A review of machine learning models for algal bloom monitoring in freshwater systems

Authors: BUSARI, IBRAHIM - Clemson University; SAHOO, DEBABRATA - Clemson University; Harmel, Daren; HAGGARD, BRIAN - University Of Arkansas

Submitted to: Journal of Natural Resources and Agricultural Ecosystems
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/9/2023
Publication Date: 3/1/2024
Citation: Busari, I., Sahoo, D., Harmel, R.D., Haggard, B.E. 2024. A review of machine learning models for algal bloom monitoring in freshwater systems. Journal of Natural Resources and Agricultural Ecosystems. 1(2):63-76. https://doi.org/10.13031/jnrae.15647.
DOI: https://doi.org/10.13031/jnrae.15647

Interpretive Summary: Harmful algal blooms (HABs) are detrimental to the lives of livestock, humans, pets, wildlife, and the global economy, which calls for a robust approach to their management. Decision support tools such as process-based models facilitate the understanding of HAB-enabling conditions and the analysis of the impact of watershed activities on HAB occurrence. The rapid breakthrough in computing prowess and statistical advances is triggering the development of data-driven techniques such as machine learning (ML) models. The application of ML models for HAB monitoring is favored by the improved sensor technologies that provide continuous data. This review describes ML models currently used for predicting and forecasting HABs in freshwater ecosystems and provides an overview of the model structures and their application for predicting algal blooms and related toxins. The use of hybrid ML models for improved prediction accuracy was also shown by reviewing studies that combine more than one ML model for HAB prediction. The present work also presents a case study comparing performance of several ML models using chlorophyll-a concentration as the index. This review serves as a guide for policymakers and researchers for the implementation of ML models for HAB prediction and reveals the potential of ML models for decision-support and early prediction for HAB management.

Technical Abstract: Harmful algal blooms (HABs) are detrimental to the lives of livestock, humans, pets, wildlife, and the global economy, which calls for a robust approach to their management. Decision support tools such as process-based models facilitate the understanding of HAB-enabling conditions and the analysis of the impact of watershed activities on HAB occurrence. The rapid breakthrough in computing prowess and statistical advances is triggering the development of data-driven techniques such as machine learning (ML) models. The application of ML models for HAB monitoring is favored by the improved sensor technologies that provide continuous data. This review describes ML models currently used for predicting and forecasting HABs in freshwater ecosystems and provides an overview of the model structures and their application for predicting algal parameters and related toxins. The use of hybrid ML models for improved prediction accuracy was also shown by reviewing studies that combine more than one ML model for HAB prediction. The present work also presents a case study comparing performance of decision tree, random forest, multilayer perceptron, support vector regression, long short-term memory, and gated recurrent unit models using chlorophyll-a concentration as the index. This review serves as a guide for policymakers and researchers for the implementation of ML models for HAB prediction and reveals the potential of ML models for decision-support and early prediction for HAB management.