A Bayesian spatiotemporal model evaluation of forecasting cyanobacterial harmful algal bloom events.
Citation:
Meyers, K., B. Schaeffer, O. Cronin-Golomb, AND W. Salls. A Bayesian spatiotemporal model evaluation of forecasting cyanobacterial harmful algal bloom events. ASLO 2024, Madison, WI, June 02 - 07, 2024.
Impact/Purpose:
The Bayesian spatiotemporal model over-predicts positive events, the potential management expense of which may be time and costs related to additional confirmation practices, such as field sampling or additional remote sensing. The purpose of this work is to evaluate model deficiencies. Once improved, this forecasting model can complement existing monitoring efforts and become part of a more comprehensive management toolbox. Federal, state, and municipal stakeholders might include a 7-day forecast that informs managers to watch a subset of lakes with high forecasted probabilities of a bloom event; this could be done by using daily satellite data and then complementary field observations for toxin analysis if imagery confirms the event is occurring.
Description:
The U.S. Harmful Algal Bloom and Hypoxia Research Control Act calls for robust approaches to forecasting cyanobacterial harmful algal blooms (cyanoHABs). Accurate forecasting technology could save local communities healthcare costs through the early detection of cyanoHABs and therefore faster advisory warnings. However, most existing forecasting models require time-consuming parametrization and/or are limited to well-sampled individual lake systems. An Integrated Nested Laplace Approximation hierarchical Bayesian spatiotemporal model forecasted weekly lake exceedance of 12 μg/L chlorophyll-a, the World Health Organization’s recreation Alert Level 1 threshold, for 2192 satellite resolved lakes. Model deficiencies were evaluated to improve the functionality of the forecast. We investigate if temporally short events may commonly occur as false negatives, and if reoccurring annual events are prematurely forecasted, resulting in a false positive. We also consider the impacts of lake type and geographic location on these predictions. This evaluation identifies key targets for model improvement with the goal of making the forecasts operational moving forward.
