Screening Inorganic Arsenic in Rice by Visible and Near-Infrared Spectroscopy
Rafael Font,1* Dinoraz Vélez,2 Mercedes Del Río-Celestino,1 Antonio De Haro-Bailón,1
Rosa Montoro2
1Instituto de Agricultura Sostenible (CSIC). Alameda del Obispo s/n. 14080, Córdoba,
Spain.
2Instituto de Agroquímica y Tecnología de Alimentos (CSIC), Apartado 73, 46100,
Burjassot (Valencia), Spain.
*Corresponding author: telephone (+34) 957499211; fax (+34) 957499252; e-mail:
/tt/file_convert/577cce5a1a28ab9e788dd64e/document.doc
1
Abstract
The potential of near infrared spectroscopy (NIRS) for screening the inorganic arsenic
(i-As) content in commercial rice was assessed. Forty samples of rice were freeze-dried
and scanned by NIRS. The i-As contents of the samples were obtained by acid
digestion-solvent extraction followed by hydride generation atomic absorption
spectrometry, and were regressed against different spectral transformations by modified
partial least square (MPLS) regression. Second derivative transformation equation of
the raw optical data, previously standardized by applying standard normal variate
(SNV) and De-trending (DT) algorithms, resulted in a coefficient of determination in
the cross-validation (1-VR) of 0.65, indicative of equations useful for a correct
separation of the samples in low, medium and high groups. The standard deviation
(S.D.) to standard error of cross-validation (SECV) ratio, shown by the second
derivative equation, was similar to those obtained for other trace metal calibrations
reported in NIRS reflectance. Spectral information related to starch, lipids and fiber of
the rice grain, and also pigments in the caryopsis, were the main components used by
MPLS for modeling the selected prediction equation. This pioneering use of NIRS to
predict the i-As content in rice represents an important reduction in labor input and cost
of analysis.
Key words
Near-infrared spectroscopy (NIRS); inorganic arsenic; brown rice; milled rice
2
Introduction
Rice is the dominant staple food crop in developing countries, particularly for the humid
tropics across the globe [1]. Almost 96% of rice is produced and consumed in in
developing countries [1] and contributing over 70% of energy to the daily intake [2].
The component of protein in rice, at 7 – 9% by weight, is relatively low [3], but it is a
major source of protein (50%) for these rice-consuming people [2].
For a food with such a high consumption it is crucial to have information about toxic
trace levels, in order to establish the potential effects on human health. Arsenic (As) and
its chemical species As(III) and As(V), collectively known as inorganic As (i-As), are
the contaminants of interest in this work. Total diet studies indicate that As
concentrations in rice are higher than those in other products of vegetable origin [4,5].
Natural processes and human activities are the two principal factors responsible for the
introduction of arsenic into the rice-growing environment. Natural processes of
introduction involve soil/water chemistry and climate. Farm management activities such
as fertilization practices, crop rotation and herbicidal/insecticidal uses also act to
introduce arsenic. For inorganic arsenic, soil properties and uses of pesticides are
assumed to be the most important interactive influences that determine its final
concentration [6].
Data on As contents in samples of rice collected in arsenic-endemic areas such as
Taiwan, West Bengal and Bangladesh show that the As content ranges between 0.04
and 0.76 µg/g [7-9]. In non-arsenic-endemic areas, the highest value reported is 0.776
µg/g [2,10,11]. The very few studies on i-As contents in rice show concentrations that
vary between 0.021 and 0.560 µg/g [5,6,7,10,12]. Evaluating the contribution of rice to
i-As intake is, in our view, a necessary task in order to make a more realistic assessment
3
of the risk of exposure to this toxin, especially in arsenic-endemic areas and developing
countries.
The standard methodologies for trace metal determination offer a high level of precision
but have some handicaps, such as high cost of analysis, slowness of operation,
destruction of the sample, and use of hazardous chemicals. In contrast, Near Infrared
Spectroscopy (NIRS) is a valuable technique that offers speed and low cost of analysis,
and also the sample is analyzed without using chemicals. The spectral information can
be used for simultaneous prediction of numerous constituents and parameters of the
samples, once appropriate calibration equations have been prepared from sets of
samples analyzed by both NIRS and conventional analytical techniques. After
calibration, the regression equation permits accurate analysis of many other samples by
prediction of results on the basis of the spectra.
