Assessing diesel contamination in soils under bare and vegetated conditions using hyperspectral data, volatile organic compound measurements, and machine learning

Diesel contamination poses a threat to soil condition and overall security. Hurricanes frequently cause chemical pollution in Florida by damaging facilities and releasing hazardous substances, such as diesel, into the environment. This study explores the use of hyperspectral point (350–2500 nm) and imaging (350–1000 nm) data to estimate volatile organic compound (VOC) headspace concentrations, a rapid indicator of recent diesel contamination in Florida soils, under bare and partially vegetated conditions. Soil samples were collected from Duval, Alachua, and Collier counties to represent diverse soil types commonly found in hurricane-affected areas of Florida. Spectral data were analyzed in a laboratory setting to assess how soil composition and diesel concentration influenced reflectance over time, resulting in 360 observations. The most important diesel-specific absorption features were observed between 1000 and 2500 nm, particularly around 2060 nm, which is critical for detecting hydrocarbons, while less distinct patterns observed with the hyperspectral camera (350–1000 nm). Two machine learning models, Cubist and Random Forest, were evaluated. The point spectroradiometer consistently outperformed the hyperspectral camera, and Cubist achieved the highest predictive accuracy for both sensors (spectroradiometer: R² = 0.806; camera: R² = 0.727).