Probabilistic forecasting of Cyanobacterial concentration in riverine systems using environmental drivers

Abstract

Toxic cyanobacteria blooms such as Anabaena, Aphanizomenon, Microcystis and Oscillatoria are of critical concern for public health and environmental system globally. An algal bloom is largely influenced by factors that jointly characterize the climatology (e.g., water temperature), hydraulics (e.g., water velocity) and nutrient concentrations (e.g., phosphorus and nitrogen). While a wide range of efforts has been made to predict a cyanobacterial bloom, there is still a need for computational tools to characterize the bloom concentration effectively. Here, we present a short-term cyanobacteria forecasting model that not only predicts the occurrences of algal bloom but also provides their concentration conditional on the selected dominant environmental variables. The prediction model operates in two stages. In the first stage, cyanobacterial occurrences are predicted using a first-order Markov model conditioned on a few selected environmental variables. On occasions where a cyanobacterial occurrence is predicted, the second stage predicts cyanobacterial cell counts again conditional on the selected environmental variables. In an application using data for four major rivers in South Korea, a minimum Threat Score of 0.56 (56% forecasting accuracy) with a single environmental variable, temperature, is attained. This simple model provides one week ahead probabilistic prediction of cyanobacteria occurrence and cell concentration making it easier to prioritize proactive measures based on the probability changes caused by relevant changes in the conditioning environmental variables.