Background/Question/Methods: Water bodies are chemical data loggers that contain tens of thousands of molecules that are derived throughout a watershed. While many compounds are widespread and originate from multiple sources, we hypothesize that the distribution of chemicals across the landscape is not random. Instead, different sources are expected to contain highly unique suites of chemicals or chemical ratios. Our aim is to extract non-polar organic compounds from water samples at/near pollution sources, and then use high-resolution mass spectrometry (HRMS) data in conjunction with machine learning tools to identify the chemical signatures that are diagnostic of each source. If these diagnostic signatures, or fingerprints, are detected in environmental samples, we can confirm the presence of source-specific discharge in receiving bodies of water. This is particularly advantageous for pollution source tracking (e.g., nitrate) where compounds are generated from multiple sources. Grab samples were collected and analyzed in triplicate from different sources including headwater streams, agricultural field runoff, animal manure, municipal wastewater, and urban/suburban road runoff.
Results/Conclusions: Data from artificial intelligence models indicate that each sample can be correctly identified with high accuracy based on its chemical composition and that ~100 non-target chemical compounds can be isolated as diagnostic chemical fingerprints. Our next step is to test existing non-target samples from previously collected surface water samples for each of the fingerprints. This workflow will be open source and freely available to help managers identify pollutant sources present in receiving water bodies, which will help direct limited resources to projects that maximize water quality improvements.