Abstract
Laser Ablation-Inductively Coupled Plasma-Mass Spectrometry (LA-ICP-MS), particularly in its time-of-flight (TOF) configuration, enables rapid, high-resolution elemental imaging across complex geological materials, offering spatial and chemical insights at the micrometer scale. However, quantitative accuracy is often limited in fine-grained or mineralogically heterogeneous matrices due to the failure of global normalization strategies, such as 100 wt % oxide assumptions, to account for mixed-phase compositions. Here, we present a workflow that leverages Uniform Manifold Approximation and Projection (UMAP) for unsupervised dimensionality reduction and k-means clustering to segment mineralogical phases directly from per-pixel elemental concentration maps. Cluster compositions are matched to known minerals based on stoichiometric similarity, enabling pixel-wise, phase-specific normalization (e.g., oxides vs carbonates). Validated with dawsonite-bearing sandstones from Mt. Amiata, Italy, this approach significantly reduces quantification errors, correcting systematic over- or underestimations of up to 60%. The method also enables a consistent, phase-resolved geochemical comparison across depth profiles. This study establishes UMAP not only as an exploratory tool but also as a practical guideline for accurate and interpretable quantification in multielemental imaging.