Brazilian Coffee Blends: A Simple and Fast

Method by Near-Infrared Spectroscopy for the

Determination of the Sensory Attributes Elicited
in Professional Coffee Cupping

New Horizons in
Food Research
Michel Rocha Baqueta, Aline Coqueiro, and Patrı́cia Valderrama

Abstract: The diversity of compounds and variations in the aroma and flavor of ground and roasted coffee make
the sensory evaluation by the “cupping test” a complex task to be performed. A total of 217 commercial coffee
samples classified as different beverage type and with different roast degrees were evaluated by official cuppers in the
“cupping test” and the responses for sensory attributes were used to verify the correlation to the near-infrared (NIR)
spectra. Chemometric models based on partial least squares (PLS) were built for the powder fragrance, drink aroma,
acidity, bitterness, flavor, body, astringency, residual flavor, and overall quality. The parameters of merit such as accuracy,
fit, linearity, residual prediction deviation, sensitivity, analytical sensitivity, limits of detection, and quantification were
evaluated. All sensory attributes were predicted with adequate values according to the parameters of merit. The proposed
method, when compared to the “cupping test,” is an alternative to the determination of the coffee sensory attributes. The
results demonstrated that the use of NIR associated with chemometrics is efficient and recommended for the prediction
of sensorial attributes of coffee by means of the direct analysis of roasted and ground samples, and without any additional
preparation, it is a promising tool for the coffee industry.

Keywords: Chemometric, cupping test, near-infrared spectroscopy, partial least squares, sensory attributes

Practical Application: This study has shown potential use of near-infrared (NIR) spectroscopy coupled with a chemo-
metric tool for the prediction of sensory attributes of commercial coffees. Prediction models for powder fragrance,
drink aroma, acidity, bitterness, flavor, body, astringency, residual flavor, and overall quality were built and showed good
predictive capacity. The use of NIR allows rapid analysis (1 min or less per sample), and it was possible to evaluate all
sensory attributes directly in roasted and ground coffee, without beverage preparation.

Introduction attributes and score the powder fragrance (by smelling the dry
Coffee is among the three most consumed beverages in the milled sample) and aroma (by smelling the water infusion) and
world, together with water and tea, and the evaluation of the sen- after tasting the drink, they score other attributes such as the fla-
sory attributes of coffee beans allows the differentiation of Coffea vor, acidity, astringency, bitterness, body, flavor, and residual flavor
arabica (C. arabica) from Coffea canephora (known as C. robusta) ac- (Bressanello et al., 2017). The sum of the individual scores of all
cording to its characteristics (Luca et al., 2016; Marcucci, Dias, attributes constitutes the final score, which represents the overall
Almeida, & Benassi, 2017). Brazil is the largest green coffee pro- quality of the coffee (Toledo, Pezza, Pezza, & Toci, 2016). Ac-
ducer and exporter, which makes the sensory analysis of coffee a cording to Cupping Protocol of the SCAA (2018), this sensory
rule in determining the quality of the beverage (Ribeiro, Boralle, testing is done for three reasons: to determine the actual sensory
Pezza, Pezza, & Toci, 2017; International Coffee Organization, differences between samples, to describe the flavor of samples, and
2018). to determine the preference of products.
Sensorial coffee evaluation is complex due to the variations of “Cupping test” is the most popular professional method of sen-
aromas, flavors, and the diversity of chemical compounds formed sory analysis of coffees (Associação Brasileira da Industria de Café,
during the roasting process (Alvarado, & Linnemann, 2010). The 2017; Bressanello et al., 2017; SCAA, 2018). In this method,
coffee beverage quality is based on the “cupping test,” an inter- expert tasters are adequately trained to identify, define, and under-
national standard recommended by the Specialty Coffee Associa- stand the sensory aspects that determine the quality of the coffee
tion of America (SCAA) for the classification of coffee beverages in the cup. In conducting these assessments, year after year and
(Lingle, 2011). In this method, accredited tasters evaluate sensory for a long period of time, coffee tasters become very sensitive
to the sensory changes that can occur in coffee (Feria-Morales,
2002). Although these professionals develop an acute sensibility
JFDS-2018-1373 Submitted 9/1/2018, Accepted 3/22/2019. Authors are with for the sensory evaluation of coffees, unfortunately this test has
Universidade Tecnológica Federal do Paraná (UTFPR), CEP, 87301–899, P.O. Box been criticized due to its subjectivity, and often the tasters are the
271, Campo Mourão, Paraná, Brazil. Direct inquiries to author Valderrama (E-mail: only professionals to evaluate the sensorial attributes of coffees in
pativalderrama@gmail.com).
the industries. In addition, other negative aspects associated with

