Quantitative evaluation of hydrocarbon contamination in soil using hyperspectral data-a comparative study of machine...

Area of Science:

• Environmental Science
• Geochemistry
• Data Science

Background:

• Soil contamination by hydrocarbons poses significant environmental risks.
• Rapid and accurate assessment of hydrocarbon levels is crucial for effective remediation.
• Traditional methods for soil analysis can be time-consuming and labor-intensive.

Purpose of the Study:

• To evaluate machine learning (ML) and deep learning (DL) techniques for predicting hydrocarbon contamination in soils.
• To assess the effectiveness of hyperspectral imaging in conjunction with ML/DL for soil analysis.
• To compare the performance of different models, including XGB regressor and neural networks, for hydrocarbon prediction.

Main Methods:

• Synthetically contaminated soil samples (clayey, silty, sandy) with crude oil, diesel, and gasoline (0-10,000 mg/kg).
• Hyperspectral imaging to capture spectral signatures.
• Analysis using XGB regressor and neural network models, with Gas Chromatography-Mass Spectrometry (GC-MS) for reference values.
• Model performance evaluation using R-squared and Root Mean Squared Error (RMSE).

Main Results:

• Models achieved high predictive accuracy, with an R-squared of 0.96 and RMSE of 600 mg/kg.
• Performance varied by petroleum type and soil matrix; gasoline showed lower accuracy.
• Feature selection (spectral bands) improved model performance by reducing overfitting.
• XGB regressor demonstrated a strong balance of accuracy and robustness.

Conclusions:

• Hyperspectral analysis combined with ML/DL models is effective for soil contamination assessment.
• Ensemble models like XGB are suitable for practical spectral applications in environmental monitoring.
• This approach offers a rapid and reliable method for detecting hydrocarbon pollution in soils.