Harmful algal blooms (HABs)—responsible for environmental damage, mass fish die-offs, economic downfalls, and even human deaths—are increasing in frequency and severity as the Earth warms. While some computer models can forecast potential blooms, their accuracy is limited by the number of algae species that can bloom harmfully under different environmental triggers, as well as how different species may overlap with one another. However, an international team has demonstrated that coupling three models and accounting for how different algae species interact can significantly improve predictions.
The researchers, led by Fumito Maruyama , a professor with the Center for Planetary Health and Innovation Science at Hiroshima University's The IDEC Institute , published their work in the March issue of Ecological Informatics .
"Harmful algal blooms are like ecological conversations, where species interactions and environmental signals continuously shape outcomes, rather than being driven by a single dominant factor," Maruyama said. "This study shows that integrating physical processes, ecological interactions, and machine learning approaches can improve prediction accuracy. Hybrid, context-specific modeling frameworks offer a more robust way to understand and forecast harmful algal blooms across environments."
$1B lost to harmful algal blooms in the past decade
Algae, minuscule plants, are a key part of marine ecosystems, serving as food for plankton and other water life. Heat or excessive nutrients from fertilizer runoff can trigger the algae to grow out of control, which depletes oxygen in the water and knocks the fragile ecosystem out of sync. This has led to major economic impact in Chile, the world's second-largest producer of salmon and one of the largest exporters of mussels. Harmful algal blooms have plagued the country in recent decades, with an estimated loss of $1 billion in the last 10 years alone, the researchers said.
"Fish and shellfish farmers are more likely to benefit from short-term harmful algal bloom forecasts of one or two weeks, as the ability to plan and close fish cages in advance of a bloom can protect their stock and increase profitability," Maruyama said. "However, there is a trade-off: False positive predictions can lead to premature harvesting and loss of revenue."
The Science and Technology Research Partnership for Sustainable Development - Monitoring of Algae in Chile (SATREPS-MACH) project evolved from a collaborative effort between Chile and Japan, which relies on Chile for three-quarters of its salmon imports, to improve the understanding and prediction of algal blooms to prevent food waste. Maruyama explained that, in this study, the team presented and evaluated three models developed under SATREPS-MACH. The first model, Parti-MOSA, simulates the physical movement of algae through specific environments, accounting for weather and other factors. The second is an artificial intelligence model based on long short-term memory, meaning it continues to learn and remember based on accumulated data, so with more data, it can better understand how different factors will influence behaviors. The third is an empirical dynamic model that incorporates long-term community data to predict how things change depending on factors interacting.
Accounting for plankton interactions sharpened forecast accuracy
Using more than 30 years of observational data from three environmentally different sampling sites around Chile, with a focus on two specific plankton species groups, the researchers evaluated how closely the models predicted the harmful algal bloom species dynamics. Model performance varied across the locations and algae species, but when the researchers included interaction among plankton species in their model data, prediction accuracy significantly improved.
"Individual models can capture important aspects of harmful algal bloom dynamics, but each has limitations," Maruyama said, explaining that the individual models cannot account for how environmental conditions and plankton species interactions influence harmful algal bloom dynamics. "Together, these models address critical gaps in forecasting harmful algal bloom dynamics in the highly complex and understudied Chilean Patagonian environment. This study shows that integrating physical processes, ecological interactions, and machine learning approaches can improve prediction accuracy."
Next, the researchers said they plan to refine the approach by incorporating additional environmental variables and extending the frameworks to broader regional contexts, including coastal systems in Japan.
"Ultimately, our goal is to develop reliable, operational harmful algal bloom forecasting tools for effective early warning and risk reduction," Maruyama said.
Maruyama is also affiliated with Hiroshima University's Center for HOlobiome and Built Environment . In addition to Maruyama, other authors affiliated with Hiroshima University are: Ishara Uhanie Perera, So Fujiyoshi, Kyoko Yarimizu, and Milko A. Jorquera. Other authors are Daiki Kumakura and Shinji Nakaoka, Graduate School of Life Science at Hokkaido University in Japan; Carolina Medel and Pablo Reche, Instituto de Fomento Pesquero, CTPA Putemún, Chile; Osvaldo Artal and Jacquelinne J. Acuña, Universidad de La Frontera, Chile; Oscar Espinoza-González and Leonardo Guzman, Instituto de Fomento Pesquero, Centro de Estudios de Algas Nocivas, Chile; Felipe Tucca and Alexander Jaramillo-Torres, Instituto Tecnológico del Salmón, Chile; Satoshi Nagai, Coastal and Inland Fisheries Ecosystems Division, Fisheries Technology Institute, Japan Fisheries Research and Education Agency. Perera is also affiliated with Yamaguchi University, Japan; Fujiyoshi with Toyama Prefectural University, Japan; Kumakura with RIKEN, Japan; Artal with University de Concepción, Chile; Jorquera with Universidad de la Frontera; and Nakaoka with Hokkaido University.
The Japan Society for the Promotion of Science, and the Science and Technology Research Partnership for Sustainable Development supported this work.
