FT-NIR Analysis of Wine

Importance of the topic

The chemical and physical composition of wine (ethanol, sugars, acids, density, pH, volatile acids, sulfites, etc.) determines product quality, legal compliance, taxation and process control. Rapid, robust multi-parameter analytics are therefore valuable in winery QA/QC and process monitoring. FT‑NIR spectroscopy offers non-destructive, container-friendly sampling and the ability to predict multiple constituents from a single spectrum, reducing analysis time compared with traditional wet-chemistry, hydrometry or chromatographic methods.

Objectives and overview of the study

This application note evaluates the Thermo Scientific Antaris FT‑NIR analyzer for simultaneous quantitative analysis of common wine quality parameters. The study aimed to develop and validate partial least squares (PLS) calibrations for multiple analytes (ethanol, °Brix, density, total and volatile acids, sugars, pH, sugar-free extract, etc.), demonstrate spectral regions useful for calibration, and characterize precision, accuracy and model stability using TQ Analyst chemometrics and RESULT software for data handling.

Used instrumentation

• Thermo Scientific Antaris FT‑NIR analyzer (FT technology, pinned-in-place optics, dynamically aligned interferometer)
• InGaAs transmission detector with C attenuation screen
• Three-position heated cuvette holder, 1 mm glass cuvettes, cell temperature 40 °C (samples degassed and pre-heated as needed)
• Spectral acquisition: 4000–10000 cm⁻¹, 100 co‑averaged scans, 4 cm⁻¹ resolution, 30 s pre‑collection delay
• Software: RESULT for acquisition/workflows and TQ Analyst for chemometric calibration (PLS)

Methodology

Samples: Representative wine standards were prepared (e.g. 124 standards for ethanol, 133 for density) covering the expected concentration ranges. Samples were analyzed in transmission through 1 mm cuvettes at 40 °C to minimize CO₂ interference.

Spectral treatment and chemometrics: Spectra were collected between 4000 and 10000 cm⁻¹; highly absorbing water bands were excluded. Calibration spectral regions identified as best for quantitation included ~4100–4600, 5700–6000 and 6450–7700 cm⁻¹ (with caution for water interference in the highest region). Partial Least Squares (PLS) regression was used for all models. Spectral pre‑processing included operations such as second derivative, baseline correction and selection of multiple regions when beneficial. Model selection used PRESS statistics and cross‑validation to avoid overfitting; automatic and manual region selection in TQ Analyst supported optimization.

Main results and discussion

Multiple quantitative models were developed with strong statistical performance. Key calibration outcomes (representative):

• Ethanol (% v/v): 124 standards, 4 PLS factors, 2nd derivative preprocessing, correlation coefficient R = 0.9984, RMSEC = 0.23% v/v, RMSECV = 0.26% v/v. PRESS plots indicated a robust minimum and a conservative 4‑factor model to avoid overfitting.
• Density (g cm⁻³): 133 standards (0.987–1.076 g cm⁻³), two spectral regions selected, one‑point baseline anchored at 8300 cm⁻¹, 6 PLS factors, R = 0.9993, RMSEC ≈ 0.0007 g cm⁻³, RMSECV ≈ 0.0008 g cm⁻³. PRESS behavior indicated model stability.
• Brix refraction (°Brix): 6 PLS factors, R = 0.9998, RMSECV ≈ 0.08 °Brix.
• Total sugars (two value ranges): R = 0.9968–0.9995, RMSECV ≈ 1.5–1.7 g L⁻¹ (lower range) and 1.2–1.5 g L⁻¹ (higher range).
• Total acids: R ≈ 0.9872, RMSECV ≈ 0.4 g L⁻¹; volatile acids: R ≈ 0.9788, RMSECV ≈ 0.06 g L⁻¹.
• pH: R ≈ 0.9505, RMSECV ≈ 0.08 pH units (poorer performance relative to major constituents).

Spectral regions around 4400 cm⁻¹ show clear ethanol features; water bands were avoided to reduce baseline and saturation effects. PRESS and cross‑validation statistics demonstrated good model generalizability for most analytes. The Antaris FT‑NIR combined with TQ Analyst produced rapid predictions with errors comparable to or better than routine methods for many components.

Benefits and practical applications of the method

• High throughput: single spectral measurement yields simultaneous predictions for many analytes in seconds versus hours for classical wet chemistry or chromatography.
• Non‑destructive and container‑friendly sampling: long NIR pathlengths allow transmission through glass or plastic, enabling at‑line and QC lab workflows.
• Good analytical performance for major wine parameters (ethanol, density, °Brix, sugars, acids) with low RMSECV and high correlation coefficients.
• Repeatability and stability: FT interferometer design, pinned optics and consistent beam path yield excellent scan‑to‑scan repeatability and straightforward method transfer potential.
• Accessible operation: RESULT workflows permit operators without spectroscopy expertise to run validated acquisition and prediction routines.

Future trends and potential applications

• At‑line and online integration: implementing FT‑NIR for real‑time process control during fermentation and blending to optimize yield and quality.
• Model transfer and networked calibration: leveraging transfer algorithms and standardized workflows to distribute calibrations across sites and instruments.
• Data fusion and advanced chemometrics: combining FT‑NIR with mid‑IR, Raman, GC or MS data and applying machine learning to improve prediction of minor or structurally similar analytes (e.g., individual phenolics, low‑level volatiles).
• Robustness and regulatory acceptance: continued validation against reference methods to expand regulatory and taxation use cases (e.g., ethanol declaration).
• Miniaturization and inline probes: development of fiber‑optic or immersion probes for continuous monitoring in tanks and pipelines.

Conclusion

The Thermo Scientific Antaris FT‑NIR analyzer, together with RESULT acquisition workflows and TQ Analyst chemometrics, provides a reliable, rapid method for simultaneous quantification of key wine parameters. The method demonstrated excellent statistical performance for ethanol, density, °Brix and several other quality metrics, enabling faster QA/QC decisions and potential process control applications. While some parameters (e.g., pH) are less accurately predicted by NIR, the technique offers substantial practical advantages for routine winery analytics when properly calibrated and validated.

Reference

Application Note 50813: FT‑NIR Analysis of Wine, Thermo Fisher Scientific (Jeffrey Hirsch, Ladislav Tenkl, Martin Hollein), 2007. Instrumentation and software described: Thermo Scientific Antaris FT‑NIR analyzer, RESULT software, and TQ Analyst chemometrics.