(k) for
traditional baseline offset correction or a linear function
of the form mk þ b).
Detrending
Detrending is a spectral correction technique, similar to
and often used in conjunction with SNV, that attempts
to reduce nonlinear trends by fitting a polynomial
equation to each spectra that is later used for baseline
correction. The detrended spectra are calculated as the
difference between the original spectrum and the poly-
nomial equation describing the new baseline. Other
data pretreatment or preprocessing techniques are also
available, such as various forms of normalization (e.g.
minimum/maximum spectral normalization [125]),
but for the sake of brevity they will not be discussed
here. Following the acquisition, preprocessing and
model building exercises, a technique known as Cross
Validation (CV) is implemented to optimize the predict-
ive capability of the multivariate model.
Cross validation
CV is often used to determine the optimum number of
principle components to include in a multivariate
model. There is currently no consensus on when to
apply a particular CV procedure. The default CV proced-
ure in ProSensus MultiVariate is to split the calibration
data into seven segments or groups with k samples
per segment (i.e. the data is divided such that
approximately 14% of the data is in each group). The
CV procedure then leaves out one of the seven seg-
ments (k samples) for validation of the regression
model, which is built using the remaining six segments
(i.e. 86% of the data calibrates the model). Other
software such as CAMO Unscrambler X allows for the
specification of the number of segments and samples.
Hames et al. [132] used a full CV procedure, a leave-
one-out technique (i.e. each sample is its own segment
or group), which is not recommended, as this procedure
tends to perform well for calibration but poorly for pre-
diction [133,134]. As the number of samples increases,
the calibration model appears to perform better under
the leave-one-out CV procedure because there are
increasingly more samples used for calibration but the
same number of samples (i.e. one) used for validation.
(6)
Thus, the leave-one-out method often results in overfit-
ting of the data and an underestimation of true predict-
ive error [134]. Alternatively, a larger number of samples
to be left out for CV, nv, could be selected such that

nv =n ! 1 as n ! 1 [133]. However, the leave-nv-out
procedure can be computationally taxing in its search
for an optimal number of components with such a
limited number of samples from which to build the
calibration model. An alternative method called
Monte Carlo CV was developed and described by
Xu and Liang [134]; however, this particular CV tech-
nique has not been extensively used for studies involv-
ing lignocellulose. In fact, a recent study by Krongtaew
et al. [125] used FT-NIR (wet-chemistry as reference),
PLS and leave-one-out CV to predict and monitor the
chemical pretreatment and delignification of wheat and
oat straw. The leave-one-out technique was selected
due to the small number of samples used for the cali-
bration set. The authors were able to extract quantita-
tive information regarding total residual lignin content,
enzymatically released reducing sugars, total solids,
volatile solids, and biogas yield using the FT-NIR PLS
regression model [125].
As a rule of thumb, it is generally advisable to leave
approximately 10% of the data out of the predictive
model for CV purposes. Clearly, there is an inherent vari-
ability in multivariate model building since so many
data-preprocessing techniques and CV protocols cur-
rently exist, each with its own strengths and weak-
nesses. Thus, it is imperative that future studies
involving lignocellulose and chemometrics thoroughly
describe and support the methods used in the develop-
ment of the model in addition to reporting the Root
Mean Square Error of Estimation/Calibration (RMSEE)
Figure 4. General framework for the development of a multivariate model.
CRITICAL REVIEWS IN BIOTECHNOLOGY 9
and Root Mean Square Error of Prediction (RMSEP) – the
latter being the best indicator of predictive perform-
ance. Figure 4 provides a summary of the approach for
producing a PCA or PLS compositional analysis using
spectroscopy. A valuable mini-review was presented by
Kalivas [135] on the development of multivariate cali-
bration models and the types of problems one might
encounter.
Calibration model development
Similarly, to univariate techniques, the performance of
multivariate regression models is largely determined by
the quality of the calibration samples. Thus, multivariate
calibration should be implemented using ASTM
E1655–05 [136], as well as ASTM E2056–04 [137] if sur-
rogate mixtures are used to build the calibration set. If
the multivariate model involves three variables or less
in Y, the response matrix, then there should be a min-
imum of 24 sample spectra in X used for calibration
purposes (following outlier removal) [136]. Alternatively,
if the model is developed using more than three varia-
bles (k > 3) in Y, then X should contain at least 6k spec-
tra following outlier removal and 6(k þ 1) spectra if the
dataset is mean centered [136]. Some types of outliers
can be difficult to identify. Raw and pre-processed data
should always be examined, but in some cases the PCA
model itself can provide further information, often
through closer examination of the scores and the resid-
uals. A detailed discussion on data processing and out-
liers in PCA is provided in Bro and Smilde [138].
10 D. J. KRASZNAI ET AL.
Surrogate mixtures can be used to model a small subset
of the chemical constituents in a material, for example,
cellulose, hemicellulose, and lignin for lignocellulosic
biomass. The surrogate mixture method vastly simplifies
experimental design by reducing the number of sam-
ples required for model calibration. The use of surrogate
mixtures for calibrating a PLS model based on FT-NIR
spectra of pharmaceutical tablets was studied by
Cournoyer et al. [139] It showed that an accurate pre-
dictive model could be constructed for tablets contain-
ing various concentrations of acetylsalicylic acid,
caffeine, microcrystalline cellulose, sodium lauryl sulfate
and starch, thus affording a new avenue for PAT in the
production process. Despite the success of surrogate
mixtures for calibrating chemometric models, the use of
these techniques for lignocellulose biomass applications
has been limited to date. Thus, NREL wet-chemical
methods have been used for calibration, whereby spec-
tral data is regressed against wet-chemical data to form
a predictive regression model [132]. Alternatively, the
detergent fiber methods of Van Soest [53] and Van
Soest and Wine [54] have been recently used by Chen
et al. [62] to calibrate a PLS model of FTIR spectra on
wood samples. Although the authors were able to
develop an adequate calibration model for the predic-
tion of cellulose, hemicellulose and lignin, the incorpor-
ation of nine principal components in the final model
could be considered to be overfitting the model. It is
generally advisable that a minimum number of principal
components be selected to maximize the modeled
trend (and minimize modeled error). Overfitting of prin-
cipal component models can be avoided by using
proper regularization, where the amount of regulariza-
tion can be determined using cross-validation or prob-
abilistic methods from which the extra penalty terms
can be derived naturally from probabilistic models.
This topic is discussed comprehensively by Ilin and
Raiko [140].
Recent applications of chemometrics in
lignocellulosic biomass research
Pyrolysis molecular beam mass spectrometry (pyMBMS)
was studied by Tuskan et al. [118] as a rapid method for
the compositional analysis of lignocellulosic biomass.
The authors showed that PLS regression could be
applied to pyMBMS data to obtain accurate and rapid
estimates of specific gravity, extractives, galactan,
mannan, xylan, alpha cellulose, and lignin contents in
loblolly pine. Arabinan concentrations could not be
quantified due to the low concentrations present in lob-
lolly pine as well as difficulties in estimating arabinan
using standard wet-chemical practices [118]. Full CV
was applied to the pyMBMS PLS regression model to
determine the optimal number of components. Data
preprocessing was not mentioned. While Tuskan and
coworkers successfully used pyMBMS in their study,
infrared spectroscopy tends to be the predominating
technique described in literature for the compositional
analysis of lignocellulosic materials.
