A large-scale analysis of U.S. water quality data reveals that most toxic chemicals remain poorly characterized or undetected in routine monitoring. This is largely due to sparse risk assessment data, as well as detection limits that are too high to capture ecologically relevant concentrations, researchers report. The findings suggest that the true scale of chemical risk to biodiversity and ecosystems may be significantly underestimated. Chemical pollution is widely recognized as a major threat to biodiversity, human health, and the stability of ecosystems worldwide. However, the accelerating rate at which new chemicals are introduced into the environment outpaces the current ability to fully assess their ecological risks. Large-scale risk assessments depend on both knowing where chemicals are present and understanding how harmful they are to living organisms, yet for most substances, such data are lacking. While new computational and lab-based approaches can estimate toxicity, they are still constrained by limitations in environmental monitoring, especially when extremely toxic substances are present in concentrations too low to be reliably detected.

To better understand how gaps in monitoring data affect risk assessments, Sascha Bub and colleagues analyzed 112 million chemical monitoring records for nearly 2,000 substances in U.S. surface waters spanning 62 years, alongside 78 million records of environmental conditions. Bub et al. compared these data with established toxicity thresholds for over 170,000 chemicals, derived from laboratory and computation studies, that indicate concentrations likely to cause ecological harm. According to the findings, large-scale assessments of chemical risks in U.S. surface waters are primarily constrained by the lack of monitoring data. While regulatory toxicity thresholds are available for over 170,000 chemicals and span a wide range of potencies, only a small fraction of these substances – less than 1% – have corresponding environmental monitoring records. What’s more, routine water monitoring programs are often unable to detect many chemicals, including highly toxic and widespread agricultural pesticides, because the detection thresholds are set too high relative to the concentrations known to be ecologically damaging. These shortcomings suggest that a large portion of chemical risks may remain hidden, especially for substances that are highly potent at low doses.

To find the best catalyst for green ammonia, researchers were staring down 8000 lab experiments. With AI, they only needed 28.