*2.1. Samples*

Traditional and BBL maple syrup samples were kindly provided by a local maple syrup farm in Jefferson, OH (*n* = 12 (traditional), *n* = 8 (BBL)) and were purchased from local grocery stores in Columbus, OH (*n* = 7 (traditional), *n* = 5 (BBL)) that consisted of traditional maple syrups (*n* = 19), including dark, amber, and golden grades, and BBL-aged maple syrup (*n* = 13). In addition, table syrups (corn *n* = 2, cane *n* = 2, and mixture, consisting of cane, maple, and agave syrups *n* = 1) (*n* = 5) were purchased from grocery stores in Columbus, OH, USA for generating training models. An independent external validation set, consisting of traditional (*n* = 4) and BBL (*n* = 4) maple syrups, was purchased from an online vendor (Amazon.com, Inc. Seattle, WA, USA). All samples were stored in the refrigerator at 4◦C and were equilibrated at room temperature before spectroscopic measurements and reference analyses.

#### *2.2. Reference Analyses*

2.2.1. ◦Brix

◦Brix of each sample was measured with the heat-controlled refractometer (RX 5000i ATAGO, Bellevue, WA, USA). The syrup sample (~0.3 mL) was carefully pipetted onto the prism of the refractometer without creating any air bubbles, and measurement at 22 ◦C was recorded.

#### 2.2.2. High-Performance Liquid Chromatography

Concentrations of sucrose, fructose, and glucose were measured with high-performance liquid chromatography (HPLC) (Shimadzu, Columbia, MD, USA). The HPLC was equipped with a SIL-20AHT autosampler, a CTO-20A oven, an LC-6AD pump, a CBM-20A controller, and a RID-10A refractive index detector. The syrup sample (~0.5 g) was weighed into a 15 mL centrifuge tube and diluted with (~7 mL) HPLC grade water. The actual weights of syrup and water were recorded. The mixture was vortexed for 40 sec and was filtered through the 0.2 μm filter (Phenomenex®, Torrance, CA, USA), and then filled into an HPLC vial. Isolated sugars were segregated by a Rezex RCM-Monosaccharide Ca+ 300 × 7.8 mm column (Phenomenex®). Sugars were eluted under the isocratic condition at 80 ◦C, using HPLC grade water as a mobile phase at a 1 mL/min flow rate for 20 min. LC Solutions software (Version 3.0, Shimadzu, Columbia, MD, USA) was used to integrate chromatograms automatically. The standard curve with concentration ranges from 10 to 50 mg/mL (>99% purity, Fisher Scientific, Fair Lawn, NJ, USA) was plotted to calculate each sugar content.

#### 2.2.3. Total Phenolics

Total phenolic contents of maple syrups were determined with Folin–Ciocalteu (FC) method described by Waterhouse with some modification [19]. The syrup sample (~0.8 g) was weighed into a microcentrifuge tube and diluted with deionized (DI) water (~0.4 mL). The actual weights of syrup and water were recorded, and the diluted sample was vortexed for 40 s. The diluted sample (50 μL) was pipetted into a 96-well plate, followed by 200 μL DI water and 20 μL FC reagent. The mixture was mixed thoroughly by pipetting and incubated for 7 min at room temperature. The sodium carbonate solution (100 μL) was added to the mixture and incubated for 2 h under dark conditions at room temperature. The equilibrated sample's absorbance was measured at 765 nm. A standard curve constructed with gallic acid standard with concentration ranges from 125 to 800 μg/mL was used to quantify total phenolics. Results were expressed as micrograms of gallic acid equivalent (GAE) per 1 mL of distilled water.

