Rapid Onsite Identification of Oil Contaminated Soils Using Visible near Infrared Spectroscopy.

In the United States, petroleum extraction, refinement, and transportation present countless opportunities for spillage mishaps. A methodology for rapid field appraisal and mapping of hydrocarbon contaminated soils for environmental cleanup purposes is needed. Visible and near-infrared (VNIR, 400–2500 nm) diffuse reflectance spectroscopy (DRS) is a rapid, non-destructive, proximal-sensing method that has proven adept at quantifying different soil properties in-situ. The objective of this study was to determine the prediction accuracy of VNIR DRS in quantifying the amount of hydrocarbons in contaminated soils. Forty-six soil samples (contained both contaminated and reference samples) were collected from five different parishes in Louisiana. Each soil sample was scanned with VNIR-DRS, with a spectral range of 350 to 2,500 nm, at three different combinations of moisture content and pretreatment: field-moist, intact aggregates; air-dried, intact aggregates; and air-dried ground and sieved (2mm sieve). The VNIR spectra were used to predict total polycyclic hydrocarbon (TPH) of the soil using partial least squares (PLS) regression. The PLS model was validated with 30% of the samples that were randomly selected and not used in the calibration model. Field-moist intact scans proved to be more efficient in predicting TPH content (r²=0.54) than air-dried intact (r²=0.40) and air-dried ground and sieved (r²=0.48) scans. Increased soil processing caused an increase in the RMSD (Root Mean Square Deviation). Since DRS was  found to be an acceptable technique for rapidly measuring soil hydrocarbon content, additional research will be conducted with more sophisticated analytical tools like wavelet analysis, cubist data mining and spatial variability analysis evaluating its effectiveness on a greater diversity of soils and a wider range of soil properties.

Keywords: Spillage, Visible and near-infrared, Total polycyclic hydrocarbon, Partial least squares, Wavelet, Cubist data mining