Liu et al. [130] developed a broad FT-NIR model for
the compositional analysis of three different lignocellu-
losic feedstocks: corn stover, switchgrass and wheat
straw. Glucan, xylan, galactan, arabinan, mannan, lignin,
and ash were selected as the analytes of interest and a
calibration model was developed using PLS regression
with EMSC data pretreatment to remove physical inter-
ferences in the spectra such as particle size effects.
The experimental procedure used by Liu et al. [130] is
illustrated in Figure 5.
Samples of corn stover and switchgrass were sub-
jected to conventional wet-chemical fractionation and
analysis as well as FT-NIR spectroscopy. The calibration
set comprised the wet chemical data while the predic-
tion set comprised the FT-NIR spectra. Wheat straw was
used to verify the robustness of the model that was
developed using two dissimilar plant species: corn
stover and switchgrass. Independent samples of corn
stover and switchgrass that were not used in the cali-
bration model were also employed as validation stand-
ards. In addition, leave-one-out cross validation was
implemented. The model could adequately predict the
three major components in each of the samples, glucan,
xylan, and lignin with relative errors of 1.99%, 2.37%,
and 3.62% respectively. Ash and the other minor sugars
exhibited much higher relative predictive errors, which
was attributed to the inability of inorganic compounds
in ash to absorb in the NIR range, as well as the rela-
tively low concentrations of the minor sugars present
within the samples. Despite this, the authors demon-
strated that a robust PLS model could be generated
from FT-NIR data that could apply to samples with dis-
similar compositions to those within the calibration set.
In an early study by Wallb€acks et al. [129], the
changes in the chemical composition of pulp samples
were analyzed over the duration of the birch Kraft pulp-
ing process using 13 C Cross Polarization Magic Angle
Spinning (CP/MAS) NMR, FTIR, and NIR coupled with
PLS regression. In addition to producing an accurate
regression model capable of predicting glucose, xylose,
and Klason lignin content, the authors showed that the
combination of spectroscopic datasets from different
instruments can sometimes yield models of greater pre-
dictive capability compared to a model built with only
one dataset. The authors also suggested that NIR, which
was demonstrated to have the best predictive capability

Figure 5. Experimental procedures used for the PLS model developed using FT-NIR spectra of corn stover, switchgrass, and wheat
straw (Reproduced with permission from Liu et al. [130]).
of the three techniques, should be implemented for on-
line measurement of samples using robust fiber optic
probes.
More recently, Brink et al. [141] developed an on-line
apparatus for the prediction of aspen and birch bark
contents in unbleached hardwood pulp using NIR spec-
troscopy and multivariate techniques (PCA and PLS).
The authors produced a semi-quantitative model that
could be used to monitor the effect of hardwood pulp
composition on final product quality at a paperboard
mill. Temperature and moisture sensitivities were cited
as the two most significant hindrances to the develop-
ment of the model. As such, a meticulous preparation
of the reference system was suggested for future work.
The hybrid grass of Miscanthus sinensis and
Miscanthus sacchariflorus, i.e. Miscanthus 3 giganteus,
was analyzed by Haffner et al. [142] using FT-NIR spec-
troscopy of dried and ball-milled plant material. The
authors implemented PLS regression analysis to predict
the main components of the sample: extractives, glu-
can, xylan, arabinan, acetyl, Klason lignin, total ash, and
ash after extraction. The predictive ability of the PLS
regression model was greatly improved by milling the
sample to a uniform size.
Spectra of Miscanthus samples were acquired using
visible and near infrared radiation (400–2500 nm) by
Hayes [143]. The effect of particle size (wet-chopped;
air-dried; dried and ground; and dried and sieved) and
moisture were investigated, and those models
CRITICAL REVIEWS IN BIOTECHNOLOGY 11
constructed using dry and sieved samples gave the
most accurate models. The author suggested that des-
pite the inaccuracies involved in detecting minor con-
stituents in wet samples, the wet sample model still
gave good predictions of glucose, xylose, and Klason
lignin.
Near infrared and visible spectroscopy was used to
estimate the gross calorific value of ground Miscanthus
and two coppice willow stem samples by Everard and
coworkers [144]. Various data pretreatments were inves-
tigated, including second derivative, smoothing, multi-
plicative scatter correction (MSC), and standard normal
variate (SNV). The final model utilized second derivative
and MSC data pretreatments. The NIR spectral range of
880–1680 nm was more accurate than the visible wave-
length range of 450–950 nm. The authors suggest that
the NIR-based estimation of gross calorific value could
be used as an alternative method of biomass pricing,
especially since NIR systems are robust and can be used
on-line at a biorefinery in harsh conditions.
Sugarcane lignocellulose was analyzed using diffuse
reflectance near-infrared spectroscopy and multivariate
calibration by Chong et al. [145]. Ash, lignin, and carbo-
hydrate composition were estimated on a total dry
weight and extractives-free basis. Extractives, such as
sugars, proteins, and waxes were found to contribute to
spectral interferences, leading to poor calibration per-
formance against extractive-free standards. Direct
Orthogonal Signal Correction (DOSC) was applied to the
12 D. J. KRASZNAI ET AL.
NIR dataset to improve the calibration by removing the
interferences caused by the extractives in the sample.
The work of Fong Chong et al. [145] underlines the
importance of exploring several data pretreatment tech-
niques to reduce the effect of interferences and incon-
sistencies between the reference and sample data.
Feedstock characteristics, such as chemical and phys-
ical composition, play an important role in selecting
the type and conditions of biomass pretreatment. For
example, biomass type, particle size, and concentrations
may limit the application of ultrasonic pretreatment to
specific biorefineries [146]. When ultrasonic pretreat-
ment is found to be a feasible technique in a particular
biorefinery, physical and chemical characteristics of the
biomass need to be carefully determined and moni-
tored to allow for the ultrasonic conditions to be tuned
for optimal efficiency. FTIR can be used to compare the
crystallinity of cellulose during chemical pretreatment
using the absorbance ratios of 1426/896 cm1 for lateral
order index; 1373/2917 cm1 for total crystallinity index
[147]; and 3336/1336 cm1 for hydrogen bond intensity
[148]. These absorbance ratios can be used to monitor
crystallinity changes during ionic liquid treatment proc-
esses, since lower crystallinity tends to favor enzymatic
digestibility. However, caution needs to be employed
when interpreting FTIR data. There is increasing concern
regarding FTIR reliability for crystallinity measurements,
although several studies comparing various methods
have reported good agreement between FTIR and other
methods. FTIR has the advantage of being simple to
use, but it has limitations, for example, that it can really
only provide relative and not absolute values since both
amorphous and crystalline regions contribute to the
spectra [149]. Other issues related to FTIR reliability for
measuring crystallinity have been noted by Karimi and
Taherzadeh [149]. Multivariate calibration of spectro-
scopic data is complementary to pretreatment proc-
esses because of the need to continually tune the
conditions to fit the characteristics of the feedstock and
associated matrix materials.