C 2019 Institute of Food Technologists

 R

doi: 10.1111/1750-3841.14617 Vol. 84, Iss. 6, 2019 r Journal of Food Science 1247
Further reproduction without permission is prohibited
 Brazilian coffee blends . . .
cupping include lack of qualified professionals and lack of repro-
ducibility of results, variation in sensory perception among tasters,
inadequate conclusions about samples, external factors that can
influence coffee preparation, time-consuming process, and signif-
icant waste after analysis (Esteban-Dı́ez, González-Sáiz, & Pizarro,
2004; Feria-Morales, 2002; Ribeiro, Augusto, Salva, Thomaziello,
& Ferreira, 2009).
New Horizons in
Food Research
The coffee production is dependent on many factors including
the quality of the green bean, roaster process, the extraction pro-
cedure, and water type used during brewing. Furthermore, drink’s
aromatic profile is much related to the binomial time–temperature
used during the roasting process. If this process is not adequately
performed, roast defects can arise and indicate problems within the
roasting process, directly resulting in the presence of off-flavors in
the coffee brew (Yang et al., 2016). Variations in time–temperature
profiles within the roasting process will directly impact the rate of
moisture loss, internal bean temperature, and chemistry of beans.
This will regulate the rate at which caramelization, Maillard reac-
tion, oxidation, and pyrolysis occur, and the resultant development
of color and flavor in the final roasted coffee bean (Sunarharum,
Williams, & Smyth, 2014).
Water type used during coffee brewing is an important aspect
to consider since drinking water is not chemically pure H2 O
and contains small concentrations of minerals, gases, and organic
substances (Sheibani & Mohammadi, 2018). The compositions
of water depend on the origin of water (geology and ecology),
chemicals, which are added during water treatment, and other
aspects that take place during the distribution and the storage
(Dietrich, 2006). For instance, research showed brewing coffee in
softened water results in overextraction and gives them salty flavor
(McGee, 2004), whereas coffee prepared with distilled water is
too sour (Pangborn, Trabue, & Little, 1971). In espresso coffee,
hard water slows the extraction of flavor, causes clouding in the
brew, and clogs the pipes in the machine (McGee, 2004). Addi-
tionally, the presence of metals such as iron in water may react with
tannins in coffee, which affects taste (Dvorak, Prasai, Skipton, &
Wildt, 2014). Similarly, the presence of minerals (Pangborn, 1982),
cations, and disinfectants (Navarini & Rivetti, 2010) in water can
also affect the flavor of coffee.
In recent years, the coffee industry has been searching for com-
plementary methods in the quantitative determinations of sensory
attributes (Sunarharum et al., 2014). For this, near-infrared (NIR)
spectroscopy has attracted attention because it is a fast technique,
nearly universal application, low cost, easy to perform, requires
minimal sample preparation, does not generate residues, and can
be used for the characterization and quantification of properties
in solid samples (Barbin, Felicio, Sun, Nixdorf, & Hirooka, 2014;
Bertone, Venturello, Giraudo, Pellegrino, & Geobaldo, 2016; Luca
et al., 2016; Mees et al., 2018). Nevertheless, few studies have
emphasized the determination of the coffee sensory attributes by
alternative methods such as NIR spectroscopy and only a few of
these studies are based on the chemometric data analysis with the
objective of correlating the coffee sensory attributes.
Different analytical methods and multivariate analysis have
been used to study coffee sensory properties. Gas chromatography
associated with partial least squares (PLS) was used by Ribeiro
et al. (2009) and Bressanello et al. (2017). NIR spectroscopy
(Esteban-Dı́ez et al., 2004; Ribeiro, Ferreira, & Salva, 2011)
and diffuse reflectance infrared Fourier transform spectroscopy
(Belchior, Franca, & Oliveira, 2016) also were combined for
coffee evaluation, but no association with the “cupping test” was
studied.
1248 Journal of Food Science r Vol. 84, Iss. 6, 2019
A number of studies, some of which involved a limited analysis
of sensory attributes of coffees, were conducted to investigate the
relationship between the sensory perception given by a cup of
coffee (Lazim, & Suriani, 2009; Mendes, Menezes, Aparecida, &
Silva, 2001; Monteiro, Minim, Silva, Chaves, & Cardello, 2005;
Ribeiro et al., 2014; Ribeiro, Augusto, & Ferreira, 2010). How-
ever, this number is limited and does not involve a complete
analysis of all sensory attributes recommended by SCAA in the
“cupping test.”
Thus, to the best of our knowledge, NIR spectroscopy coupled
with PLS chemometric tool has not been tested for the determi-
nation of sensory attributes of commercial coffee classified as the
different type of beverages and grades of toast as an alternative to
the “cupping test.” Therefore, the aim of this study was to evalu-
ate NIR spectroscopy associated with PLS analysis for determining
the sensory attributes of commercial coffee produced by a coffee
industry in the Parana state, Brazil. The performance of the PLS
models was evaluated by parameters of merit such as accuracy, fit,
linearity, residual prediction deviation, sensitivity, analytical sensi-
tivity, inverse of the analytical sensitivity, and limits of detection
and quantification.
Materials and Methods
Green coffee
Green coffee from the Brazilian crop of 2016/2017 was provided
by a coffee industry from the Paraná state (Brazil). According
to Normative Instruction 01/2011, page 6, item 3, of the Pro-
Rectory of Undergraduate and Professional Education and Pro-
Rectory of Research and Post-Graduation of the Technological
Federal University of Paraná, it is not allowed to mention partner
companies, since the company that provided the samples did not
allow the disclosure of their identity. The coffee used is from
the Coffea arabica and Coffea canephora (Robusta) species cultivated
in the Espı́rito Santo, Minas Gerais, and Paraná states of Brazil.
Figure 1 shows the states where the samples were cultivated. Paraná
is located in the south of Brazil, and Minas Gerais and Espı́rito
Santo are located in the southeast region of Brazil.
Postharvest treatment and green coffee classification
Green coffee varieties classified such as Hard, Rioysh, Rio, and
Robusta were acquired by industry. Postharvest treatments and dry
system were performed by industry. Green coffee beans were dried
in a mechanical dryer until they reached 12% moisture. Coffee
husks and impurities were separated from healthy endosperms and
green coffee was classified by grain shape. Benefited green coffee
was separated by sieve and number of defects, in the case of the su-
perior beverage. The classification by sieves was performed by a set
of sieves that separates the grains by shape and size. The set of sieves
have sieves of various sizes and two different shapes: they can be
oblong, to separate the “moca” coffee, or circular, to separate the
coffees with a flattened format. Flattened coffee is grains with suit-
able development and contains two seeds in the fruit. On the other
hand, the “moca” coffee represents an inadequately developed cof-
fee, where there is only one seed per fruit that has developed by
occupying the space of the other seed and has a round and longer
shape. The sieve screen measurements are given in fractions of 1/64
inch and the sieve number corresponds to the numerator of the
fraction, for example, 20 = 20/64-inch sieve. The hard coffee va-
riety was separated by 17/18 sieves. The Hard/Rioysh variety was
separated by 15/16 sieves. The Rioysh/Rio variety was separated
by the sieve 9/10. The variety of Robusta grains was acquired by
Brazilian coffee blends . . .
Figure 1–Brazilian regions where coffee samples were collected: (1) Paraná, (2) Minas Gerais, and (3) Espı́rito Santo.
the number of defects and quality beverage (Franca, Mendonca,
& Oliveira, 2005). The type 5/6 corresponds to a Robusta coffee
with up to 86 defects (for example, black beans and others such as
sour, shells, unripe, dark green, bored, broken, and quakers) in 300
g of sample, and beverage was classified as excellent (coffee that
has a neutral flavor and medium acidity). The same classification
method by type of defects was also used in the acquisition of hard
drink to produce the superior coffee because the same variety of
coffee can present different quality and number of defects from
one producer to the other. Therefore, the industry acquired Hard
and Robusta varieties coffees with specific aspects of number de-
fects to produce superior coffee. Only the drink Rioysh/Rio had
beans and classified as “moca.” All other varieties were classified as
flattened.
Industrial roasting and milling, and sample collection
Industrial equipment such as roaster with the capacity of
1,000 kg/hr, roller mill, rotary sieve, hammer mill, silos, and
pneumatic and mechanical conveyors belong to the coffee
industry and are all from the same company that manufactures
equipment for the coffee industries of the whole world (Cia Lilla).
Green coffee beans were roasted at temperature between 200 and
260 °C, under different time intervals. Roasting time and temper-
ature approximate of each coffee are described below: traditional
Hard/Rioysh was roasted at 240 °C for 15 min, Hard/Rioysh
extra strong was roasted at 260 °C for 15 min, Rioysh/Rio extra
strong was roasted at 255 °C for 15 min, and Hard type 5/6–
17/18/Robusta type 5/6 was roasted at 210 °C for 20 min. A total
of 217 of the roasted and ground coffee samples were collected
and the sensory attributes were determined by the “cupping test.”
The samples of roasted and ground coffee have several varia-
tion factors: producing region, species, variety, postharvest process,
drying conditions, transport, storage, coffee composition, roasting
process, coffee blend, and intra- and interlot variations. To include
these variations, a total of 217 samples were collected in the indus-
try during 5 months of production (January through May 2017).
New Horizons in
Food Research
Each day, the samples were collected, and an attempt was made to
obtain the greatest possible representativeness.
Classification by roasting degrees
The color of the ground and roasted coffee was measured
using a Minolta colorimeter (CHROMA METER CR-410,
Minolta Co., Ltd, Japan) previously calibrated in a white stan-
dard (CR-A44, Minolta), with standard D65 illumination, and
normal colorimetric observer angle of zero degrees was used in
the color measurements. The readings were performed using the
CIELAB system (L∗ , a∗ , and b∗ ) and value of luminosity (L∗ ) was
used to determine the roasting degrees of the samples, as suggested
by Belchior, Franca, and Oliveira (2016).
Panelists
Two expert tasters, also called official cupping tasters, with more
than 20 years of experience in evaluating coffee beverage, partic-
ipated in this study. In the cupping test, traditionally, the same
sample of coffee has been tried several times. Only one or both
tasters are responsible for the test, being able to taste more than
200 cups/day. These professionals develop a unique ability to rec-
ognize the quality of coffee drinks, the presence of defects, and
the distinction of sensory attributes, due to the experience gained
over several years. These professionals are hired by coffee indus-
tries to taste the coffees (Feria-Morales, 2002). The professionals
of this study are industry employees that provided the samples and
responsible for the sensory analysis of all coffee production and
the release of finished coffee lots.
Test location
All the tests were conducted at the sensory evaluation facilities
of the coffee industry, situated in Paraná, Brazil. The room had
adequate lighting, air conditioning (about 22 °C), and had no
strange odors.
Vol. 84, Iss. 6, 2019 r Journal of Food Science 1249
New Horizons in Brazilian coffee blends . . .
Food Research

