In the midst of hurricane season, the need for accurate, timely and trustworthy tools to handle natural disaster forecasting and management is ever-present. In the United States alone, extreme weather events, including hurricanes, floods and wildfires, “kill hundreds of people and cause billions of dollars in damage” every year, with their strength and frequency slated to continue increasing.
A team of Louisiana State University scientists recently came out with a new machine learning framework, called Prediction-to-Map, or P2M, that “produces accurate flooding predictions more than 100,000 times faster than the sophisticated numerical models” that the team had previously developed, and “can be carried out on a laptop and finish a 72-hour simulation in 4 seconds.” The sophistication and accuracy of the framework result from the use of both numerical modeling and artificial intelligence (AI) together, training the tool on “information from a process based numerical model combined with observational data from a specific area, to create rapid, accurate flooding predictions, for up to a six hour timeframe.”
The team of researchers tested the framework’s accuracy by recreating the events of Hurricane Nicholas and comparing it to the outputs of “a more energy- and time-intensive numerical model,” finding that P2M slightly surpassed the numerical model in prediction accuracy. The P2M framework also demonstrated promising results in a modeling technique that can be used to predict compound flooding, or what can happen during a hurricane. P2M is both more efficient and accurate than traditional methods.
There are also several other use cases with similar success results in terms of AI use in prediction modeling/forecasting and response measures. During 2024’s Hurricane Milton, Houston-based meteorologist Matt Lanza noted that a lot of different types of models, including AI, predicted where it would most likely land, but the AI modeling “picked up on that potential outcome a good 12 to 18 hours before a lot of the other modeling.” Speed and accuracy are key in the weather prediction industry, and AI integration has been a positive step towards lifesaving measures. So much so that the National Hurricane Center (NHC) Deputy Director Jamie Rhome recognized that “the presence of AI in their workflow is a pillar of the hurricane forecaster’s success.”
Meteorologist Lanza notes that the models are not yet perfect. At the moment, they “tend to underestimate the intensity of hurricanes and sometimes struggle with gauging precipitation.” Because of the possibility of errors such as these, there is a broad consensus across the natural disaster prediction community that AI should not and will not take over the job of a human. Lanza continues, “We’re not turning the reins over to these things and just saying, ‘make me a forecast and I’ll just regurgitate it. You have to still look at the broader spectrum of tools available to you.”
In line with this thinking, AI is just one of many tools being used in conjunction with one another to improve speed, accuracy and response. Professor Z. George Xue, one of the LSU researchers behind the P2M framework, also emphasized the “strong need for robust numerical models, as well as the observational data provided by federal research infrastructure, such as buoys, gauges and weather models.”
Building trust in these AI models is essential to greater adoption, and transparency throughout the relaying of AI “data, training, evaluation, and limitations,” and continued experimentation are key to this effort. The powerful applications of AI continue to be discovered daily, and by leading with a pro-innovation mindset, Americans can continue to reap the life-altering benefits.