Research from the National Weather Service and the National Hurricane Center show that approximately 90% of deaths from the US annual tropical system are caused by water-related incidents, primarily owning.
From 1963 to 2012, more than 2,500 Americans were killed by tropical cyclones.
In other words, flooding from tropical systems is very fatal, but can be difficult to predict when those coastal flooding occurs, especially on a very local scale.
David Munoz is an assistant professor at Virginia Tech in the Faculty of Civil Engineering and Environmental Engineering. He says there are multiple factors that contribute to flooding when hurricanes approach coastal areas – rising sea levels, land subsidence, storm surges, torrential rains.
“This combination of flood drivers produces an interesting phenomenon called synergy, because different drivers synergistically and synergistically have a greater effect than any of those flood drivers.”
Now, Munoz says the National Hurricane Centre is generating hurricane predictions five days ago.
“For example, it creates paths of hurricanes, wind, pressure. It produces all the parameters needed to run a fluid mechanics model, a custom model. Therefore, the US has different models on the Gulf and Atlantic coasts.
Peter is a Virginia Tech instrument
However, he says that these models solve complex physics equations.
“It takes a lot of time to simulate it, especially on the Gulf or Atlantic coast. It takes a lot of time and requires a supercomputer to do that.”
So Samuel Daramola, a graduate student at Vrije Universiteit Brussel in Belgium, and other collaborators, are working on potential solutions.
“So we trained models to learn from previous events, so our study focused on hurricanes that hit the Atlantic coast for about 40 years. [flooding.]”
A long-term long-term memory station approximation model, or deep learning model known as LSTM-SAM, attempts to identify patterns and examines conditions such as wind, pressure, rainfall, and wave height. Munoz said the model requires a large sample size, allowing them to “learn” more about current conditions.
“OK, I've seen this similar situation in the past, so this is what you'd expect. Obviously, you need to see if it actually works. In our study, we basically select some events and then train the model, then we'll be able to [used] There are a few other events just to test whether it can be captured. And the model worked very well. ”
Munoz emphasizes that traditional predictions that utilize physics equations I mentioned are always important as they produce the most important predictions. But Munoz hopes that the models his team has developed could become a key tool for predictors in the coming years.
“You can run fluid mechanics models and have hybrid models that have deep learning models of those. So, deep learning models will become future screening tools that “you can expect floods here.” And you can back it up [other] Model. ”
The team's model is quick and produces results in minutes – and only gets better.
“Once hurricane season passes, all we do is retrain the model, because obviously there is more data and we can use it for the next hurricane season.
According to Munoz, the model also features what is called “transfer learning.”
“What it does is [say]”OK, learn information from nearby stations. If there is no information in the coastal community in between, simply transfer that information to the area and you can predict it.”
The team will continue to use the LSTM-SAM framework during this hurricane season. Test the storm rolls near real time. We also made the source code available free of charge to small towns and communities in developing countries to access tools where detailed environmental information may not be available.
Thank you for checking out this edition of CommonWX – Radio IQ Weather and Climate Newsletter. Please use this link To send a newsletter to your inbox.
