Applications

Wine Analysis

Wine has been produced for thousands of years. The earliest evidence of wine is in Georgia from 6000 BC, Iran from 5000 BC, and Sicily from 4000 BC.

Introduction

Wine is one of the most ancient beverages produced and is made by transforming sugars into alcohol during fermentation of the grape must. It is composed mainly of water, ethanol, sugars, and acids. However, there are other compounds that even in very small concentrations can greatly influence the sensory properties of the final product. There is a strong need for simple, rapid, and cost-effective techniques for objectively measuring properties of wine. Measurement of grape characteristics such as maturity assessment is important for vineyard improvement and optimizing different styles of wine. At the present time, most analysis is limited to total soluble solids (TSS expressed as °Brix), acidity, visual assessment, and tasting after vinification. More complex analyses of grapes and wine are currently unattainable on a large scale because of the time and cost involved. Tasting is subjective and while it does give wine producers a good idea about the final product when done by experts, taste alone is not enough to fully assess quality. Real-time feedback during the fermentation process of the substrates can improve quality and reduce costs. Off-line classification of yeast strains, vine tissue analysis, quality profiles, and blend analysis would also be a useful tool for analyzing wine quality. Fortified wine is made by grape spirit distillation and it is important to minimize the presence of methanol during distillation, which can be made if there is bacterial contamination.

As is the case with many valuable food and beverage products, adulteration is an issue and it is difficult to assess if a high-quality wine has been adulterated with a cheaper brand by visualization and taste. The demand for wine and the need for fast, cost-effective, real-time monitoring of the parameters has created the need for methods that can replace expensive, laborious, and time-consuming wet chemistry methods. One such method that has been studied is NIR spectroscopy.

Analytes

  • Yeast Strain Classification
  • Grape Color (Total Anthocyanin)
  • Total Soluble Solids (°Brix)
  • pH
  • Grape Fungal Disease Detection and Quantification
  • Malvidin-3-Glucoside
  • Pigmented Polymers
  • Tannins
  • Quality Assessment
  • Methanol
  • Ethanol
  • Variety Identification and Product Authenticity
  • Nitrogen
  • Potassium
  • Phosphorus
  • Condensed Tannins (CT)
  • Dry Matter (DM)
  • Volatile Compounds
  • Glucose
  • Fructose
  • Glycerol
  • Total Phenolics
  • Total Anthocyanins
  • Total Flavonoids
  • Total Acidity
  • Total Sugars
  • Calcium

Summary of Published Papers, Articles, and Reference Materials

One study conducted in Australia documented differentiating yeast strains of Saccharomyces cerevisiae. Consumer demand is pushing the need to produce different and novel wine styles with particular characteristics. Changes in the yeast genome can influence potential flavor metabolites. Classification results showed the potential for using NIR spectroscopy as a screening tool for discriminating between yeast strains as well as grouping strains with deletions in genes that disturb different metabolic pathways. Two comprehensive review papers of applications investigated by the Australian Wine Research Institute (AWRI) discussed a number of studies documenting grape, wine, yeast, and leaf analysis. The primary focus was on total anthocyanins (expressed as color), which coupled with the traditional analyses of total soluble solids (expressed as °Brix) and pH could provide a new method for grape quality assessment and vineyard management. Other applications discussed in these two papers are fungal diseases in grapes, phenolic compounds in red wine fermentation, wine quality grading, grape spirit distillation, variety identification and product authenticity, and vine tissue analysis. Another paper from Australia documented measuring condensed tannins (CT) and dry matter (DM) in red grape homogenates. Tannins are important phenolic compounds for sensory properties but the standardized tests are expensive and unable to be performed on a large scale for vineyards. Dry matter measurement helps optimize harvest data and can potentially maximize juice extraction. Volatile organic compounds such as esters, higher alcohols, and fatty acids are created during fermentation from complex microbial and biochemical reactions. These compounds contribute to a wide range of physiochemical properties in wine and can only be measured at the present time by complex and expensive methods such as GC-MS. Two different studies examined the feasibility of measuring these compounds using NIR spectroscopy; one using apple wines in China and one using white wine of the Vinhos Verdes appellation in Portugal. Both studies showed good correlation and demonstrated the potential for replacing traditional methods for measuring volatile compounds in wine using NIR spectroscopy. One study compared using NIR and MIR spectroscopy for various carbohydrate concentrations, fermentation products, and phenolic compounds during red wine fermentation. Both methods showed superb results and proved the feasibility of monitoring fermentation using spectroscopic methods. Volumic Mass/Density is the most important compound to measure in white wine fermentation and this was examined using a miniature Vis/NIR spectrometer with good correlation. Calcium must be kept below a certain level in sparkling wines and NIR spectroscopy was examined as a method for this measurement in both white grape must and wine. Feasibility was demonstrated and the potential exists to use NIR spectroscopy for this measurement instead of current expensive and time-consuming reference methods.

