Quantification of mineral carbon capture with Raman spectroscopy and machine learning

Carbon mineralization converts CO₂ into stable carbonate minerals through reactions with magnesium- and calcium-rich feedstocks derived from geological and industrial sources. Despite its promise for long-term carbon storage, large-scale deployment remains limited by the lack of rapid, non-destructive, and scalable methods for quantifying carbon uptake under realistic conditions. Conventional analytical techniques, including X-ray diffraction, thermogravimetric analysis, and combustion-based measurements, are accurate but slow, destructive, and unsuitable for in situ or high-throughput monitoring. To address this need, the present work implements Raman spectroscopy to quantify mineral carbon capture with multivariate machine learning (ML) regression. We compare conventional single-laser Raman spectroscopy with dual-laser Shifted-Excitation Raman Difference Spectroscopy (SERDS), which suppresses fluorescence to enable Raman measurements in chemically-complex samples. In samples of carbonated brucite (Mg(OH)₂), we show that the chemical complexity of brucite carbonation requires the use of multivariate machine learning quantification. Our results show that single-laser, conventional Raman spectroscopy can accurately predict CO₂ uptake in low-fluorescence samples, but the performance of conventional Raman spectroscopy degrades significantly under fluorescence interference. SERDS recovered chemically informative spectral features and improved cross-validated prediction performance in the presence of fluorescence. This work establishes a Raman strategy for monitoring mineral carbon capture with applications in both geological and industrial settings.