Improving hurricane modeling through physics-based machine learning – Watts Up With That?

The algorithm reconstructs the wind field quickly and accurately with less observation data.

With Eurekalert, maybe this will work, maybe not. Worth a try.

Washington, November 19, 2024 — A hurricane or tropical cyclone can cause devastating natural disasters, leveling entire cities and claiming hundreds or thousands of lives. A key aspect of their destructive potential is their unpredictability. Hurricanes are complex weather phenomena and it is difficult to estimate their intensity or landfall location.

In a paper published this week fluid physicspublished by AIP Publishing, two City University of Hong Kong researchers used machine learning to more accurately simulate the boundary layer wind fields of tropical cyclones.

In atmospheric science, the atmospheric boundary layer is the region closest to the Earth's surface.

“We humans live in this boundary layer, so understanding it and accurately modeling it is critical for storm forecasting and disaster preparedness,” said author Qiusheng Li.

However, modeling the air in the boundary layer is particularly challenging because it interacts with land, oceans, and everything else at the surface. Traditional storm forecasting methods involve large numerical simulations run on supercomputers that contain large amounts of observational data, but they still often result in inaccurate or incomplete forecasts.

In comparison, the authors' machine learning algorithm, equipped with atmospheric physics equations, produces more accurate results faster and with less data.

“Unlike traditional numerical models, our model uses an advanced entity information machine learning framework,” said author Hu Feng. “Our model requires only a small amount of real-world data to capture the complex behavior of tropical cyclone wind fields. The model's flexibility and ability to integrate sparse observations allows for more accurate and realistic reconstructions.

Being able to reconstruct a tropical cyclone's wind field can provide valuable data that experts can use to determine the storm's severity.

“The wind field of a tropical cyclone contains information about the storm's intensity, structure, and potential impact on coastal areas,” Li said.

By understanding wind conditions in more detail, disaster authorities can better prepare before a storm makes landfall.

“As hurricanes become more frequent and intense due to climate change, our model can significantly improve the accuracy of wind field predictions,” Hu said. “This advancement can help improve weather forecasts and risk assessments, provide timely warnings and empower coastal communities and Infrastructure resilience.”

The authors plan to continue developing their model and use it to study different types of storms.

“We plan to incorporate more observational data sources and improve the model's ability to handle the temporal evolution of winds,” Hu said. “There are also plans to extend the application to more storm events around the world and integrate the model into real-time forecast systems to enhance its utility in weather forecasting and risk management.”

###

The article “Using Machine Learning to Reconstruct Tropical Cyclone Boundary Layer Wind Fields from Physical Information” was written by Hu Feng and Li Qiusheng. it will appear in fluid physics November 19, 2024 (DOI: 10.1063/5.0234728). After this date, it will be accessible at https://doi.org/10.1063/5.0234728.

Relevant