#### 2.2.4. Gas Chromatography—Mass Spectrometry

The volatile composition of the samples was identified using gas chromatography– mass spectrometry (GC-MS) (Agilent 7820A GC connected to a 5977B MS, Agilent Technologies, Santa Clara, CA, USA). A total of 1 g maple syrup sample was placed into a 20 mL clear screw-tread glass headspace vial (Restek, Bellefonte, PA, USA), and the vial was sealed with an 18 mm screw-tread PTFE/silicone septa vial cap (Restek, Bellefonte, PA, USA). The vial with the sample was placed onto a heating plate at 40 ◦C for 30 min to equilibrate the volatile compounds. A preconditioned SPME fiber (50/30 μm DVB/CAR/PDMS coated) (Supelco, Sigma-Aldrich, Bellefonte, PA, USA) assembly was inserted in the vial through the septa of the cap, and the volatiles were trapped by the fiber for 15 min. After the trapping, fiber assembly was removed from the vial and directly inserted through the GC-MS injection port. Compounds were desorbed at 250 ◦C for 1 min in splitless mode, followed by a 30 s purge flow (50 mL/min) to clean the fiber. A quality control (QC) sample was prepared by pooling 100 μL of each sample to monitor the performance of the method and identify qualified peaks. 2,3-hexanedione (Sigma-Aldrich, St. Louis, MO, USA) was prepared at 10 ppm concentration with distilled water and used as an internal standard (IS) to correct the variation through the run. A 40 μL of IS solution was added to each sample. The volatile compounds were separated on a DB-Wax column (60 m × 250 μm × 0.25 μm) (Agilent Technologies, Santa Clara, CA, USA). The oven was held at 60 ◦C for 5 min then ramped to 130 ◦C at 5 ◦C/min. This was followed by the second ramp of 5 ◦C/min to 240 ◦C, where it was held for 8 min. The MS acquisition was performed in scan mode between masses 25–300 m/z at a 2.7 scans/s rate. Data were extracted in the Agilent Masshunter Quantitative Analysis software. The spectral background was corrected, and only peaks that had a signal-to-noise (S/N) ratio higher than the detection limit (S/N > 5) were conserved. All compounds were tentatively identified using the NIST 14.L database by a Mass Spectral Library search.

#### 2.2.5. Statistics of Reference Analysis

All reference laboratory analyses were performed in duplicate, and their range, minimum, maximum, mean, and standard deviation (SD) are determined. In addition, the standard error of laboratory (SEL) was calculated according to the method of Kovalenko et al. [20].

#### *2.3. Vibrational Spectroscopy*

#### 2.3.1. Mid-Infrared Analysis

The mid-infrared data were collected with portable FT-IR spectroscopy (Agilent Technologies, Santa Clara, CA, USA) attached with a triple-reflection diamond Attenuated Total Reflectance (ATR) crystal. The ATR crystal has a sampling surface of 2 mm diameter and a 200 μm active area and provides ~6 μm depth of penetration. In addition, the FT-IR system is also attached with a deuterated triglycine sulfate (DTGS) detector and a Zinc Selenide (ZnSe) beam splitter. Spectra were collected from 4000 to 700 cm<sup>−</sup><sup>1</sup> with a resolution of 4 cm<sup>−</sup>1. Sixty-four spectra were co-added in each sample collection to increase the signalto-noise ratio. A spectral background was taken in between every measurement to reduce the environmental changes. Approximately 0.2 g of syrup sample was directly applied to the sampling surface of the ATR crystal, confirming that full coverage of the sample was achieved. Spectra for each sample were collected in triplicate, and collected spectral data were documented by Agilent MicroLab PC software (Agilent Technologies, Santa Clara, CA, USA).

#### 2.3.2. Raman Analysis

About 3 mL of syrup sample was filled in a quartz cuvette (Hellma Analytics, Mulheim, Germany) with a 10 mm light path and measured with a compact benchtop Raman spectrometer WP 1064 (Wasatch Photonics, Durham, NC, USA). The Raman spectrometer was coupled with a laser operating at 1064 nm and an Indium Gallium Arsenide (InGaAs)

detector. Spectra were collected from 350 to 1500 cm<sup>−</sup><sup>1</sup> with a resolution of 4 cm<sup>−</sup>1. In addition, three scans were co-added and averaged to increase the signal-to-noise ratio of the spectrum, which has an integration time of 3 s. A spectral background was taken in between every measurement to reduce the environmental changes. The spectral collection was performed in triplication for each sample, and collected spectral data were documented by EnlightenTM software (Wasatch Photonics, Durham, NC, USA).

#### *2.4. Multivariate Data Analysis*

FT-IR and Raman spectra were exported and analyzed using Pirouette® multi-variate statistical analysis software (version 4.5, Infometrix Inc., Bothell, WA, USA). The mean spectrum of the three replicates was used for the statistical analysis. The collected FT-IR and Raman data were preprocessed with mean-centering to reduce micro multicollinearity and transformed with the Savitzky–Golay (SG) algorithm (35-point polynomial filter) in soft independent modeling of class analogy (SIMCA) and partial least squares regression (PLSR) models [21]. The SG algorithm was used to resolve overlapping spectroscopic signals and to improve their properties, also surpassing the instrument noise [22]. A 35-point smoothing filter was found as an optimal window length for our data set. The optimum window length was chosen to resolve essential details in the collected spectra and lessen signal noise. Mean centering and SG algorithms were chosen after evaluating the preprocessing quality of the spectral data with other options, including smoothing, normalization, and divide by; however, they were all outperformed by the combination of mean-centering and SG. An additional data transformation step of normalization (2-norm × 100) was applied in the case of PLSR analysis.