The majority of high-throughput techniques for the
analysis of biomass are now spectroscopy-based, owing
to the robustness and versatility of these instruments
[150]. A commonality between spectroscopic techni-
ques is the need for calibration with a standard assay,
which is often based on wet-chemistry. Alternatively,
quantitative Nuclear Magnetic Resonance (NMR) spec-
troscopy can also be used [151] to calibrate other spec-
troscopic models, since discrete peaks represent a
specific number of atoms in that particular magnetic
environment. The increased use of 2 D NMR experi-
ments will further complement multivariate calibration
since multidimensional compositional data can be
acquired for each sample.
Menon and Rao [152] summarize the cellulose, hemi-
cellulose, and lignin composition of several types of
lignocellulosic biomass in their review paper. It is appar-
ent that there is a very high variability in the chemical
composition of different types of biomass. Furthermore,
they provide a summary of various methods for pre-
treating biomass to facilitate its conversion into higher
value products. Of the methods listed, only mechanical
diminution, mineral acids, alkali, liquid hot water,
Organosolv, and steam explosion have been success-
fully implemented at the pilot scale. The review also dis-
cusses various methods of enzymatic hydrolysis of
pretreated biomass and the subsequent fermentation of
the liberated sugars into higher value materials. The
inherent variability of raw biomass and subsequent
material streams was mentioned as a major obstacle to
the bioconversion of these materials since physical and
chemical processes are strongly influenced by the
physical and chemical composition of the biomass.
Multivariate calibration of data acquired by spectro-
scopic sensors stands out as a promising PAT for biorefi-
neries, which should aspire to be flexible processes due
to the variability in composition and volatility of price
of lignocellulosic biomass.
The conversion of lignocellulosic biomass into high
value or commodity products can be considered a
multivariate process. For example, the enzymatic
hydrolysis and fermentation of lignocellulosic biomass
depends on variables such as the chemical composition
of the substrate before pretreatment, degree cellulose
crystallinity, macromolecular structure characteristics,
pretreatment conditions, hydrolysis conditions, produc-
tion of inhibitors, enzyme activities, and fermentation
conditions [153]. Further, variation in the type or matur-
ity of the lignocellulosic substrate can affect optimal
pretreatment conditions [58]. Because the use of che-
mometrics and infrared spectroscopy, namely, FT-NIR
with fiber optic probes or FTIR with Attenuated Total
Reflectance (ATR) accessories, have been shown to
afford accurate, in situ, inexpensive, and real-time ana-
lysis of polymer systems [154,155], it is easy to see the
promise in applying these techniques in future lignocel-
lulosic biorefineries [156–158]. For more robust PAT
applications, NIR spectroscopy might be considered
more advantageous than FTIR, Raman, or NMR due to
the high availability and industrial use of inexpensive
NIR hardware [159,160]. In addition, the near infrared
region is generally less sensitive to water and thus more
robust for samples with high water content. However,
the added robustness of NIR devices comes at a cost to
ability to resolve complex mixtures, in which case FTIR,
 CRITICAL REVIEWS IN BIOTECHNOLOGY 13

Table 2. Summary of wet-chemical methods for the compositional analysis of lignocellulosic biomass.
Method of analysis Fraction measured Limitations of method Reference(s)
Proximate, Weende, or Crude fiber Portion of plant cell wall, complete cellulose Most non-cellulosic polysaccharides and lignin [15]
recovery removed [97]
May underestimate fiber content by more than [98]
50%
Neutral detergent fiber (NDF) Incompletely digestible feed fraction, almost Pectin almost completely removed, protein and [15]
complete recovery of grass cell walls starch removal can be problematic
Acid detergent fiber (ADF) Portion of plant cell wall, complete cellulose A significant portion of lignin is solubilized, [15]
recovery especially for grasses (50% or more [52]
dissolved)
ADF minus ADL Cellulose Suffers from limitations of ADF and ADL [15]
methods
NDF minus ADF Hemicellulose Suffers from limitations of NDF and ADF [15]
methods
Acid detergent lignin (ADL) Lignin Lignin Solubilization at ADF step, especially [15]
grasses
Dietary fiber Complete recovery of cell wall polymers Protein and starch removal can be problematic [15]
Uppsala dietary fiber Total cell wall recovery and composition of Complex and laborious [15]
the cell wall A significant portion of the cell wall protein can [52]
remain, potentially inflating Klason lignin
values
Crampton and Maynard Cellulose Small amount of xylan contamination on cellu- [15]
lose [99]
Significant cellulose degradation [39]
Klason lignin Lignin Possible protein and carbohydrate contamin- [15]
ation [22]
Might underestimate/overestimate lignin con- [53]
tent based on ratio of acid soluble/insoluble
lignin (especially in legumes)
Monoethanolamine (MEA) Cellulose after lignin and other carbohy- Time consuming and hemicellulose components [8]
drates removed might not be adequately removed [48]
[39]
[41]
[7]
[42–47]
USDA FPL Reducing sugars and lignin in woody Limitations of reducing sugar assays [30]
samples Choice of secondary hydrolysis conditions are [100]
not always clear
UK Plant cell wall non-starch polysaccharides Non-specific towards polysaccharides [76]
Lignin not quantified
Trifluoroacetic acid (TFA) Carbohydrate polymers: cellulose and Incomplete hydrolysis is likely to occur (can [8]
hemicellulose take days for sugars to be released) [69]
DFRC Lignin Xylan degradation and contamination can be [81]
an issue at elevated temperatures (> 50  C)
NREL LAP TP-510-42618 Structural carbohydrates and lignin in Underestimated cellulose content due to sul- [8]
biomass fonation of cellulose at high (72%) sulfuric
acid concentration

Raman, and especially NMR will provide much more
detailed chemical spectra [161–165].
Summary and recommendations
Tables 2 and 3 summarize the use of various wet-chem-
ical and chemometric techniques for the compositional
analysis of lignocellulosic biomass, respectively.
Applying chemometrics to spectroscopy can be trouble-
some, despite the many advantages over traditional
chemical routes. As a result, future work should address
and improve the following:
1. Careful selection and reporting of data preprocess-
ing, modeling, and validation parameters, as these
can greatly influence the efficacy and utility of the
multivariate technique.
2. The accuracy of the chemometric method, which
is only as good as its reference data, is often
determined using conventional, laborious wet-
chemical routes, therefore, surrogate mixtures
should be investigated for simplified model build-
ing exercises.
3. Large datasets that span a wide concentration or
property range are required to produce an accur-
ate, predictive multivariate model; perhaps further
development of broad-based databases could aid
in experimental design and verification. A data-
driven modeling approach, such as the Wiki-based
management of chemometric projects described
by Alsberg and Clare [94], appears to offer
both an avenue towards collaboration and stand-
ardization in the compositional analysis of
lignocellulose.
14 D. J. KRASZNAI ET AL.

Table 3. Analytical method, sample type, and parameters used in studies involving spectroscopy and multivariate techniques.
