Improving hurricane modeling with physics-informed machine learning

In a paper published this week in Physics of Fluids, by AIP Publishing, a pair of researchers from the City University of Hong Kong employed machine learning to more accurately model the boundary layer wind field of tropical cyclones.

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

"We human beings are living in this boundary layer, so understanding and accurately modeling it is essential for storm forecasting and hazard preparedness," said author Qiusheng Li.

However, because air in the boundary layer interacts with land, the ocean, and everything else at surface level, modeling it is especially challenging. Conventional approaches to storm forecasting involve large numerical simulations run on supercomputers incorporating mountains of observational data, and they still often result in inaccurate or incomplete predictions.

In contrast, the author's machine learning algorithm is equipped with atmospheric physics equations that can produce more accurate results faster and with less data.

"Unlike traditional numerical models, our model employs an advanced physics-informed machine learning framework," said author Feng Hu. "Only a small amount of real data is required by our model to capture the complex behavior of the wind field of tropical cyclones. The model's flexibility and ability to integrate sparse observational data result in more accurate and realistic reconstructions."

Being able to reconstruct a tropical cyclone's wind field provides valuable data that experts can use to determine how severe the storm will be.

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

With a more detailed picture of what that wind field looks like, disaster authorities can better prepare for storms before they make landfall.

"With more frequent and intense hurricanes due to climate change, our model could significantly improve the accuracy of wind field predictions," said Hu. "This advancement can help refine weather forecasts and risk assessments, providing timely warnings and enhancing the resilience of coastal communities and infrastructure. "

The authors are planning to continue to develop their model and employ it to study different types of storms.

"We are planning to incorporate more observational data sources and improve the model's capability to handle the time evolution of winds," said Hu. "Expanding the application to more storm events across the world and integrating the model into real-time forecasting systems is also planned to enhance its utility for weather prediction and risk management."