A new study has demonstrated that AI can accurately predict extreme storm surges, offering a powerful new tool for assessing coastal flood risks in a world increasingly affected by climate change. Storm surges are temporary rises in sea level caused primarily by storms and are among the leading causes of coastal flooding. As global sea levels continue to rise, the combination of higher baseline water levels and more intense coastal storms poses a growing threat to the more than 10% of the world’s population living in low-lying coastal regions.
Understanding future coastal flood risks is essential for policymakers, urban planners, engineers, and emergency management agencies. Decisions about infrastructure design, flood defenses, land-use planning, and disaster preparedness all depend on accurate projections of extreme sea level events. However, predicting these events remains a major scientific challenge because storm surges result from complex interactions between tides, atmospheric conditions, ocean circulation, and local coastal characteristics. These processes are highly nonlinear, meaning small changes in model assumptions can produce significantly different outcomes, especially when forecasting rare but severe events.
Traditional physics-based models are currently the primary tools used to simulate coastal processes. While these models provide detailed and physically realistic representations of storm surge dynamics, they require substantial computing power and time. As a result, exploring a wide range of future climate scenarios and uncertainties can be prohibitively expensive and time-consuming.
Researchers are increasingly turning to AI to address these limitations. Because AI models can generate predictions much more quickly than traditional physics-based models, they make it possible to evaluate large numbers of future scenarios and better estimate the likelihood of rare but catastrophic coastal flooding events. This capability is particularly valuable for risk assessments, where understanding extreme outcomes is crucial for protecting lives, infrastructure, and economic assets.
The study, published in Earth’s Future, investigated whether AI models could accurately reproduce the results of physics-based simulations while operating far more efficiently. Rather than replacing traditional models, the researchers trained their AI system using outputs from the Global Tide and Surge Model (GTSM), a widely used physics-based coastal simulation framework. This approach combines the physical realism of traditional modelling with the speed and flexibility of machine learning.
The researchers focused on the New York City coastal region as a case study because of its high exposure to storm surge flooding. The area contains dense populations, critical infrastructure, and significant economic assets. It has also experienced devastating storm surge events, most notably Hurricane Sandy in 2012, which caused widespread destruction, numerous fatalities, and more than $60 billion in damages.
Results showed that the AI emulator successfully learned the complex dynamics responsible for storm surge events and reproduced extreme sea level conditions with high accuracy. Importantly, the model maintained its performance when tested against future climate scenarios extending to the middle of the 21st century, even when evaluating data it had not previously encountered during training.
The findings suggest that AI-powered storm surge emulators can become valuable tools for coastal risk assessment and climate adaptation planning. By rapidly generating multiple future scenarios and conducting sensitivity analyses, these systems can help researchers better understand uncertainties and improve flood risk projections.
Despite these promising results, further work is needed. Researchers plan to test the model across a broader range of climate conditions and geographic regions while integrating it into operational risk assessment platforms such as the Aqueduct Flood Risk Analyzer and the Copernicus Climate Data Store. Future research will also focus on improving the model’s ability to quantify uncertainty, evaluate its performance under unprecedented climate conditions, and ensure its reliability for real-world decision-making as coastal risks continue to increase.