Shedding Light on the Past: Temporal Classification of Zoological Specimens from Museum Collections with Portable NIR Sensors and Multivariate Error Modeling

Museum scientific collections preserve invaluable biological archives that provide insights into historical biodiversity and environmental change. Determining the age of specimens often relies on destructive, labor-intensive, and costly methods, limiting their use on rare or valuable materials. In this study, we present a fully nondestructive and rapid approach for classifying the temporal origin of zoological skeletal specimens using portable near-infrared spectroscopy combined with an advanced chemometric framework, exemplified by red squirrel (Sciurus vulgaris) bones. Two compact NIR instruments, covering distinct wavelength ranges, were employed to analyze bone samples collected from two temporal groups: "historical" (1916-1923) and "modern" (2005-2021). To extract chemically meaningful information while accounting for instrumental and physical variability, we implemented a maximum likelihood principal component analysis-logistic regression (MLPCA-LR) strategy that explicitly incorporates the measurement error covariance structure. The resulting models achieved perfect or near-perfect discrimination, validated through cross-validation, independent test sets, and bootstrap analysis. Compared to the widely used partial least-squares discriminant analysis (PLS-DA), the MLPCA-LR framework demonstrated superior robustness and interpretability. This study suggests that NIR spectroscopy with portable sensors, combined with MLPCA-LR, offers a nondestructive and accessible approach for temporal classification of skeletal specimens, enabling practical in situ screening in museums without invasive sampling or expert operators.