How Open Science and AI Are Advancing Hurricane Research
32:21 Share

Every hurricane season, from June to November, meteorologists keep a close eye on satellite imagery of the Atlantic Ocean, the birthplace of hurricanes. These storms will start off the coast of Africa as something, you know, relatively small, just a group of clouds. Then they'll develop some circulation, and then just the warm water will create these huge storms.

That's Dean Legidagus. Every hurricane season, he's looking at the satellite data too. Because Dean is a hurricane hunter pilot. He works for NOAA, the National Oceanic and Atmospheric Administration. Once a storm gets close to the U.S., it's his job to fly directly into it.

So first off you get two reactions, like that's really cool or you're really dumb for doing that, right? And then the second question they ask is, so you guys, you fly around the storm? I said no, I mean, to get the data that we need, we have to fly right through the storm multiple times in multiple flights to collect the actual data that can't be collected via satellites. Once a hurricane spins up, satellites can't see through the cloud cover.

The only way to measure the storm's intensity and forecast its path is to get scientific instruments inside the hurricane itself. And that means flying through it. We have two aircraft at NOAA, the NOAA Aircraft Operations Center in Lakeland, Florida. And we fly into the storm almost 24 hours a day until the storm hits landfall. This is NASA's Curious Universe.

Today on the show, we're riding along with Dean to see what it's like to fly inside a hurricane to collect the data that makes forecasting possible. And once we land, we'll follow the data that NOAA's hurricane hunters collect.

We'll learn how NASA and NOAA share our findings to ensure scientific research on disasters stays freely accessible to everyone, saving lives. And stick around, because near the end, we'll talk about how you, yes you, can get involved in NASA's Open Science Program. I would say about 40, 50, 60 miles out from penetrating the eye wall, we do what's called Set 5. And that's just, hey, you know, you're on American Airlines and they hit the seatbelt light.

That's like everybody get in, strap in, you know, we're about to go through the eye. That's the eye of a hurricane. We're aboard Kermit, one of NOAA's two P-3 Orion turboprop aircraft. This plane and its sister aircraft, named Miss Piggy, are state-of-the-art flying laboratories. They're full of high-tech instruments and a crew of meteorologists.

Now, brace yourself, because we're about to fly directly into the ring of fire, the storm's eye wall. We call it the ring of fire because we're looking at a radar. Right at that eye wall, you're getting red and pink on the radar, right? And I'm a pilot, right? I don't know, especially coming to this, I didn't know a lot about radar reflectivity, but I knew flying into red on a radar was really bad.

Don't worry, Dean has done this hundreds of times. The team takes every precaution to keep themselves and their aircraft safe. But we're about to experience some next-level turbulence. So grab a seat, and hey, the light's still on, so keep that seatbelt fastened. And we're just going to do the best we can. We're going to ride the storm, right? Because you really can't fight it. You have to go with the ups and the downs. People say it's like riding on a roller coaster, you know, through a car wash, right? I'm going to grab my phone real quick.

You gotta keep these pockets zipped. Lots of lightning. You're probably getting hit by lightning multiple times flying through the storm. Lots of rain. It's loud in the flight station. It's not like deafening loud, but it being loud for three or four hours, it exhausts you. When you hear rain on a tin roof, right? But it's that times, you know, a hundred. It's hot.

Dean tries to keep the plane below about 8,000 feet, which is much lower than a normal commercial flight. That's the level where water freezes into ice. You're looking at a huge wall of water, a lot of precipitation,

If that's hitting the aircraft, it's not a big deal. But if it's frozen and it's hitting the aircraft, it's bad news. As we pass through the eye, the ride gets even rougher. You can't see anything outside. And the plane can drop thousands of feet in just seconds. You can't see anything. You're just like, you know, I'm looking at my attitude gauge and you're just trying to keep the wings level. That's when the hair on the back of my neck is like, okay, this can stop now, please, right? You know?

Loose equipment flies across the floor. There's a Kermit the Frog doll hanging in the windshield, and it's flying around on its string. We're breaking through the eye, right? And it's super bumpy, and what it can have is these what's called mesovortices in the eye wall, and that is mini tornadoes. So we're trying to avoid those as we go through the eye wall.

You're bouncing around everywhere, you know, you're trying to maintain the wings level, you can't see anything, and then you break out and it's just like, you know, it's quite beautiful, right? This is gonna change some lives, y'all. Change some lives.

Think from like, you know, the surface to 50,000 feet, you're getting this wall of clouds. And it looks like you're sitting inside of a stadium, right? It is the hand of God, pretty much, if you could imagine looking at that. Now, in the calm eye of the storm, the actual hurricane hunting begins. The term hurricane hunting comes from actually hunting the very center of the storm, looking for that zero wind speed point.

