Scientists from the Autonomous University of Madrid (UAM) and Complutense University of Madrid (UCM) have created an early warning system based on artificial intelligence to predict massive outbreaks of cyanobacteria in fresh waters. Using machine learning and deep learning models, the system could help protect aquatic ecosystems and improve water management.
A team of researchers from the Autonomous University of Madrid (UAM) and the Complutense University of Madrid (UCM), in collaboration with experts in microbiology, physics and data analysis, has developed an early warning system to predict the massive proliferation of cyanobacteria.
The results, published in the journal Water Research, represent a significant advance in the prevention of these outbreaks, favoring the preservation of aquatic ecosystems and more efficient water management.
Cyanobacteria, which in many cases are toxic, are often the main microorganisms responsible for blooms, or massive proliferations of microalgae in fresh waters. These blooms affect both the balance of aquatic ecosystems and water quality, compromising its recreational use and potability. Therefore, early warning systems are crucial to detect these threats at an early stage and mitigate the associated risks.
For their study, the researchers used data collected from a floating platform installed in the Cuerda del Pozo reservoir in Soria, Spain. For six years, sensors mounted on an automatic profiler have been monitoring the entire water column, providing a valuable database for the development of the predictive system.
"We have developed a simple but extremely robust method that makes it possible to predict the timing and intensity of cyanobacterial blooms," explains Claudia Fournier, a researcher in the UAM Department of Biology. "To do this, we only need data on water temperature, the concentration of chlorophyll-a, which is a pigment present in all algae, and phycocyanin, a pigment specific to cyanobacteria in freshwater."
The methodology employed included flexible data preprocessing and the use of predictive models of varying complexity, including machine learning and deep learning techniques, such as neural networks with short- and long-term memory (LSTM).
The effectiveness of the models was evaluated with prediction periods ranging from 4 to 28 days, and the LSTM model achieved 90% accuracy in predicting warning levels for both short (4 days) and longer (28 days) prediction horizons.
Bibliographic reference:
Claudia Fournier, Raůl Fernandez-Fernandez, Samuel Cirés, José A. López-Orozco, Eva Besada-Portas, Antonio Quesada (2024). "LSTM networks provide efficient cyanobacterial blooms forecasting even with incomplete spatio-temporal data", Water Research (2024), doi: https://doi.org/10.1016/j.watres.2024.122553
More scientific culture in UAM Gazette