Limitations

This tool has significant limitations that anyone using it should understand.

• The exact prediction is highly variable: It comes from a machine-learning model with a median error of 3.04 ppb in testing, so an individual fish's actual PFAS can be much higher or lower. The 95% and 99% confidence values are deliberately conservative upper bounds, not measured concentrations.
• Limited input variables: Real PFAS contamination depends on many additional factors not captured here, including local industrial activity, water chemistry, species-specific bioaccumulation rates, and site-specific hydrology.
• United States only: The models were built from U.S. samples and rely on U.S. military installation and U.S. Census urban-area data. Estimates for locations outside the United States are not meaningful.
• Reference data limitations: The military installation and urban area coordinates are static reference sets and may not reflect every PFAS source, nor confirm that contamination actually occurred at any given site. Species averages for relative fish length come from the study dataset and may not represent every population.
• No ground-truth calibration: Outputs have not been cross-checked against state or federal fish consumption advisories.
• Not a substitute for testing: Only laboratory analysis of actual fish tissue samples can determine real PFAS concentrations.
• Not a health authority: This tool does not replace guidance from your state Department of Fish and Wildlife, Department of Health, or the EPA.