This is Rebecca Waddington. She's another hurricane hunter pilot. A lot of times when I tell people that we're hurricane hunters, they're like, aren't they pretty easy to find? I mean, we've got satellites. You can see them. You know where the storm is. But we're looking for that actual center point because that's going to be the initialization of the model. You need to know where it is to know where it's going. And a difference of a couple miles can make a huge difference on a five-day forecast.

Knowing where the center of the storm is allows meteorologists on the ground at the National Hurricane Center to predict the storm's path toward land. The planes fly through a hurricane multiple times on a given mission. Each time, they also drop these weather instruments through the floor. They're called dropsondes. You can think of them as weather balloons in reverse. They collect data as they fall down to the sea that helps scientists determine whether the storm is getting worse or calming down.

And while Dean flies into the storm aboard Kermit or Miss Piggy, he's not alone. Rebecca Waddington is high above him on NOAA's Gulfstream IV jet.

Also named, of course, for a Muppet, Gonzo. The P3s are like monster trucks, right? They are hefty, they are stout, they are built to take a beating. The G4 is like a luxury sports car. It is sleek, it is pretty, it is fast. And that's one of the things I love most about it. If you're trying to envision it, it's basically the plane that you see in movies that are flying around celebrities. It is a business jet.

But unlike a business jet that's used for, you know, business, this plane is stripped inside. There's no fancy couches, just science stations. And it has a huge radar in its nose. Hence the name Gonzo.

So that makes the nose look a little funny. And then in the back, we have a tail Doppler radar. So it extends the tail underneath the rudder. So it just looks like we've got a little junk in our trunk. But that radar is really important. It acts like an MRI, so you're getting a full vertical picture of the storm. The G-4 jet flies high above the storm and fast, covering thousands of miles at a time.

So we're looking at the storm itself, but we're also looking at the environment that the storm is moving into. And that's something that the P3 doesn't get because their bread and butter is the storm itself. So they fly directly out to the storm, they do their mission in the storm, and then they fly home. We're out there doing lawnmower patterns, you know, for four hours before we get to the storm or after we leave the storm because we want to know what the environment is and how it's going to affect the track.

Now, hurricanes love wet air and hate dry conditions. So knowing what lies ahead of the storm is super important to determining if it'll get stronger or weaker. Essentially, our mission is to collect data in the data sparse regions over the ocean. That way we can replace assumptions in forecast models with actual information.

And Dean and Rebecca can see the impact of their efforts right away. If you're a weather nerd, you may have heard of spaghetti models. These are the initial hurricane forecast models that predict the storm tracks.

Before the planes go out, these plots look like a messy bowl of spaghetti, with noodles predicting the hurricane's path going out in many different directions. One of the things we always look forward to is seeing the next iteration of the spaghetti plot, because all those noodles will get closer together. And then we go and fly a second flight, and they get closer together. So that is really cool to see, and we've seen that with every storm system that we've been in.

Taken together, the data from Gonzo, Miss Piggy, and Kermit make accurate hurricane forecasts possible. And for the pilots, this work can be personal. We live in Tampa and Lakeland, Florida, so every storm season it's inevitable that we're flying into a storm that's going to hit where we live. In fact, when Hurricane Milton made landfall in Florida in October 2024, it passed just south of the Hurricane Hunter's home base and actually damaged some of the team members' homes.

Dean's mom lives in Pensacola on the coast of Florida. Hurricanes have damaged her house several times. Yeah, you know, I've landed from a storm fight. And my mom called me and said, hey, you know, I'm up to my chest in water. Like, yeah, I don't know what's going to happen, that type of thing. And you're just like, man, OK, well, I'm supposed to fly tomorrow in this storm. Like, what do I do? Now, Dean's mom was OK. That time she was evacuated by the National Guard. But this is why hurricane hunters do what they do.

They have their families in mind, and the families of everyone who lives in the path of hurricanes. And they know the difference their work makes. Us flying the storm, 15 to 20% increase in track and forecast accuracy is what we do, right? So, I mean, that's the difference of not evacuating Pensacola or evacuating Pensacola. They know that meteorologists are counting on them to collect data that they can't get any other way. They are relying on our data, right? It's pretty awesome because we do...

a post-flight brief, and the forecasters will send us a message back saying, hey, we really appreciate the data. Off of the data that you just gave us, we made it a hurricane, or we made it a Cat 5, or we know that this thing's going to swing to the east or the west. I think that is extremely beneficial to say, hey, okay, we flew into this environment that's terrible from 2 to 6, 7 a.m. this morning. I've been up since last night. I didn't get any sleep, but this is why we're doing it, right?

