Retrieving and modelling dielectric permittivity of diesel contamination in Terap Red and sandy soil using ground penetrating radar

Abstract

Ground Penetrating Radar (GPR) is a non-destructive, full-wave electromagnetic (EM) measurement tool for quantitative imaging to describe dielectric permeability distributions. It is an efficient technique for detecting diesel contamination in soil tomography problems. However, dielectric permittivity relies entirely on variance moisture content facilitated by diesel fuel reaction soil, which determines GPR velocity. Difficulties in interpreting GPR reflection configuration are complex qualitative features limited to noisy or nonlinear relations problems. Consequently, positioning and depth determination would be misleading due to severe polarization and velocity mismatch in traveling-wave typically in Terap Red soil as silty-clay soil. Therefore, this study aims to determine the mathematical model for dielectric permittivity prediction and investigate the GPR signal segmentation algorithm model to map the diesel contamination plume in Terap Red soil. The calibration icon function of the GPR signal was quantified by dielectric permittivity prediction. The research approach was divided into 4 phases. The investigation commenced with an evaluation of the GPR propagation signal from a simulated diesel contamination plume of Terap Red and sandy soils concerning the results of geotechnical measurements using BS 1337: 1992. Next, the dielectric permittivity using the GPR velocity in modeling the empirical relationship between dielectric permittivity and moisture content was determined using statistical analysis. Additionally, cross-validation was performed using existing literature, Vector Network Analyzer (VNA) and in-situ measurements before the GPR signal images were segmented and categorized using a Support Vector Machine (SVM). Finally, ten-fold cross-validation and Logistic Regression (LR) classification were used to evaluate the spatial distribution classification mapping of GPR signals. The result shows the best prediction on Terap Red soil from third-order polynomial using ANOVA yielded a strong positive correlation (R2=0.9892, N=24, P <0.05) and a standard error of 0.076. The accuracy of dielectric permittivity in terms of root mean square error (RMSE) and mean absolute error (MAE) was obtained at 9.772E-14 and 0.049, respectively. The best-fitting relationship does exhibit some degree of textural bias that should be considered in the choice of petrophysical relationship with uncertainty mean differences via VNA validation for Terap Red and sandy soil were only 2.706 % and 1.985 %, compared to over 3.608% and 15.990 % for the existing model. The accuracy of the spatial distribution classification map generated by the SVM classifier is encouraging, with RMSE of 0.139, kappa statistics of 0.888, and correct instances classified (CIC) of nearly 100 % for both SVM and LR. In conclusion, the study results on dielectric permittivity prediction of contaminated soils for Terap Red and sandy soils indicate that the empirical relationship model is only applicable to specific soils with similar properties. Additional supervised data is recommended to achieve better classification outputs.