Hiroshima University researchers have unveiled a groundbreaking hybrid AI model that significantly enhances the accuracy of harmful algal bloom (HAB) forecasting by incorporating the overlooked role of tiny plankton. Published in the journal Ecological Informatics on April 9, 2026, this study addresses a critical challenge in marine ecology amid rising climate-driven HAB events worldwide.
Harmful algal blooms, often called red tides, occur when certain microalgae proliferate uncontrollably, depleting oxygen, releasing toxins, and devastating fisheries. In Japan, where aquaculture supports a vital seafood industry, HABs pose ongoing threats to coastal economies. This innovation from Hiroshima's Center for Planetary Health and Innovation Science promises more reliable early warnings, protecting both ecosystems and livelihoods.
The Growing Threat of Harmful Algal Blooms in Japan and Beyond
Harmful algal blooms represent a complex interplay of nutrient enrichment, warming waters, and microbial dynamics. Globally, their frequency has surged due to climate change, with economic losses exceeding billions annually. In Chile, a key partner in this research, HABs have cost the salmon industry over $1 billion in the past decade alone.
Japan, heavily reliant on imported salmon—75% from Chile—feels these ripples directly. Domestic HAB events, like those caused by Karenia mikimotoi, have led to massive fish kills, with monitoring data since the 1970s revealing persistent risks to East Asian coasts. For instance, a 2015 bloom in the Yatsushiro Sea triggered significant fishery losses, highlighting the need for precise forecasting.
In Japanese waters, species like Heterosigma akashiwo and Chattonella species have historically damaged yellowtail farms, costing hundreds of millions of yen. Recent statistics show Karenia mikimotoi blooms expanding spatially, affecting shellfish and fish mortality rates. Hiroshima University's work bridges this gap, leveraging international collaboration to safeguard Japan's seafood supply chain.
The study's context underscores higher education's role in tackling planetary health crises. Universities like Hiroshima are at the forefront, blending microbiology, AI, and oceanography to foster sustainable solutions.
Hiroshima University's Leadership in Planetary Health Research
At the heart of this advancement is Hiroshima University's Center for Planetary Health and Innovation Science within The IDEC Institute. Professor Fumito Maruyama, the project's lead, specializes in microbial genomics and ecology. His team has pioneered holobiome approaches—studying entire microbial communities—to decode HAB triggers.
Maruyama's lab, the Microbial Genomics and Ecology Group, has a track record in HAB dynamics, including prior studies on algae interactions in coastal environments. This latest publication builds on February 2025 research using empirical dynamic modeling to map species interactions, demystifying bloom unpredictability.
Hiroshima U emphasizes interdisciplinary training, equipping students with skills in AI, environmental DNA (eDNA) analysis, and predictive modeling. This aligns with Japan's push for innovation in aquaculture sustainability, positioning the university as a hub for global environmental research.

SATREPS-MACH: A Japan-Chile Partnership Against HABs
The research stems from the SATREPS-MACH project (Monitoring of Algae in Chile), a Japan Science and Technology Agency (JST)-funded initiative. Launched to support Chile's aquaculture amid HAB threats, it develops monitoring kits and forecast systems.
Goals include characterizing HAB holobiomes—phytoplankton, bacteria, viruses—and integrating data with machine learning for predictions. Partners span Chilean institutes like IFOP, SERNAPESCA, and INTESAL, fostering academia-industry ties. Outputs feature the EZ holobiome monitoring kit and 'HAB Weather News' bulletins.
For Japan, reliant on Chilean salmon exports (valued at hundreds of millions USD yearly), MACH ensures supply stability. The prototype model tested here couples three tools from the project, demonstrating hybrid power.
Unpacking the Hybrid AI Model: A Three-Pronged Approach
The hybrid model fuses physics-based simulation, machine learning, and empirical dynamics for superior HAB forecasting. Here's how it works step-by-step:
- Parti-MOSA (Physical Model): Simulates algae particle dispersion via ocean currents, weather, and environmental factors like temperature and salinity. It tracks bloom transport without biological growth details.
