Predictable Coastal Floods: How AI Anticipates Tidal Surges
A study from the University of Central Florida shows that recurring coastal floods follow predictable patterns. By combining predictive models, neural networks, and satellite data, researchers pave the way for more accurate warnings for coastal populations.
Each year, storm waves and high tides inundate low-lying areas of American coasts, causing property and human damage. Yet, research at the University of Central Florida reveals that these floods are not random: they occur at temporal intervals that can be anticipated through artificial intelligence. This discovery changes the game for coastal risk management, offering the possibility to plan protective measures well before waters reach the critical threshold.
What the study reveals: identifiable temporal cycles
Ivan Haigh and his team analyzed decades of coastal flood events using Copernicus atmospheric data archives and tide gauge records. They found that, far from being random, floods often recur at specific times of the year, notably during the combination of spring tides and Atlantic-origin depressional systems. This recurrence, detected in over a hundred cases, suggests the existence of a 'vulnerability calendar' unique to each coastal front.
The key point of the study is the reduction of forecast uncertainty: by identifying windows of maximum risk, models can focus their resolution efforts on critical periods. Thus, coastal flood forecasts gain accuracy, translating into more reliable alerts for local authorities and exposed populations.
How: AI, neural networks, and satellite data for forecasting
To extract these patterns, researchers trained a predictive model based on deep neural networks, fed by satellite data from the Sentinel-1 mission and ECMWF reanalyses. The machine learning process integrated variables such as wave height, atmospheric pressure, sea surface temperature, and tide levels. By combining these atmospheric data with flood histories, the model could identify the strongest temporal correlations.
The system draws on technologies already used in advanced weather forecasting, like GraphCast or Pangu-Weather, which use spatial graphs to capture complex interactions between atmosphere and ocean. By adapting these approaches to the coastal context, the team created a tool capable of projecting the probability of flooding up to ten days in advance, while quantifying forecast uncertainty through ensemble forecasting.
Scientific explanation of the phenomenon
Coastal floods result from a combination of astronomical and meteorological effects. Spring tides, which occur approximately every two weeks, coincide with peak high tides and can exacerbate coastal flooding when they coincide with storm surges or high waves. Additionally, low-pressure systems originating in the Atlantic can cause significant coastal flooding by pushing large volumes of water onto the coast.
