EURECOM’s AI for climate resilience From energy efficiency to the anticipation of extreme weather patterns

How can artificial intelligence help us face the climate crisis rather than contributing to it? At EURECOM, a Sophia-Antipolis based leading research institute and engineering school in digital science, the answer lies in designing smarter, leaner, and more purposeful AI systems. From energy-efficient models that fight climate disinformation online to sophisticated simulations that can help anticipate storm surges on the French Riviera, EURECOM’s climate-focused AI research is both technically ambitious and deeply grounded in real-world impact. Whether it’s reducing the carbon footprint of algorithms or predicting the next extreme weather event, AI is also emerging as a tool of innovation towards climate resilience.

Frugal AI for fighting climate disinformation

To promote energy-conscious AI, the Frugal AI Challenge—organized by France’s Ministry for Ecological Transition and Hugging Face—tasked teams with building low-energy, high-performance models addressing climate-related issues. Held during the AI Action Summit in Paris (Feb 10–11, 2025), it included three tasks: detecting illegal deforestation from audio recordings, identifying wildfires from satellite images, and identifying climate related disinformation in textual news.

QuotaClimat provided news articles labeled by types of false climate narratives, such as denial of human responsibility. Each team’s solution was judged not only on accuracy but also energy efficiency as the energy consumption of each model was a decisive factor in the evaluation process.

EURECOM’s team, led by Prof. Raphaël Troncy and PhD student Youri Peskine, won in the textual data category of the Frugal AI Challenge thanks to a meticulous trade-off between model accuracy and energy efficiency. Rather than using large, resource-heavy architectures, the team developed a lean solution built around a compact language model and a streamlined six-layer neural network. This architecture minimized computational overhead while preserving strong classification performance. This achievement stemmed from EURECOM’s experience in prior research projects focused on disinformation detection. such as the European-funded CIMPLE project, coordinated by R. Troncy on designing AI systems to analyze sensitive narratives around topics such as COVID-19, geopolitical conflict, and climate change.

The ClimateSense Project

More recently this topic will also be addressed in ClimateSense, a European research project launched in March 2025 looking at how climate change misinformation spreads online. An ambitious project, which brings together research teams from the UK, France, Lithuania, and the Czech Republic.

ClimateSense takes a multidimensional approach to climate resilience. On one hand, it focuses on detecting and understanding extreme climate events using data science and geospatial tools. On the other, it investigates how climate misinformation proliferates across digital platforms in the wake of such events—ranging from the denial of human-caused climate change to conspiracy theories targeting scientific institutions.

EURECOM contributes its AI and machine learning expertise to develop tools for automatic classification of climate-related content, particularly disinformation. One experimental effort involves AI-generated counter-narratives delivered by bots like Misinfome, which interact with online content using humor, poetry, or memes to subtly challenge misleading posts. This approach draws from behavioral sciences and social psychology to explore non-confrontational methods of engagement—an area still under-researched but increasingly vital in combating disinformation fatigue. EURECOM’s researchers are helping build robust, open-access tools that can be used by journalists, educators, and policymakers. In doing so, they position AI as a digital advocate in the fight for informed climate action.

Towards probabilistic AI for predicting the impact of storm surges and coastal flood risk

Another major climate-related research axis focuses on using simulation technologies to model natural disasters such as storm surges, which are expected to become more frequent and severe as oceans warm and sea levels rise. For example, how will climate change affect coastal flooding in the coming decades?

This question is at the heart of a project led by Professor Motonobu Kanagawa at EURECOM, in collaboration with the Interministerial Delegation for Digital Transition of the Principality of Monaco. As a member of EURECOM's consortium (GIE), the Prince’s Government aims to address this question by simulating storm surges along the French Riviera and beyond. These simulation technologies have also been showcased as part of a European cross-border observatory project involving French public organizations (Région Sud, Nice Côte d’Azur Metropolis) and Italian partners (the Regions of Piedmont and Liguria, as well as the metropolitan areas of Turin and Genoa).

Storm surges—sea level rises caused by storm-driven winds and pressure drops—pose serious threats to coastal communities. Climate change is expected to exacerbate these events by rising sea levels and altering storm behavior. However, the assessment of flood risks assessments at very precise and localized levels remains a very difficult thing to do.

Using the GeoClaw simulator, the team is working to reconstruct past events such as Storm Alex (October 2020), a recent extreme event in the region. A major challenge lies in the uncertainty of model parameters (e.g., storm intensity and pressure fields) and the limited availability of historical data.

To address this, the project aims to incorporate AI-based methods to systematically quantify and propagate uncertainties, including those from numerical discretization, simplified storm models (e.g., Holland model), and unknown future storm behavior under climate change. Professor Kanagawa’s expertise in probabilistic AI—such as Bayesian quadrature and probabilistic numerical methods—can provide a principled foundation for this effort. The Principality of Monaco is integrating the simulation tools and methodologies developed in partnership with EURECOM into its public data platform “data.gouv.mc” as part of its Extended Monaco digital strategy.

Accelerating satellite-based climate monitoring with AI

Prof. Kanagawa is also in collaboration with Japan’s National Institute for Environmental Studies (NIES), applying AI to improve greenhouse gas monitoring using data from the GOSAT satellite program. These satellites capture radiance spectra—light measurements altered by the presence of atmospheric gases like CO₂ and methane. While these readings indirectly reveal gas concentrations, translating them into precise measurements typically requires computationally intensive simulations based on atmospheric physics, a process known as “retrieval”.

EURECOM’s team is addressing the required computationally intensive simulations using AI models trained to approximate the retrieval algorithm. By learning from hundreds of thousands of precomputed radiance-CO₂ data pairs, these models can estimate greenhouse gas levels far more quickly, enabling timely insights for climate research and policy.

However, atmospheric conditions evolve, introducing a phenomenon known as “distributional shift”—a challenge for AI systems trained on past data. EURECOM’s research tackles this by integrating principles such as: Bayesian inference techniques, atmospheric radiative transfer physics and physics-guided modeling for robustness, directly into the learning process, improving the models’ ability to generalize and remain accurate over time. This hybrid approach, blending physics and machine learning, marks a promising direction for climate-focused satellite analytics. “AI is not a magic bullet, however. A real challenge lies in how we train it to enable good predictions for future observations, not past ones.”, as Prof. Kanagawa explains.

CIRCALIS: Powering AI responsibly

While AI methods can directly contribute to better anticipate the impact of climate change as shown above, we must recognize that some of these methods are math intensive, hence require more computing power, fueled by electricity. A responsible approach to computing is therefore necessary.

EURECOM’s CIRCALIS project embodies such a local and sustainable vision to advance AI. Launched with support from Région Sud, this new high-performance computing center on the school campus accelerates research in key areas such as health, cybersecurity, mobility, and climate resilience.

CIRCALIS integrates a major renewable energy source, the SUNrecom solar farm, which supplies about one-third of the site’s energy needs. Additionally, the infrastructure is optimized to minimize energy consumption through efficient systems and effective cooling. This collaborative platform fosters a regional ecosystem of academic institutions, industries, and startups committed to high-performance AI, directly contributing to the climate ambitions of the Côte d’Azur.

A new era of AI for climate resilience

Innovative AI projects highlight the importance of a responsible approach to artificial intelligence in addressing the climate crisis. By optimizing models to reduce their environmental impact and harnessing their potential to better predict and manage extreme climate events, responsible AI research helps build more resilient societies and a more sustainable digital future. Whether combating misinformation, simulating storm surges, AI takes a growing role as a tool in support of climate challenges.