Scientific References and Statistics

Combining Near-Infrared Spectroscopy and Multivariate Analysis as a Tool to Differentiate Different Strains of Saccharomyces cerevisiae: a metabolomic study – Cozzolino, Flood, Bellon, et al., Wiley Science, Yeast 2006: 23: 1089-1096

The purpose of this study was to examine the metabolic profiles produced by Saccharomyces cerevisiae deletion strains sourced from the Euroscarf yeast collection using NIR spectroscopy. Eight separate strains were used, each with a different gene deletion. Samples were scanned in transmission mode after centrifuging in a 1 mm cuvette. Wavelength range was from 400 nm to 2500 nm. Replicate experiments were carried out multiple times over three months. Spectra were exported for post-processing and chemometric analysis. Principle Component Analysis (PCA) was performed to visualize any significant grouping amongst the samples. ANOVA analysis showed that there were some spectral differences within the same group in the separate sample sets scanned over the three month period. The experiment did use some strains that possess mutations in the same or closely related pathways and replicate samples were derived from different starter cultures. This was done in an attempt to ensure the classification results reflected the mutation and not batch-to-batch variation. Linear Discriminant Analysis showed acceptable classification results and two strains were classified with 100% accuracy. The potential of chemometrics and NIR spectroscopy to discriminate between yeast strains and grouping strains with deletions in genes that disturb similar metabolic pathways was demonstrated. These methods may be useful in defining the functions of genes that have no obvious genotype. Using NIR spectroscopy as a high-throughput tool for yeast selection could accelerate progress in genome-based wine yeast research and allow the selection of strains for more detailed biochemical analysis.
https://onlinelibrary.wiley.com/doi/full/10.1002/yea.1418

Grape and Wine Analysis – Enhancing the Power of Spectroscopy with Chemometrics. A Review of Some Applications in the Australian Wine Industry – Gishen, Dambergs, Cozzolino, Australian Journal of Grape and Wine Research, 11, 296-305, 2005

Analysis of Grapes and Wine by Near-Infrared Spectroscopy – Cozzolino, Dambergs, Janik, et al., Journal of Near Infrared Spectroscopy, 14, 279-289, 2006

These papers summarized investigations of applications by the Australian Wine Research Institute (AWRI) for measuring parameters in grapes, wine, yeast, and vine tissues using NIR spectroscopy. The primary focus has been on rapid analysis of red grapes for color (expressed as total anthocyanins), total soluble solids (expressed as °Brix), and pH. Grape color is a strong indicator of red wine quality and rapid NIR analysis shows great promise to replace traditional methods for color and quality assessment. Other research and applications discussed are fungal diseases in grapes, phenolic compounds during fermentation, quality grading, monitoring grape spirit distillation, variety identification and product authenticity, and vine tissue analysis.

Color

Approximately two thousand three hundred homogenized red grape samples incorporating three vintages, ten regions, and ten grape varieties were scanned using a research-grade NIR spectrometer. PLS calibration models were created using the spectral data and reference values for color. A PLS model incorporating all samples showed good results but the analysis showed non-linearity in the model. Non-linearity can be corrected using local algorithms that match similar spectra in the calibrations. This frequently occurs when using large data sets. Individual models for different groups can work as well but this approach may create data sets with too few data points for a good calibration.

