Using visible and near infrared spectroscopy and machine learning for estimating total petroleum hydrocarbons in contaminated soils.

This article focuses on evaluating visible near-infrared (Vis-NIR) spectroscopy as a rapid and cost-effective method to predict total petroleum hydrocarbons (TPH) in oil-contaminated soils near the Shadegan wetlands in southern Iran. One hundred soil samples were analyzed using Vis-NIR reflectance spectroscopy combined with preprocessing techniques and three modeling approaches: partial least squares (PLS) regression, random forest (RF), and support vector machine (SVM). The study found that reflectance at wavelengths of 1725 nm and 2311 nm correlated inversely with TPH levels, and among the models tested, SVM with baseline correction preprocessing provided the most accurate TPH predictions (r² = 0.85, RMSEP = 1.59%, RPD = 2.6). The findings suggest that Vis-NIR spectroscopy, particularly when paired with machine learning models like SVM, offers a promising alternative to traditional chemical methods for monitoring petroleum pollution in soils.