Forecasting microcystin concentrations in Lake Erie using an Eulerian tracer model

Abstract

Cyanobacteria biomass models are routinely used in Lake Erie to predict the occurrence and location of algal blooms. However, current forecasts do not predict the microcystin toxins produced by these blooms. In this study, we used an extensive dataset of microcystin concentrations to generate weekly distribution maps in Lake Erie for the summers of 2018 and 2019. Using a 3D Eulerian tracer model (ETM) initialized with these maps, we simulated microcystin transport over 7 days, under two conditions: (1) the initial microcystin is mixed within the surface-mixed layer; (2) the initial microcystin is distributed throughout the entire water column. Two scenarios were tested for each condition: one incorporating microcystin production rates into hydrodynamic transport and one excluding them. Model performance was evaluated against weekly sample data in predicting whether microcystin concentrations surpassed specific thresholds (0.3, 1.0, 5.0, 10.0, and 20.0 µg/L), and in predicting trend directionality over each week. Overall, the ETM with hydrodynamics alone captured the transport of microcystins and predicted microcystin concentrations in 69% of the simulations. Incorporating microcystin production into the model increased the accuracy of forecasts by an additional 10%. Moreover, models with microcystin production successfully predicted microcystin concentrations greater than 5 μg/L during a large bloom, high-microcystin year (2019), while incorrectly forecasting concentrations above 5 μg/L during a small bloom year (2018). With limited data to initialize the ETM, no single model configuration consistently outperformed others. It is necessary to consider the full range of model configurations when utilizing their outputs for making management decisions.