Multivariate Curve Resolution of Synchronous Fluorescence Spectra Matrices of Fulvic Acids Obtained as a Function of pH ´ TAULER JOAQUIM C. G. ESTEVES da SILVA* and ROMA Centro de Investigac¸a˜o em Quı´mica, Departamento de Quı´mica, Faculdade de Cieˆncias da Universidade do Porto, R. Campo Alegre 687, 4169-007 Porto, Portugal (J.C.G.E.d.S.); and Department of Environmental Chemistry, Institute of Chemical and Environmental Research, CSIC, Jordi Girona 18–26, 08034, Barcelona, Spain (R.T.)
Synchronous fluorescence spectra (excitation wavelength range between 280 and 510 nm and wavelength interval of 25 nm) of three samples of fulvic acids (FA) were obtained as a function of the pH, in the range from 2.0 to 10.5, and as a function of the FA concentration, in the range from 20 to 180 mg/L. FA were obtained from composted livestock materials (lsFA), composted sewage sludge (csFA), and Laurentian soil (laFA). Threedimensional spectral matrices were obtained (wavelength, pH, and FA concentration) and multivariate curve resolution (MCR) was used to calculate spectra and fluorescence intensity profiles for the detected components. Cluster analysis of the calculated spectra showed the existence of similar and unique fluorescent properties in the three FA samples. Some of the calculated fluorescence intensity profiles have a shape compatible with acid–base species distribution diagrams, which allowed pKa values to be estimated, namely, a well-defined acid–base equilibrium with pKa 5.7 6 0.2 (lsFA), 6.9 6 0.4 (csFA), and 5.5 6 0.2 (laFA); and other acid–base systems not well defined with pKa at about 3.0 and 8.6. Other spectral variations revealed the existence of inner-filter effects or self-quenching as the concentration of FA increases. Index Headings: Synchronous fluorescence; Fulvic acids; Acid–base equilibrium; Multivariate curve resolution; MCR; Alternating least squares; ALS.
INTRODUCTION Fulvic acids (FA) are among the most reactive structures of soil organic matter, participating in acid–base reactions and interacting with metal ions and organic substances.1 Molecular synchronous fluorescence (SyF) spectroscopy is a particularly useful technique for the study of the chemical equilibrium properties of FA because it is a highly sensitive analytical technique, allowing measurements to be made at natural environmental concentrations, and measurements are nondestructive.2–17 Moreover, SyF spectra of FA are characterized by a great amount of information,5,12,15–17 which is an important property for FA studies because they are a complex mixture of substances with somewhat similar properties. The SyF spectra of FA are quite sensitive to the sample pH.6,8–16 The origin of these spectral variations with the pH was attributed to different acid–base and spectroscopic properties of the most important pH-reactive structures. Selfmodeling curve resolution methods based on factor analysis have been proposed for the reduction of the SyF spectral variations in the spectra and concentration profiles (from which pKa values could be estimated) of a small number of components.8,10,13,15,16 These methods were based on the decomposition of one two-way data matrix composed of SyF spectra (wavelength) collected as a function of the pH. In order to increase the information about FA samples using Received 7 June 2006; accepted 1 September 2006. * Author to whom correspondence should be sent. E-mail:
[email protected]. pt.
Volume 60, Number 11, 2006
the pH-induced SyF spectra variations, a further experimental factor can be changed, namely the concentration of FA. This methodology originates three-way data structures for one FA sample (wavelength 3 pH 3 concentration). The increase in the concentration of FA will originate modifications in the SyF spectra as a consequence of filter effects due to the relatively high optical density of the organic matter.18,19 Multivariate curve resolution (MCR), which is a constrained interactive alternating least squares procedure, is suitable for the analysis of these three-way data matrices that deviate from the true trilinear model.20–25 Indeed, the data matrices under analysis in this work hardly follow a trilinear model, mainly because of the following points: although the spectra collected as a function of the concentration show some distortions due to filter effects, a high colinearity degree is expected; and exact pH value reproduction in different titrations is almost impossible to achieve. The objective of the work described in this paper is to propose a new methodology for the study of the acid–base properties of FA based on these three-way data matrices and in the use of MCR to reduce the experimental data in the SyF spectra, pH, and FA concentration profiles of the detected acid– base systems. This paper shows the results of the proposed methodology for three samples of FA: lsFA, FA extracted from compost livestock; csFA, FA extracted from composted sludge from a wastewater treatment plant; and laFA, a sample of Laurentian FA.
