Indiana University Researcher Using Machine Learning to Study Hurricane Intensity Prediction

Newswise — Chanh Kieu is an associate professor in earth and atmospheric sciences in The College of Arts and Sciences at Indiana University Bloomington. His research currently focuses on the application of machine learning to understand hurricane intensity and prediction, trying to answer the question: How far in advance can you predict hurricane intensity? He’s also using satellite and climate data to build machine learning models to predict the chance of hurricane formation in different climates.

“Hurricane Milton is unique, in that it's the first one that has formed inside the Gulf of Mexico entirely, and we have not seen in the past any Category 4 or 5 hurricanes forming in the Gulf of Mexico that are this strong. The National Hurricane Center has done very well in predicting this hurricane -- They predicted the formation of Hurricane Milton several days in advance, as well as the track, and the intensity. So far, they’ve done a great job of predicting the hurricane intensity and its track: That is never easy. Over time, predictions have improved quite substantially.”

“I think Hurricane Milton is a very unique case. Its strength is really what everybody is talking about. However, people have had plenty of time to prepare: 4-5 days in advance – That is such progress as compared to the past. This is a very outstanding job by NOAA National Hurricane Center and operational centers.

“Unlike most other storms, which impact either one side of Florida or the other, Milton is impacting both coastal areas at the same time when it makes a landfall over the Peninsula, which is very different from other hurricanes. It’s also going to be very strong upon impact.”

 “The National Hurricane Center has been so accurate on both Hurricane Helene and Hurricane Milton. This is not a trivial thing. It's such a big achievement for not only the National Weather Service, but also other hurricane modelling systems, as 10 different models could all say the same thing. This is a whole suite of improvements that can produce such a good accurate prediction from the observation and modeling standpoint. Both Hurricane Helene and Milton have been well predicted, and so I feel so confident in the prediction by the National Hurricane Center, with a long-range prediction for 5-7 days forecast."

“I am following Milton closely because I want to see how my AI models capture the intensity fluctuation during Milton’s development. Will the intensity go up or down, and how precise can our predictions be for that intensity? There’s still some uncertainty in modeling, which is what I’m currently studying. Can I bring that uncertainty further down, or - What is the limit in the uncertainty that we can achieve for future hurricane predictions?

“Hurricane prediction requires many different components. You need to have good observation, you need to have good modeling, and you also need to have good theoretical understanding of what is going on during each hurricane development. Hurricanes are different from each other, so we try to understand systematic behavior so we can identify the bias in each hurricane intensity forecast and apply that for the future.

“And in the meantime, we also need to quantify the limit in our accuracy, so that when we give a warning to the public, we know our accuracy in advance. This is why for Hurricane Milton, we could say several days in advance that it was going to make landfall at a range of area, not one particular point. As it gets closer to land, we can get a more precise location, and the key to my research is to try to reduce such uncertainty of track/intensity prediction as much as possible."

More about Kieu’s research:

“My research focuses on modeling and understanding hurricane development and more recently, the application of machine learning for hurricane intensity prediction.

“I try to answer the question - how far in advance can you predict hurricane intensity and how it changes in different environments and climate conditions. So essentially, when you try to predict hurricane intensity, you want to see if it can be predicted three days in advance, five days or seven days in advance. And it's not that easy. Our main aim is to quantify the maximum accuracy we can get for hurricane intensity prediction and how it changes in different environments and climates."

“This is important because people want to know in advance how strong a hurricane could be, so they can have preparation. We are using satellite data and climate data in the past to train a machine learning model and try to understand what aspect of the climate that the machine learning can detect from the data and use that information to predict the chance of hurricane formation in the future."