GeoAI-based 3D spatial distribution modeling of PAHs in industrial contaminated soils.

Soil pollution threatens human health and food security, particularly in industrial legacy sites. Accurate three-dimensional distribution modeling of soil contamination is crucial for understanding pollutant migration and guiding targeted remediation. Yet, the strong heterogeneity of contaminants limits the performance of traditional methods. We proposed a GeoAI-based approach, the three-dimensional deep kriging neural network (3D-DKNN), which combines deep learning with geostatistical principles to enhance interpolation accuracy in heterogeneous environments. Applied to polycyclic aromatic hydrocarbons at a typical industrial site, 3D-DKNN was benchmarked against traditional three-dimensional ordinary kriging (3D-OK) and inverse distance weighting (IDW). Cross-validation shows that 3D-DKNN achieved the lowest RMSE and MAE, and the highest correlation coefficient. Relative to IDW, it reduced RMSE by 36 %-80 %, MAE by 40 %-58 %, and increased correlation by over 19 %. Based on risk thresholds, contamination hotspots were identified in the northern and northwestern areas, particularly manufacturing and warehouse areas, which were recognized as key risk and remediation areas. This study demonstrates the potential of GeoAI for modeling complex pollutants and improving soil risk assessment.