Time is of the essence in tropical cyclone prediction: The more warning time a community has, the better prepared its members will be when a storm makes landfall. Currently, the path and nature of tropical cyclones can be predicted up to only five days in advance.
But new research published in Journal of Geophysical Research looked at a new way to forecast these storms that could increase that lead time to two weeks.
The researchers created a hybrid model for longer-range tropical cyclone forecasting, combining the strength and high resolution of the physics-based Weather Research and Forecasting (WRF) Model with the large-scale circulation and storm path modeling capabilities of a machine learning model called Pangu-Weather (Pangu).
WRF divides Earth’s surface into a grid with squares as small as 2 kilometers on a side, simulating the processes at play in a tropical cyclone’s evolution. However, physics-based models like WRF have some limitations in predicting storm intensity levels because they don’t always capture changing environmental factors such as sea surface temperatures or interactions with other storms.
Machine learning prediction models are better at forecasting tropical cyclone paths, but the 25-kilometer resolution of Pangu means it can miss smaller-scale variability within a developing storm.
To reduce these limitations, the researchers combined the approaches of the WRF and Pangu models. They conducted six experiments over the course of two weeks to test their modeling design. After adjusting their models, they tested their approach on 2023’s Freddy—the longest-lasting tropical cyclone on record—as a case study.
They found that the hybrid approach significantly improved the tracking and intensity forecasts compared with using only one modeling method. Their approach also extended accurate predictions from five to seven days and closely matched the path and intensity of Freddy for the whole two weeks.
The authors note that with more testing on tropical cyclone cases, their modeling approach could extend warning time to beyond two weeks, helping at-risk communities to better prepare in the face of major storms.
More information:
Liu et al, A hybrid machine learning/physics-based modeling framework for 2-week extended prediction of tropical cyclones. Journal of Geophysical Research: Machine Learning and Computation (2024). DOI: 10.1029/2024JH000207. agupubs.onlinelibrary.wiley.co … 10.1029/2024JH000207
This story is republished courtesy of Eos, hosted by the American Geophysical Union. Read the original story here.
Citation:
Physics meets machine learning for better cyclone predictions (2024, August 21)