Method or technique
FT-NIR and PCA
pyMBMS and PLS regression
FTIR and PCR/PLS regression
FT-NIR/FTIR and PLS regression
FT-NIR and PLS regression
FTIR ATR and PCA
FTIR comparison study
FTIR ATR, NIR and PLS regression
CP/MAS NMR, FTIR DRIFT, FT-NIR with
PLS regression
On-line FT-NIR and PCA and PLS
regression
FT-NIR and PLS regression
FT-NIR and PLS regression
VIS spectra with PCR and PLS
regression
Sample type Description/parameters Reference
Wood First, second derivative, MSC [124]
Hybrid poplar and loblolly pine PLS applied to pyMBMS data; full CV applied to [126]
pyMBMS PLS model
Milled wood lignins Phenolic hydroxyl group content [119]
Corn stover Full CV [140]
Swedish pine, Swedish spruce, Polish pine MSC applied to NIR spectra [109]
Wood, fibers, and particles of Beech and First derivative applied to FTIR ATR spectra [111]
Grand Fir
30 wood species Two spectral zones analyzed: 1800–1100 cm1 [154]
and 900–800 cm1
Various wood species Baseline corrected, second derivative infrared [155]
spectra
Birch Kraft pulp 60–115 ppm region used for NMR, 2000- [137]
715 cm1 region used for FTIR DRIFT, MSC
applied to both FTIR and NIR spectra
Aspen and birch bark First derivative preprocessing used on NIR spec- [150]
tra; CV with 25% of data left out
Corn stover, switchgrass, and wheat straw EMSC applied to FT-NIR spectra; full CV used in [138]
addition to external validation standards
Wheat and oat straw Second derivative, second order polynomial fit, [133]
or minimum/maximum absorbance normal-
ization applied to FT-NIR spectra; full CV
Unbleached Kraft pulps of Scots pine, VIS data was mean centered prior to multivari- [127]
Norway spruce, and Silver birch ate analysis; CV applied to model as per
Wold, 1978

Conclusions and future outlook analysis on a representative number of biomass sam-

This historical perspective on the compositional analysis
of lignocellulosic biomass was presented in approxi-
mate chronological order with respect to the develop-
ment of each method or technique. The compositional
analysis of lignocellulosic biomass has evolved over the
past 150 years, ranging from laborious wet-chemical
fractionation techniques originally aimed at relating
chemical composition with feedstock digestibility to
more recent noninvasive and nondestructive spectro-
scopic techniques involving the emerging fields of che-
mometrics and PAT.
It is possible that the limitations of the previously
discussed standard methods can be attributed to the
large number of variables involved in the complicated
fractionation process. Slight variations in season, plant
species, and spatial location within the plant as well as
fractionation conditions such as temperature, hydrolysis
time, acid type, acid concentration, and even HPLC con-
figuration can lead to irreproducible and contradictory
results when attempting to compare compositional
data across interdisciplinary research groups. Laborious
and expensive techniques that result in compositional
data that are difficult to compare with other research
groups can be considered “standard” only insofar as a
single, standardized laboratory does all of the analytical
work. When the compositional analysis of a single sam-
ple can cost anywhere between $800-$2000 [132] and
can take days or weeks to complete, the cost associated
with the elaboration of a thorough compositional
ples can become laborious and expensive and is pro-
hibitive to a profitable biorefinery.
The rapid prediction of biomass quality and compos-
ition can substantially reduce the cost of refining ligno-
cellulosic materials. Furthermore, because infrared and
Raman instruments are relatively inexpensive and
robust, they can be implemented as PAT in the biorefi-
nery. It is likely that these robust methods will soon
become commonplace for the analysis of lignocellulosic
material in research and industrial settings.
Funding
The authors would like to acknowledge the National Science
and Engineering Council (NSERC) of Canada, the Canada
Research Chairs program (PC), and the Ontario Research
Chairs program (MFC) for financial support.
References
[1] Giger-Reverdin S. Review of the main methods of cell
wall estimation: interest and limits for ruminants.
Anim Feed Sci Technol. 1995;55:295–334.
[2] Scheller HV, Ulvskov P. Hemicelluloses. Annu Rev
Plant Biol. 2010;61:263–289.
[3] Vanholme R, Demedts B, Morreel K, et al. Lignin bio-
synthesis and structure. Plant Physiol.
2010;153:895–905.
[4] Ahmad M, Taylor CR, Pink D, et al. Development of
novel assays for lignin degradation: comparative ana-
lysis of bacterial and fungal lignin degraders. Mol
BioSyst. 2010;6:815–821.

[5] Barton FE. Chemistry of lignocellulose: methods of
analysis and consequences of structure. Anim Feed
Sci Technol. 1988;21:279–286.
[6] FitzPatrick M, Champagne P, Cunningham MF, et al.
A biorefinery processing perspective: treatment of
lignocellulosic materials for the production of value-
added products. Bioresour Technol. 2010;101:
8915–8922.
[7] Foyle T, Jennings L, Mulcahy P. Compositional ana-
lysis of lignocellulosic materials: evaluation of meth-
ods used for sugar analysis of waste paper and
straw. Bioresour Technol. 2007;98:3026–3036.
[8] Sluiter JB, Ruiz RO, Scarlata CJ, et al. Compositional
analysis of lignocellulosic feedstocks. 1. Review and
description of methods. J Agric Food Chem. 2010;58:
9043–9053.
[9] Henneberg W, Stohmann E. €
Uber das
Erhaltungsfutter vollj€ahrigen Rindviehs. J [29]
Landwirtsch. 1859;3:485–551.
[10] Van Soest PJ. Nutritional Ecology of the Ruminant.
2nd ed. Ithaca, NY: Comstock Publishing; 1994.
[11] Gorham J. Chemical analysis of Indian corn. N Engl J
Med Surg. 1820;4:320–328.
[12] Horsford EN. Value of different kinds of vegetable
food, based upon the amount of nitrogen. Phil Mag
Ser 3. 1846;29:365–397.
[13] AOAC. AOAC Official Method 978.10 Official Methods
of Analysis of AOAC International. 13th ed.
Washington, DC: Association of Official Analytical
Chemists; 1980.
[14] Jung HJG. Analysis of forage fiber and cell walls in
ruminant nutrition. J Nutr. 1997;127:S810–S813.
[15] Ralph J. Lignin structure: recent developments.
Proceedings of the 6th Brazilian Symposium
Chemistry of Lignins and Other Wood Components.
Guaratingueta, Brazil, 1999, 97–112.
[16] Rzedzicki Z, Sykut-Domanska E, Popielewicz J. Quality
of wheat breakfast cereals available on the polish
market. Polish J Food Nutr Sci. 2008;58:307–312.
[17] Marton J, Adler E. Preface, Dedication. In: Marton J,
editor. Lignin structure and reactions. Washington,
DC: American Chemical Society; 1996. p. vii–xiii.
[18] Klason P. The determination of lignin. In: Brauns FE,
editor. The chemistry of lignin. New York, NY:
Academi Press; 1952. p. 151.
[19] Browning BL. Methods of wood chemistry. New York,
NY: Wiley-Interscience; 1967.
[20] TAPPI. TAPPI test methods. Acid-insoluble lignin in
wood and pulp. Atlanta, GA: Technical Association of
the Pulp and Paper Industry; 2006.
[21] Sjo€stro
€m E, Alen R. Analytical methods in wood
chemistry, pulping, and papermaking. New York, NY:
Springer Verlag; 1999.
[22] TAPPI. TAPPI Test Method T222 om-02. Acid-insoluble
lignin in wood and pulp. Atlanta, GA: Technical
Association of the Pulp and Paper Industry Press;
1988.
[23] TAPPI. TAPPI Test Method UM 250. Acid-soluble lig-
nin in wood and pulp. Atlanta, GA: Technical
Association of the Pulp and Paper Industry Press;
1985.