Partial Least Squares (PLS) Calibration Model for Color R2= 0.90 RMSEP= 0.14 mg/g
Local Weighted Regression Algorithm for Color R2= 0.96 RMSEP= 0.09 mg/g

Calibration results were good and improved using the local weighted regression algorithm in the model. Measuring color along with already proven measurable analytes using NIR spectroscopy such as pH, °Brix, reducing sugars and lactic acid could prove to be a valuable tool for vineyard management and optimization of harvest time.

Fungal Diseases in Grapes

Mold contamination in harvested grapes can be difficult to assess visually. NIR spectroscopy was explored to detect powdery mildew in Chardonnay grapes. Samples were first visually classified for mildew contamination, homogenized, and scanned in reflectance mode using a NIR spectrometer from 400 nm to 2500 nm. The homogenates were analyzed for powdery mildew DNA content and the analysis matched well with the visual classification. Strong spectral differences were observed correlating to contamination level and it was confirmed that these differences were not related to pH and °Brix, eliminating the possibility that any classification analysis could be based on these parameters and not mildew contamination. Classification discriminant analysis was able to accurately classify 92% of the samples based on infection level. A PLS regression model could sufficiently discriminate between no infection and the lowest infection level samples (1% to 10%). While results were good, it must be noted that this was a small data set and more work is necessary to confirm the feasibility of detecting mildew in grapes in a real-time setting.

Malvidin-3-Glucoside R2= 0.91 RMSEP= 28.0 mg/L
Pigmented Polymers R2= 0.87 RMSEP= 5.9 mg/L
Tannins R2=0.83 RMSEP = 131.1 mg/L

Modeling results showed good correlation between the NIR spectra and the major anthocyanins. Visual analysis of the spectral data showed similar changes occur in both Cabernet Sauvignon and Shiraz during the fermentation itself and maturation after fermentation. Many simultaneous changes occur during fermentation and more work is necessary to define the specificity of the calibrations, but the results here do offer a possibility for real-time monitoring of phenolic compounds during red wine fermentation.

Wine Quality Grading

Wine quality in terms of sensory characteristics is often a subjective measure that can be biased by individual preferences and day-to-day variation. An objective quality measurement system would be useful and NIR spectroscopy was examined for this purpose. While many flavor compounds are below the detection level for NIR analysis, some of the more abundant organic compounds do affect quality and provide a basis for examining the feasibility of this measurement. Samples of Cabernet Sauvignon were scanned from 400 nm to 2500 nm and the reference quality scores were segmented from 1 to 5 with 1 being the lowest quality score.

Quality Score R2= 0.76 RMSEP= 0.6

Acceptable results were achieved, especially when considering that the theoretical minimum error was 0.5 as all reference values were whole numbers and the calibration can predict fractions that have to be rounded off. It must be noted that calibration models were created for both the full wavelength range and from 400 nm to 700nm which showed similar results. The smaller and lower range is an absorbing area of the spectrum for anthocyanins and polymerized pigments, two notable parameters in wine quality. This supports the validity of the calibration model and shows the correlation is corresponding to parameters affecting wine quality.

Grape Spirit Distillation

Grape spirit is produced by distillation of wine or wine and grape process waste and is used to make fortified wine. Methanol concentration can be high due to the presence of mold or bacteria in the raw product. In order to operate continuous stills, rapid feedback of methanol concentration is necessary for fine-tuning. Samples of wine were scanned in transmission mode and GC was used as the reference method to determine methanol and ethanol concentration.

Methanol R2=0.99 RMSEP= 0.06 g/L
Ethanol R2=0.96 RMSEP=0.08 % v/v

Correlation coefficients for both methanol and ethanol were high and comparable to the error in the reference method, proving the validity of the calibration models. The results here show the possibility of real-time grape spirit distillation monitoring using NIR spectroscopy.

Variety Identification and Product Authenticity

Few studies have been conducted for identifying wine adulteration using NIR spectroscopy. However, there have been studies for classifying grape and wine varieties. One study was able to classify Merlot, Tempranillo, and Grenache red grape varieties grown in Spain with 100% discrimination. Similar results were achieved with two white grape varieties – Viura and Chardonnay. Australian Riesling and Chardonnay white wine varieties were classified with 95% accuracy. The results here show potential for adulterant identification as well as blend analysis for grapes and wine.