Figure 2–Sensory evaluation sheet.

Sensory analysis by “cupping test” the industry to the university, where the equipment is available.
Coffee beverages were prepared by the percolation process using The NIR equipment located in university laboratories operates in
100 g of roasted and ground coffee and 1,000 mL of tap filtered the wavelength from 900 to 1,650 nm. The spectral measurements
water heated up to 92 °C using a digital thermometer (Incoterm, were performed at room temperature and the diffuse reflectance
model 9791.16.1.00) for temperature control. The beverage was mode in an equipment microNIR (MicroNIR, JDSU). The spec-
filtered through filter paper no. 103 (Mellita). A filter holder and tra were obtained from the samples in the solid phase. The blank
domestic teapot were used. The water was added continuously in was measured using an NIR reflectance standard (SpectralonTM )
the center of the powder to the end, without interruptions. Cof- with a diffuse reflection coefficient of 99%, while a dark reference
fee was served 15 min after preparation in white porcelain cups, (zero, to simulate nonreflection) was obtained with the lamp off.
free of strange odors. Before each sample and after analyzing the According to Barbin et al. (2014), carbohydrates, water, lipids,
beverage, mineral water was served to tasters. The trained tasters proteins, aromatics, chlorogenic acid, and caffeine present in cof-
evaluated the samples sensory attributes and provided a note for fees show absorption of groups CH, C = C-H, H2 O, CH3 , CH2 ,
each attribute on a scale ranging from 1 to 10; the average of RNH, and ROH in the region from 900 to 1,650 nm.
the values is obtained for the overall quality. This methodology
is recommended by Brazilian Association for the Coffee Industry
(ABIC, Associação Brasileira da Indústria de Café), and Brazilian PLS models
recommendation (Brasil, Ministério da Agricultura, Pecuária e For the PLS models development, the X matrix (NIR spec-
Abastecimento, 2010) and is based on professional cupping guides tra) was correlated to a y vector (y block), which contained the
described by Lingle (2011), SCAA (2018), and Bessanello et al. sensory attributes determined by the “cupping test” in a PLS1
(2017). The tasters evaluated the following sensory attributes: correlation. All spectra preprocessing and PLS models were per-
powder fragrance, drink aroma, acidity, bitterness, flavor, body, formed in MATLAB R2007B (The MathWorks Inc., Natick,
astringency, and residual flavor using the sensory evaluation sheet USA). The reflectance spectra were transformed into absorbance,
presented in Figure 2. In this research, a specific terminology was passed through the multiplicative scattering correction (Isaksson
used to evaluate the type of beverage. This work was approved by & Naes, 1988), had the baseline corrected through the baseline
the Ethics Committee in the Research of UTFPR under protocol algorithm from PLS Toolbox, and were smoothed by the Savgol
number 89924917.6.0000.5547. algorithm (Savitzky & Golay, 1964).
A total of 150 samples were employed in the calibration step,
whereas 67 samples were used in the external validation step, all of
Granulometry standardization
them selected by the Kenston algorithm (Kennard & Stone, 1969).
The granulometric analysis was performed by an automatic
Data were mean centered and the number of latent variables (LVs)
sieve (BERTEL Caieiras, BR). Sieves with apertures of 0.85 and
was chosen based on the lowest value found for the Root Mean
0.60 mm were used. The procedure was performed using 100 g
Square Error of Cross Validation (RMSECV) (Brereton, 2000) by
of coffee powder. The analyses were carried out for 10 min. The
continuous blocks cross-validation of 10 samples.
maximum frequency of the automatic sieve corresponds to 2 mm
of amplitude. Thus, 50% of the maximum frequency was used.
Parameters of merit
NIR spectra Multivariate models were validated by the determination of
An aliquot of the roasted and ground coffees was stored in an the parameters of merit such as accuracy, fit, linearity, residual
amber flask at room temperature for a maximum of 2 hr to obtain prediction deviation, sensitivity, analytical sensitivity, and lim-
the NIR spectra. This time was required to take the samples from its of detection and quantification, according to the description

