A contaminant-adaptive evidence chain for source attribution in multi-operator contaminated land

Source attribution is difficult when successive on-site operations overlap and candidate sources occur beyond parcel boundaries. We developed a contaminant-adaptive evidence chain and applied it to 307 soil records from 72 locations at an electroplating-to-recycling site. Risk screening under future sensitive land use prioritized Cr(VI), total Cr, Ni, Zn, Pb and total petroleum hydrocarbons (TPH), while Cu provided a strongly reconstructed processing fingerprint. Process and material-use histories, correlation, principal component analysis (PCA), and positive matrix factorization (PMF) resolved Cu, Zn and total Cr as electroplating-associated signals; Cr(VI) was co-interpreted with total Cr as a redox-specific fraction. Pb and TPH remained multi-source candidates spanning the internal electroplating-recycling overlap and adjacent enterprises. For these residuals, internal maxima and distance-decay, runoff, groundwater-gradient and pipeline-connectivity evidence reduced the plausibility of evaluated off-site inputs. Within the parcel, performance-gated machine learning with Shapley additive explanations (SHAP) bounded diffuse TPH to the internal overlap, whereas unstable Pb prediction routed Pb to HYDRUS-1D transport testing. Near-surface Pb was compatible with at least one tested surface-input matrix-transport history, whereas deep Pb and the SA20 TPH maximum were not reproduced by the tested one-dimensional histories and were localized for verification. By turning routine investigation data into a bounded attribution hierarchy, the framework resolves coherent source domains, separates competing evidence tracks and identifies where targeted confirmation and forensic evidence collection are required.