Intelligent classification of oil-contaminated areas based on geoecological monitoring data

Abstract

Relevance. Oil pollution of the soil cover has a complex spatial pattern and may persist for a long time in industrial and adjacent areas. Its assessment should not be limited only to petroleum hydrocarbon content, since site conditions also depend on terrain, moisture, granulometric composition, vegetation, surface runoff, and position relative to pollution sources. Aim of the study. The study aims to develop an approach for intelligent classification of oil-contaminated areas based on integrated processing of geoecological monitoring data, laboratory soil analysis, remote sensing, and GIS modeling. Research methods. Field survey results, laboratory measurements of petroleum hydrocarbons in soil samples, remote sensing materials, a digital surface model, and GIS indicators of terrain, moisture, vegetation cover, distance to pollution sources, and proximity to runoff zones were used. The k-nearest neighbors, Support Vector Machine, Gradient Boosting, and Random Forest algorithms were applied for classification. Model performance was evaluated by accuracy, F1-score, stability of class separation, and factor interpretability. Results. The most significant indicators were petroleum hydrocarbon content, soil moisture, NDVI, clay fraction, distance to the source, surface slope, and proximity to runoff elements. Random Forest provided the best results, with an accuracy of 0.84–0.87 and F1-score of 0.78–0.90. Background, slightly contaminated, moderately contaminated, heavily contaminated, and extremely contaminated areas were identified. The last two classes occupy about 8–12% and 3–5% of the territory but form the main environmental risk.