THEORY For each titration, all the SyF spectra measure after each addition of titrant were arranged in a data matrix D. This matrix D has m rows (number of SyF spectra recorded as function of the pH) and n columns (number of wavelengths of the SyF spectra: in this work n was always equal to 47). At a determined pH value the SyF spectra of FA is the sum of the individual SyF spectra of the fluorescent constituents of FA multiplied by the correspondent concentration. If the fluorescent constituent has no acid–base properties its concentration is constant throughout the titration. On the contrary, if the constituent undergoes ionization in the pH window under analysis, its concentration profile has the shape of an acid–base species distribution diagram. The bilinear decomposition of the matrix D can be described by the following equation: D ¼ CST þ E
ð1Þ
where C is the matrix describing the concentration (or fluorescence intensity) profiles of every detected component, ST is the matrix describing the individual ‘‘pure’’ SyF spectrum
0003-7028/06/6011-000000$2.00/0 Ó 2006 Society for Applied Spectroscopy
APPLIED SPECTROSCOPY
of these components, and E is the residual matrix describing the variance not explained by CST. In this work, several D matrices were obtained for each sample of the three FA samples under analysis by independent titration of different concentration solutions (ns) of the FA (six or seven solutions in the range 20 to 180 mg/L): matrices D1, D2,. . ., Dns. These matrices were arranged in an augmented column-wise data matrix. This augmented data matrix has a number of rows equal to the total number of acquired SyF spectra (m 3 ns) and it has a number of columns equal to the number of wavelengths of the SyF spectra (n ¼ 47). This column-wise data matrix augmentation assumes that common vectors span the column vector spaces of the different individual data matrices, i.e., that common SyF spectra are present in the four data sets. The linear model given by Eq. 1 can be easily extended to the augmented data matrix as follows: Daug ¼ Caug ST þ E
ð2Þ
In this work three augmented column-wise data matrices are obtained, D1aug, D2aug, and D3aug, one for each FA sample (lsFA, csFA, and laFA). The resolution power of the MCR–alternating least squares (ALS) method improves considerably if several experiments, giving data matrices obtained for the same chemical system under different experimental conditions, are simultaneously analyzed. Moreover, interrelations between different FA samples can be obtained. In order to evaluate the fitting error in the reproduction of the original matrix using the solutions found either by principal components analysis (PCA) or by MCR-ALS, a percentage of lack of fit (lof) value was calculated using the following equation: vffiffiffiffiffiffiffiffiffiffiffiffi ffi uX 2 u e ij u u i; j ð3Þ lofð%Þ ¼ 100 3 uX 2 t dij i; j
where eij are the elements of the residuals matrix E and dij are the elements of the data set (D or Daug).
