Detecting and Quantifying Diesel and Gasoline Contaminants in Sandy Soil by Multivariate Analysis Based on Nuclear Magnetic Resonance T2 Spectrum

Nuclear magnetic resonance (NMR) offers the distinct advantage of swiftly detecting petroleum hydrocarbon contaminants in situ. However, when confronted with petroleum hydrocarbon contaminants within the pores of sandy soil, the signals of water and contaminants within the sandy soil overlap in the NMR
${T}_{2}$
spectrum, making direct distinction unfeasible. In this article, multivariate analysis based on the
${T}_{2}$
spectrum with a small sample size is proposed to detect and quantify diesel and gasoline contaminants in sandy soil with certain soil conditions. The classification models are established according to the
${T}_{2}$
spectral features extracted by principal component analysis (PCA) for the detection of contaminants. The regression models are established based on the interval segmentation of the
${T}_{2}$
spectrum for the quantification of contaminants. Laboratory-prepared samples include sandy soil (WA), sandy soil contaminated with diesel (DI), and sandy soil contaminated with gasoline (GA). The
${T}_{2}$
spectrum data were acquired through measurements and inversion using a low-field NMR instrument. Models such as PCA-ANN, PCA-SVM, and PCA-KNN were established to classify the samples, while partial least square (PLS), interval PLS (iPLS), and synergy iPLS (siPLS) models were developed to quantify fluid content in each sample. The PCA-ANN has the highest accuracy on the test set, reaching 100%. The RMSEP values for the full-PLS are 0.0681, 0.0265, and 0.0606 mL for WA, DI, and GA, respectively. The siPLS has better predictive performance than full-PLS. For WA, the optimal NMR spectral region spans from 31.6 to 316 ms, whereas for DI and GA, the range from 100 to 1000 ms proves optimal.