- LSTM Neural Network (AI Component): Long Short-Term Memory networks analyze time-series data from eDNA metabarcoding and sensors. LSTM 'remembers' patterns, predicting influences on bloom behavior as data accumulates.
- Empirical Dynamic Model (EDM): Non-parametric tool using convergent cross-mapping (CCM) to detect causal species interactions from long-term community data. Multivariate S-map forecasts targets like Pseudo-nitzschia using causal plankton.
The prototype couples these: physical transport feeds into AI learning and EDM causality, creating a robust framework. This addresses single-model limitations in complex fjord systems like Patagonia.
Photo by Annie Spratt on Unsplash

Tiny Plankton: The Unsung Drivers of HAB Dynamics
Picophytoplankton and other non-motile tiny plankton were key. The study focused on two Pseudo-nitzschia groups (toxin-producers linked to amnesic shellfish poisoning). Causal species like Ceratium (dinoflagellate) and Leptocylindrus (diatom) consistently predicted blooms across sites.
CCM revealed these interactions: ρ up to 0.733 for P. seriata in Metri (p<0.0001). Tiny plankton influence via competition, predation, or nutrient sharing—'ecological conversations' per Maruyama. Ignoring them leads to inaccurate forecasts; including boosts reliability, especially for bloom onset.
In Japan, similar dynamics occur with Karenia mikimotoi, where microbial interactions terminate blooms via algicidal bacteria in Hiroshima Bay.
Data-Driven Insights: 30 Years of Observations Fuel Accuracy
Using PROMOFI dataset (1992-2020) from Metri, Quellón, Melinka—diverse Patagonian sites—the team preprocessed biweekly phytoplankton counts. Frequent species (>15% presence) underwent CCM for causality, then S-map predictions.
Performance: Strongest in Metri (r=0.733, RMSE=0.994); weaker elsewhere due to variability. EDM excelled for presence/absence, aiding short-term (1-2 week) alerts. Funded by JSPS and SATREPS, this validates hybrid viability.
Results Spotlight: Precision Gains from Species Interactions
Without interactions, models faltered; with them, correlations soared. Ceratium and Leptocylindrus proved reliable predictors for Pseudo-nitzschia. Site-specific: Metri's stable environment yielded best results; fjord variability challenged Quellón/Melinka.
Hybrid coupling mitigates weaknesses—Parti-MOSA handles transport, LSTM patterns, EDM causality. This multi-lens view captures HAB complexity, outperforming silos.
Implications for Japan's Fisheries and Aquaculture
Japan's $2B+ salmon imports from Chile risk HAB disruptions. Accurate forecasts enable preemptive cage closures, minimizing losses like Chile's $1B decade tally. Domestically, applying to coasts could curb Karenia impacts, protecting yellowtail and shellfish farms.
Higher ed benefits: Trains students in AI-ecology, boosting Japan's research edge. Maruyama: "HABs are conversations shaped by interactions and signals—hybrid models listen."
Stakeholder Views and Expert Insights
Fumito Maruyama emphasizes context-specific hybrids: "Integrating physical, ecological, ML improves accuracy amid climate flux." Co-author Milko Jorquera notes Chile's gains; Satoshi Nagai eyes Japan ops.
Fisheries experts praise early warnings for ROI, despite false positive costs. Governments seek operational tools like 'HAB Weather News'.
Photo by Armand Mckenzie on Unsplash
Future Horizons: From Chile to Japanese Waters
Next: Add variables (nutrients, viruses), out-of-sample validation, real-time eDNA. Extend to Japan coasts via Fisheries Agency data. Hiroshima aims for global HAB platform, advancing planetary health curricula.
This exemplifies university-driven innovation, blending AI with ecology for sustainable futures.
Hiroshima U's Role in Japan's Higher Ed Landscape
Hiroshima fosters interdisciplinary excellence, with labs like Maruyama's training PhDs in genomics-AI. Amid Japan’s uni reforms, it leads env tech, attracting intl talent.
Explore research positions or Japan opportunities at AcademicJobs.