Vine Tissue Analysis

Work has been conducted to analyze nutrients in grape petioles using NIR spectroscopy. Petioles are the stalks that attach a leaf blade to the stem. Samples were dried and ground before scanning. Reference tests were conducted for nitrogen, potassium, and phosphorus.

Nitrogen R2=0.997
Potassium R2=0.99
Phosphorus R2=0.996

The results here show promise for using NIR spectroscopy as a tool for vineyard management and soil nutrient analysis. However, the sample set was limited and there were not enough samples to carry out a validation analysis. More work will be necessary to fully validate the results shown here:
https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1755-0238.2005.tb00029.x
https://journals.sagepub.com/doi/abs/10.1255/jnirs.679?journalCode=jnsa

Measurement of Condensed Tannins and Dry Matter in Red Grape Homogenates Using Near-Infrared Spectroscopy – Cozzolino, Cynkar, Dambergs, et al., Journal of Agricultural and Food Chemistry, 2008, 56, 7631-7636

Condensed Tannins (CT) and Dry Matter (DM) are two very important parameters in wine grapes.  Tannins are one type of phenolic compound that contributes to sensory properties like color, flavor, and bitterness.  There is a standardized test for tannins known as the methylcellulose precipitable (MCP) tannin assay, but it is time-consuming and expensive to implement.  DM has a direct effect on the stability and quality of all foods. A direct measurement of DM would help optimize harvest date and maximize the potential extraction of juice.  Six hundred twenty samples of red grape cultivars were procured for this study covering four grape types, four harvests, and eight different growing regions.  Samples were collected as whole berries and stored frozen for up to six months before analysis.  Thawed samples were homogenized and scanned from 400 nm to 2500 nm in reflectance mode at 2 nm intervals.  Thirty-two scans were collected for each sample and averaged into one spectrum. Reference tests for CT and DM were performed after the samples were homogenized.

Condensed Tannins (CT)R2=0.86RMSEP= 0.46 mg/g Epicatechin Equivalents
Dry Matter (DM) R2= 0.92RMSEP= 0.83% w/w

Calibration models for CT and DM showed good correlation and the validity of the models was verified by validation data sets. Analysis of the wavelengths used to build the calibration models showed the models built correlation from absorbing areas of the parameters of interest. This study proved that CT and DM can be measured using NIR spectroscopy, offering a suitable and efficient tool for measuring these parameters in homogenized grape samples in addition to parameters that have been proven in other studies, such as total anthocyanins (color), °Brix, and pH.
https://www.ncbi.nlm.nih.gov/pubmed/18707119

Rapid Detection of Volatile Compounds in Apple Wines Using FT-NIR Spectroscopy – Ye, Gao, Li, et al., Food Chemistry 190 (2016) 701-708
Volatile organic compounds found in apple wines are created during fermentation from complex microbial and biochemical reactions. Esters, higher alcohols, and fatty acids are of particular interest and contribute to a wide range of distinct physiochemical properties, such as volatilities, polarity, boiling points, and sensory threshold. Analysis of these compounds is typically performed using GC-MS which is expensive, time-consuming, and requires highly skilled technicians. Seventy-two apple wine samples were procured to test the feasibility of measuring volatile compounds using FT-NIR spectroscopy. All samples were made in the laboratory using micro-fermentation trials. GC-MS was used as the reference method for the samples. Samples were scanned in transmission mode from 12000 cm-1 to 4000 cm-1 using a quartz cuvette with 1 mm pathlength. Fifty-two samples were used to build calibration models and the remaining twenty were used for a validation set.