NIRS has been applied to analysis of metal content mostly in the environmental field,
and to a lesser extent in the agro-food fields. In environmental studies various authors
have reported the analysis of heavy metals in lake sediments [13], studies concerning
the chemical characterization of soils [14], and the determination of heavy metals and
arsenic by NIRS in plant tissues [15,16]. Recently, in the agro-food field the feasibility
of this technique for measuring K, Na, Mg, and Ca in white wines was demonstrated
[17]. In the speciation field, NIRS has been used for predicting mercurial species in the
membrane constituents of living bacterial [18] cells, and i-As in crustaceans of
commercial interest [19]. So far, however, no reports have been published on the use of
NIRS for predicting arsenic species in rice.
The objectives of this work were: (i) to test the potential of NIRS for predicting the i-As
content in rice samples, and (ii) to provide a mechanism to explain why NIRS is capable
of predicting i-As in this species.
4
Experimental
Samples. Samples of commercial rice were selected at different markets in Valencia
(Spain) on the basis of the type of rice (brown or milled, long or medium grain). This
criterium was based on the fact that rice is marketed usually without regard to
geographic origin and specific cultivar type. In addition, previous studies demonstrated
that any differences in concentrations of arsenic are not anticipated to be distinctive
enough to establish geographic origin, rice variety, or other source attributes produced
under normal growing circumstances [20]. Rice samples were ground and freeze-dried
before determination of the i-As content by the reference method, and NIRS analysis.
Determination of inorganic arsenic. The methodology applied was developed
previously by Muñoz et al. [21]. Deionized water (4.1 mL) and concentrated HCl (18.4
mL) were added to 0.5 g of freeze-dried sample. The mixture was left overnight. After
reduction by HBr and hydrazine sulfate, the inorganic arsenic was extracted into
chloroform, and back-extracted into 1 mol/L HCl. The back-extraction phase was dry-
ashed and the i-As was quantified by flow injection-hydride generation atomic
absorption spectrometry (FI-HG Perkin Elmer FIAS-400; AAS Perkin Elmer Model
3300). The analytical characteristics of the method were: detection limit = 0.013 μg/g
dry weight (dw); precision = 3-5%; recovery As(III) 99% and As(V) 96%.
NIRS equipment and software. Near infrared spectra were recorded on an NIRS
spectrometer model 6500 (Foss-NIRSystems, Inc., Silver Spring, MD, USA) in
reflectance mode equipped with a transport module. The monochromator 6500 consists
of a tungsten bulb and a rapid scanning holographic grating with detectors positioned
for transmission or reflectance measurements. To produce a reflectance spectrum, a
ceramic standard is placed in the radiant beam, and the diffusely reflected energy is
measured at each wavelength. The actual absorbance of the ceramic is very consistent
5
across wavelengths. In this work, each spectrum was recorded once from each sample,
and was obtained as an average of 32 scans over the sample, plus 16 scans over the
standard ceramic before and after scanning the sample. The ceramic and the sample
spectra are used to generate the final Log (1/R) spectrum. The whole time of analysis
took about 2 min., approximately. Mathematical transformations of the spectra and
regressions performed on the spectral and laboratory data were obtained by using the
GLOBAL v. 1.50 program (WINISI II, Infrasoft International, LLC, Port Matilda, PA,
USA).
NIRS procedure: recording of spectra and processing of data. Freeze-dried, ground
samples of rice were placed in the NIRS sample holder (3 cm diameter) until it was full
(weight 3.50 g), and were then scanned. Their NIR spectra were acquired at 2 nm
intervals over a wavelength range from 400 to 2500 nm (visible plus near infrared
regions).
