Forensic scientists are frequently called to conduct analyses on pre-existing data collected for purposes not directly related to the forensic investigation such as data designed to support the remedial investigation, risk assessment, or remedial design. This often means that practitioners must use a dataset limited in scope of samples, analytes, and/or geographic coverage. In many such cases, it is not practical or feasible to collect additional high-resolution data for forensic analysis. This is especially true at legacy or complex multi-party sites. While imperfect datasets present roadblocks in performing a forensic analysis for source identification, a holistic weight of evidence approach can complement the forensic analysis and allow the scientist to draw meaningful conclusions.

When a forensic evaluation is done in a vacuum without an understanding of the site history or the conceptual site model, gaps in the analytical record can stall the analysis or lead to erroneous conclusions. In our experience, some analysts may dismiss useful (but imperfect) data without reaching any meaningful conclusions, while others may speculate on potential sources that would be readily rejected based on a review of site history. To avoid these extremes, we recommend that practitioners employ other available lines of evidence to support reaching coherent conclusions in line with a well-reasoned conceptual site model. Techniques that forensics practitioners may overlook include: an understanding of known or likely historical contaminant discharge pathways, marker contaminants for a particular operation or process, data from other site media, or data and findings from comparable sites in the academic literature. In this presentation we discuss common pitfalls related to use of imperfect datasets in forensic analysis without regard for other available evidence. We also offer examples of solutions using a holistic weight of evidence approach for source identification and how these solutions can complement an imperfect analytical dataset.