EXPERIMENTAL Reagents. Anthropogenic FA were extracted from two samples of livestock waste (mixed with vegetable residues) (lsFA) and composted solid wastes derived from sewage sludge (csFA). FA were isolated by a procedure recommended by the International Humic Substances Society.26 The characterization of the anthropogenic FA samples has been done previously.15,16 A commercial FA sample extracted from a Laurentian soil (Laurentian FA, laFA) was obtained from Fredriks Research Products (The Netherlands). Solutions of FA ranging from 20 to 180 mg/L were prepared in 0.1 M potassium nitrate. The initial pH of the FA solutions was adjusted to pH ¼ 2.0 with a standard solution of nitric acid (0.1 M). A solution of decarbonated potassium hydroxide (about 0.05 M) was used as the titrant and the total amount added at the end of the titration did not cause significant dilution of the FA solution (less than 5%). Instruments and Procedures. Potentiometric titrations with pH measurement were conducted with a PC-controlled system, assembled with a Crison MicropH 2002 pHmeter and a Crison
Volume 60, Number 11, 2006
MicroBu 2030 microburette. The experiments were done under nitrogen at 25 6 0.2 8C. The cell was calibrated with three buffer solutions with ionic strength adjusted to 0.1 M.27 Synchronous fluorescence measurements were made with a Perkin-Elmer LS-50 luminescence spectrometer with a flow cell. A Gilson Minipuls 2 peristaltic pump forced the displacement of the titrated solution into the flow cell after each addition of potassium hydroxide. During pH and SyF spectra measurements the pump was turned off. The spectra were recorded with the following settings: between 280 and 510 nm; 0.5 nm acquisition interval between points; 7.5 nm excitation and emission slit widths; wavelength difference of 25 nm; and scan rate of 200 nm min1. The resolution of the raw spectra was reduced to 5 nm before data analysis (47 points per spectrum) by eliminating intermediate raw spectral data points. Programs. The calculations associated with the MCR-ALS method were performed using several programs implemented in MATLAB and obtained from http://www.ub.es/gesq/mcr/ mcr.htm. Hierarchical cluster analysis was performed with SPSS (Statistical Package for the Social Sciences) using the single linkage and Ward methods, and Euclidean distances were calculated. Multivariate Curve Resolution–Alternating Least Squares Constraints. The MCR-ALS interactive procedure can be subjected to several constraints according to the physico-chemical properties of the data structure under analysis.20–25 In this work, non-negativity and equality constraints were tested. Non-negativity constraints on both the spectra and the concentration profiles (fluorescent intensity profiles) resulted from the physical meaning of a fluorescence spectrum because the intensity must always be positive. Equality constraints were applied when the fluorescence intensity profile of a component has the shape of a species distribution diagram as function of the pH (S or Z curves).15 This shape can be observed by evolving factor analysis8,10,11,15,16,28,29 or by the MCR-ALS unconstrained solution. If the fluorescence intensity profile corresponds to an ionization of the acid species (decreasing trend), then the concentration at the end of the titration is zero (at this point the acid is completed ionized). If the fluorescence intensity profile corresponds to the formation of the conjugate base species (increasing trend), then the concentration at the beginning of the titration is zero. If the fluorescence intensity profile corresponds to the conjugate base (first ionization) and acid species (second ionization) of a diprotic acid (increasing followed by a decreasing trend), then the concentration of this species is zero at the beginning and at the end of the titration. In this work a harder equality constraint (equal to practically zero) and a softer equality (unequality) constraint (equal to or lower than a threshold value equal to one) were tested.
RESULTS AND DISCUSSION Preliminary Analysis of the Synchronous Fluorescence Matrices. Some typical SyF spectra of the three FA samples and the effect of varying pH on their intensities are shown in Fig. 1. The comparative analysis of these spectra shows that the pH induces marked spectral variations on the SyF spectra of csFA and minor variations on the SyF spectra of lsFA and laFA samples. However, the pH-induced variation of these two samples apparently is more complex because the intensity of
FIG. 2. SyF spectra of (top) lsFA, (middle) csFA, and (bottom) laFA at pH 3 and at four concentrations: () 20 mg/L; (3) 80 mg/L; (m) 110 mg/L; and (&) 140 mg/L.
FIG. 1. SyF spectra of 140 mg/L solutions of (top) lsFA, (middle) csFA, and (bottom) laFA at four pH values: () 2.0; (3) 5.3; (&) 7.5; and (m) 10.5.