Samples of rice were recorded as an NIR file, and were checked for spectral outliers
spectra with a standardized distance from the mean (H) > 3 (Mahalanobis distance), by
using principal component analysis (PCA). The objective of this procedure was to
detect and, if necessary, remove possible samples whose spectra differed from the other
spectra in the set [22].
In the second step, laboratory reference values for i-As, as obtained from the reference
method, were added to the NIR spectra file. Calibration equations were computed in the
new file by using the raw optical data (log 1/R, where R is reflectance), or first or
second derivatives of the log 1/R data, with several combinations of segment
(smoothing) and derivative (gap) sizes. The use of derivative spectra instead of the raw
optical data to perform calibration is a way of solving problems associated with
overlapping peaks and baseline correction [23]. A first-order derivative of log (1/R)
6
results in a curve containing peaks and valleys that correspond to the point of inflection
on either side of the log (1/R) peak, while the second-order derivative calculation results
in a spectral pattern display of absorption peaks pointing down rather than up, with an
apparent band resolution taking place [24]. In addition, the gap size and amount of
smoothing used to make the transformation will affect the number of apparent
absorption peaks.
To correlate the spectral information (raw optical data or derived spectra) of the samples
and the i-As content determined by the reference method, modified partial least squares
(MPLS) was used as regression method, using wavelengths from 400 to 2500 nm every
8 nm. Standard normal variate and De-trending (SNV-DT) transformations [25] were
used to correct baseline offset due to scattering effects (differences in particle size and
path length variation among samples).
Cross-validation. Cross-validation is an internal validation method that like the external
validation approach seeks to validate the calibration model on independent test data, but
it does not waste data for testing only, as occurs in external validation. This procedure is
useful because all available chemical analyses for all individuals can be used to
determine the calibration model without the need to maintain separate validation and
calibration sets. The method is carried out by splitting the calibration set into M
segments and then calibrating M times, each time testing about a (1/M) part of the
calibration set [26]. In this work, the different calibration equations were validated with
7 cross-validation segments, as this was the optimum number of groups automatically
selected by the software as a function of the number of samples employed.
The prediction ability of the equations obtained was determined on the basis of their
coefficient of determination in the cross-validation (r2) [27] (eq. 1) and standard
deviation (S.D.) to standard error of cross-validation (SECV) ratio (RPD) [28] (eq. 2).
7
r2= eq. 1
where: = NIR measured value; = mean “y” value for all samples; = lab reference
value for the ith sample.
RPD = eq. 2
where: = lab reference value for the ith sample; = NIR measured value; N=
number of samples, K= number of wavelengths used in an equation; S.D.= standard
deviation.
The statistics shown in eq. 1 and eq. 2, give a more realistic estimate of the applicability
of NIRS to the analysis than those of the external validation, as cross-validation avoids
the bias produced when a low number of samples representing the full range are
selected as validation set [27,28]. The SECV method is based on an iterative algorithm
which selects samples from a sample set population to develop the calibration equation
and then predicts on the remaining unselected samples. This statistic indicates an
estimate of the standard error of prediction (SEP) that may have been found in an
external validation [29], and as occurred with SEP is calculated as the square root of the
mean square of the residuals for N-1 degrees of freedom, where the residual equals the
actual minus the predicted value.
In this work, cross-validation was computed on the calibration set for determining the
optimum number of terms to be used in building the calibration equations.
Results and Discussion
8
Population boundaries and identification of spectral outliers for rice samples.
Population boundaries for spectra of samples of rice were determined by PCA
performed over the entire population (Figure 1). By using twelve PCs, calculated on the
second derivative (2, 5, 5, 2; SNV+DT) of the raw spectra, the 98.54 % of the whole
spectral variability in the data was explained. The global H (GH) of the sample
population extended from 0.25 to 2.12 with a mean distance of 0.96.
One sample was shown to be a GH outlier in PCA. After carefully examination of the
commercial description of the product, it was decided to eliminate it from the
calibration set as product composition was in doubt.
