1：Experimental Design and Method Selection:

The study employs spectroscopic techniques (MIR, NIR, EEM, 1H-NMR) for non-targeted analysis and data fusion strategies (mid-level data fusion and P-ComDim) to classify PDO wine vinegars. Chemometric methods like PCA, PARAFAC, MCR, PLS-DA, and NPLS-DA are used for data analysis and classification.

2：Sample Selection and Data Sources:

Sixty-five PDO wine vinegar samples from three Spanish PDOs (Vinagre de Jerez, Vinagre de Condado de Huelva, Vinagre de Montilla-Moriles) were provided by local wineries and analyzed in duplicate. Samples include different categories based on aging and sweetness.

3：List of Experimental Equipment and Materials:

Spectrometers and software for MIR (Bruker Vertex 70 FTIR with ATR accessory, OPUS 7.0, OMNIC), NIR (ABB Bomen IR spectrometer with cuvette, GRAMS/AI software), EEM (Varian Cary-Eclipse fluorescence spectrophotometer with Cary-Eclipse software), and 1H-NMR (Bruker AVIII 700 spectrometer with Bruker TopSpin 3.0 and Chenomx NMR Suite 7.0). Chemicals include TMSP-2,2,3,3-d4 in D2O for NMR.

4：0, OMNIC), NIR (ABB Bomen IR spectrometer with cuvette, GRAMS/AI software), EEM (Varian Cary-Eclipse fluorescence spectrophotometer with Cary-Eclipse software), and 1H-NMR (Bruker AVIII 700 spectrometer with Bruker TopSpin 0 and Chenomx NMR Suite 0). Chemicals include TMSP-2,2,3,3-d4 in D2O for NMR. Experimental Procedures and Operational Workflow:

4. Experimental Procedures and Operational Workflow: Samples were analyzed directly without pre-treatment for MIR, NIR, and EEM; for NMR, samples were prepared with TMSP addition. Spectral data were acquired under specified conditions (e.g., wavelength ranges, scans, temperature). Data preprocessing included mean centering, SNV, scattering correction, alignment, and baseline correction. Data were split into training and test sets using the Duplex algorithm, with five repetitions for model validation.

5：Data Analysis Methods:

Decomposition methods (PCA for MIR/NIR, PARAFAC for EEM, MCR for NMR) were applied for feature extraction. Classification used PLS-DA and NPLS-DA with LDA on scores. Data fusion involved concatenating features from different techniques with scaling (autoscaling or block-autoscaling). P-ComDim was used for multiblock analysis, and performance was evaluated based on classification error rates.