1250 Journal of Food Science r Vol. 84, Iss. 6, 2019

Brazilian coffee blends . . .
presented in the scientific literature (Beltrame et al., 2016; Santos, ated quality are consequently cheaper and more consumed by the
Lima, Março, & Valderrama, 2016).
Results and Discussion
Sensory attributes evaluated by “cupping test”
For each sample analyzed by trained taster, cupping sensory
attributes were obtained with a score for each attribute. A total of
1,953 sensory responses were obtained (217 samples evaluated for
nine sensory attributes) and due to the dimensions of this table,
these results are presented in Table S1.
In the “cupping test,” each sensory attribute reveals a charac-
teristic of the coffee and, for this reason, it is necessary to per-
form the sensorial analysis of the coffees by experienced tasters.
The acidity is related to the perception caused by substances such
as chlorogenic, citric, malic, and tartaric acids that produce acid
taste. The astringency translates to the sensation of dryness in the
mouth left after drinking. Bitterness is revealed by the taste per-
ception caused by substances such as caffeine, trigonelline, caffeic
and quinic acids, and another phenolic compound that produces
the bitter taste of coffee. The powder fragrance induces the olfac-
tory perception caused by the volatile compounds released from
the roasted and ground coffee, while the drink aroma reveals the
same olfactory perception, but after the beverage preparation. The
attribute body expresses the perception of oiliness and viscosity in
the mouth. The flavor is related to the sensation caused by the
chemical compounds of the beverage when consumed. Finally,
the residual flavor indicates the persistence of the taste sensation
after the coffee beverage tasting (Brasil, Ministério da Agricultura,
& Pecuária e Abastecimento 2010; Lingle, 2011).
The Brazilian Association for the Coffee Industry (ABIC) certi-
fies the coffee quality through the “cupping test” sensorial analysis.
ABIC considers the following coffee quality categories and then
determines the overall quality (OQ) score range on a scale from 1
to 10: Traditional and Extra Strong (OQ ࣙ 4.5 and <5.9), Supe-
rior (OQ ࣙ 6.0 and <7.2), and Gourmet (OQ from 7.3 to 10.0)
(ABIC, 2017). Our trained tasters differentiated the commercial
coffee samples according to the sensory attributes, even in coffees
with similar grain blends and with a different roast degree. The
scores obtained for the sensory attributes and OQ showed that
87 samples were classified as traditional beverage Hard/Rioysh,
42 as extra strong Hard/Rioysh, 85 as extra strong Rioysh/Rio,
and three as superior Hard type 5/6-17/18/Robusta type 5/6.
In general, by analyzing the grades obtained for each attribute
and for overall quality in Table S1, it was possible to observe
that coffee sample with superior quality, such as Hard type 5/6-
17/18/Robusta type 5/6, had higher scores for all attributes, as
expected. On the other hand, the coffee samples with low quality,
such as the Rioysh/Rio, had lower notes when compared to the
other coffee samples, as suggested by ABIC.
According to Brazilian official classification (COB, Classificação
Oficial Brasileira), a coffee named Hard coffee has an acid taste,
astringent, rough, and without sweetness, featuring a full-bodied
and pleasant drink. Rioysh coffee has a slight taste, similar to iod-
oform or phenic acid, and Rio coffee has an unpleasant taste, very
accentuated, which resembles even more iodoform or phenic acid
(Farah et al., 2006). Using the combination of these varieties of
coffee, the coffee industries, as in the case of this study, produce
coffees with different qualities and different prices. It is possible
to verify that less classified samples with higher quality were col-
lected, because coffees of superior quality are produced once or
twice during a year by the industry and the coffees of less appreci-
population in general.
Roasting color
The four groups of beverages discriminated by sensory attributes
on “cupping test” also had the roasting degrees determined by col-
orimetric analysis to complement the description of the sample,
New Horizons in
Food Research
as suggested by Reis, Franca, and Oliveira (2013). The luminosity
(L∗ ), ranging from 0 (black) to 100 (white), was used for determi-
nation of the roasting degrees (light, medium, dark, and very dark).
These results are present in Table S1 (Supporting Information).
In this evaluation, three samples of superior Hard type 5/6-
17/18/Robusta type 5/6 beverage were classified as light roasted
(L∗ between 50 and 60). Note that 87 samples of traditional
Hard/Rioysh beverage were classified as medium roasted (L∗ be-
tween 45 and 55) and 42 samples of extra strong Hard/Rioysh
beverage were classified as very dark roasted (L∗ between 35 and
45). Finally, 85 samples of extra strong Rioysh/Rio were classified
as dark roasted (L∗ between 40 and 50). Analyzing the results of
roasting degrees of all the samples, it was possible to observe that
coffee blends formulated with beans of superior sensorial quality,
as in the case of the superior Hard type 5/6-17/18/Robusta type
5/6, have lower roasting degrees than the coffees of inferior qual-
ity or with lower overall quality, such as other beverages. In other
words, the more roasted the coffee, the lower were the points of
the sensory attributes, possibly associated with more intense roast-
ing of these samples. In the literature, studies have shown that
higher temperatures and roasting times affect the formation and
degradation of volatile compounds and other markers of the sen-
sorial quality of coffee, such as bitter taste (Ribeiro et al., 2009). In
addition, coffee variety also implies sensory quality (Franca et al.,
2005). In this study, the coffee blend with lighter roasting has a
higher quality classification. This blend is roasted by the indus-
try at a milder temperature to preserve the volatile compounds,
flavor, and other markers of the superior quality of this coffee
blend. However, color is not the best measurement of roasting
degree, since color also depends on the bean original composition
and roasting temperature evolution. Many different methods have
been proposed to determine the roasting degree, such as color
generation, weight loss, moisture content, degradation of chloro-
genic acid, or the ratio of free amino acids (Baggenstoss, Poisson,
Kaegi, Perren, & Escher, 2008). Nevertheless, the nature of the
roasting process is very complex, and no clear universally accepted
definitions exist. Color, although imprecise, is therefore currently
used as the industry standard (Yang et al., 2016).
Granulometry analysis
The particle size values of the roasted and ground coffee samples
are shown in Table S2. In the studied coffee samples, the particle
size is very similar, with a prevalence of particles in the range of
0.85 to 0.60 mm. The results in Table S2 show how much of the
sample was classified with 0.60 mm of particle size. Approximately
80% to 92% of each sample had these granulometry values.
PLS models
The aim of the present work was to investigate an alternative
to determine the sensory attributes of commercial coffee classi-
fied with different beverage types by NIR spectroscopy coupled
with PLS. The results demonstrated that the sensory attributes can
be predicted by these techniques associated with the PLS regres-
sion and may be an alternative to the “cupping test” for coffee
evaluation.
Vol. 84, Iss. 6, 2019 r Journal of Food Science 1251
 Brazilian coffee blends . . .

elimination in all models. The total number of outliers eliminated

in the calibration and validation models is described in Table 1.
The optimization of the calibration and validation sets after
the outlier elimination resulted in 150 calibration samples for all
the attributes, except for the body attribute that had 13 outliers
removed during the calibration step, resulting in 137 calibration
samples. For the validation sets, outlier evaluation resulted in 49
New Horizons in
Food Research

validation samples for the body attribute; 65 for flavor, acidity, and
bitterness; 66 for residual flavor, powder fragrance, drink aroma,
and overall quality; and 67 for the astringency.

