Analytica Chimica Acta 588 (2007)

Analytica Chimica Acta 588 (2007) 224 230 Effect of temperature variation on the visible and near infrared spectra of wine and the consequences on the partial least square calibrations developed to measure chemical composition D. Cozzolino a,b,,l.liu a,b,c,1, W.U. Cynkar a,b,1, R.G. Dambergs a,b,1, L. Janik a,b,1, C.B. Colby c, M. Gishen a,b,1 a The Australian Wine Research Institute, Waite Road, Urrbrae, PO Box 197, Adelaide, SA 5064, Australia b Cooperative Research Centre for Viticulture, PO Box 154, Adelaide, SA 5064, Australia. c School of Chemical Engineering, Engineering North Building, The University of Adelaide, Adelaide, SA 5005, Australia Received 29 November 2006; received in revised form 29 January 2007; accepted 31 January 2007 Available online 6 February 2007 Abstract Many studies have reported the use of near infrared (NIR) spectroscopy to characterize wines or to predict wine chemical composition. However, little is known about the effect of variation in temperature on the NIR spectrum of wine and the subsequent effect on the performance of calibrations used to measure chemical composition. Several parameters influence the spectra of organic molecules in the NIR region, with temperature being one of the most important factors affecting the vibration intensity and frequency of molecular bonds. Wine is a complex mixture of chemical components (e.g. water, sugars, organic acids, and ethanol), and a simple ethanol and water model solution cannot be used to study the possible effects of temperature variations in the NIR spectrum of wine. Ten red and 10 white wines were scanned in triplicate at six different temperatures (25 C, 30 C, 35 C, 40 C, 45 C and 50 C) in the visible (vis) and NIR regions (400 2500 nm) in a monochromator instrument in transmission mode (1 mm path length). Principal component analysis (PCA) and partial least squares (PLS) regression models were developed using full cross validation (leave-one-out). These models were used to interpret the spectra and to develop calibrations for alcohol, sugars (glucose + fructose) and ph at different temperatures. The results showed that differences in the spectra around 970 nm and 1400 nm, related to O H bonding were observed for both varieties. Additionally an effect of temperature on the vis region of red wine spectra was observed. The standard error of cross validation (SECV) achieved for the PLS calibration models tended to inverse as the temperature increased. The practical implication of this study it is recommended that the temperature of scanning for wine analysis using a 1 mm path length cuvette should be between 30 C and 35 C. 2007 Elsevier B.V. All rights reserved. Keywords: Near infrared spectra; Temperature; Wine; Principal components; Partial least squares; Spectral changes; Hydrogen bonding 1. Introduction The near infrared (NIR) spectra of agricultural products arise from overtones and combinations of vibrations of molecular bonds of the organic components [1]. In order to relate spectral properties to chemical information, several chemometric methods such as multiple linear regression (MLR), principal component regression (PCR), partial least squares regression Corresponding author at: The Australian Wine Research Institute, Waite Road, Urrbrae, PO Box 197, Adelaide, SA 5064, Australia. Fax: +61 8 8303 4373. E-mail address: Daniel.Cozzolino@awri.com.au (D. Cozzolino). 1 Fax: +61 8 8303 6601. (PLS), locally weighted regression (LWR) and neural networks are frequently used [2 10]. In recent years, two-dimensional (2D) correlation spectroscopy was used to examine the effect of temperature induced changes in different materials [11 14]. In routine food analysis using NIR spectroscopy methods, the main drawback of calibration models developed for a specific food or beverage is their lack of robustness [6,10]. The lack of robustness in a calibration can be due to several causes such as variable sample temperature, spectrophotometer temperature or ambient stray light, among others [1]. The external factor that has been most widely reported as affecting calibration robustness is sample temperature. Vibrational spectra of liquid samples are not only primary molecular features such as chemical structure and functional groups, but also intramolecular features such as 0003-2670/$ see front matter 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2007.01.079

D. Cozzolino et al. / Analytica Chimica Acta 588 (2007) 224 230 225 hydrogen bonding. These weaker forces that influence molecular bond and therefore their vibration modes are themselves affected by conditions such as temperature and pressure [2]. Many authors have demonstrated the sensitivity of NIR spectra to sample temperature [1 14]. It is well known that NIR spectra of alcohols are very sensitive to temperature because the self-associated forms of the alcohols dissociated into small oligomers, dimers and monomers as a function of temperature [5,6]. Hansen et al. [8] demonstrated that temperature changes affect the vibration intensity of molecular bonds, so the spectrum will change according to the temperature variation. Swierenga et al. [9] reported that an increase in temperature results in a decrease in the number of hydroxyl groups