It's a big responsibility. Thanks to NASA's commitment to open science, once NASA and NOAA satellites and planes collect hurricane data, meteorologists can access it right away. They plug that data into complicated equations that explain the physics of how our atmosphere works as best as we understand it right now. Now, it's complex stuff, and it takes some really serious computing power. And then those models project atmospheric conditions into the future.

Bit by bit, with each new batch of data, the forecasts get better. And then, communities on the ground get a better sense of how much rainfall and what wind speeds to expect. And whether they'll be hit by flooding or landslides. The problem is, our understanding of physics and of hurricanes is not complete

Scientists still have big questions about how they form in the first place and the factors that can stop them in their tracks. Like, for example, a mysterious layer of Saharan dust that circles the world in the atmosphere. Not to mention climate change, which is warming the ocean, adding more energy to the system, and causing storms to spin up more frequently and more quickly than ever before.

So while NOAA and the National Hurricane Center handle operational forecasts for hurricanes using the tried-and-true traditional methods, it's NASA's job to explore new ways to understand storms. To do that, we're plugging all of our data into a technology you might have been hearing a lot about in the news: artificial intelligence and neural networks. You know, instead of using physics, basically you are learning things from the data itself.

And you're hoping that the network can actually learn the underlying physics and things that we even don't know about well, that it can do those predictions in the future. That's Rahul Ramachandran. He leads NASA's AI work from NASA's Office of the Chief Science Data Officer. Now, NASA has collected Earth science data for more than 60 years. So we have a pretty huge archive.

We make that data accessible to everyone for free to benefit humanity. Rahul's job is building tools to make that sharing easier. One is a new AI model released in fall 2024 that looks for patterns in NASA's Earth data. It all started a few years ago. You've probably heard about ChatGPT, the Generative AI Large Language Model.

It's built around what's called a foundation model, which is a large-scale base model that can be adapted to chat with you, write computer code, and so on. So the base model is the same, but you can adapt it to do different kinds of tasks. Now, NASA has been experimenting with AI for decades, mainly to schedule missions for our faraway planetary rovers and to sift through images of exoplanets. But these foundation models are a new technology.

So Rahul teamed up with IBM to start building a language model for NASA to handle the complicated scientific terms that researchers here use. Then he said, "Wait a minute.

Language models look at sequences of words to make sentences. So why wouldn't the same process work with sequences of measurements of, say, temperature or pressure? You could make a foundation model that works on NASA data instead of words. If you're like us, you've been hearing a lot about AI recently, but less about how it actually works. So let's pull back the curtain a bit and get into it. Language models need to be trained on lots of written text to work well.

To train a language model, you feed it sentences and cover up or mask one word in that sentence. The model looks at the context of the sentence and it tries to figure out what the missing word is. You do that over and over with different sentences and missing words. And the model learns how sentence structure works. It basically learns to read and write. On the science side, it's very much the same thing.

It's the same approach, but instead of words, what you do is you make patches in your data. So you black out patches in your data and it has to learn, you know, the missing patches in your data set. For example, you might block out some temperature data in a town in Florida. And based on what the AI knows about the surrounding area, it can fill in the missing temperature records. That turns out to be good training for the real world where there are always big gaps in the data.

That's how these models are trained, by feeding it large amounts of data and then it going through this process of learning. And once it has learned enough, then you can adapt it for your particular task. NASA released its model, which is officially called the Prithvi Weather and Climate AI Model, in September 2024.

This base foundation model, with all its knowledge from NASA data, can now be adapted to work on entirely different problems with just a few examples to tell it what you're looking for. Since the model is open access, anyone can use it no matter where they are in the world and participate in NASA science. To test the model, the team chose hurricanes. How well does the model do in predicting out these extreme phenomena, both in terms of track and intensity, right?

The nice thing about hurricanes is it's very well observed. There's actually a really good existing data set where we actually have observations of different hurricanes with their tracks as well as their intensity with actual observations. So there's a really good benchmark for any model to use that data to test against those observations. And that, of course, is thanks to the hurricane hunters.

It's their data that makes validating this model possible. In the team's tests, the NASA General Purpose AI model did a better job at forecasting hurricanes than other AI models designed specifically to do forecasts. AI models aren't being used to forecast hurricanes at this point. This NASA climate model is just a research tool.

