Determination of petroleum hydrocarbon contamination in soil using VNIR DRS and PLSR modeling

Abstract

Visible near infrared diffuse reflectance spectroscopy (VNIR DRS) is being proposed as a rapid and cheaper alternative to conventional soil analysis. This approach to soil analysis will be especially useful when conducting an environmental risk management for petroleum contamination in soil. This study evaluated the potential of VNIR diffuse reflectance spectra for rapid non-destructive quantitative analysis of extractible total petroleum hydrocarbon (ETPH) in soils. It also assessed the effect soil organic carbon (SOC) has on the performance of partial least square regression (PLSR) models developed for characterizing ETPH in soils. Model performance was evaluated based on the coefficient of determination (R2), ratio of performance to deviation (RPD) and the root means square error (RMSE). Result show that VNIR DRS can be a potentially viable analytical tool for petroleum hydrocarbon contamination in soils. However, model quality was found to be affected by spatial variations within soil samples. Models developed from contaminated soils from highly variable geological origins had fair but promising model statistics (R2 = 0.72, RPD = 1.4) as against excellent predictions obtained from contaminated soils with similar geology (R2 = 0.97, RPD = 4.5) implying that the VNIR DR approach to characterizing petroleum hydrocarbon contamination in soils will be better suited to development of local prediction models. PLSR models developed for soil groups with SOC range (0.94–26.5% OC) gave quite robust prediction (R2 = 0.90–0.97, RPD = 2.7–4.5), though a high SOC content slightly lowered PLSR model statistics. These results suggest that VNIR DRS can be quite useful for rapid characterization of petroleum hydrocarbon contamination especially when low budgets and reduced timelines are desirable for remediation purposes.