several overlapped bands oscillates, particularly for the natural FA sample (laFA). The shape of the SyF spectra of FA may be affected by the concentration because of inner-filter effects due to the strong absorption of these substances.30–33 A red shift is observed when humic substance concentration increases as a consequence of the attenuation of the fluorescence of the fluorescence emission in the shorter wavelength region.30 Indeed, several fluorophores present in the FA samples show different concentration dependence as a consequence of the inner-filter effects.33 Figure 2 shows typical SyF spectra of the three FA samples at pH ¼ 3 at four concentrations. The analysis of these spectra shows that spectral distortions and nonlinearities are observed in the low wavelength range, around 300 nm. This effect is
particularly marked for the lsFA and laFA samples because the global spectral intensity of these samples is smaller than that observed for csFA. Principal Components Analysis of the Synchronous Fluorescence Matrices. The three sets of SyF spectra were subjected to PCA, either the individual concentration matrices or the augmented data matrices, in order to obtain an estimation of the intrinsic number of components of each FA sample. In this case, each component represents a fluorophore present in the FA sample whose fluorescent properties vary with the pH. Acid–base chemical reactions and physical processes (for example, quenching and inner-filter effects) may be responsible for these spectral variations. Table I shows the PCA results of the 140 mg/L individual concentration matrix and of the augmented data matrices. Although the analysis of these results is not straightforward, three types of information can be obtained: (1) three or four components are necessary to describe the data set variance; (2) the lsFA and laFA samples require a higher number of components than the csFA sample; and (3) the augmented data
APPLIED SPECTROSCOPY
TABLE I. PCA results of SyF matrices corresponding to 140 mg/L and augmented data matrices.a lsFA
csFA
%V
%CV
140 mg/L matrices 1 3.8 3 104 2 2.9 3 101 3 5.2 3 100 4 2.2 3 100 5 7.5 3 101
99.90 0.08 0.01 0.01 0.00
99.90 99.98 99.99 100.00 100.00
4.1 3.5 1.5 1.9 9.5
3 3 3 3 3
Augmented 1 2 3 4 5
85.94 11.21 1.97 0.25 0.13
85.94 87.15 99.11 99.37 99.50
3.2 1.1 3.3 3.5 1.1
3 10 3 10
Nc
a
EV
matrices 4.0 3 10 5.3 9.2 3 101 1.2 3 101 6.2 3 102
EV 104 102 101 100 101
3 101 3 101
laFA
%V
%CV
99.12 0.83 0.01 0.00 0.00
99.12 99.95 99.99 100.00 100.00
67.23 24.24 6.96 0.75 0.24
67.23 91.46 98.43 99.18 99.42
%V
%CV
104 101 101 101 101
99.71 0.21 0.07 0.00 0.00
99.71 99.92 99.99 99.99 99.99
3.5 3 10 8.5 2.3 4.8 3 101 9.9 3 102
74.94 18.17 4.88 1.03 0.21
74.94 93.11 97.98 99.01 99.22
EV
1.5 3.2 1.0 5.0 3.1
3 3 3 3 3
Nc: number of eigenvalues; EV: eigenvalue; %V: percentage of variance of each eigenvalue; %CV: cumulative percentage of variance.
matrices require more components than the single concentration matrices. This last result supports our previous discussion about the deviations of the trilinear model of the three-way data structures under investigation as a consequence of spectral distortions due to inner-filter effects and to the impossibility of exact pH value reproduction in different experiments. Multivariate Curve Resolution–Alternating Least Squares of the Synchronous Fluorescence Matrices. The three augmented data matrices were analyzed by MCR-ALS forcing the number of components between three and five and using non-negativity constraints (concentrations and spectra) and equality constraints (concentration or fluorescent intensity profiles). From the analysis of the MCR-ALS results, the SyF data set of the lsFA, csFA, and laFA samples are best described using four, three, and four components, respectively. Table II shows the fitting quality parameters of the best MCR-ALS model observed; only non-negativity in both the spectra and concentration were used. When equality (equal to practically zero) or unequality (equal to or lower than one) constraints were used, the worst fittings were found. Nevertheless, the calculated spectra and fluorescence intensity profiles show a global shape similar to that obtained when only non-negativity constraints were used. Consequently, equality constraints were not considered in further analysis. Figures 3, 4, and 5 show the calculated spectra and fluorescence intensity profiles (concentration and pH) for the three FA samples. Analysis of the Multivariate Curve Resolution–Alternating Least Squares Results. All the calculated spectra
(Figs. 3a, 4a, and 5a) have a shape compatible with a SyF spectrum, i.e., the spectral peaks are not too narrow or as large as a typical emission spectrum. The analysis of the fluorescence intensity profiles of the lsFA sample (Fig. 3a) shows three different types of fluorescence intensity variation with the concentration and pH. Type i. The component represented by (3) shows no marked relative variation either by the concentration or the pH. This component probably corresponds to a background signal common to all individual concentration data sets. This
TABLE II. Statistical quality parameters of the best MCR-ALS model.a Matrix
Nc
PCA (%lof)
ALS (%lof)
R2
Daug (lsFA) Daug (csFA) Daug (laFA)
4 3 4
0.004570 0.001593 0.002518
0.7740 1.4494 1.0424
99.994 99.979 99.989
a
Nc: number of components; PCA (%lof): percentage of lack of fit values between the experimental data matrix and the reproduced data matrices by PCA and ALS; ALS (%lof) percentage lack of fit values between the experimental data and ALS fitting; R2: percentage of variance explained at the optimum.