Esters:
7 Total
Highest:Ethyl AcetateR2=0.9450RMSEP = 37.50 mg/L
Lowest:Ethyl Caprylate R2=0.8967RMSEP = 0.120 mg/L
Higher Alcohols
5 Total
Highest:HexanolR2=0.9497RMSEP = 0.153 mg/L
Lowest:3,4,5-Trimethyl-4-Heptanol R2=0.8844RMSEP = 0.181 mg/L
Fatty Acids
3 Total
Highest:Hexanoic Acid R2=0.9179RMSEP = 0.746 mg/L
Lowest:Decanoic Acid R2=0.9007RMSEP = 0.409 mg/L

A large number of volatile compounds were detected from the GC-MS analysis and those chosen for the calibration models were present in most of the samples. Various data pre-treatments and selective wavelength ranges were checked to optimize the results. Wavelengths corresponding to the first overtone and the stretch vibration for C-H, O-H, and C=O were used for many of the calibrations. This study indicates that NIR spectroscopy can be used to determine volatile compounds in apple wine and results from the NIR technique were comparable to the GC-MS reference method. However, it must be noted that the concentration of the measured volatile compounds is very small. While promising, further work will be needed to prove that the calibration models are correlating to the parameters of interest and not some indirect correlation that appears to measure the volatile compounds concentration. Further investigation would also be necessary to determine the feasibility of measuring volatile compounds that were not detected in this study.
https://www.ncbi.nlm.nih.gov/pubmed/26213028

New PLS Analysis Approach to Wine Volatile Compounds Characterization by Near Infrared Spectroscopy – Genisheva, Quintelas, Mesquita, et al., Food Chemistry 246 (2018) 172-178
The aim of this study was to examine the potential of NIR spectroscopy for measuring ten of the most relevant volatile compounds in Portuguese Vinhos Verdes white wine. The normal method for measuring these compounds is using gas chromatography coupled to at least one detector, such as a flame ionization detector or mass spectrometer. These methods are time-consuming, expensive, and require skilled technicians to operate. Seven white wine grape varieties comprising one hundred five samples were used for the study, all designated “Appelation of Origin Vinhos Verdes” that are produced in Northern Portugal. Vinification was performed on the grapes according to the traditional technology of the wine designation. Reference tests were performed using GC and different detector methods. Wine samples were scanned in transmittance mode from 14000 cm-1 to 600 cm-1 using a flow cell with 0.7 mm pathlength at 8 cm-1 resolution. Thirty-two scans were collected per sample and averaged into one spectrum.

Volatile Compounds Analyzed:

Ethyl Acetate
Methanol
2-Methyl-1-Butanol
3-Methyl-1-Butanol
2-Phenylethanol
3-Methylbutyl Acetate
Ethyl Lactate
Ethyl Octanoate
Diethyl Succinate
Diethyl Malate
All R2 Correlation Coefficients ranged from 0.94 – 0.97

After all samples were scanned and the reference methods performed, boxplot and Principle Component Analysis (PCA) were performed for cluster identification and outlier analysis. A selective wavelength range from 6357 cm-1 to 5435 cm-1 was used for all Partial Least Squares (PLS) calibration models. An iterative approach was used for PLS models using the cluster identification to reduce the dataset and then correlate the spectral data in the wavelength range of interest to construct the calibrations. The predictive capability of the models was shown by an independent validation set. As was the case with the apple wine study, further work will be needed to prove that the calibration models are correlating to the parameters of interest and not some indirect correlation that appears to measure the volatile compounds concentration. It can be stated that the results are promising enough to warrant further work to fully demonstrate the potential to replace traditional expensive and time-consuming methods for these measurements.
https://www.sciencedirect.com/science/article/pii/S0308814617318162

NIR and MIR Spectroscopy as Rapid Methods to Monitor Red Wine Fermentation – Di Egidio, Sinelli, Giovanelli, et al., Eur Food Res Technology (2010) 230: 947-955

Fifteen micro-fermentation trials were conducted during a single vintage harvest in the Valtellina region of Northern Italy for the purpose of analyzing parameters in wine fermentation using both NIR and MIR spectroscopy.  Sampling was conducted at five subsequent times during each fermentation trial from initial crushing of the grapes until approximately thirty days after fermentation began for a total of seventy-five samples.  NIR transmission spectra were collected from 12500 cm-1 to 3600 cm-1 for each sample with a 1 mm pathlength flow cell using 8 cm-1 spectral resolution.  Sixteen scans were collected and averaged for each spectrum.  MIR spectra were collected for each sample on an ATR crystal background from 4000 cm-1 to 700 cm-1 using 16 cm-1 spectral resolution and thirty-two scans per average.  Standard chemical methods were used to obtain reference values for sugars (glucose and fructose), alcohols (ethanol and glycerol), and phenolic compounds (total phenolics, total anthocyanins, and total flavonoids).  Both sets of spectral data were transformed using different pretreatments. Classification analysis was performed using algorithms to determine the feasibility of separating samples based on fermentation stage, divided into four steps for the purpose of the analysis.  Partial Least Squares (PLS) regression models were created correlating the reference values of the parameters of interest to the spectral data. 

