AI Wave Models in Coastal Engineering

Coastal engineers are increasingly utilizing surrogate artificial intelligence (AI) wave models to rapidly and accurately predict seawall overtopping and storm-surge shoreline erosion. Traditional physics-based numerical models require immense computational power and time to simulate complex fluid dynamics, making real-time forecasting difficult during approaching storms. By training machine learning algorithms on historical weather data, physical wave flume experiments, and high-fidelity simulations, engineers create lightweight "surrogate" models. This article explores how these AI tools work, their specific applications in predicting structural overtopping and coastal erosion, and how they transform real-time coastal hazard management.

Understanding Surrogate AI Wave Models

Traditional coastal modeling relies on solving complex fluid dynamics equations, such as the Navier-Stokes or Boussinesq equations. While accurate, a single storm simulation can take hours or days to run on a supercomputer.

Surrogate AI models bypass these heavy calculations. Engineers first run thousands of traditional, physics-based simulations covering various storm scenarios. They then feed the inputs (wind speed, offshore wave height, wave period, and bathymetry) and the resulting outputs into a machine learning algorithm, such as a Deep Neural Network (DNN) or a Gaussian Process. Once trained, the AI acts as a stand-in—or surrogate—for the physics model, recognizing patterns and delivering identical predictions in milliseconds rather than hours.

Predicting Seawall Overtopping

Seawall overtopping occurs when waves exceed the crest of a defensive structure, causing coastal flooding and potential structural failure. Predicting the exact volume of water that will breach a wall is notoriously difficult due to the chaotic nature of breaking waves.

Coastal engineers use AI surrogates—specifically Artificial Neural Networks (ANNs) and Gradient Boosting Machines—to predict overtopping discharge rates. These models are trained on massive datasets, such as the European CLASH database, which contains thousands of physical model tests of different seawall geometries.

To predict overtopping, engineers input specific parameters into the AI model:

• Structural Geometry: Crest freeboard (height of the wall above water), slope angle, and surface roughness.
• Hydraulic Conditions: Significant wave height, peak wave period, and water depth at the toe of the structure.

The surrogate model instantly calculates the mean overtopping discharge. This allows engineers to quickly test hundreds of seawall designs to find the safest and most cost-effective geometry, or to predict flooding volumes hours before a hurricane makes landfall.

Forecasting Storm-Surge Shoreline Erosion

Storm surges temporarily elevate sea levels, allowing large waves to attack higher up the beach profile, leading to rapid dune destruction and shoreline retreat. Traditional morphodynamic models (like XBeach) simulate this sediment transport well, but they are too slow for large-scale, real-time probabilistic forecasting.

To solve this, engineers deploy Long Short-Term Memory (LSTM) networks and Convolutional Neural Networks (CNNs) as surrogates for sediment transport models. These AI architectures are particularly good at processing time-series data and spatial grids.

Engineers feed the surrogate model time-varying storm inputs, including:

• Surge Hydrographs: The rise and fall of water levels over time.
• Wave Energy: The power and direction of incoming waves during the storm's peak.
• Pre-Storm Morphology: LiDAR data of the existing beach slope, dune height, and sediment grain size.

The AI model processes these inputs to predict the post-storm beach profile. It maps out exactly how much sand volume will be lost and how far the shoreline will retreat. Because the AI runs instantly, engineers can run thousands of Monte Carlo simulations to account for uncertainties in a storm's path, generating probabilistic erosion maps that show the likelihood of dune breaches across hundreds of miles of coastline.

Real-Time Early Warning Systems

The ultimate value of surrogate AI wave models lies in their speed. Because predictions are generated in seconds, these models are now the backbone of operational Early Warning Systems (EWS).

As a storm approaches, meteorological forecasts are continuously updated and fed directly into the coastal AI surrogates. Emergency managers receive real-time updates on which specific seawalls are at risk of overtopping and which neighborhoods face severe erosion. This enables targeted evacuations, the strategic deployment of temporary flood barriers, and more resilient long-term coastal urban planning.