involved in hydrogen bonding, and consequently the absorption band of free hydroxyl increases. The same authors reported that the second overtone absorption band of the hydroxyl group of ethanol and iso-propanol (around 970 nm) increases as the temperature increases. The hydroxyl group gives a sharper band for the free O H group and a broader one for the stretch mode of hydrogen-bonded OH groups [2,5,6]. With increasing temperature, the broad band, that can be seen as an overlay of many bands that belong to different cluster sizes of molecules formed by the hydrogen bonding, is shifted towards lower energies (longer wavelengths) as the degree of hydrogen bonding decreases [2]. The existence of temperature dependent behaviour of the bands due to rotational isomerism of the monomer terminal free O H groups, weakly hydrogen bonded O H groups and hydrogen bonded O H of the self-associated species is evident from spectroscopic analysis [5,6]. Similar phenomena were described by various authors using mixtures of water and ethanol, ethanol and iso-propanol, consequently, sample temperature can affect the result of either classification or calibration models when a NIR spectrum is used [1 15]. Wine is a complex mixture of chemical components (e.g. sugars, ethanol, organic acids, metal ions). These components can themselves affect hydrogen bonding, therefore a simple ethanol and water model solution should not be used to study the effect of temperature on the NIR spectrum of wine. Although some studies have been done using NIR spectroscopy to either classify wines or to predict wine chemical composition, little is known about the effect of different temperatures on the NIR spectrum of wine and subsequent effect on calibration performance [16 19]. The objective of this study was to investigate the effect of temperature on the visible (vis) and NIR spectra of wine and on the predictive ability of calibration models for the measurement of wine chemical composition. 2. Materials and methods 2.1. Wine samples and chemical analysis Ten white and 10 red wine samples were collected randomly from the AWRI Analytical Service laboratory. All samples were commercially available bottles of Australian wine. The red wine sample (labelled as Rws) set was a composite of Cabernet Sauvignon (n = 5), Shiraz (n = 2), Pinot Noir (n = 1), a blend of Cabernet Sauvignon and Shiraz (n = 1) and Rose (n = 1) wines. The white wine sample (labelled as Wws) set was a composite of Chardonnay (n = 4), Pinot Gris (n = 1), Riesling (n = 1), Semillon (n = 2), Sauvignon Blanc (n = 1) and Verdelho (n = 1) wines. Each bottle was analysed for alcohol content, ph, titratable acidity (TA), and glucose plus fructose (G + F) using a mid-infrared spectrophotometer (Foss WineScan TM FT 120; Foss, Hillerød, Denmark). Note that the aim of this study was not to develop NIR calibrations for wine compositional parameters but rather to test the effect of the temperature on such calibration models. Therefore, the calibrations developed in this study were only used as an indication of the effect of the different temperature treatments. 2.2. Spectroscopic measurements Both MilliQ (deionised) water and wine samples were scanned in the vis and NIR wavelength regions (400 2500 nm) using a scanning monochromator FOSS NIRSystems6500 (FOSS NIRSystems, Silver Spring, MD, USA). Spectral data collection was performed with Vision software (version 1.0, FOSS NIRSystems, Silver Spring, USA). Samples were analysed in transmission mode using a 1 mm path length cuvette after equilibration in the instrument at 30 C, 35 C, 40 C, 45 C, 50 C, and at room temperature ( 25 ± 1 C) for 2 min. Spectral data were stored as the logarithm of the reciprocal of transmittance [log (1/T)] at 2 nm intervals (1050 data points). Each sample was scanned in triplicate (repack) and the spectra were averaged for chemometric analysis. Air was used as reference (empty sample holder). 2.3. Data analysis and interpretation Spectra were exported from the Vision software in NSAS format to The Unscrambler software (Version 9.1, CAMO ASA, Oslo, Norway) for chemometric analysis. Principal component analysis (PCA) was performed to examine the dominant patterns in the spectral data. Calibration models between chemical composition and NIR spectra were developed using partial least square (PLS) with full cross validation [20]. The spectra were transformed using the second derivative (Savitzky Golay transformation, 10 point smoothing and second order filtering) before calibration models were developed. In order to evaluate the effect of the temperature on the NIR calibration for alcohol, ph, TA and G + F, the resulting standard error in cross validation (SECV) of the calibration was compared using a Fisher s test (F value) [21]. The F value was calculated as F = SECV 2, where SECV 1 < SECV 2 SECV 1 The calculated F value was compared with the confidence limit F critical (1 α, n 1 1, n 2 2 ), obtained from the distribution F table, where α is the test significance level (α = 0.05 in this experiment), n 1 the sample number measured at the first temperature, n 2 the sample number measured at the second temperature (n 1, n 2,...= 10 in this experiment). The differences between the SECV are significant when F > F limit.