Linear Discriminant Analysis (LDA) for NIR Data:
Correct Classification of Fermentation Stages 1-4: 91.1%
PLS Calibration Models for NIR Data:
Glucose: R2=0.99 RMSEP=1.11 g/L
Fructose: R2=0.99 RMSEP=4.68 g/L
Ethanol: R2= 0.99 RMSEP=1.96 g/L
Glycerol: R2=0.99 RMSEP=0.41 g/L
Total Phenolics: R2=0.99 RMSEP=217 mg/L
Total Anthocyanins R2=0.97 RMSEP=17.7 mg/L
Total Flavonoids R2=0.97 RMSEP=213 mg/L

Results were excellent for both classification and regression analysis. MIR data showed very similar results and those results are not shown here, but the feasibility of measuring these parameters from the spectral data for both sets was proved. In the case of LDA analysis, the NIR spectra showed a 100% correct classification for the beginning Stage 1 and ending Stage 4 of fermentation, indicating that the spectra can be used to determine when fermentation is complete. The PLS calibration models created here could be used to monitor fermentation in an on-line, real-time setting for carbohydrate concentrations, fermentation products, and phenolic compounds. It must be noted that the concentration for total anthocyanins is very small and it is likely that the calibration model is not measuring such a low concentration directly. However, anthocyanins are directly related to color and it has been proven in numerous studies that absorption at a visible wavelength can be correlated to NIR spectra. An indirect correlation is acceptable for NIR spectra calibration modeling but the analysis needs to be carefully examined and validated. Such analysis is time-consuming and expensive using traditional methods, especially in an on-line setting. NIR spectroscopy can be used as a quality tool to optimize fermentation in red wine and assure product quality at all stages of the process.
https://link.springer.com/article/10.1007%2Fs00217-010-1227-5

Feasibility of Using a Miniature NIR Spectrometer to Measure Volumic Mass During Alcoholic Fermentation – Fernandez-Novales, Lopez, Gonzalez-Caballero, International Journal of Food Sciences and Nutrition, June 2011 62(4): 353-359

There is one major difference in grape preparation for white and red wine fermentation. For white wine, the “must” obtained by crushing and pressing grapes is sent for fermentation, which includes the skins, seeds, and stems of the fruit.  In contrast, red wine grapes are first destalked and fermentation takes place with maceration of skins and seeds.  A consequence of this is that the most important component to monitor during white wine fermentation is must volumic mass (density).  A miniature NIR spectrometer was procured to measure must samples of wine grapes for volumic mass. One hundred twenty-four samples were used comprising six different varieties of white wine grapes and six different varieties of red wine grapes collected during fermentation trials over three consecutive harvests.  The white grapes were combined and placed into fermentation tanks and the same was done for the red grapes.  Samples were taken at random during the fermentation process: sixty-six for the white grapes and fifty-eight for the red grapes.  The traditional reference method aerometry was used to determine volumic mass values.  Each sample was scanned for NIR spectra from 200 nm to 1100 nm at 0.5 nm intervals averaging three hundred scans per spectrum. 

Volumic Mass:R2=0.96 RMSEP=5.85 g/dm3
Wavelength Range:800 nm to 1050 nm

Different spectral pre-treatments and selective wavelength ranges were applied as well as two modeling algorithms. The best results used the wavelength range from 800 nm to 1050 nm and the Partial Least Squares algorithm. The results were especially good considering the different grape varieties and the model combined both the white and red grape data. Monitoring volumic mass during fermentation in real-time using NIR spectroscopy can avoid stuck fermentation and potential refermentation, both of which can lead to quality deficiencies in both physical and chemical characteristics in the final product of wine.
https://www.tandfonline.com/doi/full/10.3109/09637486.2010.533161