CRITICAL REVIEWS IN BIOTECHNOLOGY 15
[24] Giger S. Revue sur les me thodes de dosage de la lig-
nine utilis
ees en alimentation animale. Ann Zootech.
1985;34:85–122.
[25] Sun R, Song X, Sun R, et al. Effect of lignin content
on enzymatic hydrolysis of furfural residues.
Bioresour. 2011;6:317–328.
[26] Robertson JA, I'Anson KJA, Treimo J, et al. Profiling
brewers' spent grain for composition and microbial
ecology at the site of production. LWT – Food Sci
Tech. 2010;43:890–896.
[27] Xu ZA, Yang XS. Ethanol production from corn stover
using soaking pretreatment. Adv Mater Res.
2011;171–172:261–265.
[28] Templeton DW, Scarlata CJ, Sluiter J, et al. Biomass
Compositional Analysis Method Variability. NREL, 31st
Symposium on Biotechnology for Fuels and
Chemicals, San Francisco, CA; 2009.
Saeman JF, Bubl JL, Harris EE. Quantitative saccharifi-
cation of wood and cellulose. Madison, WI: U.S. Dept.
of Agriculture, Forest Service, Forest Products
Laboratory; 1994.
[30] Shaffer PA, Somogyi M. Copper-iodometric reagents
for sugar determination. J Biol Chem. 1933;100:
695–713.
[31] Saeman JF, Bubl JL, Harris EE. Quantitative saccharifi-
cation of wood and cellulose. Ind Eng Chem Anal
Ed.. 1945;17:35–37.
[32] Tsubaki S, Ozaki Y, Azuma J. Microwave-assisted
autohydrolysis of prunus mume stone for extraction
of polysaccharides and phenolic compounds. J Food
Sci. 2010;75:C152–C159.
[33] Hipsley EH. Dietary “fibre” and pregnancy toxcemia.
Int J Food Sci Nutr. 1953;7:168–173.
[34] DeVries JW, Prosky L, Li B, et al. A historical perspec-
tive on defining dietary fiber. Cereal Foods World.
1999;44:367–369.
[35] Saeman JF, Moore WE, Mitchell RL, et al. Techniques
for the determination of pulp constituents by quanti-
tative paper chromatography. TAPPI J. 1954;37:
336–343.
[36] Houghton TP, Stevens DM, Pryfogle PA, et al. The
effect of drying temperature on the composition of
biomass. Appl Biochem Biotechnol. 2009;153:4–10.
[37] Nelson L, Leming JA. Evaluation of monoethanol-
amine method of cellulose determination for agricul-
tural residues. TAPPI J. 1957;40:846–850.
[38] Wise LE, Peterson FC, Harlow WM. The action etha-
nolamine on woody tissue. Ind Eng Chem Anal Ed.
1939;11:18–19.
[39] Reid JD, Nelson GH, Aronovsky SI. Determination of
cellulose in fibrous agricultural wastes: a rapid
method using monoethanolamine. Ind Eng Chem
Anal Ed. 1940;12:255–259.
[40] Norman AG, Jenkins SH. A new method for the
determination of cellulose, based upon observations
on the removal of lignin and other encrusting mate-
rials. Biochem J. 1933;27:818–831.
[41] K€urschner K, Hoffer A. A new quantitative cellulose
determination. Chem Zeit. 1931;55:161–181.
[42] Cunningham RL, Detroy RW, Carr ME. Thermal hydro-
pulping of wheat straw. Trans Illinois Acad Sci.
1982;75:283–287.
16 D. J. KRASZNAI ET AL.
[43] Cunningham RL, Detroy RW, Bagby MO, et al.
Modifications of wheat straw to enhance cellulose
saccharification by enzymatic hydrolysis. Trans Illinois
Acad Sci. 1981;74:67–76.
[44] Cunningham RL, Carr ME. Pretreatments of wheat
straw for separation into major components. Biotech
Bioeng Symp. 1984;14:96–103.
[45] Cunningham RL, Carlson KD, Bagby MO. Extracted
sweet sorghum substrates as a source of fermentable
sugars. Appl Biochem Biotechnol. 1988;17:117–124.
[46] Claus I, Kordsachia O, Schro €der N, et al.
Monoethanolamine (MEA) pulping of beech and
spruce wood for production of dissolving pulp.
Holzforschung. 2004;58:573–580.
[47] Banerjee S, Sen R, Pandey RA, et al. Evaluation
of wet air oxidation as a pretreatment strategy for
bioethanol production from rice husk and process
optimization. Biomass Bioenergy. 2009;33:1680–1686.
[48] Tasman JE, Berzins V. The permanganate consump-
tion of pulp materials: I. Development of a basic pro-
cedure. TAPPI J. 1957;40:691–695.
[49] Tasman JE, Berzins V. The permanganate consump-
tion of pulp materials: III. The relationship of the
kappa number to the lignin content of pulp materi-
als. TAPPI J. 1957;40:699–704.
[50] Hatfield R, Fukushima RS. Can lignin be accurately
measured? Crop Sci. 2005;45:832–839.
[51] Fukushima RS, Hatfield RD. Comparison of the acetyl
bromide spectrophotometric method with other ana-
lytical lignin methods for determining lignin concen-
tration in forage samples. J Agric Food Chem.
2004;52:3713–3720.
[52] Van Soest PJ. Use of detergents in the analysis of
fibrous feeds. I. Preparation of fiber residues of low
nitrogen content. J Assoc Off Anal Chem. 1963;46:
825–829.
[53] Van Soest PJ. Use of detergents in the analysis of
fibrous feeds. II. A rapid method for the determin-
ation of fiber and lignin. J Assoc Off Anal Chem.
1963;46:829–835.
[54] Van Soest PJ, Wine RH. Use of detergents in the ana-
lysis of fibrous feeds. IV. Determination of plant cell-
wall constituents. J Assoc Off Anal Chem.
1967;50:50–55.
[55] Goering HK, Van Soest PJ. Forage fiber analyses
(apparatus, reagents, procedures, and some applica-
tions). Agric Handb. 1970;379:1–20.
[56] Van Soest PJ. Nonnutritive residues: a system of ana-
lysis for the replacement of crude fiber. J Assoc Off
Agric Chem. 1966;49:546–551.
[57] Jung HG, Mertens DR, Payne AJ. Correlation of acid
detergent lignin and Klason lignin with digestibility
of forage dry matter and neutral detergent fiber. J
Dairy Sci. 1997;80:1622–1628.
[58] Dien BS, Jung HJG, Vogel KP, et al. Chemical compos-
ition and response to dilute-acid pretreatment and
enzymatic saccharification of alfalfa, reed canarygrass,
and switchgrass. Biomass Bioenergy. 2006;30:
880–891.
[59] Cherney DJR, Patterson JA, Cherney JH. Use of 2-
ethoxyethanol and {alpha}-amylase in the neutral
detergent fiber method of feed analysis. J Dairy Sci.
1989;72:3079–3084.
[60] Mertens DR, Allen M, Carmany J, et al. Gravimetric
determination of amylase-treated neutral detergent
fiber in feeds with refluxing in beakers or crucibles:
collaborative study. J AOAC Int. 2002;85:1217–1240.
