Automated Full-Spectrum Multivariate Calibration for Quantitative Glow Discharge Optical Emission Spectrometry

Glow discharge optical emission spectrometry (GD-OES) enables rapid bulk and depthresolved multi-elemental analysis of solid materials with minimal sample preparation. Conventional quantification, however, relies on preselected analyte emission lines and, if sputtering-rate variations must be compensated, on matrix reference lines or sputtering-rate correction, requiring matrix composition or sputtering rates as input. Modern CCD-based instruments simultaneously record broadband spectra containing thousands of spectral variables, yet most of this information remains unused in conventional calibration based on a few selected lines. Here, we investigate a full-spectrum calibration strategy based on Variable Importance in Projection-selected Partial Least Squares (VIP-PLS) regression for the quantitative determination of individual alloying elements. Al-Si and Cu-Zn alloys were selected as metallurgically and spectrochemically distinct model systems, with Si and Zn as target analytes, respectively. The VIP-PLS models predicted Si and Zn with Root Mean Square Error of Cross-Validation (RMSECV) values of 0.38 wt% and 0.50 wt%, compared with 1.05 wt% and 1.41 wt% for univariate calibration of a single preselected line, and 0.24 wt% and 0.63 wt% for normalized ratio-based calibration using a reference line and the matrix mass fraction. VIP-PLS thus achieved performance comparable to the ratio-based calibration, implicitly compensating for sample-specific intensity variations, such as the sputtering rate, without matrix composition as input and without manual line selection. Time-resolved evaluation further showed that model performance reflects the transient and steady-state stages of the glow discharge, while VIP analysis revealed which spectral features contribute most strongly to the calibration and how they evolve during sputtering. Full-spectrum multivariate calibration therefore provides an automated, spectrum-driven and interpretable route for matrix-specific quantitative GD-OES that exploits information already contained in broadband measurements.