Characterizing risk for harmful algae blooms on the Ohio River using time series data
Citation:
Nietch, C. Characterizing risk for harmful algae blooms on the Ohio River using time series data. 2020 American Fisheries Society Annual Meeting, Virtual, September 14 - 25, 2020.
Impact/Purpose:
Research conducted as part of a RARE grant with Region 3 and 5 and partners in ORSANCO will be highlighted. The research serves to characterize and communicate the risk of harmful algae blooms in the Ohio River using time series water data.
Description:
Two large harmful cyanobacteria blooms (CyanoHABs) have occurred on the Ohio River: One in 2015, affecting most of the river’s length, the other in 2019 covering 300 miles from Ashland to Louisville, KY. The blooms have threatened drinking water for millions of people, prompted state recreation advisories, and caused the cancellation of significant recreation events in Cincinnati, OH, and Louisville, KY. Two statistical methods were used for explaining and predicting CyanoHAB incidence on the Ohio River based on water data time series. The first method utilized conceptual knowledge of river hydrodynamics as they relate to algal ecology to identify predictors of bloom occurrence with a Bayesian regression model, and the second is classification of functional flow data into bloom or non-bloom membership. The functional classification method utilizes a proximity measure between year-to-date functions of daily average flow exceedances and a kernel estimator of posterior probabilities. Then, nonparametric techniques were used to compare the two functional objects. The results of the method applications are presented within a web-based risk characterization tool to help water quality professionals in their effort to monitor, sample, and communicate risk. The web app provides results in real time, with risk probability plots that can be visualized for 20 sites covering the entire river’s length. Current flow conditions are quantified relative to those that produced blooms in the past. The Bayesian regression model proved useful for signaling similarities leading up to a bloom while the application of the classification of functional flow data potentially helps in managing the sampling and risk communication effort during a bloom event. The later may prove useful to analyzing several common water quantity and quality variables as it can be generalized to predict any dichotomous outcome from time series data.