PLS models and parameters of merit

The outlier identification allows the construction of models
with higher predictive capacity (Valderrama et al., 2007). How-
ever, the exclusion of these samples may be performed up to a
limit. Some recommendations from Brazilian (Botelho, Mendes,
& Sena, 2013) and international guides (ASTM International,
2005) show that samples classified as outliers can be removed up
to 22.2% as a maximum of the total number of samples. Even for
the body sensory attribute that had the largest number of outliers
identified in the calibration and validation sets, this, as well as the
other attributes, did not exceed the number of samples removed as
recommended by these guides. The values obtained for parame-
ters of merit such as accuracy, fit, linearity, and residual prediction
deviation are presented in Table 2.
The accuracy of the chemometric models can be evaluated
Figure 3–NIR spectra of coffee samples. through the root mean square error of calibration (RMSEC) and
root mean square error of prediction (RMSEP) (Valderrama et al.,
2007). When these values are close to each other, this indicates
A vector of instrumental responses was obtained for each roasted
that the values predicted by the model have a good agreement
and ground coffee sample by using NIR spectroscopy. The spec-
with the reference method (responses of “cupping test”) and that
tra obtained in the NIR region for all samples are presented in
the number of LVs was properly chosen, there is no adjustment or
Figure 3.
subadjustment (Santos, Março, & Valderrama, 2013).
PLS models were constructed individually for each sensory at-
RMSEP and RMSEC express models accuracy but these pa-
tribute (PLS1 models) by using five LVs for residual flavor, acid-
rameters can incorporate systematic and random errors. Therefore,
ity, bitterness, powder fragrance, flavor, drink aroma, and overall
the correlation coefficient is another accuracy indicator. This pa-
quality; 11 LVs for astringency; and 20 LVs for body (Table 1),
rameter consists of the correlation coefficient for the adjustment
considering the RMSECV values and the percentage of vari-
between the reference values (sensory responses obtained by “cup-
ance explained in the sensorial responses obtained by the tasters
ping test”) and the sensorial responses predicted by the PLS model.
(y block). In order to optimize the PLS models, the outlier samples
For each attribute modeled, the results were as follows: 0.7988 for
were evaluated. Outliers correspond to the samples with different
the body, 0.7724 for flavor, 0.8398 for astringency, 0.7986 for
behavior from the bulk data set (Valderrama, Braga, & Poppi,
acidity, 0.8351 for bitterness, 0.7514 for powder fragrance, 0.7533
2007). Outliers were identified in the calibration set for the body
drink aroma, 0.7469 for the residual flavor, and 0.7357 for the
attribute. These outliers were detected based on the leverage and
overall quality. Astringency and bitterness presented the highest
residues on the dependent variable. These outliers were elimi-
correlation coefficients, around 0.8, whereas for other attributes
nated, and the calibration model was rebuilt. An optimization step
the correlation coefficients are in the range of 0.73 to 0.79. For
was also applied in the validation sets for outlier identification and
multivariate calibration models elaborated from results obtained
with high variability (as in this case), a correlation coefficient up
Table 1–Outlier information. to 0.7 can be considered satisfactory (Beltrame et al., 2016; Santos
et al., 2016). In addition, it is known that the correlation coeffi-
Before opti- After opti-
mization mization Total cient is directly related to the errors embedded in the execution
Sensory quantity of of the reference method, in this case, the “cupping test.” This
attributes C. S. V. S. C. S. V. S. outliers LVs method of analysis requires preparation of the coffees by percola-
Body 150 67 137 49 31 20 tion of roasted and ground coffee for further analysis and demands
Flavor 150 67 150 65 02 05 tasters with a high level of experience in coffee tasting, capable
Astringency 150 67 150 67 00 11 of distinguishing and quantifying the sensory attributes. There-
Acidity 150 67 150 65 02 05
Bitterness 150 67 150 65 02 05 fore, if this method presents high levels of analytical errors, they
Power fragrance 150 67 150 66 01 05 will possibly be inserted in the modeled multivariate calibration
Drink aroma 150 67 150 66 01 05 models, consequently decreasing the correlation coefficient val-
Residual flavor 150 67 150 66 01 05 ues. Finally, this study is a real application, without laboratory
Overall quality 150 67 150 66 01 05
simulation, with all the scores of the sensory attributes acquired
C. S., calibration set; V. S., validation set; LVs, latent variables. during the production of industrial coffee.

1252 Journal of Food Science r Vol. 84, Iss. 6, 2019

Brazilian coffee blends . . .

Table 2–Performance parameters of the best PLS models.

Power Drink Residual Overall

Parameters of merit Body Flavor Astringency Acidity Bitterness fragrance aroma flavor quality
Accuracya RMSEC 0.2536 0.1581 0.1300 0.1467 0.1289 0.1618 0.1733 0.1732 0.1698
RMSEP 0.2849 0.1313 0.1339 0.1104 0.0996 0.1493 0.1633 0.1545 0.1594
Correlation coefficientb 0.7988 0.7724 0.8398 0.7986 0.8351 0.7514 0.7533 0.7469 0.7357
Jarque-Bera testc JBSTAT 1.4881 1.3041 2.685 1.206 0.5366 3.157 1.7907 3.0236 3.2268

New Horizons in
CRITVAL 5.1203 5.1795 5.2009 5.1795 5.1795 5.1904 5.1904 5.1904 5.1904

Food Research
RPDcald 1.1771 1.8505 1.8529 1.7291 1.9422 1.7521 1.7772 1.7115 1.7167
RPDvale 1.8627 1.2305 1.0856 5× 0.7381 1.0549 1.0152 0.8942 1.0225
10−14
a
Accuracy: This parameter reports the closeness of agreement between the reference value and the value found by the calibration model. RMSEC is the Root Mean Square Error of
Calibration and RMSEP is the Root Mean Square Error of Prediction. The proximity of these values indicates that the number of latent variables was properly chosen and the
model does not present over-fit or under-fit.
b
Correlation coefficient: The correlation coefficient provides information on the goodness of fit of the model. Values ranging from 0 to 1.0 can be attained, with 0 indicating no
correlation and 1.0 revealing perfect linear relationship of predicted and reference values, respectively.
c
Jarque-Bera test: When JBSTAT< CRITVAL, the null hypothesis (residuals are normally distributed with unspecified mean and standard deviation) can be accepted at significance
level of 95%.
d
RPDcal = Residual Prediction Deviation of the calibration. Values between 2.4 and 1.5 are considered satisfactory for good prediction ability in the model.
e
RPDval = Residual Prediction Deviation of the validation. Values between 2.4 and 1.5 are considered satisfactory for good prediction ability in the model.