226 D. Cozzolino et al. / Analytica Chimica Acta 588 (2007) 224 230 Four wavelength regions were analysed in order to spectrally evaluate the effect of the temperature on the NIR wine spectrum. These regions were, 950 1000 nm, 1410 1470 nm related with O H overtones (water and ethanol), 1660 1706 nm and 2250 2360 nm related with C H combinations (ethanol, sugars, organic compounds). The spectra in the region of 1800 2000 nm were off scale, and therefore were not used in the analysis. The heights of each absorption band were exported to Microsoft Excel for linear regression analysis of the temperature change and absorbance variation. To quantify the proportions of the total spectral variability explained by temperature and wine variety, scores from the PCA were analysed statistically as follows. After PCA analysis, the scores from the first four principal components (PCs), which accounted for more than 95% of the total spectral variability in the raw spectra, were analysed using ANOVA (Systat, USA). The sum of variances of a specific factor (e.g. temperature and variety) or an interaction term (e.g. temperature variety) of the PCs can be interpreted as the expected variance of future samples taken from the whole population [20,22]. 3. Results and discussion A number of changes were observed in the vis and NIR spectra of wines analysed at different temperatures. It is well known that water shows typical NIR absorption bands around 970 nm (O H stretch second overtone), at 1450 nm (O H stretch first overtone) and at 1900 nm (O H stretching and deformation vibrations) [11]. These bands are affected by sample composition and temperature. Hydrogen bonding lowers the frequency of water stretching vibrations while increasing the frequency of bending vibrations, therefore at higher temperatures the stretching vibrations increase in frequency [11]. Fig. 1 shows the NIR spectra of both red and white wine samples scanned between 1400 nm and 1500 nm at six different temperatures. It was observed that the absorbance at low frequency (longer wavelengths) with respect to the position of band maxima decreased with the temperature increase. Opposite changes were observed at the high frequency (shorter wavelengths), where it was observed that the absorption is increased with the temperature. A close examination of the isobestic point between the two sub-ranges reveals that for white wine samples analysed this absorbance is located at 1446 nm while for red wine samples is located at 1448 nm. In addition, red wine samples have a distinctive absorption band that is affected by temperature changes in the vis region, around 540 nm related to the absorption of wine pigments, principally anthocyanins (data not shown) [23]. Water is the major component of wine, but ethanol also contributes to O H absorption bands [24 26]. Absorption bands of alcohols in the region between 1390 nm and 1640 nm were reported to be related to the first overtones of the stretching mode of the free O H group of the monomer and of the terminal free and hydrogen bonded O H groups of the self-associated species [5,6,25,26]. This is also observed in the spectra of the wine samples analysed. Between 2270 nm and 2300 nm (C H combinations), an increased absorbance in response to temperature has been observed. According to other authors it is believed that the amount of absorbed light in the region between 2200 nm and 2300 nm decreases significantly as sample temperature increases [5,6,11]. This can be of particular importance for the glucose, fructose and ethanol band in this wavelength region [11]. At short wavelengths, both wine types had an absorption band at 978 nm when analysed at 30 C, following temperature increase, shifting to 972 nm at 50 C. Additionally, it was observed that the absorption bands for MilliQ water was shifted from 974 nm to 970 nm from 30 Cto50 C. A close examination of the lowest points of the second derivative spectra showed that the absorption bands of all wine samples and water at six different temperatures were at 966 nm. Plotting the height of the second derivative absorption bands against the six temperatures of each wine sample, a linear relationship was observed (Fig. 2). At wavelengths around 1410 1470 nm a similar shifting trend was observed, but the extent of the observed shifting was even larger than that observed for the short wavelengths. Absorption bands shifted from 1454 nm at 30 C to 1444 nm at 50 C. After second derivative transformation, the 1420 nm absorption band shifted from 1420 nm to 1418 nm as temperature increased (data not presented) [24]. However, the absorption band at 1460 nm did not shift. The second derivative at 1420 nm decreased with temperature (increased absorbance), while at 1460 nm increased. These phenomena might be explained by the increase of free hydroxyl groups with temperature as reported by other authors [2,11]. The broad band, that can be seen as an overlay of many bands that belong to different cluster sizes formed by the hydrogen bonding, is shifted towards lower energies (longer wavelengths) Fig. 1. Near infrared spectrum of red (A) and white (B) wine samples scanned at six different temperatures. Arrows indicate direction of temperature increase.

