In a game-changing climate study, researchers have harnessed artificial intelligence (AI) to predict extreme Atlantic Niño and Benguela Niño events up to 3–4 months in advance, offering new hope for protecting marine ecosystems and coastal communities that rely on them.
Led by Marie-Lou Bachèlery at the CMCC (Euro-Mediterranean Center on Climate Change), this research marks a major leap in our ability to forecast complex oceanic phenomena that were once considered nearly impossible to predict. Published in Science Advances, the study introduces a deep learning model trained on 90 years of ocean temperature data, delivering unmatched accuracy and early warnings for climate anomalies in the tropical Atlantic.
Why It Matters
The Tropical Atlantic, particularly the Angola-Benguela Upwelling System, supports one of the world’s most productive marine ecosystems. This region is vital for local fisheries, marine biodiversity, and the livelihoods of coastal communities across West Africa and even impacts weather and economies far beyond.
But this ecosystem faces serious threats from Atlantic Niño and Benguela Niño events, which involve sudden warming or cooling of ocean waters. These disrupt fish populations, rain patterns, and marine food chains. Until now, predicting these events has been extremely difficult.
“Overfishing alone can’t explain the changes in fish stocks,” says Bachèlery. “Climate extremes play a huge role, and forecasting them has become a major scientific challenge.”
How AI Changed the Game
Bachèlery and her team developed a convolutional neural network (CNN)—a deep learning algorithm that can detect subtle patterns in vast datasets. Trained on almost a century of temperature records, the AI learned to recognize early signals, like slow-moving ocean waves, that precede major climate events.
Unlike traditional climate models, which often struggle with biases and inaccuracies, this AI system consistently delivered early warnings up to 4 months in advance and in peak seasons, even 5 months.
One standout moment: the model accurately predicted the strong 2021 Atlantic and Benguela Niño events four months before they occurred. Standard forecasting tools missed them entirely.
“Our AI model picks up signals that older models overlook,” says Bachèlery. “It’s not just about better forecasts this gives us a new lens for understanding extreme ocean events.”
What’s Next?
The team is working on a web platform to share these forecasts directly with those who need them most: fishers, marine managers, and coastal planners. The goal is to make the data accessible, understandable, and actionable.
“It’s crucial our science supports decision-making on the ground,” says Bachèlery. “This is about making real-world impact.”
Future plans include expanding the AI system to predict other variables like oxygen levels and fishery productivity factors critical for managing marine life and food security.
The study was developed during Bachèlery’s Marie Curie fellowship at the University of Bergen, in collaboration with NERSC, University of Cambridge, and LEGOS in France. Now based at CMCC, she continues to build on this pioneering work.