The linearity of the developed multivariate calibration models
was evaluated by the error distribution using the statistical test of
Jarque-Bera. In this test, when the JBSTAT values are lower than
those of CRITVAL, it is considered that the residues exhibit a
random behavior at a 95% significance, and this is an indication of
linearity for the PLS models (Beltrame et al., 2016). Analyzing the
obtained results, all the models presented JBSTAT values inferior
to those of CRITVAL confirming that the errors present a random
behavior and are appropriate for the PLS model.
Chemometric models can use the residual prediction deviation
(RPD) to evaluate the predictive capacity of these models. RPD
values can be considered satisfactory between 1.5 and 2.4 and
indicate a good predictive quality above this value (Botelho et al.,
2013; Porep, Kammerer, & Carle, 2015). In the evaluation of the
RPD values for the calibration sets, the body sensory attribute
was the one with less predictive capacity (1.17), whereas for other
attributes, the RPD was higher than 1.7, indicating that the PLS
models present satisfactory predictive capacity. In contrast, in RPD
values for the validation sets, the body sensory attribute was the one
that obtained the highest value (1.86), whereas the other attributes
had lower RPD values. These results suggest that PLS models have
an excellent forecasting capability, which is fundamental when the
multivariate models are applied in quality and process control.
The sensitivity results of the models are presented in Table 3.
Sensitivity and analytical sensitivity of models were evaluated ac-
cording to Beltrame et al. (2016), demonstrating that the models
are highly sensitive to predict sensory responses. The inverse of
the analytical sensitivity (analytical sensitivity−1 ) determines the
smaller difference of response measure between the samples, which
can be distinguished by the method, considering the instrumen-
tal random error as the only error source (Beltrame et al., 2016).
Thus, the PLS models showed that it is possible to distinguish
coffee samples by NIR spectroscopy with a difference in sensory
responses up to 0.2169, 0.0201, 0.0656, 0.0278, 0.0242, 0.0199,
0.0201, 0.0238, and 0.0205 for body, flavor, astringency, acid-
ity, bitterness, powder fragrance, drink aroma, residual flavor, and
overall quality, respectively.
The limit of detection is the lower concentration of the sub-
stance of interest that can be detected, but not necessarily quan-
tified and the limit of quantification represents the lowest con-
centration of the substance of interest that can be quantified with
reliability (Bóque, & Rius, 1996; Valderrama et al., 2007). The
performance of all models in the detection and quantification of
sensory attributes are described in Table 4. The minimum sensory
responses that can be reliably detected by the NIR spectroscopy
model were 0.7159, 0.0662, 0.2165, 0.0918, 0.0799, 0.0656,
0.0663, 0.0785, and 0.0675 for the attributes body, flavor, astrin-
gency, acidity, bitterness, powder fragrance, drink aroma, residual
flavor, and overall quality, respectively, indicating that samples with
lower scores for each sensory attribute cannot be detected. The
limits of detection were greater for the attributes flavor, acidity, bit-
terness, powder fragrance, drink aroma, residual flavor, and overall
quality than for the attributes body and astringency. This may be
associated with the greater number of outliers identified in these
models as shown in Table 1, which possibly decreases detection
performance. For other attributes, detection limits in the order of
0.06 to 0.09 can be evaluated by the models and properly detected.
The quantification limits showed that samples of coffees with
sensory responses for body, flavor, astringency, acidity, bitterness,
powder fragrance, drink aroma, residual flavor, and overall qual-
ity lower than 2.1693, 0.2005, 0.6561, 0.2771, 0.2422, 0.1987,
0.2008, 0.2378, and 0.2047, respectively, may not be accurately
quantified, because they exceed the limit of quantification of the
PLS models. Similarly, there was a lower quantification capacity
of the PLS models for the body attribute (2.17) and astringency
(0.66), while for the other attributes, better quantification per-
formances were obtained (between 0.19 and 0.27). The body
attribute presented the lowest performance in the quantification,
but still achieved satisfactory results because the coffee does not
have sensory notes lower than 4.6 for both samples.
In the exception of the body attribute, the limits of quantifi-
cation had values ranging from 0.2 to 0.6, far away from the
values obtained by the reference sensorial method (“cupping
test”) for the different attributes (scores between 4.6 and 7.0),
demonstrating that the PLS models presented a high performance
for discrimination of the sensory attributes of coffees classified
as different beverage types and roasting degrees, allowing the
implementation in the routine of coffee industries. Cupping tests
typically distinguish 0.5 points for each attribute. Thus, except
for “body,” our method is well within the range of cupping tests.
Although classifying coffees with sensory grades lower than 4.5
in sensory classification is not interesting for quality inspection
purposes, as they are below the minimum quality, we would like
to mention the performance of the models in the quantification,
even in coffees with inferior quality. In addition to allowing the
classification of coffees with regular and superior quality, the tool

Vol. 84, Iss. 6, 2019 r Journal of Food Science 1253

 Brazilian coffee blends . . .

Table 3–Parameters of sensitivity of the PLS models.

Power Drink Residual Overall

Parameters of merit Body Flavor Astringency Acidity Bitterness fragrance aroma flavor quality
Sensitivitya 2.4 × 10−4 0.0226 0.0018 0.0169 0.0195 0.0228 0.0222 0.0193 0.0223
Analytical sensitivityb 4.6098 49.865 15.242 35.963 41.289 50.328 49.805 42.049 48.862
Analytical sensitivity −1c 0.2169 0.0201 0.0656 0.0278 0.0242 0.0199 0.0201 0.0238 0.0205
a
New Horizons in

This parameter is the fraction of analytical signal that is due to the increase of the concentration of a particular analyte at unit concentration.
Food Research

b
It
is the ratio between sensitivity and the standard deviation of instrumental noise.
c