D. Cozzolino et al. / Analytica Chimica Acta 588 (2007) 224 230 227 Fig. 3. Linear relationship of the second derivative heights at 1460 nm for red wine samples. Table 1 Analysis of variance of the PCA scores of the visible and near infrared raw spectra of red and white wines analysed together PC1 PC2 PC3 PC4 Temperature 25 C 0.03 a 0.50 a 0.22 a 0.19 a 30 C 0.05 a 0.58 b 0.05 b 0.17 b 35 C 0.04 a 0.26 c 0.008 bc 0.10 c 40 C 0.06 a 0.16 d 0.03 b 0.22 a 45 C 0.08 a 0.39 e 0.13 cd 0.16 b 50 C 0.06 a 0.75 f 0.17 d 0.02 d Fig. 2. Linear relationship of the second derivative heights at 962 nm for red and white wine samples. relative to the free O H stretch. Therefore, increasing the temperature of the sample, decreases the average cluster size and increases the relative absorbance of free groups [2,5,6,11]. A similar phenomenon was observed in this study. The existence and temperature dependent behaviour of the bands due to rotational isomerism of the monomer, terminal free O H groups, O H groups weakly hydrogen bonded and hydrogen bonded O H of the self-associated species might explain the observable changes in the NIR spectra of the wines analysed. In this study, the NIR spectra of the wine samples analysed did not show any shifting as a result of temperature changes in both the raw spectra and after the second derivative transformation in the region between 1660 nm and 1710 nm [25] (Fig. 3). The results from the one-way analysis of variance, in order to test the effect of temperature and variety on the scores of the vis and NIR spectra, are shown in Tables 1 and 2. When comparing the overall effect of temperature on the spectra of both wine varieties, no statistically significant differences were observed in PC1, while statistically significant differences were observed for the other three PCs. When the overall effect of the variety was analysed, statistically significant differences were observed in PC1 and PC3 but none between the other PCs analysed. When the effect of temperature was analysed in relation to an individual variety, it was observed that no statistically significant Type Red 1.82 a 0.02 a 0.004 a 0.01 a White 1.85 b 0.04 a 0.19 b 0.002 a Effect of temperature and type of wine. PC: principal component. Levels in the column not connected by same letter (a f) are significantly different (p < 0.05). Table 2 Analysis of variance of the PCA scores of the visible and near infrared raw spectra for red and white wine samples with respect to temperature analysed separately PC1 PC2 PC3 PC4 Red wine 25 C 1.75 a 0.49 a 0.22 a 0.25 a 30 C 2.01 a 0.57 b 0.09 b 0.15 b 35 C 1.75 a 0.32 c 0.10 b 0.22 b 40 C 1.70 a 0.15 d 0.04 c 0.23 a 45 C 1.90 a 0.43 e 0.07 c 0.09 c 50 C 1.84 a 0.74 f 0.15 d 0.06 d White wine 25 C 1.69 a 0.56 a 0.02 a 0.15 a 30 C 1.63 ab 0.64 b 0.19 b 0.17 c 35 C 1.65 c 0.26 c 0.07 b 0.02 d 40 C 1.64 bc 0.12 d 0.15 bc 0.25 b 45 C 1.53 d 0.32 e 0.37 d 0.23 c 50 C 1.54 d 0.72f 0.39d 0.006 e PC: principal component. Levels in the column not connected by same letter (a f) are significantly different (p < 0.05).