But Rahul says in the future, AI could supplement traditional forecasting methods. After the model is trained, it takes a lot fewer resources to run than physics models. So anybody can use it on their own computer. You don't need a supercomputer. And these AI models think differently than human meteorologists. I think it is coming. You know, AI models will supplement the physics-based models that we have, right? So we'll have data-driven models.

as well as our known physics model and they will be feed one into each and the other. And things that you learn from the physics based model can be used to improve the AI based model. It's going to be yet another tool in the tool belt for scientists and researchers, you know, having both physics based models and AI based models.

In a world where greenhouse gas emissions have changed the climate, warming the oceans so much that they supercharge hurricanes, we need all the tools we can get. And what's better than one that can think on its own, finding patterns in the data, and learning the physics we already know, and maybe even some we've missed? Now, once a hurricane has made landfall, there's one more way NASA's Earth Science Data and Open Science approach come into play: response and recovery.

Keri Roller works for NASA's Disasters Program, but she started out as a meteorologist. She was a TV weathercaster, and she even interned with the NOAA hurricane hunters in college. And as somebody who I would not say flying is my favorite thing, I would say that flying into hurricanes is not something that I would jump at to do again, but I'm very glad that I got the experience. If you're thinking to yourself, NASA has a disasters program? Well, you're not alone.

A lot of people don't think of disasters at all when they think of NASA. And being from Florida, I think of rockets. So I completely understand that perspective. I grew up watching rocket launches and space shuttle launches on the Space Coast. Now, that Space Coast is right in Hurricane Central. NASA is at the mercy of natural disasters just like everyone else. The Kennedy Space Center in Florida has weathered its fair share of storms.

Most recently, that included Hurricane Milton, which delayed the launch of our Europa Clipper mission. But we also have unique Earth observations that can help people respond to those disasters. That's the job of the Disasters Response Coordination System, or DRCS.

So the DRCS and the Disasters Program works with governments and nonprofit organizations across the globe. And a lot of our focus is on the people aspect. We are looking to help people after disasters, which is their most vulnerable hour. And we want to make sure that we're getting them the data products that they need when they need it. We are seeking to provide the best possible Earth observation data to people who need it most.

So here's how it works. When a disaster strikes, like a hurricane, an earthquake, even an oil spill, state governments, federal agencies like FEMA, or nonprofits like the Red Cross can request that NASA activate its Disasters Response System. Then, the NASA team can provide the data and maps they need.

This happened during Hurricane Beryl, for example, which made landfall in July 2024. Hurricane Beryl was a rarity in that it was the earliest Category 5 in the Atlantic Basin ever, which is a scary thought. And we were only in early July to have a huge Category 5 storm, you know, barreling towards coastlines with lots of people. And we were able to activate for this following a request from the Texas Department of Emergency Management.

Folks in Texas wanted to know where power outages were after the storm, which turns out to be something that NASA satellites can see at night. And this nighttime power product actually shows us where the power is still out. So we can lay that on top of census data. We can actually pinpoint areas where populations are most vulnerable and target or assist organizations in that area in targeting aid to those specific locations.

Carey and the rest of the team heard the data came in handy. One of our requesters actually took our data to the commissioner's office and said, look at this. The power companies cannot tell us where the power is out. But NASA can. NASA data can. And we're utilizing this to tell us where we need to send aid, basically. Hurricane Beryl also stirred up tornadoes. Carey's team provided satellite imagery that helped the National Weather Service in Louisiana study them.

And the Hurricane Hunters team played a role in Beryl too. The data they brought in proved that the hurricane was a Category 5. It's now the earliest Category 5 hurricane ever. And although Beryl broke records, it won't be the last storm to do so. Since April 2023, global sea surface temperatures have consistently stayed higher than any other time in recorded history. And that means storms that are stronger and happen more frequently.

And we're very cognizant of the fact that disasters are happening at an increasing rate. We're seeing stronger hurricanes. We're seeing earlier stronger hurricanes. Just that increasing rate is really alarming for those of us who are working on the program. When it hit Florida in October 2024, Hurricane Milton spurred 126 tornado warnings, which is a record for Florida. It was also the quickest ever to intensify to a Category 5 in the Gulf of Mexico.

In the aftermath, NASA's Disasters Program sprang into action again, working with FEMA and state emergency management agencies. They used satellite imagery and flew uncrewed aircraft over Florida to detect power outages and map flooding. As our climate continues to change, NASA's Open Science approach will make sure forecasters and local first responders have the data they need to keep people safe.

Now, we have covered a lot of ground in this episode. Hurricane hunters, weather forecasts, AI models, and disaster damage maps. And you might have picked up on a common thread that runs through all of that. Open science. All of the data that NASA collects, from satellites in space to field stations on the ground, it's available for free to the public. So let's dive into the world of open science and how you can get involved with Kevin Murphy. ♪

Kevin is NASA's Chief Science Data Officer. He leads AI and data science efforts for the Science Mission Directorate and makes sure everyone can access our science. He also leads a program called TOPS, which stands for Transform to Open Science. Hey, Kevin.

