You are here:
Forecasting Cyanobacteria Harmful Algal Blooms Utilizing Spatio-Temporal Integrated Nested Laplace Approximation within the Contiguous United States.
Citation:
Salls, W., B. Schaeffer, N. Von Tress, AND H. Ferriby. Forecasting Cyanobacteria Harmful Algal Blooms Utilizing Spatio-Temporal Integrated Nested Laplace Approximation within the Contiguous United States. National Monitoring Conference, Virginia Beach, VA, April 24 - 27, 2023.
Impact/Purpose:
Cyanobacterial harmful algal blooms (cyanoHABs) have remained a major water quality concern, impacting human, animal, and ecological health as well as recreational and aesthetic value of lakes and the areas surrounding them. Tools for effectively monitoring cyanoHABs are in place, but a major priority for federal, state, and local lake management stakeholders remains knowing when and where cyanoHABs. This effort presents a national forecasting model with high accuracy, advancing our ability to prepare for and respond to cyanoHAB occurrences in lakes across the US.
Description:
Cyanobacterial harmful algal blooms (cyanoHABs) are detrimental not only to human and ecosystem health, but also to local economies that rely on surface water for consumption, tourism, and recreation. Previous work has been done to predict cyanoHABs in Florida at the lake and state levels as an initial case study. Because cyanoHABs are a threat to health throughout the contiguous United States (CONUS), the need for a large-scale prediction model is apparent. To address this need, we used a Bayesian model with spatio-temporal integration to estimate the probability of a bloom occurring in a given week. Specifically, we used an Integrated Nested Laplace Approximation (INLA) model with the R-INLA package because it provided a complex and computationally efficient forecasting model that can work with missing data and irregular lake sampling. The INLA model used cyanobacteria presence data derived from Sentinel-3 Ocean Land Color Instrument (OLCI) along with a suite of environmental predictor variables from 2016 to 2022 in satellite resolvable lakes across CONUS. The prediction results were compared to independent previous cyanobacteria presence satellite imagery to provide the accuracy, precision, Brier score, and Cohen’s kappa. Utilizing a hierarchical Bayesian spatio-temporal modeling method like R-INLA provides water managers, public health officials, and local governments a useful prediction for near future cyanoHAB bloom probabilities.