Inorganic arsenic contents in the rice samples. Samples of rice used to conduct this
work showed mean content and S.D. of 110.37 and 49.80 ng/g dw, respectively (Table
1). The range of i-As found in the samples extended form 13.0 to 268.0 ng/g dw, these
values being similar to those contents previously reported in white rice from the United
States of America [6]. Inorganic arsenic contents were normally distributed in the
occurrence range (Figure 2).
Spectral data pre-treatments and equation performances. The application of the
second derivative and SNV+DT algorithms to the raw spectra (Log 1/R) (Figure 3),
resulted in substantial correction (Figure 4) of the baseline shift caused by differences in
particle size and path length variation. Peaks and troughs in Figure 4 correspond to the
points of maximum curvature in the raw spectrum, and it has a trough corresponding to
each peak in the original. The increase in the complexity of the derivative spectra
resulted in a clear separation between peaks which overlap in the raw spectra.
The use of the second derivative transformation (2, 5, 5, 2; SNV+DT) of the raw optical
data performed over the entire segment (400-2500 nm), yielded a higher prediction
ability equation in cross-validation than any other of the various mathematical
9
treatments used. MPLS regression resulted in an equation that presented four terms and
showed a low standard error of calibration (SEC = 20.19 ng/g dw) and high coefficient
of determination in the calibration (R2 = 0.80) (Table 1). In cross-validation the selected
equation showed an r2 of 0.65 (meaning that the 65% of the chemical variability in the
data was explained), which was indicative of equations useful for a correct separation of
samples with low, medium and high contents [27] (Figure 5). In accordance with the
RPD value (1.67) shown by the highest prediction ability equation obtained, and
considering the limits for RPD recommended by Chang et al. [30], and Dunn et al. [31],
this equation was acceptable for i-As prediction in rice.
The use of the coefficient of determination in the evaluation of an NIR equation
involving trace elements and mineral species has received some criticism [15,32]. In
addition, the interpretation of the value of the coefficient of determination as it was first
reported by Shenk and Westerhaus [27] for agricultural products, probably needs to be
revised for element analysis. On the other hand, while much effort has been applied to
the development of calibration of quality components in the agro-food field, no critical
levels of the RPD statistic have been set for trace elements and mineral species in these
products. By this reason, those studies reported on mineral composition of soils [30,31]
show a special relevance at the time of establishing suitable limits of RPD.
But in spite of the above considerations, authors currently researching NIRS for
environmental analysis and food safety still base their decisions on these statistics for
rapid field and laboratory measurements [19,33,34], to relate chemistry and apparent
absorption of NIR spectra.
Brown and milled rice reflectance spectra. The average second derivative (2, 5, 5, 2;
SNV+DT) spectra of those samples that were clearly identified as brown (n= 16) and
milled rice (n= 14) were obtained. As it can be observed in Figure 6, milled rice showed
10
higher absorptions than brown rice at wavelengths 914 and 984 nm, which have been
assigned to C-H stretching third overtone of CH2 groups and O-H stretching second
overtone of starch [35], respectively. The relative higher starch content of milled rice
(78%) in comparison to that of brown rice (66%) [36], as a consequence of removing
the bran and embryo fractions in the abrasive milling, explain these differences in
apparent absorption between both spectra.
The same phenomenon, but of inverse sign, can be observed at wavelengths 1778 and
2348 nm, related to C-H stretching first overtone of cellulose, and CH2 symmetric
stretching plus =CH2 deformation [35,37] groups of oil and fiber (Figure 6). Most non-
starch constituents are removed during milling, with fiber showing the most dramatic
drop, followed by other nutrients except protein [36]. Results reported on distribution of
nutrients in brown rice, supports the idea that only a 27% of the total cellulose, a 21% of
the lignin and about a 20% of the non-starch lipids (ether-soluble) are present in the
milled rice [38], being the rest of them removed during milling.
The visible segment of the spectrum similarly showed absorption bands that differed in
intensity between brown and milled rice. The fact that pigments in coloured rices are
located in the pericarp or the seed coat, which are removed during milling, explains
these differences shown by spectra (Figure 6). The conspicuous band at 668 nm is
displayed by both types of rice, but with little higher intensity in brown than in milled
rice. This band, which has been related previously to some bran component [39] is
difficult to explain here as being caused only by the outer layers of the grain because its
ubiquity in the different types of rice.
