Quantifying Soil Total Inorganic Carbon Using Shifted-Excitation Raman Difference Spectroscopy (SERDS) and Machine Learning

Abstract

Accurate quantification of soil inorganic carbon, which is stored predominantly in carbonate minerals, is essential for understanding the soil chemistry, agricultural productivity, and carbon sequestration processes. Traditional analytical techniques, such as coulometric titration, Fourier transform infrared (FTIR) spectroscopy, thermogravimetric analysis (TGA), and X-ray diffraction (XRD), while precise, are often labor-intensive, costly, and destructive. To address these limitations, we present a nondestructive, rapid, and reliable approach combining shifted-excitation Raman difference spectroscopy (SERDS) with advanced machine learning techniques for predicting soil carbonate content. SERDS effectively mitigates fluorescence interference that frequently obscures conventional Raman spectra, providing clearer and more resolved signals from carbonate minerals. Using a data set of 212 soil samples collected from Saskatchewan, Canada, we compared the performance of SERDS-based spectroscopy against conventional Raman spectroscopy, employing partial least-squares regression (PLSR) and extreme gradient boosting (XGBoost) models. The SERDS-XGBoost model achieved the best test-set performance, with an R2 of 0.93 and an RMSE of 0.16 wt % C, compared with an R2 of 0.63 and an RMSE of 0.40 wt % C for the corresponding conventional Raman-XGBoost model. This represents a 48% relative increase in R2 and a 60% reduction in the RMSE. While the method effectively quantified total inorganic carbon, overlapping Raman peaks among carbonate phases limited the precise mineral identification, requiring complementary techniques for qualitative assessment. The combined use of SERDS for quantification and XRD for qualitative mineralogical assessment provides a robust and scalable approach to evaluate soil health and improve environmental management.