The inverse of the analytical sensitivity allows one to establish a minimum difference between the samples that can be distinguished by the method, considering instrumental
random error as the only source of error.

Table 4–Limits of detection and quantification of the PLS models.

Power Drink Residual Overall

Parameters of merit Body Flavor Astringency Acidity Bitterness fragrance aroma flavor quality
Limit of detectiona 0.7159 0.0662 0.2165 0.0918 0.0799 0.0656 0.0663 0.0785 0.0675
Limit of quantificationb 2.1693 0.2005 0.6561 0.2781 0.2422 0.1987 0.2008 0.2378 0.2047
a
Lower sensory response that can be detected, but not necessarily quantified with accuracy.
b
Lower sensory response that can be quantified with reliability.

proposed in this study also allows identifying a coffee of inferior
quality that can be commercialized by the coffee producer or other
coffee industries. Many coffees are sold without purity labels and
without any quality certification. In these cases, the proposed tool
can be tested in new samples to identify if the quality of these un-
known coffees falls within the regular, superior, or inferior quality.
Although the prediction of sensory attributes of coffee samples
has already been studied, there are insufficient reports in the liter-
ature that report the use of NIR spectroscopy associated with the
PLS method for the sensorial evaluation of commercial roasted
and ground coffee and there are a lot of studies where roasting
trials and laboratory coffee production are performed. Moreover,
differentiating the sensory attributes of coffees classified with dif-
ferent types of beverages and roasting degrees is a task that requires
experienced tasters (Esteban-Dı́ez et al., 2004) and for this rea-
son, there must be a difficulty in carrying out studies with this
objective. Most studies report on the evaluation of C. arabica or
C. canephora (Robusta), with few studies involving the analysis of
commercial blends of these species, as in this study.
Over the months, many coffee growers deliver their product to
industry, which tries to deal with variability in the raw material
received. In these 5 months of collection, many lots of green cof-
fee benefited by the dry route were received and used to produce
coffee blends. In this way, even with the variability in the raw
material, the industry seeks to retain the same sensorial quality
among coffee industrialization. A fundamental aspect of this re-
search was that the prediction models were constructed from the
spectra of 217 samples of industrial coffee beans acquired in dif-
ferent Brazilian producing regions sampled during the 5 months
of the production of the coffee industry. This feature is crucial to
demonstrate the viability of the NIR to predict sensory attributes
of roasted and ground coffee since it provides a realistic variability
in the PLS models.
Conclusion
The obtained results showed that the sensory attributes from
commercial coffee, classified as different beverage types, can be
predicted by the NIR spectroscopy coupled to the chemometric
regression PLS. The models were validated by the parameters of
merit showing that the PLS models were sensitive to differentiate,
detect, quantify, and predict the sensory attributes of coffees sam-
ples. The proposed methodology, when compared to the “cupping
test,” is an alternative to the traditional coffee sensory analysis,
which does not require the sample preparation, does not generate
residues, and contributes to cost reduction, being a promising tool
for the coffee industry. On the basis of results of our studies, in the
future, it will also be possible to evaluate other varieties of coffee
produced in Brazil and in other countries, contributing to rapid
and analytical discrimination of sensory attributes of coffees with
different qualities.
Acknowledgments
The authors thank Coordenação de Aperfeiçoamento de Pessoal
de Nı́vel Superior (CAPES) for the master scholarship (M. R.
Baqueta) and Postdoctoral scholarship (Dr. A. Coqueiro).
Author Contributions
P. V. designed the study. M. R. B. collected test data. P. V.
and M. R. B. constructed the prediction models and interpreted
the results. M. R. B. wrote the manuscript and P. V. and A. C.
critically revised the manuscript. All authors read and approved
the final manuscript.
References
Associação Brasileira da Industria de Café. (2017). Programa de Qualidade do Café. Norma da
Qualidade Recomendável e Boas Práticas de Fabricação de Cafés Torrados em Grão e Cafés Torrados e
Moı́dos, 29. Retrieved from http://abic.com.br/ Accessed 10 March 2018.
ASTM International. (2005). Annual book of ASTM standards, standard practices for infrared multi-
variate quantitative analysis - E1655-05. West Conshohocken, PA: Author.
Alvarado, R. A., & Linnemann, A. R. (2010). The predictive value of a small consumer
panel for coffee-cupper judgment. British Food Journal, 112(9), 1023–1032, 2010. https://
doi.org/10.1108/00070701011074372
Baggenstoss, J., Poisson, L., Kaegi, R., Perren, R., & Escher, F. (2008). Coffee roasting and
aroma formation: Application of different time-temperature conditions. Journal of Agricultural
and Food Chemistry, 56(14), 5836–5846. https://doi.org/10.1021/jf800327j
Barbin, D. F., Felicio, A. L. S. M., Sun, D. W., Nixdorf, S. L., & Hirooka, E. Y. (2014). Application
of infrared spectral techniques on quality and compositional attributes of coffee: An overview.
Food Research International, 61, 23–32. https://doi.org/10.1016/j.foodres.2014.01.005
Belchior, V., Franca, A. S., & Oliveira, L. S. (2016) Potential of diffuse reflectance infrared fourier
transform spectroscopy and chemometrics for coffee quality evaluation. International Journal of
Food Engineering, 2(1), 1–8. https://doi.org/10.18178/ijfe.2.1.1-8
Beltrame, K. K., Souza, A. M., Coelho, M. R., Winkler, T. C. B., Souza, W. E., & Valderrama, P.
(2016). Soil organic carbon determination using NIRS: Evaluation of dichromate oxidation
and dry combustion analysis as reference methods in multivariate calibration. Journal of the
Brazilian Chemical Society, 27, 1527–1532. https://doi.org/10.5935/0103-5053.20160031
Bertone, E., Venturello, A., Giraudo, A., Pellegrino, G., & Geobaldo, F. (2016). Si-
multaneous determination by NIR spectroscopy of the roasting degree and Arabica/
Robusta ratio in roasted and ground coffee. Food Control, 59, 683–689. https://doi.org/
10.1016/j.foodcont.2015.06.055