[61] Holechek JL, Vavra M. Comparison of micro-and
macro-digestion methods for fiber analysis. J Range
Manag. 1982;35:799–801.
[62] Chen H, Ferrari C, Angiuli M, et al. Qualitative and
quantitative analysis of wood samples by Fourier
transform infrared spectroscopy and multivariate
analysis. Carbohydr Polym. 2010;82:772–778.
[63] Moore WE, Johnson DB. Determination of pentosans
in pulp. In: Procedures for the Chemical Analysis of
Wood and Wood Products (as Used at the U.S. Forest
Products Laboratory). Forest Products Laboratory
Forest Service, U.S. Department of Agriculture; 1967.
[64] Moore WE, Johnson DB. Determination of wood sug-
ars. In: Procedures for the Chemical Analysis of Wood
and Wood Products (as Used at the U.S. Forest
Products Laboratory). Forest Products Laboratory
Forest Service, U.S. Department of Agriculture; 1967.
[65] Fengel D, Wegener G. Hydrolysis of polysaccharides
with trifluoroacetic acid and its application to rapid
wood and pulp analysis. In: Brown RD, Jurasek L, edi-
tors. Hydrolysis of cellulose: mechanisms of enzym-
atic and acid catalysis: Based on a Symposium Jointly
Sponsored by the ACS Cellulose, Paper, and Textile
Division, the Institute of Paper Chemistry of TAPPI;
1978 May 17–19; Appleton, WI. Volume 181 of
Advances in Chemistry Series No 181. American
Chemical Society; 1979. p. 145–158.
[66] Morrison IM, Stewart D. Plant cell wall fragments
released on solubilisation in trifluoroacetic acid.
Phytochemistry. 1998;49:1555–1563.
[67] Windeisen E, Strobel C, Wegener G. Chemical
changes during the production of thermo-treated
beech wood. Wood Sci Technol. 2007;41:523–536.
[68] Grohmann K, Himmel M, Rivard C, et al. Chemical-
mechanical methods for the enhanced utilization of
straw. Biotechnol Bioeng Symp. 1984;14:137–157.
[69] Niu K, Chen P, Zhang X, et al. Enhanced enzymatic
hydrolysis of rice straw pretreated by alkali assisted
with photocatalysis technology. J Chem Technol
Biotechnol. 2009;84:1240–1245.
[70] Prosky L, Asp NG, Furda I, et al. Determination of
total dietary fiber in foods, food products, and total
diets: interlaboratory study. J Assoc Off Anal Chem.
1984;67:1044–1052.
[71] Prosky L, Asp NG, Furda I, et al. Determination of
total dietary fiber in foods and food products: collab-
orative study. J Assoc Off Anal Chem. 1985;68:
677–679.
[72] Theander O, Westerlund EA. Studies on dietary fiber.
3. Improved procedures for analysis of dietary fiber.
J Agric Food Chem. 1986;34:330–336.
[73] Theander O, Westerlund E, Andersson R, et al.
Determination of total dietary fiber and its individual
components by the uppsala method. In: G.A. Spiller,
editor. CRC handbook of dietary fiber in human

nutrition. 3rd ed. Boca Raton, FL: CRC Press; 2001. p.
87–110.
[74] Marlett JA. Measuring dietary fiber. Anim Feed Sci
Technol. 1989;23:1–13.
[75] Theander O, Aman P, Westerlund E, et al. Total diet-
ary fiber determined as neutral sugar residues, uronic
acid residues, and Klason lignin (the Uppsala
method): collaborative study. J AOAC Int. 1995;78:
1030–1044.
[76] Milne TA, Chum HL, Agblevor F, et al. Standardized
analytical methods. Biomass Bioenergy. 1992;2:
341–366.
[77] Del Rıo JC, Marques G, Rencoret J, et al. Occurrence
of naturally acetylated lignin units. J Agric Food
Chem. 2007;55:5461–5468.
[78] Hatfield RD, Grabber J, Ralph J, et al. Using the acetyl
bromide assay to determine lignin concentrations in
herbaceous plants: some cautionary notes. J Agric
Food Chem. 1999;47:628–632.
[79] Sluiter A, Hames B, Ruiz R, et al. Determination of
structural carbohydrates and lignin in biomass.
Golden, CO: NREL; 2004.
[80] Sluiter J, Sluiter A. Summative Mass Closure
Laboratory Analytical Procedure (LAP) Review and
Integration. Technical Report NREL/TP-510-48087
(Task No. BB072230). July 2011.
[81] ASTM. Standard test method for determination of
carbohydrates in biomass by high performance liquid
chromatography. West Conshohocken, PA: ASTM;
2001.
[82] Wolfrum EJ, Sluiter AD. Improved multivariate calibra-
tion models for corn stover feedstock and dilute-acid
pretreated corn stover. Cellulose. 2009;16:567–576.
[83] U.S. Department of Energy- Energy Efficiency and
Renewable Energy Biomass Program. Biomass feed-
stock composition and property database. NREL;
2006. Available from: http://www1.eere.energy.gov/
biomass/printable_versions/feedstock_databases.html
[84] Lee SH, Doherty TV, Linhardt RJ, et al. Ionic liquid-
mediated selective extraction of lignin from wood
leading to enhanced enzymatic cellulose hydrolysis.
Biotechnol Bioeng. 2009;102:1368–1376.
[85] Zhang YHP, Ding SY, Mielenz JR, et al. Fractionating
recalcitrant lignocellulose at modest reaction condi-
tions. Biotechnol Bioeng. 2007;97:214–223.
[86] Sannigrahi P, Miller SJ, Ragauskas AJ. Effects of orga-
nosolv pretreatment and enzymatic hydrolysis on cel-
lulose structure and crystallinity in Loblolly pine.
Carbohydr Res. 2010;345:965–970.
[87] Humbird D, Mohagheghi A, Dowe N, et al. Economic
impact of total solids loading on enzymatic hydroly-
sis of dilute acid pretreated corn stover. Biotechnol
Progress. 2010;26:1245–1251.
[88] Xu J, Cheng JJ, Sharma-Shivappa RR, et al. Lime pre-
treatment of switchgrass at mild temperatures for
ethanol production. Bioresour Technol. 2010;101:
2900–2903.
[89] Shafiei M, Karimi K, Taherzadeh MJ. Pretreatment of
spruce and oak by N-methylmorpholine-N-oxide
(NMMO) for efficient conversion of their cellulose to
ethanol. Bioresour Technol. 2010;101:4914–4918.
CRITICAL REVIEWS IN BIOTECHNOLOGY 17
[90] Brijwani K, Oberoi HS, Vadlani PV. Production of a
cellulolytic enzyme system in mixed-culture solid-
state fermentation of soybean hulls supplemented
with wheat bran. Process Biochem. 2010;45:120–128.
[91] Ververis C, Georghiou K, Danielidis D, et al. Cellulose,
hemicelluloses, lignin and ash content of some
organic materials and their suitability for use as
paper pulp supplements. Bioresour Technol. 2007;98:
296–301.
[92] Scarlata CJ, Hyman DA. Development and validation
of a fast high pressure liquid chromatography
method for the analysis of lignocellulosic biomass
hydrolysis and fermentation products. J Chromatogr
A. 2010;1217:2082–2087.
[93] Templeton DW, Scarlata CJ, Sluiter JB, et al.
