Scientists developed a statistical approach (model) to use both readily available measurements, such as wind speed, and laboratory measurements, such as cyanobacteria toxin gene counts, to predict levels of a single toxin and—for the first time—a mixture of toxins. Models are based on data from water samples collected from Kabetogama Lake during May–September over 2 years.

Predictions from models for a single toxin and for toxin mixture matched measured toxin values equally well, and the toxin-mixture models produced no false negatives, i.e., they did not fail to predict an exceedance of the drinking-water guideline when such an exceedance actually occurred.

Although models using readily measured variables did not explain as much variability in the measured cyanotoxin data as models that also incorporated laboratory measurements, models that use data that can be measured in real-time may be more useful as early-warning indicators. Such models could help health-agency personnel alert visitors as to when boating, fishing, and swimming could result in skin contact, ingestion, or inhalation of tiny water droplets containing cyanotoxins.

Although the models were developed for Kabetogama Lake, the approach used could be applicable to other lakes or beaches where harmful algal blooms occur. The research was supported by the U.S. Geological Survey–National Park Service Partnership Program.