Remote Sensing and Machine Learning Based on Environmental Conditions for Optimised Cyanobacteria Bloom Management

Cyanobacteria blooms are a threat to human health due to their toxicity and also generate taste and odour compounds. Changes in cyanobacteria concentrations are usually rapid, and high frequency sampling and analysis are therefore required. Most traditional methods are unsuited to such analysis and also be highly onerous. In-situ fluorescence probes are currently employed by a number of water utilities in order to obtain near real time water quality monitoring of waterbodies. Nevertheless, to date fluorescence probes can only provide total cyanobacteria estimates without discretising bloom composition in terms of species. This challenge emerges due to the fact that cyanobacteria species do not have a clear fluorescence spectral difference at the specific wavelengths monitored by these sensors, which ultimately results in the inability of fluorescence probes per se to provide bloom species’ composition information. However, the literature already shows that certain species of cyanobacteria bloom within certain thresholds of environmental factors such as water temperature, pH, and nutrition status and inorganic carbon content for a specific site. Understanding of the bloom species’ composition can be helpful since some species are well known toxin and/or taste and odour producers. In this context, this study aims to add the ability to predict bloom species composition to existing cyanobacteria monitoring fluorescence probes, through the analysis and further machine learning of historical water quality data and other environmental conditions. Preliminary analysis of historical data from an Australian drinking source reservoir were conducted and showed patterns of dominance and appearance of key species according to environmental conditions. Further, directed laboratory work combined with fluorescence monitoring of the reservoir indicates potential to couple data-driven species predictions with real time total cyanobacteria monitoring. The development of such a model with species differentiation in real time could be useful for decision makers to evaluate risk and establish proper responses to different scenarios. Such options may include restriction of recreational water use during high toxicity risk, and remedial measurements implemented by the water utility at reservoir, intake or treatment level in order to remove toxins and taste and odour compounds.