Compositional analysis of lignocellulosic feedstocks.
2. Method uncertainties. J Agric Food Chem. 2010;58:
9054–9062.
[94] Alsberg BK, Clare A. Wiki based management of che-
mometric research projects. J Chemometrics. 2010;
24:408–417.
[95] Workman J, Koch M, Lavine B, et al. Process analytical
chemistry. Anal Chem. 2009;81:4623–4643.
[96] Naumann D. FT-infrared and FT-Raman spectroscopy
in biomedical research. Appl Spectrosc Rev. 2001;
36:239–298.
[97] Mackanos MA, Contag CH. Fiber-optic probes enable
cancer detection with FTIR spectroscopy. Trends
Biotechnol. 2010;28:317–323.
[98] Teodor ES, Teodor ED, Virgolici M, et al. Non-destruc-
tive analysis of amber artefacts from the prehistoric
Cioclovina hoard (Romania). J Archaeol Sci.
2010;10:2386–2396.
[99] Yonenobu H, Tsuchikawa S, Oda H. Non-destructive
near infrared spectroscopic measurement of antique
washi calligraphic scrolls. J Near Infrared Spectrosc.
2003;11:407.
[100] Dayal BS, MacGregor JF, Taylor PA, et al. Application
of feedforward: neural networks and partial least
squares regression for modelling kappa number in a
continuous Kamyr digester: how multivariate data
analysis might help pulping. Pulp Paper Canada.
1994;95:26–32.
[101] Antti H, Sjo€stro
€m M, Wallb€acks L. Multivariate calibra-
tion models using NIR spectroscopy on pulp and
paper industrial applications. J Chemometrics.
1996;10:591–603.
[102] Benito MTJ, Ojeda CB, Rojas FS. Process analytical
chemistry: applications of near infrared spectrometry
in environmental and food analysis: an overview.
Appl Spectrosc Rev. 2008;43:452–484.
[103] Mu €ller G, Scho€pper C, Vos H, et al. FTIR-ATR spectro-
scopic analyses of changes in wood properties dur-
ing particle- and fibreboard production of hard- and
softwood trees. BioResources. 2009;4:49–71.
[104] Duran N, Angelo R. Infrared microspectroscopy in the
pulp and paper-making industry. Appl Spectrosc Rev.
1998;33:219–236.
[105] Schulz H, Baranska M. Identification and quantifica-
tion of valuable plant substances by IR and Raman
spectroscopy. Vib Spectrosc. 2007;43:13–25.
18 D. J. KRASZNAI ET AL.
[106] Naumann A, Peddireddi S, Ku €es U, et al. Fourier
transform infrared microscopy in wood analysis. In:
Ku€es U, editor. Wood production, wood technology,
and biotechnical impacts. Go€ttingen:
Universit€atsverlag Go
€ttingen; 2007. p. 179–193.
[107] Kacurakova M, Wilson RH. Developments in mid-
infrared FT-IR spectroscopy of selected carbohy-
drates. Carbohydr. Polym. 2001;44:291–303.
[108] Adapa PK, Karunakaran C, Tabil LG, et al. Qualitative
and Quantitative analysis of lignocellulosic biomass
using infrared spectroscopy. CSBE/SCGAB Annual
Conference. Prince Edward Island: The Canadian
Society of Bioengineering; 2009.
[109] Gierlinger N, Schwanninger M. Chemical imaging of
poplar wood cell walls by confocal Raman micros-
copy. Plant Physiol. 2006;140:1246–1254.
[110] Argyropoulos DS. P-31 NMR in wood chemistry: a
review of recent progress. Res Chem Intermed.
1995;21:373–395.
[111] Faix O, Bo €ttcher JH. Determination of phenolic
hydroxyl group contents in milled wood lignins by
FTIR spectroscopy applying partial least-squares (PLS)
and principal components regression (PCR).
Holzforchung. 1993;47:45–49.
[112] Tsuchikawa S. A review of recent near infrared
research for wood and paper. Appl Spectrosc Rev.
2007;42:43–71.
[113] Muteki K, MacGregor JF. Optimal purchasing of raw
materials: a data-driven approach. Aiche J.
2008;54:1554–1559.
[114] Muteki K, MacGregor JF, Ueda T. Rapid development
of new polymer blends: the optimal selection of
materials and blend ratios. Ind Eng Chem Res.
2006;45:4653–4660.
[115] MacGregor JF, Yu H, Garcıa Mun ~oz S, et al. Data-
based latent variable methods for process analysis,
monitoring and control. Comput Chem Eng. 2005;29:
1217–1223.
[116] Nkansah K, Dawson-Andoh B, Slahor J. Rapid charac-
terization of biomass using near infrared spectros-
copy coupled with multivariate data analysis: Part 1
yellow-poplar (Liriodendron tulipifera L.). Bioresour
Technol. 2010;101:4570–4576.
[117] Sellick CA, Hansen R, Jarvis RM, et al. Rapid monitor-
ing of recombinant antibody production by mamma-
lian cell cultures using fourier transform infrared
spectroscopy and chemometrics. Biotechnol Bioeng.
2010;106:432–442.
[118] Tuskan G, West D, Bradshaw HD, et al. Two high-
throughput techniques for determining wood prop-
erties as part of a molecular genetics analysis of
hybrid poplar and loblolly pine. ABAB. 1999;
77–79:55–65.
[119] Malkavaara P, Alen R. A spectroscopic method for
determining lignin content of softwood and hard-
wood kraft pulps. Chemom Intell Lab Syst.
1998;44:287–292.
[120] Wold S, Sjo €stro
€m M, Eriksson L. PLS-regression: a
basic tool of chemometrics. Chemom Intell Lab Syst.
2001;58:109–130.
[121] Boysworth MK, Booksh KS. Aspects of multivariate
calibration applied to near-infrared spectroscopy. In:
Burns DA, Ciurczak EW, editor. Handbook of near-
infrared analysis. Boca Raton, FL: CRC Press; 2008. p.
209–239.
[122] Nomikos P, MacGregor JF. Monitoring batch proc-
esses using multiway principal component analysis.
Aiche J. 1994;40:1361–1375.
[123] Dhanoa MS, Lister S, Sanderson R, et al. The link
between multiplicative scatter correction (MSC) and
standard normal variate (SNV) transformations of NIR
spectra. J Near Infrared Spectrosc. 1994;2:43–47.
[124] Savitzky A, Golay MJE. Smoothing and differentiation
of data by simplified least squares procedures. Anal
Chem. 1964;36:1627–1639.
[125] Krongtaew C, Messner K, Ters T, et al.
Characterization of key parameters for biotechno-
logical lignocellulose conversion assessed by FT-NIR
spectroscopy. Part II: Quantitative analysis by partial
least squares regression. BioResources. 2010;5:
2081–2096.
[126] Helland IS, Naes T, Isaksson T. Related versions of the
multiplicative scatter correction method for prepro-
cessing spectroscopic data. Chemom Intell Lab Syst.
1995;29:233–241.
[127] Martens H, Nielsen JP, Engelsen SB. Light scattering
and light absorbance separated by extended multi-
plicative signal correction. application to near-infra-
red transmission analysis of powder mixtures. Anal
Chem. 2003;75:394–404.