228 D. Cozzolino et al. / Analytica Chimica Acta 588 (2007) 224 230 differences were observed for PC1 in red wine samples, while statistically significant differences were observed in PC1 for the white wine samples analysed. It is interesting to highlight that both wine types had different behaviour on the PCs in response to temperature changes. The highest eigenvectors observed in PC1 were related to pigments (visible region) for the red wine samples and to O H vibration bands for white wine samples. Generally, PC score distributions indicate the degree of similarities between the sample spectra. If the scores for some components are spread irregularly as a function of a temperature, it is very likely that the PC is capturing only noise or other non-deterministic variations and does not describe physically significant changes. It is interesting to note that the scores were arranged in a linear way against temperature. When the scores of the samples scanned at different temperatures were plotted it was observed that samples scanned at the same temperature were clustered together (Fig. 4). Additionally, it was observed that for the case of white wines, samples clustered from left to right as temperature increased along PC1, while for red wines the samples were clustered from bottom to top along PC2, which seems to explain the effect of temperature. It was observed that PC1 captures all the dominant changes in the spectra related to temperature in the set of white wine samples while PC2 did the same for the red wine samples. Comparing Fig. 5. Eigenvectors for the first two principal components for the effect of temperature in red wine samples. the eigenvectors from the PCA, it can be observed that PC2 of the red wine is similar to PC1 of the white samples. On the other hand, the PC1 in red wines is mainly related to the vis region, around 540 nm corresponding to wine pigments (anthocyanins) [23]. Both PC2 in the case of red wines and PC1 in the case of white wine samples explain the spectra temperature related changes, and are specifically related to the observed changes in the NIR region around the O H bonds. However, the most important finding of this study was that for red wines, PC1 was also related to temperature changes in electronic transitions of wine pigments, or to shifts in anthocyanin absorption related with co-pigmentation [23,27]. Looking at the eigenvectors for both varieties (Figs. 5 and 6), one can obtain the temperature profile of the absorption changes developed in the system. In order to quantify the effect of temperature on the spectra of wine and on the accuracy of NIR calibrations models for alcohol, sugars (glucose plus fructose), ph and titratable acidity, the SECV values for different NIR calibrations models for spectra collected at different temperature and the chemical composition, were statistically compared. Fig. 4. Score plot of the first two principal components for red wines showing effect of temperature variation. Fig. 6. Eigenvectors for the first two principal components for the effect of temperature in white wine samples.

Table 3 Sample characteristics and chemical composition of white wine samples analysed D. Cozzolino et al. / Analytica Chimica Acta 588 (2007) 224 230 229 Sample code Variety Alcohol (%) ph TA (g L 1 ) G+F(gL 1 ) Ww1 CH 13.75 3.36 6.55 1.80 Ww2 CH 13.37 3.41 6.73 4.10 Ww3 CH 13.43 3.26 6.54 0.70 Ww4 PG 13.76 3.13 7.48 1.60 Ww5 R 13.21 3.18 6.52 8.0 Ww6 SB 12.99 3.27 6.26 2.10 Ww7 S 11.70 3.27 6.3 0.70 Ww8 S 11.09 3.18 7.05 2.60 Ww9 UCH 13.55 3.36 6.32 9.30 Ww10 Verdelho 12.90 3.41 6.91 4.10 Mean 12.98 3.28 6.67 3.50 S.D. 0.89 0.10 0.39 2.97 Min 11.09 3.13 6.26 0.70 Max 13.76 3.41 7.48 9.30 S.D.: standard deviation; Min: minimum; Max: maximum; TA: titratable acidity; G + F: glucose plus fructose; CH: Chardonnay; PG: Pinot Gris; R: Riesling; UCH: unwooded Chardonnay; S: Semillon; SB: Sauvignon Blanc. Table 4 Sample characteristics and chemical