Researchers at Virginia Tech and Vrije Universiteit Brussel have developed an innovative deep learning model to improve the prediction of coastal flooding during hurricane season. The new model, called Long Short-Term Memory Station Approximated Models (LSTM-SAM), uses artificial intelligence and transfer learning techniques to forecast storm surge and coastal flooding more accurately even in locations with limited or no water level observation stations like tide gauges.

The traditional methods for forecasting coastal floods require dense observational networks and extensive historical data, which many vulnerable regions lack. LSTM-SAM addresses this issue by learning from data-rich areas and applying that knowledge to data- poor regions, effectively filling in the gaps in the observational network.

This model is built on Long Short-Term Memory (LSTM) neural networks, a type of recurrent neural network capable of learning temporal patterns. The researchers trained the model using data from the U.S. East Coast and Gulf Coast and found that it could accurately simulate flooding in nearby areas that lacked direct tide gauge measurements.

Key advantages of the LSTM-SAM model include:

• High accuracy in forecasting water levels during storm events.
• Reduced computational cost compared to traditional hydrodynamic models.
• Greater accessibility, especially for communities with limited technical or financial resources.

This AI-driven approach has the potential to transform emergency planning and disaster response, allowing authorities to better predict, prepare for, and respond to flooding risks during hurricanes. The researchers hope the technology will support more equitable and data-informed decision-making, particularly in underserved.

Source: Environmental News Network