[128] Martens H, Stark E. Extended multiplicative signal
correction and spectral interference subtraction: New
preprocessing methods for near infrared spectros-
copy. J Pharm Biomed Anal. 1991;9:625–635.
[129] Wallbacks L, Edlund U, Norden B, et al. Multivariate
characterization of pulp using solid-state 13C NMR,
FTIR, and NIR. TAPPI J. 1991;74:201–206.
[130] Liu L, Ye XP, Womac AR, et al. Variability of biomass
chemical composition and rapid analysis using FT-
NIR techniques. Carbohydr Polym. 2010;81:820–829.
[131] Wold S, Antti H, Lindgren F, et al. Orthogonal signal
correction of near-infrared spectra. Chemom Intell
Lab Syst. 1998;44:175–185.
[132] Hames BR, Thomas SR, Sluiter AD, et al. Rapid bio-
mass analysis. ABAB. 2003;105–108:5–16.
[133] Shao J. Linear model selection by cross-validation.
J Am Stat Assoc. 1993;88:486–494.
[134] Xu QS, Liang YZ. Monte Carlo cross validation.
Chemom Intell Lab Syst. 2001;56:1–11.
[135] Kalivas JH. Multivariate calibration, an overview. Anal
Lett. 2005;38:2259–2279.
[136] ASTM. Standard practices for infrared multivariate
quantitative analysis. West Conshohocken, PA: ASTM;
2005.
[137] ASTM. Standard practice for qualifying spectrometers
and spectrophotometers for use in multivariate anal-
yses, calibrated using surrogate mixtures. West
Conshohocken, PA: ASTM; 2004.
[138] Bro R, Smilde AK. Principal component analysis. Anal
Methods. 2014;6:2812–2831.
[139] Cournoyer A, Simard JS, Cartilier L, et al. Quality con-
trol of multi-component, intact pharmaceutical tab-
lets with three different near-infrared apparatuses.
Pharm Dev Technol. 2008;13:333–343.

[140] Ilin A, Raiko T. Practical approaches to principal com-
ponent analysis in the presence of missing values.
J Machine Learning Res. 2010;11:1957–2000.
[141] Brink M, Mandenius CF, Skoglund A. On-line predic-
tions of the aspen fibre and birch bark content in
unbleached hardwood pulp, using NIR spectroscopy
and multivariate data analysis. Chemom Intell Lab
Syst. 2010;103:53–58.
[142] Haffner FB, Mitchell VD, Arundale RA, et al.
Compositional analysis of Miscanthus giganteus by
near infrared spectroscopy. Cellulose. 2013;20:
1629–1637.
[143] Hayes DJM. Development of near infrared spectros-
copy models for the quantitative prediction of the
lignocellulosic components of wet Miscanthus sam-
ples. Bioresour Technol. 2012;119:393–405.
[144] Everard CD, McDonnell KP, Fagan CC. Prediction of
biomass gross calorific values using visible and near
infrared spectroscopy. Biomass Bioenergy. 2012;45:
203–211.
[145] Fong Chong B, Purcell D, O’Shea M. Diffuse reflect-
ance, near-infrared spectroscopic estimation of sugar-
cane lignocellulose components – effect of sample
preparation and calibration approach. Bioenerg Res.
2013;6:153–165.
[146] Bussemaker MJ, Zhang D. Effect of ultrasound on
lignocellulosic biomass as a pretreatment for biorefi-
nery and biofuel applications. Ind Eng Chem Res.
2013;52:3563–3580.
[147] Qiu Z, Aita GM, Walker MS. Effect of ionic liquid pre-
treatment on the chemical composition, structure
and enzymatic hydrolysis of energy cane bagasse.
Bioresour Technol. 2012;117:251–256.
[148] Kljun A, Benians TA, Goubet F, et al. Comparative
analysis of crystallinity changes in cellulose I poly-
mers using ATR-FTIR, X-ray diffraction, and carbohy-
drate-binding module probes. Biomacromolecules.
2011;12:4121–4126.
[149] Karimi K, Taherzadeh MJ. A critical review of analyt-
ical methods in pretreatment of lignocelluloses: com-
position, imaging, and crystallinity. Bioresour
Technol. 2016;200:1008–1018.
[150] Lupoi J, Singh S, Simmons BA, et al. Assessment of
lignocellulosic biomass using analytical spectroscopy:
an evolution to high-throughput techniques.
Bioenerg Res. 2014;7:1–23.
[151] Yuan TQ, Xu F, Sun RC. Role of lignin in a biorefinery:
separation characterization and valorization. J Chem
Technol Biotechnol. 2013;88:346–352.
CRITICAL REVIEWS IN BIOTECHNOLOGY 19
[152] Menon V, Rao M. Trends in bioconversion of lignocel-
lulose: biofuels, platform chemicals & biorefinery con-
cept. Prog Energy Combust Sci. 2012;38:522–550.
[153] Vallander L, Eriksson K. Production of ethanol from
lignocellulosic materials: state of the art. In: Fiechter
A, editor. Bioprocesses and applied enzymology. New
York, NY: Springer; 1990. p. 63–95.
[154] Dub e MA, Li L. In-line monitoring of SBR emulsion
polymerization using ATR-FTIR spectroscopy. Polym.
Plast Technol Eng. 2010;49:648–656.
[155] Puskas JE, Antony P, Kwon Y, et al. Macromolecular
engineering via carbocationic polymerization:
branched structures, block copolymers and nano-
structures. Macromol Mater Eng. 2001;286:565–582.
[156] Rubin EM. Genomics of cellulosic biofuels. Nature.
2008;454:841–845.
[157] Monro J, Burlingame B. Carbohydrates and related
food components: INFOODS tagnames, meanings,
and uses. J Food Compos Anal. 1996;9:100–118.
[158] Walford SN. Sugarcane bagasse: how easy is it to
measure its constituents? Proc S Afr Sug Technol Ass.
2008;81:266–273.
[159] Hayes DJM, Hayes MHB, Leahy JJ. Analysis of ligno-
cellulosic components of peat samples with develop-
ment of near infrared spectroscopy models for rapid
quantitative predictions. Fuel. 2015;150:261–268.
[160] Rambo MKD, Ferreira MMC, Amorim EP. Multio-prod-
uct calibration models using NIR spectroscopy.
Chemometr Intell Lab. 2016;151:108–114.
[161] Cheng G, Zhang X, Simmons B, et al. Theory, practice
and prospects of X-ray and neutron scattering for
lignocellulosic biomass characterization: towards
understanding biomass pretreatment. Energy Environ
Sci. 2015;8:436–455.
[162] Michailof CM, Kalogiannis KG, Sfetsas T, et al.
Advanced analytical techniques for bio-oil character-
ization. WIREs Energy Environ. 2016;5:614–639.
[163] Watkins D, Nuruddin M, Hosur M, et al. Extraction
and characterization of lignin from different biomass
resources. J Mat Res Technol. 2015;4:26–32.
[164] Lupoi JS, Singh S, Simmons BA, et al. Assessment of
lignocellulosic biomass using analytical spectroscopy:
an evolution to high-throughput techniques.
Bioenerg Res. 2014;7.
[165] Lupoi JS, Gjersing E, Davis MF. Evaluating lignocellu-
losic biomass, its derivatives, and downstream prod-
ucts with Raman spectroscopy. Front Bioeng
Biotechnol. 2015;3:50.