composition of red wine samples analysed Sample code Variety Alcohol (%) ph TA (g L 1 ) G+F(gL 1 ) Rw1 CS 13.61 3.53 6.63 0.30 Rw2 CS 13.08 3.57 6.04 1.80 Rw3 CS 12.49 3.43 7.35 0.40 Rw4 CS 13.21 3.49 7.78 3.90 Rw5 CS 12.92 3.36 7.22 0.20 Rw6 PN 13.51 3.62 7.31 0.50 Rw7 Rośe 13.08 3.36 6.06 4.60 Rw8 SH 13.65 3.54 6.44 0.20 Rw9 SH 14.08 3.63 6.73 0.60 Rw10 Blend of CS and SH 14.15 3.43 6.58 0.30 Mean 13.29 3.50 6.84 1.34 S.D. 0.52 0.10 0.58 1.64 Min 12.49 3.43 6.04 0.2 Max 14.15 3.63 7.78 4.6 S.D.: standard deviation; Min: minimum; Max: maximum; TA: titratable acidity; G + F: glucose plus fructose; CS: Cabernet Sauvignon; SH: Shiraz; PN: Pinot Noir. Tables 3 and 4 show the chemical composition of both red and white wine samples analysed. A wide range in composition was observed in the set of wines analysed. It was therefore considered to be a representative set of samples on which NIR calibration models could be developed in order to test the effect of temperature on both the spectra and calibration robustness. Note that the aim of this work was not to develop NIR calibrations for wine compositional parameters rather to test the effect of the temperature on such calibration models. Table 5 shows the standard error of cross validation (SECV) for the chemical parameters evaluated for calibrations developed on samples scanned at six different temperatures. Firstly, it was observed that the SECV obtained was different depending on the type of wine used (e.g. red or white). No statistically significant differences for all parameters were observed for the SECV obtained with calibrations developed on red wine samples scanned at 30 C and 35 C. While same statistically significant differences were observed for the SECV of red wine samples scanned at ambient ( 25 C), 40 C, 45 C and 50 C, respectively, no statistically significant differences were observed on Table 5 Standard error in cross validation (SECV) obtained using partial least squares as calibration models for the determination of chemical composition in red and white wines scanned at six different temperatures Alcohol (%) ph TA (g L 1 ) G+F(gL 1 ) Red wine 25 C 0.084 a 0.038 a 0.18 a 0.54 a 30 C 0.059 a 0.013 b 0.12 b 0.18 b 35 C 0.062 a 0.017 b 0.071 b 0.27 b 40 C 0.14 b 0.029 a 0.18 a 0.51 a 45 C 0.30 c 0.059 c 0.11 b 0.43 a 50 C 0.097 a 0.027 a 0.17 a 0.59 a White wine 25 C 0.077 a 0.056 a 0.19 a 0.64 a 30 C 0.070 a 0.058 a 0.17 a 0.66 a 35 C 0.074 a 0.059 a 0.22 a 0.80 a 40 C 0.12 b 0.065 a 0.23 a 1.04 b 45 C 0.069 a 0.040 a 0.17 a 0.58 a 50 C 0.23 b 0.08 b 0.24 a 2.58 b Levels in the column not connected by same letter are significantly different (p < 0.05); TA: titratable acidity; G + F: glucose plus fructose.

230 D. Cozzolino et al. / Analytica Chimica Acta 588 (2007) 224 230 the SECV of white wine samples scanned between laboratory ambient temperature ( 25 C) up to 40 C. However, at 50 C, the SECV values obtained for the calibrations for alcohol and G + F were statistically significantly different. It was observed that the NIR calibrations developed using red wine samples, were more influenced by changes in temperature than those developed using white wine samples, but for both wine types a systematic trend of the error, increasing as the temperature increased, was observed for the four parameters measured. Similar results were found by other authors where the effect of temperature was studied on alcohol calibration in beverages [4]. It was suggested that some compositional characteristics of the red wine matrix could be more affected than others (e.g. pigments, phenolic compounds) when perturbations in the vis and NIR spectra are induced by temperature. 