Correlation plot of i-As vs wavelength. The correlation plot for i-As vs wavelength
absorbance for the standardised (SNV+DT) optical data in displayed in Figure 7. Most
relevant features shown by the correlation plot were the negative correlation between i-
11
As and absorption existing in those wavelengths which have been assigned to starch
(around 984 nm, and also from 2200 to 2254 nm) and protein (2052 nm) [35,37].
Previous studies reporting element distribution in rice demonstrated a higher i-As
concentration in the brown rice than in milled rice [36]. A considerable portion of the
rice caryopsis ash is accounted for by phosphorus. Thus, milling results in loss of
different essential elements. Although several studies have been reported concerning the
element distribution in the milling fractions of rice [40], data available on arsenic
concentration in this species are mainly referred to milled rice [6].
However, because the similar biochemistry of arsenic with that of phosphorus [41,42], it
is logical to think that caryopsis account also for most arsenic in the grain. This fact
would explain by itself the negative correlation with starch shown by i-As, i.e., milled
rice has a lower concentration of i-As and a relative higher percentage of starch, and the
opposite for brown rice.
More difficult is to explain the relative high negative correlation of i-As with those
wavelengths related to protein absorption. The low difference in protein concentration
between brown (7.1-8.3 %) and milled rice (6.3-7.1%) [36] does not justify this fact. It
is probably that the multiple factors controlling the final protein content in rice, or
geographic location and farm management activities [6] relate both, protein and i-As
contents.
Positive correlations were found between i-As and absorption in wavelengths regions
related to fiber and oil (1722 and 2310 nm) and also pigments (from 472 to 506 nm),
which can be explained by the main location of these components in the outer layers of
the grain, where higher concentrations of i-As are supposed to be found.
Modified partial least square loadings. MPLS regression reduces the spectral
information of the samples by creating a much smaller number of new orthogonal
12
variables (factors), which are combinations of the original data, and which retain the
essential information needed to predict the composition. The role played by the NIR
absorbers (organic and inorganic molecules) present in the samples, in modelling the
calibration equations for i-As, can be interpreted by studying the bands of the MPLS
factors (loading plots). These loading plots show the regression coefficients of each
wavelength related to the element (i-As) being calibrated, for each factor of the
equation. The wavelengths represented in the loading plots as participating more highly
in the development of each factor are those that have greater spectral variation and
better correlation with the element in the calibration set.
It has been stated that the success of estimation via NIRS of specific mineral elements
in some grasses and legumes is usually dependent on the occurrence of those elements
in either organic or hydrated molecules [15]. At the very low concentrations in which i-
As is found in the rice samples used in this work (mean= 110 ng/g dw), any prediction
of this element has to be done on the basis of secondary correlations with plant
components [24,34]. This phenomenon is supported by data from MPLS loadings
(Figure 8) in this work for the selected equation for i-As. It can be concluded from
Figure 8, that C-H (912 nm) and also O-H (984 nm) groups of starch highly influenced
the first three MPLS loadings for this element. In addition, C-H groups of oil and fiber
(2308 and 2348 nm) also participated to model, mainly, the first term of the equation.
In the visible region of the spectrum, cromophores located in the caryopsis (absorption
at 672 nm and shorter wavelengths) also participated actively in constructing the first
terms. In spite of the low r value shown by the band at 912 nm (Figure 7), this band was
selected to highly participate in the first three terms of the equation for i-As, due to the
high variability in absorbance displayed by it (Figure 4).
13
Prediction results obtained from cross-validation showed for the first time that NIRS
can be employed with speciation purposes in rice, and that this technique is able to
predict the i-As concentration in samples of this species with sufficient accuracy for
screening purposes in spite of the low i-As levels shown in this work. Thus, NIRS can
be used for identifying those samples having low, medium and high i-As contents. In
the second step, the exact value of i-As of the samples selected by the researcher as
being of interest, can be obtained by the reference method. NIRS can, therefore,
decrease the number of analyses in the laboratory needed for monitoring the i-As
content in screening programs.
