Sugar: Brix and Total Sugar by the Phenol–Sulfuric Acid Method

• An infrared LED shines light into the fruit and the detector picks up light scattered back from the flesh — no cutting or juicing; measure on the tree or after harvest
• The instrument has built-in conversion models for each fruit, so models are sold by fruit type (grape, strawberry, apple, pear, mango, melon, watermelon, corn… over 25 kinds) and users need no calibration of their own
• The reading is a model-estimated Brix; such models are generally built with refractometer Brix of juice as the reference, so accuracy still has to be checked against the refractometer
• Strong outdoor sunlight interferes; the display shows “nnn” to prompt you to shade the fruit and remeasure (ambient-light interference detected)

• Light path: LED light enters the flesh from the outer ring through the ring lens; the scattered light is collected by the central light guide and passes through the filters to the sensor — illumination and collection are separated, which blocks direct reflection from the skin surface
• The 875 nm filter is paired with an 880 nm LED; the 1050 and 1064 nm filters share a 1064 nm LED — LED wavelength ≠ filter wavelength

• Bandpass filter: lets only a narrow band of wavelengths through, so the sensor measures one wavelength at a time

• Absorbance A=log₁₀(I₀/Iₛ): I₀ is the intensity sent in, Iₛ the intensity collected
• Subtract A1050 from all: sugar barely absorbs at 1050 nm, so it serves as a common reference that removes the overall offset from skin thickness and scattering
• Multiple regression Ŷ=b0+b1x1+…: several x (wavelength differences, two temperatures) each multiplied by a coefficient and summed; the coefficients must be fitted to refractometer Brix of many fruit (PLS in W13 is the advanced version)

• Spectrometer: Zeiss MMS-1 (enhanced VIS–NIR version), 310–1100 nm continuous spectrum, 256-pixel silicon photodiode array, about 3 nm sampling, resolution about 10 nm
• Why it is stable: the grating, fiber inlet and detector are all fixed on a single glass body, aligned at the factory; thermal drift < 0.01 nm/K, suitable for taking to the orchard
• Measurement: a tungsten lamp shines broadband light on a ring-shaped area of the skin, and after scattering in the flesh it is collected in the center (interactance) — the same concept as the ATAGO patent on the previous page
• Predicts: soluble solids (°Brix), dry matter (DM), flesh and skin color, titratable acidity

• You build the model: measure 10–200 training fruit plus destructive reference values (refractometer Brix, oven-dried DM), then use the manufacturer's Model Builder to build a model with PLS (second-derivative absorbance spectra) — exactly what W10–W13 do
• Common in research: Walsh's group at CQU in Australia built fruit sugar models with the MMS-1 as early as 2000; more recently the F-750 measured mango dry matter with an RMSEP of about 0.8 %FW (Walsh et al., 2024; this is dry matter, not °Brix)
• Note: the F-751 series (dedicated mango, avocado, etc. units) uses a Hamamatsu 640–1050 nm spectrometer with factory built-in models and is not the MMS-1

• How to judge accuracy: plot “refractometer Brix vs NIR prediction” as a scatter plot with a y = x line; calculate Bias, MAE (mean absolute error) and RMSEP (root mean square error of prediction) — explained in plain language on the next page
• 60 readings ≠ 60 fruit: 10 fruit × 2 spots × 3 replicates gives 60 readings, but there are only 10 independent samples; replicates only reflect repeatability
• Avoid data leakage: all readings from the same fruit must go together into the training set or the test set; splitting them randomly makes RMSEP look too good
• Practice: calculate Bias, MAE and RMSEP using a test set of 8 simulated fruit (test set CSV); the seven-channel intensity conversion can be checked against the simulated intensity CSV. Both are teaching simulation data, not patent measurements or instrument performance

y = refractometer reference value, ŷ = NIR predicted value, n = number of independent fruit. The same formula is called RMSEC when computed on the training set, RMSECV with cross-validation, and RMSEP only when computed on a test set that took no part in model building. Other abbreviations: NIR = near-infrared, PLS = partial least squares regression, DM = dry matter.

• Concentrated sulfuric acid causes sugars to undergo hydrolysis + dehydration, forming furfural derivatives
• These derivatives condense with phenol to give a stable orange-yellow product, read colorimetrically at 490 nm
• Measures total sugar (total carbohydrate), including sucrose and oligosaccharides (unlike copper-reduction methods, which measure only reducing sugars); note this is not the “sugars” on a nutrition label, which count only free monosaccharides and disaccharides
• Sugar alcohols do not react (sorbitol, erythritol, etc.) — the g3 game will verify this shortly
• Different sugars (e.g., fructose vs glucose) have different molar color yields, so changing the standard causes systematic bias
• Color development is not strictly stoichiometric, so you must always run your own calibration curve
• Add the concentrated sulfuric acid quickly in one go onto the liquid surface: the instant heat release makes the dehydration fast and uniform; adding slowly down the wall tends to layer and gives inconsistent color

• Maple syrup: must be diluted heavily first to fall within the calibration range
• Orange juice: measure both Brix and phenol–sulfuric total sugar and compare them
• Sports drink: often contains high-fructose corn syrup (HFCS); watch for bias from the choice of standard
• Winter melon tea: a dark sample; check whether the color after dilution still interferes with colorimetry, and subtract the background color with a sample blank without phenol

Rule of 86: 86/2% ≈ 43 : 1, consistent with the volume ratio calculated above; the video's “40:1” fits exactly if the sap is just slightly sweeter (about 2.18%) — the order of magnitude matches, so this claim survives a mass-balance check.