Volume 60, Number 11, 2006
FIG. 3. SyF (a) spectra and (b) concentration profiles obtained by MCR-ALS for the lsFA sample: (m) first, (&) second, (*) third, and (3) fourth component.
FIG. 4. SyF (a) spectra and (b) concentration profiles obtained by MCR-ALS for the csFA sample: (m) first, (&) second, and (*) third component.
background signal is probably due to an instrumental artifact usually present in the fluorescent spectra of humic substances.33 Type ii. The component represented by (m) shows a slight variation with the pH and its intensity decreases with the concentration of the FA sample. The variation (increase of fluorescence intensity) with the pH, observed in the most acid pH region (beginning of the titration), corresponds to an acid– base fluorophore. The decrease of the intensity with the concentration must be due to an inner-filter effect resulting from the increase of the FA concentration. Indeed, the maximum of the SyF spectra corresponding to this component fall in the ultraviolet (UV) spectral region (at about 320 nm), which is most attenuated by inner-filter effects.30 Type iii. The components represented by (*) and (&) show a typical variation of a monoprotic acid species (&) and the corresponding conjugated base species (*). The analysis of the fluorescence intensity profiles of the csFA sample (Fig. 4b) shows marked similarities with those calculated for the lsFA sample, namely, with the previously mentioned Types (ii) and (iii). However, the component represented by (m) shows no decrease in intensity with the increase of the concentration of the FA but its intensity remains constant (Fig. 4b). Also, besides the increase of intensity due to an acid ionization in the beginning of the titration, this increase is followed by a decrease of the intensity, probably due to an acid ionization.
FIG. 5. SyF (a) spectra and (b, c) concentration profiles obtained by MCRALS for the laFA sample: (m) first, (&) second, (*) third, and (L) fourth component.
The fluorescence intensity profiles obtained for the laFA sample are more complex to analyze and are shown in Figs. 5b and 5c. Indeed, besides the existence of fluorescence intensity profiles with marked similarities with those calculated for the lsFA and csFA sample, namely, with the previously mentioned Types (ii) and (iii), there is a fourth fluorescence intensity profile with a particular type of variation (represented by L in Fig. 5c). Type iv. This type is characterized by a marked increase of intensity, beginning at a zero value, followed by a decrease to zero intensity. This type of component probably corresponds to an acid fluorophore that becomes fluorescent upon ionization but turns into a non-fluorescent species when it undergoes the second ionization. Acid Ionization Constants. The analysis of the calculated fluorescent intensity profiles (Figs. 3b, 4b, 5b, and 5c) shows the existence of several variations compatible with acid ionizations. The variations corresponding to Type (iii) (* and
APPLIED SPECTROSCOPY
FIG. 6. Dendrograms obtained by hierarchical cluster analysis of the spectra of the FA samples calculated by MCR-ALS. The calculated SyF spectra of the components are designated by: LS1, LS2, LS3, and LS4: lsFA sample; CS1, CS2, and CS3: csFA sample; and LA1, LA2, LA3, and LA4: laFA sample.
