Short-term probabilistic microcystin prediction using Bayesian model averaging.

Area of Science:

• Environmental Science
• Ecology
• Limnology

Background:

• Microcystin, a toxin produced by cyanobacteria, poses risks to aquatic ecosystems and human health.
• Western Lake Erie experiences recurrent harmful algal blooms (HABs) dominated by Microcystis spp.
• Accurate prediction of microcystin concentrations is crucial for water resource management.

Purpose of the Study:

• To develop and validate a dynamic modeling approach for predicting high microcystin concentrations in Western Lake Erie.
• To utilize a Bayesian hierarchical modeling framework for forecasting toxin risks.
• To incorporate seasonal variations and iterative updating for short-term predictions.

Main Methods:

• An empirical model was developed, assuming microcystin concentration is proportional to Microcystis spp. biomass.
• Bayesian hierarchical modeling allowed for annual and seasonal variation in the proportionality constant.
• An iterative updating algorithm facilitated sequential model updates and short-term forecasting.
• An ensemble of four seasonal variation models was employed, with predictions weighted by accuracy.

Main Results:

• The dynamic model provides a framework for predicting microcystin risk.
• The Bayesian approach allows for adaptive forecasting as new data become available.
• Ensemble averaging of seasonal models enhances prediction reliability.

Conclusions:

• The developed dynamic modeling approach offers a robust method for forecasting microcystin concentrations in Lake Erie.
• This predictive tool can aid in managing the risks associated with harmful algal blooms.
• The iterative and ensemble-based methodology improves the accuracy and timeliness of short-term forecasts.