Non-destructive measurement of ascorbic acid content in bell peppers by VIS-NIR and SWIR spectrometry

The present study created a basis for development of a fast, non-destructive method to measure ascorbic acid content during various growth stages of three bell pepper cultivars: ‘Ever Green’, ‘No. 117’ and ‘Celica’. Fruit were tagged at the flowering stage and picked weekly during 9 weeks of growth, until fully ripe. Visible and near-infrared (VIS–NIR) spectra in the range 477–950 nm were obtained in reflectance mode from intact peppers with a USB2000 mini-spectrometer. Short-wave infrared (SWIR) spectra in the range 850–1888 nm were obtained in reflectance mode from the same sampled peppers with a Liga Spectrometer. Fruit flesh samples were cut from the area scanned for spectral measurements, and destructively analyzed for ascorbic acid content. The highest ascorbic acid content was measured in the green variety ‘Ever Green’ (148.1 mg per 100 g). Varieties ‘Ever Green’ and ‘No. 117’ accumulated significantly higher concentrations of ascorbic acid than ‘Celica’, and consistent and significant differences were found between the premature, green and mature stages of all three cultivars. Partial Least-Squares (PLS) regression was used in spectral model development. Comparisons were made between the PLS regression analysis of the reflectance spectra (R), its first derivative (D1R), the log(1/R) and its first (D1log(1/R)) and second derivative (D2log(1/R)). VIS–NIR and SWIR spectroscopy measurements showed good correlation (r2 = 0.76) with changes in ascorbic acid content. The PLS regression models were able to predict ascorbic acid content for all three cultivars, with 15.1–18.9 mg per 100 g cross-validation error (RPD = 2–2.4). The method offers potential for non-destructive detection of changes in ascorbic acid content during growth and maturation of intact bell peppers.

► Measurements of ascorbic acid contents in intact bell peppers were made using VIS–NIR and SWIR mini-spectrometers.
► Regression models were able to predict ascorbic acid content with 11.7–14.6% root mean square error of cross-validation.
► VIS–NIR spectral measurements predicted ascorbic acid contents efficiently in mixed cultivar datasets.
► A new evaluation method was suggested for a more objective comparison among complex regression models in differing spectral ranges.