) clearly have the shape of a species distribution diagram of a monoprotic acid. From the graphical observation of the plots crossover of two conjugated systems at about 50% ionization (one plot for each FA concentration under analysis), the pKa of this acid–base system can be estimated. The pKa for this acid– base system for the three FA samples are (average of the pKa values calculated for each concentration and respective standard deviation): lsFA, 5.7 6 0.2; csFA, 6.9 6 0.4; and, laFA, 5.5 6 0.2. This result shows that this component of the lsFA and laFA samples has similar acid–base properties and that of csFA is a weaker acid than the other two FA. No trend was detected in the calculated pKa induced by the increase from 20 to 180 mg/L in the FA concentration. Other variations compatible with acid ionizations can be detected in the calculated fluorescent intensity profiles but the assignment of the corresponding pKa is somewhat fuzzy and, consequently, their values should be understood as rough estimations: the variations corresponding to Type (ii) (m), and present in the three FA samples, correspond to the conjugated base of an acid–base system with a pKa about 2.8; the variations corresponding to Type (iv) (L) of the laFA sample reveal a fluorophore with two pKas at about 3.1 and 8.6. Similarity Analysis of the Calculated Synchronous Fluorescence Spectra. From the above discussion about the &
Volume 60, Number 11, 2006
fluorescence intensity profiles calculated for the three FA samples several similarities were detected. The next step was to verify whether the spectra of the components detected in the FA samples have similarities. Figure 6 shows the dendograms obtained by cluster analysis of the calculated spectra of the components (four spectra for lsFA, three spectra for csFA, and, four spectra for laFA). The analysis of Fig. 6 clearly shows the existence of two clusters: (1) Cluster A composed of the spectra cs3, la1, and ls1; and (2) Cluster B composed of the spectra ls2, ls3, la2, and la3. The other four spectra (cs1, cs2, la4, and ls4) do not cluster. Cluster A correspond to the similar fluorescence intensity profiles calculated for the three FA samples [Type (i) above] and represented by the symbol (m) in Figs. 3b, 4b, and 5c. Cluster B corresponds to the similar fluorescence intensity profiles detected in the lsFA and laFA samples of Type (ii) (* and & in Figs. 3b and 5b), which also have similar acid–base properties (pKa about 5.6). This result shows that the same fluorophores having acid–base properties exist in the lsFA and laFA samples; indeed, as discussed above, these fluorophores have similar acid–base properties with a pKa about 5.6. Two of these fluorophores have similar SyF spectra characterized by two overlapped bands with maxima at about 340 and 400 nm
and the third fluorophore has a SyF spectrum with a maximum at 320 nm. This last fluorophore is also present in the csFA sample. The other four spectra that do not cluster contain particular fluorescent properties of the three FA samples, revealing their different origins. The ls4 spectrum corresponds to the abovedefined Type (i) component that was only detected as background fluorescence of the lsFA sample. The la4 SyF spectrum has a maximum at 460 nm and corresponds to the above defined Type (iv) component that was only detected in the laFA sample. The cs1 and cs2 spectra correspond to the fluorescence intensity profiles of the Type (iii) component but are different from the spectra of Cluster B. The main differences are a pronounced maximum at 350 nm (cs1) and the band at 420 nm (cs2) besides the two characteristic bands of the Cluster B spectra (340 and 400 nm).
CONCLUSION Multivariate curve resolution–alternating least squares is a particularly useful soft-modeling data analysis technique for the interpretation of complex spectroscopic data matrices. Indeed, in the case of three-way data with small deviations from trilinearity, MCR-ALS succeeded in the resolution of the spectroscopic information. The SyF spectra of the FA from different origins collected as a function of the pH and concentration of FA were successfully resolved in the spectra and corresponding fluorescent intensity profiles (as a function of the pH and concentration) of a small number of components. Different spectral variations, induced by varying the concentration of FA or pH, and resulting from different physico-chemical mechanisms, were detected, for example, quenching, inner-filter effects, and acid–base reaction. Although the FA samples under analysis have quite different origins, similarities among the samples were detected as well as particularities. Indeed, either similar spectra or fluorescence intensity profiles were detected in the three FA samples under study, suggesting that similar chemical constituents were responsible for the fluorescent and acid–base properties. However, all three samples contain some components having unique fluorescent properties, revealing their different origins. ACKNOWLEDGMENTS Integrated Actions Portugal-Spain E26/01 has financially supported this work. Sandra L.R.R.S. Bastos is acknowledged for performing some of the experimental work.