References
14
[1] M. Hossain, Long-term prospects for the global rice economy, FAO Rice
Conference. Rome, Italy, 12-13 February 2004.
[2] T. D. Phuong, P. V. Chuong, D. T. Khiem, S. Kokot, Elemental content of
Vietnamese rice Part 1. Sampling, analysis and comparison with previous studies,
Analyst 1999, 124, 553.
[3] F. F. Shih, An update on the processing of high-protein rice products,
Nahrung/Food 2003, 47, 420.
[4] R. W. Dabeka, A. D. Mckenzie, G. M. A. Lacroix, C. Cleroux, S. Bowe, R. A.
Graham, H. B. S. Conacher, P. Verdier, Survey of arsenic in total diet food
composites and estimation of the dietary intake of arsenic by Canadian adults and
children. J. AOAC Int. 1993, 76, 14.
[5] R. A. Schoof, L. J. Yost, J. Eickhoff, E. A. Crecelius, D. W. Cragin, D. M.
Meacher, D. B. Menzel, A market basket survey of inorganic arsenic in food,
Food Chem. Toxicol. 1999, 37, 839.
[6] W. H. Lamont, Concentration of inorganic arsenic in samples of white rice from
the United Sattes, J. Food Comp. Anal. 2003, 16, 687.
[7] R. A. Schoof, L. J. Yost, E. A. Crecelius, K. Irgolic, W. Goessler, H. R. Guo, H.
Greene, Dietary arsenic intake in Taiwanese district with elevated arsenic in
drinking water, Hum. Ecol. Risk Assess. 1998, 4, 117.
[8] T. Roychowdhury, T Uchino, H. Tokunaga, M. Ando, Survey of arsenic in food
composites from an arsenic-affected area of West Bengal, India, Food Chem.
Toxicol. 2002, 40, 1611.
[9] H. K. Das, A. K. Mitra, P. K. Sengupta, A. Hossain, F. Islam, G. H. Rabbani,
Arsenic concentration in rice, vegetables, and fish in bangladesh: a preeliminary
stuy, Environ. Int. 2004, 30, 383.
15
[10] D. T. Heitkemper, N. P. Vela, K. R. Stewart, C. S. Westphal, Determination of
total and speciated arsenic in rice by ion chromatography and inductively coupled
plasma mass spectrometry, J. Anal. At. Spectrom. 2001, 16, 299.
[11] S. D´Ilio, M. Alessandrelli, R. Cresti, G. Forte, S. Caroli, Arsenic contents of
various types of rice as determined by plasma-based techniques, Microchem. J.
2002, 73, 195.
[12] M. D´Amato, G. Forte, S. Caroli, Identification and quantification of major
species of arsenic in rice, J. AOAC Int. 2004, 87, 238.
[13] D. F. Malley, P. C. Williams, Use of near-infrared reflectance spectroscopy in
prediction of heavy metal in freshwater sediment by their association with organic
matter, Environ. Sci. Technol. 1997, 31, 3461.
[14] V. P. Krischenko, S. G. Samokhvalov, L. G. Fomina, G. A. Novikova, Use of
infrared spectroscopy for the determination of some properties of soil. In: I.
Murray, I. Cowe, (Eds.), Making Light Work: Advances in Near Infrared
Spectroscopy, VCH, Weinheim, 1992, p 239.
[15] D. H. Clark, E. E. Cary, H. F. Mayland, Analysis of trace elements in forages by
near infrared reflectance spectroscopy, Agron. J. 1989, 81, 91.
[16] R. Font, M. Del Río, A. De Haro, Use of near infrared spectroscopy to evaluate
heavy metal content in Brassica juncea plants cultivated on the polluted soils of
the Guadiamar river area, Fresenius Environmental Bulletin 2002, 11, 777.