4. Conclusions This study demonstrated the importance of the effect of temperature on both vis and NIR spectra of wine and the calibrations developed for alcohol, sugars, ph and TA in red and white wine samples. It was observed that the SECV obtained was dependent on the variety used to develop the NIR calibrations and tended to be higher as the temperature of the sample increased. The NIR spectra were affected by changes in the temperature and these changes were also different depending on the variety of wine. The main changes observed in the NIR spectra of the wine samples relating to temperature were observed around 970 nm and 1400 nm (O H bonds). It was also observed that no changes occurred in the SECV when wine samples were scanned between 30 C and 35 C. The practical implication of this study is that the recommended temperature for scanning of wines using a 1 mm path length cuvette is between 30 C and 35 C. Acknowledgements We wish to thank Dr. M. Herderich for the suggestions and comments made in the manuscript. This project is supported by Australia s grapegrowers and winemakers through their investment body the Grape and Wine Research and Development Corporation, with matching funds from the Australian government, and by the Commonwealth Cooperative Research Centres Program. The work was conducted by The Australian Wine Research Institute, and formed part of the research portfolio of the Cooperative Research Centre for Viticulture. References [1] H. Abe, C. Iyo, S.J. Kawano, Near Infrared Spectrosc. 8 (2000) 209 214. [2] F. Wulfert, W.T. Kok, A.K. Smilde, Anal. Chem. 7 (1998) 1761 1767. [3] K. DebRaekeleer, F. Cuesta Sanchez, P.A. Hailey, D.C.A. Sharp, A.J. Pettman, D.L. Massart, J. Pharm. Biomed. Anal. 17 (1998) 141. [4] F.D. Barboza, R.J. Poppi, Anal. Bioanal. Chem. 377 (2003) 695 701. [5] H. Maeda, Y. Ozaki, J. Near Infrared Spectrosc. 3 (1995) 191 201. [6] H. Maeda, Y. Wang, Y. Ozaki, M. Suzuki, M.A. Czarnecki, M. Iwahashi, Chem. Intell. Lab. Syst. 45 (1999) 121 130. [7] F. Chauchard, J.M. Roger, V. Bello-Meurel, J. Near Infrared Spectrosc. 12 (2004) 199 205. [8] W.G. Hansen, S.C.C. Wiedemann, M. Snieder, V.A.L. Wortel, J. Near Infrared Spectrosc. 8 (2000) 125. [9] H. Swierenga, F. Wülfert, O.E. De Noord, A.P. de Weijer, A.K. Smilde, L.M.C. Buydens, Anal. Chim. Acta 411 (2000) 121. [10] M. Blanco, D. Valdes, J. Near Infrared Spectrosc. 3 (2004) 121 126. [11] M.A. Czarnecki, Y. Ozaki, Phys. Chem. Chem. Phys. 1 (1999) 797 800. [12] Y. Ozaki, S. Sasic, J.H. Jiang, J. Near Infrared Spectrosc. 9 (2001) 63 95. [13] B. Czarnik-Matusewicz, S. Pilorz, J.P. Hawranek, Anal. Chim. Acta 544 (2005) 15 25. [14] B. Czarnik-Matusewicz, S. Pilorz, Vib. Spec. 40 (2006) 235 245. [15] J.A. Hageman, J.A. Westerhuis, A.K. Smilde, J. Near Infrared Spectrosc. 13 (2005) 53 62. [16] D. Cozzolino, H.E. Smyth, M. Gishen, J. Agric. Food Chem. 51 (2003) 7703 7708. [17] D. Cozzolino, M. Kwiatkowski, M. Parker, M. Gishen, R.G. Dambergs, W. Cynkar, M. Herderich, Anal. Chim. Acta 513 (2003) 73 80. [18] M. Urbano Cuadrado, M.D. Luque de Castro, P.M. Pérez Juan, M.A. Gómez-Nieto, Talanta 66 (2005) 218 224. [19] M. Urbano-Cuadrado, M.D. Luque de Castro, P.M. Pérez-Juan, J. García- Olmo, M.A. Gómez-Nieto, Anal. Chim. Acta 527 (2004) 81 88. [20] T. Naes, T. Isaksson, T. Fearn, T. Davies, A User-friendly Guide to Multivariate Calibration and Classification, NIR Publications, Chichester, UK, 2002, p. 420. [21] M. Samuels, Statistics for the Life Sciences, Macmillan, Inc., 1989. [22] R.G. Brereton, Chemometrics, in: Data Analysis for the Laboratory and Chemical Plant, Wiley and Sons, 2003. [23] C. Somers, The Wine Spectrum: An Approach towards Objective Definition of Wine Quality, Winetitles, Adelaide, 1998, p. 50. [24] W.R. Hruschka, Spectral reconstruction, in: D.A. Burns, E.W. Ciurczak (Eds.), Handbook of Near-infrared Analysis, Marcel Dekker, New York, 1992, pp. 365 382. [25] B.G. Osborne, T. Fearn, P.H. Hindle, Practical NIR Spectroscopy with Applications in Food and Beverage Analysis, 2nd ed., Longman Scientific and Technical, Harlow, Essex, UK, 1993, p. 227. [26] R.G. Dambergs, A. Kambouris, I.L. Francis, M. Gishen, J. Agric. Food Chem. 50 (2002) 3079 3084. [27] R. Boulton, Am. J. Enol. Vitic. 52 (2001) 67 87.