Hurricanes are America’s most destructive natural hazards, causing more deaths and property damage than any other type of disaster. Since 1980, these powerful tropical storms have caused more than $1.5 trillion in damage and killed more than 7,000 people.
The number one cause of damage and deaths caused by hurricanes is storm surge.
Storm surge is a rising ocean water level, caused by a combination of powerful winds pushing water toward the coast and reduced air pressure inside the hurricane compared to the pressure outside it. In addition to these factors, waves breaking near the coast cause sea level to rise near the coast, a phenomenon we call wave formation, which can be an important component of storm surge.
Accurate storm surge predictions are critical so that coastal residents have time to evacuate and emergency services have time to prepare. But high-resolution storm surge forecasts can be slow.
As a coastal engineer, I study how storm surge and waves interact with natural and man-made features on the ocean floor and coastline and ways to mitigate their impact. I have used physics-based models for coastal flooding and have recently been exploring ways that artificial intelligence can improve the speed of storm surge forecasting.
How storm surges are forecast today
Today, operational storm surge forecasts are based on hydrodynamic models, which are based on the physics of water flow.
These models use current environmental conditions (such as the speed with which the storm is moving toward the coast, the speed and direction of the wind, the timing of the tide, and the shape of the seafloor and landscape) to calculate the height of the projected surge and determine which locations are most at risk.
Hydrodynamic models have improved substantially in recent decades and computers have become significantly more powerful, so that rapid low-resolution simulations over very large areas are possible. However, high-resolution simulation that provides neighborhood-level details can take several hours to run.
Those hours can be critical for at-risk communities to evacuate safely and for first responders to adequately prepare.
To forecast storm surge over a wide area, modelers divide the target area into many small parts that together form a computational grid or mesh. Image pixels in an image. The smaller the pieces or grid cells, the higher the resolution and the more accurate the forecast. However, creating many small cells over a large area requires more computing power, so forecasting storm surges takes longer.
Forecasters can use low-resolution computer grids to speed up the process, but that reduces accuracy, leaving communities with more uncertainty about their flood risk.
AI can help accelerate that.
How AI can create better forecasts
There are two main sources of uncertainty in storm surge predictions.
One involves the data entered into the computer model. The storm track and wind field of a hurricane, which determine where it will make landfall and how intense the surge will be, are still difficult to accurately forecast more than a few days in advance. Changes to the coastline and seabed, such as channel dredging or the loss of salt marshes, mangroves or sand dunes, can affect the resistance that storm surge will face.
The second uncertainty involves the resolution of the computational grid, on which the mathematical equations of surge and wave motion are solved. Resolution determines how well the model sees and accounts for changes in landscape elevation and land cover, and at what granularity the physics of storm surge and hurricane waves is resolved.
AI models can produce detailed predictions more quickly. For example, engineers and scientists have developed AI models based on deep neural networks that can predict water levels along the coast quickly and accurately using data about the wind field. In some cases, these models have been more accurate than traditional hydrodynamic models.
AI can also develop forecasts for areas with little historical data or be used to understand extreme conditions that may not have occurred there before.
For these forecasts, physics-based models can be used to generate synthetic data to train AI on scenarios that might be possible but have not actually happened. Once an AI model is trained with historical and synthetic data, it can quickly generate surge forecasts using details about wind and atmospheric pressure.
Training AI with data from hydrodynamic models can also improve its ability to quickly generate flood risk maps that show which streets or homes are likely to flood in extreme events that may have no historical precedent but could occur in the future.
The future of AI for hurricane forecasting
AI is already being used to a limited extent in operational storm surge forecasting, primarily to augment commonly used physics-based models.
In addition to improving those methods, my team and other researchers have been developing ways to use AI to predict storm surge using observed data, assessing damage after hurricanes, and processing camera images to infer flood intensity. This can fill a critical gap in the data needed to validate storm surge models at granular levels.
As artificial intelligence models rapidly spread into all aspects of our lives and more data becomes available to train them, the technology offers potential to improve the forecasting of hurricanes and storm surges in the future, giving coastal communities faster and more detailed warnings about risks along the way.
Navid Tahvildari is an associate professor of civil and environmental engineering at Florida International University.
This article is republished from The Conversation under a Creative Commons license. Read the original article.