1254 Journal of Food Science r Vol. 84, Iss. 6, 2019

Brazilian coffee blends . . .
Bóque, R., & Rius, F. X. (1996). Multivariate detection limits estimators. Chemomet-
rics and Intelligent Laboratory Systems, 32(1), 11–23. https://doi.org/10.1016/0169-7439(95)
00049-6
Botelho, B. G., Mendes, B. A. P., & Sena, M. M. (2013). Implementação de um
método robusto para o controle fiscal de umidade em queijo minas artesanal. Abordagem
metrológica multivariada. Quı́mica Nova, 36(9), 1416–1422. https://doi.org/10.1590/S0100-
40422013000900023
Brasil, Ministério da Agricultura, Pecuária e Abastecimento. (2010). Instrução normativa n° 16, de
24 de maio de 2010. Retrieved from http://sistemasweb.agricultura.gov.br.
Brereton, R. G. (2000). Introduction to multivariate calibration in analytical chemistry. The
Analyst, 125(11), 2125–2154. https://doi.org/10.1039/B003805I
Bressanello, D., Liberto, E., Cordero, C., Rubiolo, P., Pellegrino, G., Ruosi, M. R., & Bic-
chi, C. (2017). Coffee aroma: Chemometric comparison of the chemical information pro-
vided by three different samplings combined with GC–MS to describe the sensory prop-
erties in cup. Food Chemistry, 214, 218–226. https://doi.org/10.1016/j.foodchem.2016.
07.088
Dietrich, A. M. (2006). Aesthetic issues for drinking water. Journal of Water and Health, 4, 11–16.
Dvorak, B. L., Prasai, G., Skipton, S. O., & Wildt, W. (2014). Drinking water: Iron and manganese.
Retrieved from http://www.ianrpubs.unl.edu/epublic/live/g1714/build/g1714.pdf.
Esteban-Dı́ez, I., González-Sáiz, J. M., & Pizarro, C. (2004). Prediction of sensory properties
of espresso from roasted coffee samples by near-infrared spectroscopy. Analytica Chimica Acta,
525(2), 171–182. https://doi.org/10.1016/j.aca.2004.08.057
Farah, A., Monteiro, M. C., Calado, V., Franca, A. S., & Trugo, L. C. (2006). Correlation
between cup quality and chemical attributes of Brazilian coffee. Food Chemistry, 98, 373–380.
https://doi.org/10.1016/j.foodchem.2005.07.032
Franca, A. S., Mendonca, J. C. F., & Oliveira, S. D. (2005). Composition of green and
roasted coffees of different cup qualities. LWT - Food Science and Technology, 38(7), 709–715.
https://doi.org/10.1016/j.lwt.2004.08.014
Feria-Morales, A. M. (2002). Examinating the case of green to illustrate the limitations of grading
systems/expert tasters in sensory evaluation for quality control. Food Quality and Preference, 13,
355–367. https://doi.org/10.1016/S0950-3293(02)00028-9
International Coffee Organization. (2018). World coffee consumption and total production by all
exporting countries. Retrieved from http://www.ico.org/pt/Market-Report-17-18-p.asp.
Isaksson, T., & Naes, T. (1988). The Effect of Multiplicative Scatter Correction (MSC) and
Linearity Improvement in NIR Spectroscopy. Applied Spectroscopy, 42(7), 1273–1284.
Kennard, R. W., & Stone, L. A. (1969). Computer aided designing of experiments. Technometrics,
11(1), 137–148.
Lazim, M. A., & Suriani, M. (2009). Sensory evaluation of the selected coffee products using fuzzy
approach. International Journal of Mathematical, Computational, Physical, Electrical and Computer
Engineering, 3(2). https://doi.org/10.1999/1307-6892/1965
Lingle, T. R. (2011). The coffee cupper’s handbook: Systematic guide to the sensory evaluation of coffee’s
flavor (pp. 66). Long Beach, Calif.: Specialty Coffee Association of America.
Luca, S., Filippis, M., Bucci, R., Magrı̀, A. D., Magrı̀, A. L., & Marini, F. (2016). Characteri-
zation of the effects of different roasting conditions on coffee samples of different geograph-
ical origins by HPLC-DAD, NIR and chemometrics. Microchemical Journal, 129, 348–361.
https://doi.org/10.1016/j.microc.2016.07.021
Marcucci, C. T., Dias, R. C. E., Almeida, M. B., & Benassi, M. T. (2017). Antioxidant activity of
commercial soluble coffees. Beverages, 3(27), 1–7. https://doi.org/10.3390/beverages3020027
McGee, H. (2004). On food and cooking: The science and lore of the kitchen (2nd ed.). New York,
NY: Scribner.
Mees, C., Souard, F., Delporte, C., Deconinck, E., Stoffelen, P., Stévigny, C., . . . Braekeleer,
K. (2018). Identification of coffee leaves using FT-NIR spectroscopy and SIMCA. Talanta,
177, 4–11. https://doi.org/10.1016/j.talanta.2017.09.056
Mendes, L. C., Menezes, H. C., Aparecida, M., & Silva, A. P. (2001). Optimization of the
roasting of robusta coffee (C. canephora conillon) using acceptability tests and RSM. Food
Quality and Preference, 12, 153–162. https://doi.org/10.1016/S0950-3293(00)00042-2
Monteiro, M. A. M., Minim, V. P. R., Silva, A. F., Chaves, J. B. P., & Cardello, H.
M. A. B. (2005). Sensorial profile of beverage coffee (Coffea arabica l.) determined by
analysis time-intensity. Food Science and Technology, 25(4), 772–780. https://doi.org/10.
1590/S0101-20612005000400024.
Navarini, L., & Rivetti, D. (2010). Water quality for Espresso coffee. Food Chemistry, 122(2),
424–428. https://doi.org/10.1016/j.foodchem.2009.04.019
Pangborn, R. M., Trabue, I. M., & Little, A. C. (1971). Analysis of coffee, tea and artificially
flavoured drinks prepared from mineralized waters. Journal of Food Science, 36, 355–362.
https://doi.org/10.1111/j.1365-2621.1971.tb04061.x
Pangborn, R. M. (1982). Influence of water composition, extraction procedures, and holding
time and temperature on quality of coffee beverage. LWT-Food Science and Technology, 15,
161–168.
Porep, J. U., Kammerer, D. R., & Carle, R. (2015). On-line application of near infrared
(NIR) spectroscopy in food production. Trends in Food Science & Technology, 46(2), 211–230.
https://doi.org/10.1016/j.tifs.2015.10.002
Reis, N., Franca, A. S., & Oliveira, L. S. (2013). Quantitative evaluation of multiple adulterants
in roasted coffee by Diffuse Reflectance Infrared Fourier Transform Spectroscopy (DRIFTS)
and chemometrics. Talanta, 115: 563–568. https://doi.org/10.1016/j.talanta.2013.06.004
Ribeiro, M. V. M., Boralle, N., Pezza, H. R., Pezza, L., & Toci, A. T. (2017). Authenticity
of roasted coffee using 1 H NMR spectroscopy. Journal of Food Composition and Analysis, 57,
24–30. https://doi.org/10.1016/j.jfca.2016.12.004
Ribeiro, B. B., Mendonça, L. M. V. L., Assis, G. A., Mendonça, J. M. A., Malta, M.
R., & Montanari, F. F. (2014). Evaluation of the chemical and sensory characteristics
of Coffea canephora Pierre AND Coffea arabica L. blends. Coffee Science, 9(2), 178–186.
https://doi.org/10.25186/cs.v9i2.596
New Horizons in
Food Research
Ribeiro, J. S., Ferreira, M. M. C., & Salva, T. J. G. (2011). Chemometric models for the quanti-
tative descriptive sensory analysis of Arabica coffee beverages using near infrared spectroscopy.
Talanta, 83, 1352–1358. https://doi.org/10.1016/j.talanta.2010.11.001
Ribeiro, J. S., Augusto, F., & Ferreira, M. M. C. (2010). Uso de perfis cromatográficos de
voláteis de cafés arábicas torrados para a diferenciação das amostras segundo o sabor, o
aroma e a qualidade global da bebida. Quı́mica Nova, 33(9), 1897–1904. https://doi.org/
10.1590/S0100-40422010000900015
Ribeiro, J. S., Augusto, F.; Salva, T. J. G., & Ferreira, M. M. C. (2012). Prediction models for
Arabica coffee beverage quality based on aroma analyses and chemometrics. Talanta, 101,
253–260. https://doi.org/10.1016/j.talanta.2012.09.022
Ribeiro, J. S., Augusto, F.; Salva, T. J. G., Thomaziello, R. A., & Ferreira, M. M. C. (2009).
Prediction of sensory properties of Brazilian Arabica roasted coffees by headspace solid phase
microextraction-gas chromatography and partial least squares. Analytica Chimica Acta, 634,
172–179. https://doi.org/10.1016/j.aca.2008.12.028
Santos, D. A., Lima, K. P., Março, P. H., & Valderrama, P. (2016). Vitamin C determination by
ultraviolet spectroscopy and multiproduct calibration. Journal of the Brazilian Chemical Society,
27, 1912–1917. https://doi.org/10.5935/0103-5053.20160071
Santos, D. A., Março, P. H., & Valderrama, P. (2013). Multi-product calibration: Prelim-
inary studies to determine quality parameters in industrialized juices based on ultravi-
olet spectroscopy. Brazilian Journal of Analytical Chemistry, 12, 495–498. https://doi.org/
10.1007/s12161-017-1099-4
Savitzky, A., & Golay, M. J. E. (1964). Smoothing and Differentiation of Data by Sim-
plified Least Squares Procedures. Analytical Chemistry, 36(8), 1627–1639. https://doi.org/
10.1021/ac60214a047.
SCAA. (2018). SCAA protocols. Cupping specialty coffee. Retrieved from http://www.scaa.
org/PDF/resources/cupping-protocols.pdf.
Sunarharum, W. B., Williams, D. J., & Smyth, H. E. (2014). Complexity of coffee fla-
vor: A compositional and sensory perspective. Food Research International, 62, 315–325.
https://doi.org/10.1016/j.foodres.2014.02.030
Sheibani, E., & Mohammadi, A. (2018). The impacts of water compositions on sensory properties
of foods and beverages cannot be underestimated. Food Research International, 108, 101–110.
https://doi.org/10.1016/j.foodres.2018.03.024
Toledo, P. R. A. B., Pezza, L., Pezza, H. P., & Toci, A. T. (2016). Relationship be-
tween the different aspects related to coffee quality and their volatile compounds. Com-
prehensive Reviews in Food Science and Food Safety, 15, 705–719. https://doi.org/10.1111/
1541-4337.12205
Yang, N., Liu, C., Liu, X., Degn, T. K., Munchow, M., & Fisk, I. (2016). Determination of
volatile marker compounds of common coffee roast defects. Food Chemistry, 211, 206–214.
https://doi.org/10.1016/j.foodchem.2016.04.124
Valderrama, P., Braga, J. W. B., &Poppi, R. J. (2007). Validation of multivariate cal-
ibration models in the determination of sugar cane quality parameters by near in-
frared spectroscopy. Journal of the Brazilian Chemical Society, 18(2), 259–266. https://doi.
org/10.1590/S0103-50532007000200003
Supporting Information
Additional supporting information may be found online in the
Supporting Information section at the end of the article.
Table S1. Scores of sensory attributes of coffee samples obtained
by “cupping”.
Table S2. Granulometry and color of the coffee samples.
Vol. 84, Iss. 6, 2019 r Journal of Food Science 1255