Comparing post-acquisition methods for Raman-based mineral identification of particles in yellow earth pigments

Natural yellow earth pigments are heterogeneous mineral assemblages dominated by iron (hydr)oxides and accompanied by variable accessory phases such as quartz, calcite, gypsum, and aluminosilicates. Although these accessory minerals may encode geological origin, their reliable identification from bulk Raman spectra is complicated by multiphase band overlap, fluorescence background, and particle-level heterogeneity. As a result, mineral discrimination from Raman data alone remains analytically demanding. Here, we examine whether systematic post-acquisition processing can enable robust mineral-based differentiation of closely related pigments using Raman spectroscopy alone. Five yellow earth pigments (three yellow ochres and two raw siennas) were analyzed using particle-resolved confocal Raman spectroscopy, generating statistically representative datasets (>200 spectra per sample). Mineral identification was performed using three strategies, visual interpretation, automated three-component spectral library fitting (3C model), and supervised machine learning classification (XGBoost), and assessed using a binary presence/absence scheme to support comparison across heterogeneous particle populations. Both the 3C and XGBoost approaches enabled statistical separation of all five pigments, whereas K-means cluster analysis of the visual interpretation did not distinguish between the two raw siennas. Together, these results support a reproducible Raman-only workflow for preliminary mineral assignment and pigment discrimination in provenance-oriented cultural heritage studies.