[17] L. Sauvage, D. Frank, J. Stearne, M. B. Millikan, Trace metal studies of selected
white wines: an alternative approach, Anal. Chim. Acta 2002, 458, 223.
16
[18] J. C. Feo, A. J. Aller, Speciation of mercury, methylmercury, ethylmercury and
phenylmercury by Fourier transform infrared spectroscopy of whole bacterial
cells, Anal. At. Spectrom. 2001, 16, 146.
[19] R. Font, M. Del Río-Celestino, D. Vélez, A. De Haro-Bailón, R. Montoro, Visible
and near-infrared spectroscopy as a technique for screening the inorganic arsenic
content in the red crayfish (Procambarus clarkii Girard), Analytical Chemistry “in
press” 2004.
[20] S. Kokot, T. D. Phuong, Elemental content of Vietnamese rice. Part 2. Multivariate
data analysis, Analyst 1999, 124, 561.
[21] O. Muñoz, V. Devesa, M. A. Suñer, D. Vélez, R. Montoro, I. Urieta, M. L. Macho,
M. Jalón, Total and inorganic arsenic in fresh and processed fish products, J. Agr.
Food Chem. 2000, 48, 4369.
[22] J. S. Shenk, M. O. Westerhaus, Population structuring of near infrared spectra and
modified partial least squares regression, Crop Sci. 1991, 31, 1548.
[23] W. R. Hruschka, Data analysis: wavelength selection methods. In: P. C. Williams,
K. Norris (Eds.), Near-Infrared Technology in the Agricultural and Food
Industries, American Association of Cereal Chemists, Inc., St. Paul, 1987, p 35.
[24] J. S. Shenk, J. J. Workman, M. O. Westerhaus, Application of NIR spectroscopy to
agricultural products. In: D. A. Burns, E. W. Ciurczak, (Eds.), Handbook of Near
Infrared Analysis; Marcel Dekker, New York, 1992, p 383.
[25] R. J. Barnes, M. S. Dhanoa, S. J. Lister, Standard normal variate transformation
and de-trending of near-infrared diffuse reflectance spectra, Appl. Spectrosc.
1989, 43, 772.
17
[26] H. Martens, T. Naes, Multivariate calibration. John Wiley & Sons, New York,
1989, p 250.
[27] J. S. Shenk, M. O. Westerhaus, Calibration the ISI way. In: A. M.C. Davies, P. C.
Williams, (Eds.), Near Infrared Spectroscopy: The Future Waves, Nir
Publications, Chichester, 1996, p 198.
[28] P. C. Williams, D. C. Sobering, How do we do it: a brief summary of the methods
we use in developing near infrared calibrations. In: A. M.C. Davies, P. C.
Williams, (Eds.), Near Infrared Spectroscopy: The Future Waves, Nir
Publications, Chichester, 1996, p 185.
[29] J. J. Jr. Workman, Nir spectroscopy calibration basics. In: D. A. Burns, E. W.
Ciurczak, (Eds.), Handbook of Near-Infrared Analysis, Dekker Inc., New York,
1992, p 247.
[30] C. W. Chang, D. A. Laird, Near infrared reflectance spectroscopy analysis of soil
C and N, Soil Science 2002, 167, 110.
[31] B. W. Dunn, H. G. Beecher, G. D. Batten, S. Ciavarella, The potential of near
infrared reflectance spectroscopy for soil analysis -a case study from the Riverine
Plain of south-eastern Australia, Aust. J. Exp. Agr. 2002, 42, 607.
[32] D. H. Clark, H. F. Mayland, R. C. Lamb, Mineral analysis of forages with near
infrared reflectance spectroscopy, Agron. J. 1987, 79, 485.
[33] D. Cozzolino, A. Morón, The potential of near-infrared reflectance spectroscopy to
analyse soil chemical and physical characteristics, J. Agr. Sci. 2003, 140, 65.
18
[34] R. Font, M. Del Río, D. Vélez, R. Montoro, A. De Haro, Use of near-infrared
spectroscopy for determining the total arsenic content in prostrate amaranth, Sci.
Total Environ. 2004, 327, 93.
