Physics-based interpretation of NIR interaction spectroscopy in layered, structurally evolving samples: Water content...

Near-infrared (NIR) interaction spectroscopy enables non-contact, non-destructive assessment of water content in dried salt-cured whitefish (clipfish), but structural heterogeneity alters photon transport and limits calibration robustness. This study investigates how drying-related changes in photon transport govern spectra and calibration behaviour. NIR interactance spectra (761-1081nm) were acquired from 108 intact dried salt-cured whitefish (two positions per fish) across four production batches using five illumination-detection offsets from 6.5 to 18.5mm. For two batches (58 fish), measurements were repeated after removal of the surface salt layer. Monte Carlo simulations in a two-layer system investigated how absorption, scattering, salt-layer development, and sampling geometry govern spectral variation. Partial least squares (PLS) regression models were calibrated using 116 samples and tested using 99 samples from two independent batches. Spectral variation with water content was governed primarily by formation of a highly scattering salt layer rather than by changes in muscle absorption. Measurements exhibited regime-dependent behaviour consistent with muscle-dominated sampling, salt-layer development, and a stabilised salt layer limiting muscle sensitivity. For absorbance and intensity-normalised spectra, external SEP ranged from 0.69-1.19% and 1.11-1.53% for the two test batches, with biases of 1.50-3.15% and -0.05 to -3.31%, respectively. Sampling geometry revealed a trade-off between sampling depth and signal strength. Intensity-related variation, often removed through spectral pre-processing, contained physically meaningful information linked to structural evolution. These results establish a physics-based framework for interpreting NIR interactance calibration behaviour and guiding measurement strategies toward robust, transferable NIR measurements in layered, structurally evolving products.