Data-Driven Spectral Unmixing in X-ray Absorption Spectroscopy

by Dr Ilnura Usmanova (Senior Data Scientist, SDSC hub at PSI)

Thursday 19 Mar 2026, 11:30 → 13:00 Europe/Zurich

OHSA/B17

Abstract: 
Signal separation in operando X-ray Absorption Spectroscopy (XAS) is challenging when the chemical species involved are unknown and reference spectra are unavailable. Blind signal separation methods such as Multivariate Curve Resolution (MCR) are widely used, but their solutions are often non-unique and depend strongly on initialization, making the selection of starting points a time-consuming expert task. We present a data-driven hybrid approach that leverages databases of measured XAS spectra to guide the MCR decomposition through regularization toward physically realistic solutions. We also provide a way to estimate uncertainty of the obtained solution. The method requires no prior experimental knowledge or manual initialization. Its performance is demonstrated on synthetic mixtures and real operando XAS datasets.

TEAMS link