Can we see extreme events coming before they hit? At the University of Hamburg, scientists seem to think so. They’re combining observations and Earth system models with artificial intelligence techniques to improve the initial conditions of climate predictions.
When Goratz Beobide Arsuaga moved from Spain to Germany to study climate dynamics, he didn’t expect that the relationship between the ocean and the atmosphere would become the cornerstone of his research career.
“I got quite hooked into how the ocean and the atmosphere are related and how they affect each other,” he recalls. “These interactions fascinated me because they can influence extreme events like heatwaves.”
Now a postdoctoral researcher at the University of Hamburg’s Climate Modelling Department, Dr Beobide Arsuaga is actively involved in EXPECT project. His work explores how AI-based data simulation can improve seasonal and decadal forecasts, especially when observations are limited. The novel method was co-developed with Dr Christopher Kadow at DKRZ.
Learning from the ocean to predict the atmosphere
During his PhD, Dr Beobide Arsuaga investigated whether patterns in the Atlantic Ocean could help anticipate European heatwaves months in advance.
“The main question was whether we can predict a season in advance the occurrence of heatwaves by identifying certain patterns in the Atlantic Ocean,” he explains. “That led me to the dynamics behind extreme events and whether climate models can simulate those connections accurately.”
This focus on the physical processes that drive extremes naturally led him to EXPECT, where he’s now developing AI-based data assimilation techniques, a key step in improving how climate models start their simulations.
“Data assimilation determines the initial conditions of a prediction,” he says. “If we start from inaccurate data, the predictions are likely to be wrong. We’re using AI to combine observations and Earth system models, to make those initial conditions more realistic, and therefore, the forecasts more reliable.”
Bridging science and society
For Dr Beobide Arsuaga, the science is only half of the story. His background in economics keeps him thinking about the societal value of climate predictions. “Now that we know certain extreme events are at least somewhat predictable, the next step is understanding how this information can help different sectors of society,” he says.
From farmers and energy providers to policymakers, improved forecasts can guide better decisions. Dr Beobide Arsuaga recalls an American farmer who once contacted him, after reading his research on Atlantic Ocean precursors of heatwaves in Europe. “He didn’t want a yes-or-no answer about heatwaves,” Goratz smiles. “He was after probabilities, so he could plan accordingly. That’s the kind of mindset we need to cultivate in Europe as well.”
Trusting AI and verifying it
Artificial intelligence plays a central role in Dr Beobide Arsuaga’s work, but he’s the first to point out that it must be handled carefully. “AI models are very good at finding connections,” he explains, “but we always need to ask if they’re doing it for the right reasons. It takes a lot of time to create an AI model and even more to understand what it’s really doing,” he continues.
To ensure that AI-based predictions are grounded in real physics, Goratz and his team test whether the improvements produced by the algorithms reflect better representations of real oceanic and atmospheric dynamics. “We compare AI outputs with real-world processes,” he says. “Only when they align do we start to trust them.”
This rigorous validation is essential for EXPECT’s broader mission: to explain extreme events. “If we have a good prediction but don’t understand why, we can’t trust it or apply it to future events,” he adds.
Looking ahead
The promise of AI in climate science is vast, but Goratz remains pragmatic. “Combining AI with Earth system models opens up many possibilities, but also increases complexity,” he admits. His team is currently testing whether the improvements they observe are truly physical or just statistical noise.
Yet the direction is clear. As global temperatures rise and extreme events intensify, society’s need for reliable forecasts has never been greater. “The more time in advance we can predict, the more time we have to prepare and potentially reduce the impacts,” says Dr Beobide Arsuaga. “That’s what motivates me: to not just improve our predictions but to make them more useful.”
More news hot off the lab bench? Find many more EXPECT updates here.
We’re also on LinkedIn, X, and Bluesky. Follow us there if you haven’t already!