Delineation of soil contaminant plumes at a co-contaminated site using BP neural networks and geostatistics

Abstract

The delineation of contamination at a co-contaminated site is vital for designing remedial strategy and estimating costs. Backpropagation (BP) neural networks and the Nemerow pollution index (NPI) with 3-D kriging were combined in this study to delineate contaminant plumes, analyze the spatial distribution of pollutants in different layers and visualize them in three dimensions, quantify polluted areas and pollution levels, and identify hotspots of the contaminants of concern (COCs). The results of a comprehensive assessment performed using BP networks and NPI were compared. The analysis of the volumes of soil contaminated with specific COCs in different soil layers showed that arsenic (As) hardly migrated downgradient whereas benzo[a]pyrene (BaP) had a strong tendency to migrate. The tendency of fluorene (FLE), naphthalene (NAP), and phenanthrene (PHE) to migrate was between that for As and BaP. The volumes of earth contaminated with of all five of these COCs generally decreased with increasing pollution levels. The volumes of PHE, FLE, and NAP at different pollution levels exhibited similar trends, and most contaminated areas were safe. The volume of BaP at high pollution levels was markedly greater than that for the other four COCs, and the volume of As at low pollution levels was also notably greater than that for the other four pollutants. (3) The spatial patterns performed by the NPI and BP network comprehensive assessment methods were similar. However, BP networks can overcome the deficiencies of NPI, which are amplification of the effect of heavily polluted elements and narrowing the separability of polluted and unpolluted areas. After comparing performance and performing a cost-benefit analysis, we propose a model that integrates BP networks and geostatistics to delineate soil contaminant plumes at co-contaminated sites.