No abstract

The study objective was to identify factors most closely associated with the presence of cyanotoxins in 9 lakes in the Puget Sound lowlands region of western Washington, USA. Four cyanotoxins (microcystins, anatoxin-a, saxitoxin, and cylindrospermopsin), phytoplankton, and limnological parameters were monitored twice per month from June through October 2012. Microcystin (MC) was the most commonly detected cyanotoxin and was detected in every lake at least once. Nonparametric decision forests and classification trees were used to identify variables that best predicted MC categories for 2 models; (1) presence-absence of MC, and (2) MC concentrations. The best predictors of MC in concentration categories for both models were epilimnetic total nitrogen to total phosphorus (TN:TP) ratios and the abundance of potential MC-producing cyanobacteria. Model 1 showed that observations with TN:TP ratios ≤25.7 were associated with MC presence, while MC was generally absent when TN:TP ratios were >25.7 and MC-producing cyanobacteria were ≤330 cells mL −1. Model 2 showed that Microcystis abundance >1300 cells mL −1 captured moderate (>1 and ≤6 µg L −1) and high (>6 µg L −1) MC concentrations. Low MC concentrations (>0.05 and ≤1 µg L −1) were found when TN:TP was ≤28.8 or when Dolichospermum abundance was >110 cells mL −1. Because of their broad applicability, thresholds for these variables may be useful in evaluating public health risk in the absence of MC measurements from lakes in this and similar regions. Decision forest and classification tree models may be promising tools for lake mangers to identify dominant factors and threshold limnological values associated with cyanotoxins.

No abstract