Application of geographic data for spatial modeling of lead in contaminated fluvial soils

Abstract

The present study aims to determine the spatial distribution of soils with lead (Pb) content above the quality thresholds in a section of the Ogosta River valley (NW Bulgaria). The study area was contaminated with mine waste from the extraction and flotation of iron, lead-silver, and gold-bearing ores in the second half of the XX century. Predictive modeling was performed with the software Maximum Entropy Species Distribution Modeling (MaxEnt), Version 3.4.4, which uses machine learning algorithms and applies the maximum entropy method. The choice of predictors of contaminated soil distribution is consistent with the main factor for Pb dispersal within the valley floor - flooding from the Ogosta River. The following six parameters explained the environmental settings related to the accumulation of contaminated floodplain sediment: vertical distance to the river channel,  lateral distance to the Ogosta River, terrain slope, land cover (CORINE Land Cover, 2019), morphographic units of topography, and elevation. The results represent the average values of 10 replicates of the model. We evaluated the individual models by the value of the area under the relative operating characteristic curve (AUC) and the geographic logic of the obtained results. The AUC score for the test samples was 0.666 for the soil group 1 with Pb ≤120 mg/kg, 0.782 for group 2 with Pb (120-500] mg/kg, and 0.934 for group 3 with Pb>500 mg/kg. The most significant predictors for the models are the vertical and lateral distance to the river and the slope of the terrain. Lead concentrations tend to decrease with the distance from the main river and by increasing the elevation above the river channel due to lower inundation frequency and deposition rate of polluted river sediments. The soils with a Pb concentration below the permissible threshold of 120 mg/kg cover more than 58.42% of the valley floor of the studied section, and lands with Pb content above the intervention value of 500 mg/kg occupy nearly 10.82% of the investigated territory. The selected predictors describe the distribution of highly contaminated soils well and define the range of soils with lower Pb content worse. Combining clean and contaminated soil samples into one group is considered the main reason for the poor performance of MaxEnt for soils with Pb ≤120 mg/kg. However, the results prove the model's ability to predict the spatial distribution of not only biological species but also the dispersal of hazardous substances in soil.