[35] B. G. Osborne, T. Fearn, P. H. Hindle, Practical NIR spectroscopy with
applications in food and beverage analysis. Longman Scientific & Technical,
Essex, 1993, p. 13.
[36] B. O. Juliano, D. B. Bechtel, The rice grain and its gross composition. In: B. O.
Juliano, (Ed.), Rice: Chemistry and Technology, The American Association of
Cereal Chemists, Inc., St. Paul, 1985, p 17.
[37] I. Murray, P. C. Williams, Chemical principles of near-infrared technology. In: P.
C. Williams, K. Norris (Eds.), Near-Infrared Technology in the Agricultural and
Food Industries, American Association of Cereal Chemists, Inc., St. Paul, 1987, p
17.
[38] M. Leonzio, Il contenuto di pentosani nel riso e nei suoi principali sottoprodotti.
Riso 1967, 16, 313.
[39] T. Yano, K. Suehara, Y. Nakano, Determination of the content of water and rice
bran in solid media used for mushroom cultivation using near-infrared
spectroscopy. Journal of Fermentation and Bioengineering 1988, 86, 472.
[40] R. Masironi, S. R. Koirtyohann, J. O. Pierce, Zinc, copper, cadmium and
chromium in polished and unpolished rice, Sci. Total Environ. 1977, 7, 27.
[41] A. A. Carbonell-Barrachina, M. A. Aarabi, R. D. Delaune, R. P. Gambrell, W. H.
Jr. Patrick, Bioavailability and uptake of arsenic by wetland vegetation effects, J.
Environ. Sci. Health 1998, A33(1), 45.
19
[42] A. A. Meharg, M. R. Macnair, An altered phosphate uptake system in arsenate-
tolerant Holcus lanatus L., New Phytol. 1990, 116, 29.
20
Table 1. Calibration and cross-validation statistics (ng/g, dry
weight) for inorganic arsenic for the selected equations (2, 5, 5,
2; SNV+DT), performed on the range from 400 to 2500 nm.
Calibration Cross-
validation
n range mean S.D. SEC R2 RPD r2
40 13.0-268.0 110.37 49.80 20.19 0.80 1.67 0.65
n= number of samples in the calibration file; range= minimum
and maximum reference values in the calibration file; S.D.=
standard deviation of the calibration file; SEC= standard error of
calibration; R2= coefficient of determination in the calibration;
RPD= standard deviation to standard error of cross-validation
ratio; r2= coefficient of determination in the cross-validation.
21
Figure Captions
Figure 1. First two principal component plot (PC1 vs PC2) for rice samples (n= 40)
used in this work.
Figure 2. Frequency distribution of inorganic arsenic (ng/g dw) in the samples used in
the study (n= 40).
Figure 3. Raw spectra (Log 1/R) of the rice samples used in this work (n= 40), in the
range from 400 to 2500 nm.
Figure 4. Second derivative spectra (2, 5, 5, 2; SNV+DT) of the raw optical data in the
range from 400 to 2500 nm.
Figure 5. Cross-validation scatter plot of laboratory vs. predicted values by NIRS for
inorganic arsenic in rice samples (n= 40) (ng/g dw).
Figure 6. Second derivative (2, 5, 5, 2; SNV+DT) spectra of a) brown and b) milled
rice samples used in this study.
Figure 7. Correlation plot for inorganic arsenic reference values vs. wavelength
absorbance by using SNV+DT algorithms, in the range from 400 to 2498 nm (n= 40).
Figure 8. MPLS loading spectra for inorganic arsenic in the second derivative (2, 5, 5,
2; SNV+DT) transformation. From the top to bottom, panels represent loadings for
factors 1, 2 and 3, respectively.
22
Figure 1.
23
Figure 2.
24
Figure 3.
25
Figure 4.
26
Figure 5.
0 50 100 150 200 250 300 350
50
100
150
200
250
inor
gani
c ar
seni
c (p
redi
cted
)
inorganic arsenic (laboratory)
27
Figure 6.
28
Figure 7.
29
Figure 8.
30