What Lies INSIDE: Chemometric Insights on the Penetration Depth of Near-Infrared Radiation in Spectral Imaging Configurations

Understanding how deeply near-infrared (NIR) radiation penetrates into matter is essential for interpreting (hyper)spectral imaging (SI) data, yet comprehensive assessments of penetration depth remain limited. Although NIR-SI is traditionally considered a surface analytical technique, subsurface contributions may affect spectral profiles, challenging this assumption. This study presents a systematic and quantitative investigation of NIR penetration in a controlled SI setup using multilayer polymeric samples manufactured by 3D printing. Layered cubes and cylinders composed of polylactic acid (PLA) and polyethylene terephthalate glycol (PETG) were analyzed using two independent short-wave NIR-SI systems. Penetration behavior was evaluated through an integrated chemometric workflow combining exploratory analysis (principal component analysis, PCA), unmixing (classical least squares, CLS), supervised linear (partial least squares, PLS) and nonlinear (convolutional neural network, CNN) regression methods. The results demonstrated that NIR-SI can retrieve chemical information from subsurface layers to depths of similar to 1 cm. Penetration depth was strongly influenced by material composition, internal structure, and illumination conditions. PCA and CLS revealed nonlinear attenuation and scattering phenomena, while regression models successfully retrieved subsurface information across the full sample height. CNN consistently outperformed PLS, highlighting the importance of nonlinear modeling approaches. The robustness of the proposed strategy arises from the use of designed ad hoc samples, two independent instruments, and multiple complementary chemometric methods converging toward consistent results. Overall, this work challenges the conventional surface-limited view of NIR-SI and provides a robust framework for investigating NIR penetration, supporting the development of depth-resolved and potentially tomographic NIR-SI approaches.