The frequency of cyanobacteria blooms in North American lakes is increasing. A major concern with rising cyanobacteria blooms is microcystin, a common cyanobacterial hepatotoxin. To explore the conditions that promote microcystin occurrence, we analyzed the US EPA National Lake Assessment (NLA) dataset collected in the summer of 2007. We applied a multilevel Bayesian approach to estimate microcystin presence/absence within ecological regions. We used the results of random forest modeling as a means of variable selection. Model parameters under the multilevel modeling framework were eco-region specific, but they were also assumed to be exchangeable across eco-regions for broad continental scaling. The exchangeability assumption ensured that both the common patterns and eco-region specific features would be reflected in the model. Furthermore, the method incorporated appropriate estimates of uncertainty. The response variable, presence/absence of microcystin, is binary; hence, a logistic regression was developed. Trophic state and nitrogen to phosphorus ratio were used as predictors. The microcystin and nitrogen to phosphorus ratio relation was approximated by a threshold model. Trophic state was modeled using proportional odds logistic regression, appropriate for ordered categorical variables, with nitrogen, phosphorus, elevation, and secchi disk depth as predictors.
Despite prevalence of microcystin in US lakes, uncertainty still exists in our understanding of factors associated with its occurrence. Our proposed method, through probability calculus, provides an explicit expression of the amount of uncertainty in our knowledge. Our preliminary results show associations between microcystin occurrence, trophic state, and nitrogen to phosphorus ratio. The risk of microcystin occurrence increases as nitrogen to phosphorus ratio increases and decreases once nitrogen to phosphorus is beyond the threshold (~21). The risk of microcystin occurrence is greatest when the nitrogen to phosphorus ratio is approximately 21. The NLA 2012 will be used for Bayesian updating. The results will help develop management strategies to alleviate microcystin impacts and improve lake quality.