1. F. S. Stevenson, Humus Chemistry – Genesis, Composition and Reactions (John Wiley and Sons, New York, 1992). 2. F. H. Frimmel and H. Bauer, Sci. Total Environ. 62, 139 (1987). 3. M. R. Provenzano, T. M. Miano, and N. Senesi, Sci. Total Environ. 81/82, 129 (1989). 4. N. Senesi, T. M. Miano, M. R. Provenzano, and G. Brunetti, Sci. Total Environ. 81/82, 143 (1989). 5. N. Senesi, Anal. Chim. Acta 232, 77 (1990). 6. S. E. Cabaniss, Environ. Sci. Technol. 26, 1133 (1992). 7. T. M. Miano and N. Senesi, Sci. Total Environ. 117/118, 41 (1992). 8. A. A. S. C. Machado and J. C. G. Esteves da Silva, Chemom. Intell. Lab. Syst. 19, 155 (1993). 9. J. C. G. Esteves da Silva and A. A. S. C. Machado, Talanta 41, 2095 (1994). 10. J. C. G. Esteves da Silva and A. A. S. C. Machado, Anal. Lett. 28, 2401 (1995). 11. E. Casassas, I. Marqueˆs, and R. Tauler, Anal. Chim. Acta 310, 473 (1995). 12. M. J. Pullin and S. E. Cabaniss, Environ. Sci. Technol. 29, 1460 (1995). 13. A. A. S. C. Machado, J. C. G. Esteves da Silva, and J. A. C. Maia, Anal. Chim. Acta 292, 121 (1994). 14. J. C. G. Esteves da Silva and A. A. S. C. Machado, Chemom. Intell. Lab. Syst. 27, 115 (1995). 15. J. C. G. Esteves da Silva and A. A. S. C. Machado, Analyst (Cambridge, U.K.) 122, 1299 (1997). 16. J. C. G. Esteves da Silva, A. A. S. C. Machado, and M. A. B. A. Silva, Water Res. 32, 441 (1998). 17. W. Chen, P. Westerhoff, J. A. Leenheer, and K. S. Booksh, Environ. Sci. Technol. 37, 5701 (2003). 18. B. C. MacDonald, S. J. Lvin, and H. Patterson, Anal. Chim. Acta 338, 155 (1997). 19. F. Fang, S. Kanan, H. Patterson, and C. S. Cronan, Anal. Chim. Acta 373, 139 (1998). 20. R. Tauler, A. Izquierdo-Ridorsa, and E. Casassas, Chemom. Intell. Lab. Syst. 18, 293 (1993). 21. R. Tauler, A. K. Smilde, and B. R. Kowalski, J. Chemom. 9, 31 (1995). 22. R. Tauler, Chemom. Intel. Lab. Syst. 30, 133 (1995). 23. J. Saurina, S. Herna´ndez-Cassou, R. Tauler, and A. Izquierdo-Ridorsa, J. Chemom. 12, 183 (1998). 24. R. Tauler, R. Gargallo, M. Vives, and A. Izquierdo-Ridorsa, Chemom. Intell. Lab. Syst. 46, 275 (1999). 25. A. K. Smilde, R. Tauler, J. Saurina, and R. Bro, Anal. Chim. Acta 398, 237 (1999). 26. E. M. Thurman, ‘‘Isolation of soil and aquatic humic substances (Group Report)’’, in Humic Substances and their Role in the environment, A. Piccolo, Ed. (Wiley-Interscience, Chichester, UK, 1988), pp. 31–43. 27. M. T. S. Vasconcelos and A. A. S. C. Machado, Talanta 33, 919 (1986). 28. H. Gampp, M. Meader, C. J. Meyer, and A. D. Zuberbuhler, Talanta 32, 1133 (1985). 29. H. Gampp, M. Meader, C. J. Meyer, and A. D. Zuberbuhler, Talanta 33, 943 (1986). 30. J. J. Mobed, S. L. Hemmingsen, J. L. Autry, and L. B. McGown, Environ. Sci. Technol. 30, 3061 (1996). 31. B. C. MacDonald, S. J. Lvin, and H. Patterson, Anal. Chim. Acta 338, 155 (1997). 32. T. Ohno, Environ. Sci. Technol. 36, 742 (2002). 33. M. C. G. Antunes, J. C. G. Esteves da Silva, and Anal. Chim., Acta 546, 52 (2005).
APPLIED SPECTROSCOPY