Rapid Detection of Three Quality Parameters and Classification of Wine Based on Vis-NIR Spectroscopy with Wavelength Selection by ACO and CARS Algorithms – Hu, Yin, Ma, Liu, Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 205 (2018) 574-581

This study evaluated the feasibility of using a Vis-NIR spectrometer for classifying wine samples based on geographical origin and analyzing three constituents in wine: total acidity, total sugars, and alcohol. Ninety-one samples of wine were procured from local markets comprising seven different brands. Wines were stored at constant temperature before analysis. Spectra were collected at 2 nm intervals from 400 nm to 2500 nm using a 1 mm quartz sampling cell. Standard methods were used to obtain reference values for total acidity, total sugars, and alcohol. Principle Component Analysis (PCA) was performed for classification analysis. Partial Least Squares (PLS) regression models were created to correlate the spectral data to reference values using the full wavelength range as well as two algorithms for selective wavelength analysis: CARS (Competitive Adaptive Reweighted Sampling Method) and ACO (Ant Colony Optimization).

PLS-Full:
Total Acidity:R2=0.941RMSEP =.00175 mol/l
Total Sugars:R2=0.990RMSEP =0.157 g/l
Alcohol by VolumeR2=0.911RMSEP =0.242 v/v
PLS-CARS
Total Acidity:R2=0.972RMSEP =0.00116 mol/l
Total Sugars:R2=0.996RMSEP =0.102 g/l
Alcohol by VolumeR2=0.939RMSEP =0.200 v/v
PLS-ACO
Total Acidity:R2=0.987RMSEP =0.00108 mol/l
Total Sugars:R2=0.999RMSEP =0.0827 g/l
Alcohol by VolumeR2=0.942RMSEP =0.187 v/v

PCA analysis showed that most of the wines used in the study could be classified based on geographical origin. Some of the wines were very similar in acidity, sugars, and alcohol and it is likely that more detailed classification methods could separate these samples as well. PLS modeling results were excellent and the PLS-ACO method showed the best results. The ACO algorithm selected eighty-six specific wavelengths for the PLS models. The results here show promise for using Vis-NIR spectra with calibration models created using reference data and selective wavelength algorithms as a tool for classifying wine and performing quality assessment during wine fermentation and other production processes.
https://www.sciencedirect.com/science/article/pii/S138614251830708X

Predicting Calcium in Grape Must and Base Wine by FT-NIR Spectroscopy – Vestia, Barroso, Ferreira, et al., Food Chemistry 276 (2019) 71-76

Calcium content in sparkling wines cannot exceed 80 mg/L due to the risk of aggregation with alginate capsules.  It can be abundant in the grape itself as well derived from contamination in soil.  The concentration of calcium as well as other minerals is affected by maturity, variety, soil type, and climate during grape growth.  The current reference method for determining calcium content in wine is atomic absorption spectrophotometry (AAS), which is time-consuming, expensive, and requires complex operations and skilled technicians to implement.  NIR spectroscopy was examined as a method for predicting calcium content in both grape must and wine.  Calcium is a common element for NIR spectroscopic methods due to the high content in plants and its interaction with some food quality parameters.  Ninety-eight white wine samples and sixty grape must samples were procured for the study.  NIR spectra were collected using an from 1100 nm to 2300 nm at 1 nm wavelength intervals and two hundred fifty scans per average.  A probe with 2 mm pathlength was used to collect the spectra.  Calcium was determined using the AAS method and various data treatments were performed on the spectral data before creating Partial Least Squares calibration models correlating calcium to the spectra.

Grape Must:
Calcium:R2=0.935RMSEP=6.960 mg/L
Wine:
Calcium:R2=0.956RMSEP=3.311 mg/L

Correlation was good between the spectral data and calcium reference method and predictions using an external validation set proved the feasibility of the models. This validation set compromised ten samples from the following year’s vintage at the winery where the study was performed. These models could be used for a rapid and reliable technique for quantifying calcium from NIR spectra. One potential quality control improvement from such a method would be to separate grapes and must according to calcium content in order to prevent putting a large amount of high calcium grapes into one fermentation vat.
https://www.sciencedirect.com/science/article/pii/S030881461831687X