Hey, how are you doing? Great. Well, let's start at the beginning. What is open science? Open science is really the principle and the practice of making sure that everybody can participate in science, has access to the publications, the data, and the software necessary to do the scientific activities.

And generally makes that information more broadly and equitably accessible to a lot of people. So Kevin, on Curious Universe, we've talked about citizen science, which is a way that anyone can get involved in NASA science. What's the difference between citizen science and open science?

When we talk about open science, we talk about four primary pillars which enable things like citizen science to happen and be more effective. The first pillar is that the data that we collect has to be available with no period of exclusive access. So that means that, you know, we launch a mission or we do a scientific study and once that's concluded, once that data is initially collected, so we make a commitment to making that information immediately available to everybody.

And that has real-world contexts, especially in things like earth science or space weather, where people can use these data that we collect for a lot of different purposes. Everything from hurricane prediction to disaster response to forest fires to solar storms. The second really big one is that we commit to making our software open.

And that's really critical to make sure that other people can reuse it without having to do the same work. The third thing is that we commit to supporting publications or scientific publications without paywalls.

So people don't have to have a subscription to the scientific journal to see the NASA-funded research or the taxpayer-funded research. So no paywalls for scientific publications. And the last one is we commit to making sure that there is public participation in the scientific process, including the scientific meetings. And I know you guys have a curriculum for TOPS to get started. So what can I learn through that curriculum?

As people have begun to use open data and open software more readily,

You know, it's very difficult for them sometimes to figure out what tools are available, so what licenses to use, how to share things effectively, how to get credit for the things that they do to support science, whether that's develop software, write papers, format data, or even, you know, run conferences or challenges. And what the TOPS training does is really gives people the tools necessary to operate in that kind of open science world.

And is that something, you know, if I'm not a scientist, say if I'm a student or someone like that, is there something in there for me that I can learn? Absolutely. Absolutely. So the first module is called Ethos of Open Science. And that really kind of describes how people work with open science, why it's important and how we can do discovery much faster if we do it together as opposed to alone.

Then we have modules that look at how we do open software development, how we do open data publications, how we do some ethics, and how we do open collaborations in effective ways. Okay, Kevin, I've got one last thing for you. The name of our show is Curious Universe, so I have to ask you the question we ask everybody we interview. What are you still curious about? Oh, I'm curious about a lot of things, so I don't know if we have enough time left in this podcast to have that discussion.

But if I look at work, you know, I'm really interested in how we use a lot of the modern data science techniques to analyze the hundreds of petabytes of data that we have about the Earth, about the Sun, about the universe.

and really how we can push the technology forward so that we can make discoveries, even from information that we've already collected from other missions and instruments. So I'm curious to how we can harness those technologies, how we can broaden the participation in developing the scientific expertise to analyze that information and really bring that out to the public so that they can also be curious in everything that we do. Yeah, it's important work.

Kevin, thanks so much. Thank you very much for having me. If you want to learn more about NASA's Open Science and get involved yourself, search online for Open Science at NASA or head to science.nasa.gov slash open-science. This is NASA's Curious Universe. This episode was written and produced by Christian Elliott.

Our executive producer is Katie Konins. The Curious Universe team also includes me, Jacob Pinter, Mattie Olson, Michaela Sosby, and of course, Patti Boyd. Christopher Kim is our show artist. Our theme song was composed by Matt Russo and Andrew Santaguida of System Sounds. Special thanks to Scott Brown at NASA's Goddard Space Flight Center for helping us understand hurricane science. And if you want to check out NASA's disaster maps for yourself, go to maps.disasters.nasa.gov.

Also, if you're curious about the AI model that Rahul talked about, its full name is the NASA and IBM Prithvi Weather and Climate Foundation Model. So type that into your search bar. As always, if you enjoyed this episode of NASA's Curious Universe, please let us know. Leave us a review. Share the show with a friend. And remember, you can follow NASA's Curious Universe in your favorite podcast app to get a notification each time we post a new episode.

You know, it looks like a seatbelt light from the late 70s, right? And it makes a ding sound when you hit it, right? It's funny, most of the flight we're searching for times when we can take it off because, you know, eight hours, people have to go to the bathroom or they want to get a cup of coffee or you just want to stretch your legs. So you think, okay, this isn't that bad. And you'll turn it off and then you hit something big and then you turn it right back on and people are like, oh, geez. Three, two, one. This is an official NASA podcast.