Determination of organic carbon concentrations in marine sediments across different dates by visible and near-infrared spectroscopy

The application of spectral models across different monitoring dates remains a challenge in fixed-point monitoring. Here, we studied the temporal dynamics of the organic carbon concentrations in sediments from the Yellow Sea and Bohai Sea with visible and near-infrared spectroscopy using the established spectral models. In this study, 96 surface sediment samples were collected from the same sites in 2020 and 2023. These samples were prepared as dried powder samples, and their total organic carbon (TOC) concentrations were determined. Their visible and near-infrared spectra were also recorded. Firstly, TOC spectral models were established based on partial least squares regression (PLSR) and least squares support vector machine (LSSVM) using the samples collected in 2020. The results showed that these models had strong analytical capabilities. The coefficients of determination ( R 2 ) of the spectral models established by PLSR were 0.982 for the calibration set and 0.777 for the validation set. The R 2 of the spectral models established by LSSVM were 0.998 for the calibration set and 0.839 for the validation set. Subsequently, the two established models were used to analyze the TOC concentrations in the samples collected in 2023. The detection errors of the TOC in the samples collected in 2023 were large with direct use of the established models. After correction using the piecewise direct standardization (PDS) and intercept/slope ( S / B ) algorithms, the errors were significantly reduced. Specifically, after PDS and S / B correction, the R 2 increased from 0.292 to 0.710 and 0.732, while the root mean square error (RMSE) decreased from 0.302 to 0.193 and 0.186, respectively, as detected by the PLSR model. On the other hand, the R 2 increased from 0.409 to 0.745 and 0.760 and the RMSE decreased from 0.276 to 0.181 and 0.176, respectively, as detected by the LSSVM model. This study investigated the application of spectral analysis in fixed-point monitoring, providing an effective reference for high-frequency and even continuous time-series analysis in fixed-point monitoring.