On 28 August 2005, New Orleans area residents received a bulletin from the National Weather Service (NWS) office in Slidell, La., warning them of “a most powerful hurricane with unprecedented strength.” One excerpt of the chilling announcement, issued via NOAA radio and the Federal Communications Commission’s Emergency Alert Service, read,
BLOWN DEBRIS WILL CREATE ADDITIONAL DESTRUCTION. PERSONS…PETS…AND LIVESTOCK EXPOSED TO THE WINDS WILL FACE CERTAIN DEATH IF STRUCK.
POWER OUTAGES WILL LAST FOR WEEKS…AS MOST POWER POLES WILL BE DOWN AND TRANSFORMERS DESTROYED. WATER SHORTAGES WILL MAKE HUMAN SUFFERING INCREDIBLE BY MODERN STANDARDS.
Hurricane Katrina, which caused 1,833 fatalities and about $108 billion in damage (more than $178 billion in 2025 dollars), remains the costliest hurricane on record to hit the United States and among the top five deadliest.
“If we were to have a Katrina today, that [forecast] cone would be half the size that it was in 2005.”
In the 20 years since the hurricane, meteorologists, modelers, computer scientists, and other experts have worked to improve the hurricane forecasting capabilities that inform bulletins like that one.
Consider the forecast cone, for instance. Also known as the cone of uncertainty, this visualization outlines the likely path of a hurricane with decreasing specificity into the future: The wider part of the cone might represent the forecasted path 36 hours in advance, and the narrower part might represent the forecasted path 12 hours in advance.
“If we were to have a Katrina today, that cone would be half the size that it was in 2005,” said Jason Beaman, meteorologist-in-charge at the National Weather Service Mobile/Pensacola office.
How to Make a Hurricane
The ingredients for a hurricane boil down to warm water and low pressure. When an atmospheric low-pressure area moves over warm ocean water, surface water evaporates, rises, then condenses into clouds. Earth’s rotation causes the mass of clouds to spin as the low pressure pulls air toward its center.
Storms born in the Gulf of Mexico or that traverse it, as Katrina did, benefit from the body’s sheltered, warm water, and the region’s shallow continental shelf makes storm surges particularly destructive for Gulf Coast communities.
Hurricanes gain strength as long as they remain over warm ocean waters. But countless factors contribute to how intense a storm becomes and what path it takes, from water temperature and wind speed to humidity and proximity to the equator.
Because predicting the behavior of hurricanes requires understanding how they work, data gathered by satellites, radar, and aircraft are crucial for researchers. Feeding these data into computer simulations helps researchers understand the mechanisms behind hurricanes and predict how future storms may behave.
“Since 2005, [there have been] monumental leaps in observation skill,” Beaman said.
Seeing a Storm More Clearly
Many observations of the weather conditions leading up to hurricanes come from satellites, which can offer a year-round bird’s-eye view of Earth.
NOAA operates a pair of geostationary satellites that collect imagery and monitor weather over the United States and most of the Atlantic and Pacific oceans. The mission, known as the Geostationary Operational Environmental Satellite (GOES) program, has been around since 1975; the current satellites are GOES-18 and GOES-19.
When Beaman started his career just a few years before Katrina hit, satellite imagery from GOES-8 to GOES-12 was typically beamed to Earth every 30–45 minutes—sometimes as often as every 15 minutes. Now it’s routine to receive images every 5 minutes or even as often as every 30 seconds. Having more frequent updates makes for much smoother animations of a hurricane’s track, meaning fewer gaps in the understanding of a storm’s path and intensification.
For Beaman, the launch of the GOES-16 satellite in 2016 marked a particularly important advance: In addition to beaming data to scientists more frequently, it scanned Earth with 4 times the resolution of the previous generation of satellites. It could even detect lightning flashes, which can sometimes affect the structure and intensity of a hurricane.
The transition to GOES-16 “was like going from black-and-white television to 4K television.”
The transition to GOES-16 “was like going from black-and-white television to 4K television,” Beaman said.
NOAA also has three polar-orbiting satellites, launched between 2011 and 2017, that orbit Earth from north to south 14 times a day. As part of the Joint Polar Satellite System (JPSS) program, the satellites’ instruments collect data such as temperature, moisture, rainfall rates, and wind for large swaths of the planet. They also provide microwave imagery using radiation emitted from water droplets and ice. NOAA’s earlier polar-orbiting satellites had lower resolution at the edges of scans, a more difficult time differentiating clouds from snow and fog, and less accurate measurements of sea surface temperature.
“With geostationary satellites, you’re really just looking at the cloud tops,” explained Daniel Brown, branch chief of the Hurricane Specialist Unit at NOAA’s National Hurricane Center in Miami. “With those microwave images, you can really kind of see into the storm, looking at structure, whether an eye has formed. It’s really helpful for seeing the signs of what could be rapid intensification.”
NOAA’s Geostationary Operational Environmental Satellites (GOES) monitor weather over the United States and most of the Atlantic and Pacific oceans. Credit:
NOAA/Lockheed Martin,
Public Domain
Rapid intensification is commonly defined as an increase in maximum sustained wind speed of 30 or more nautical miles per hour in a 24-hour period. Katrina had two periods of rapid intensification, and they were one reason the storm was so deadly. In the second period, the storm strengthened from a low-end category 3 hurricane (in which winds blow between 178 and 208 kilometers per hour, or between 111 and 129 miles per hour) to a category 5 hurricane (in which winds blow faster than 252 kilometers per hour, or 157 miles per hour) in less than 12 hours.
New Angles
Radar technology has also made strides in the decades since Katrina. Hurricane-tracking radar works via a ground- or aircraft-based transmitter sending out a radio signal. When the signal encounters an obstacle in the atmosphere, such as a raindrop, it bounces back to a receiver. The amount of time it takes for the signal to return provides information about the location of the obstacle.
Between 2011 and 2013, NWS upgraded its 150+ ground-based radars throughout the United States with dual-polarization technology—a change a 2013 NWS news release called “the most significant enhancement made to the nation’s radar network since Doppler radar was first installed in the early 1990s.”
So-called dual-pol technology sends both horizontal and vertical pulses through the atmosphere. With earlier technology, a radar signal might tell researchers only the location of precipitation. Dual-pol can offer information about how much precipitation is falling, the sizes of raindrops, and the type of precipitation or can even help researchers identify debris being transported in a storm.
Credit:
NOAA
“That’s not something that we had back in Katrina’s time,” Beaman said. In 2005, forecasters used “much more crude ways of trying to calculate, from radar, how much rain may have fallen.”
Radar updates have become more frequent as well. Beaman said his office used to receive routine updates every 5 or 6 minutes. Now they receive updated radar imagery as often as every minute.
Hunting Hurricanes from the Skies
For a more close-up view of a hurricane, NOAA and the U.S. Air Force employ Hurricane Hunters—planes that fly directly through or around a storm to take measurements of pressure, humidity, temperature, and wind speed and direction. These aircraft also scan the storms with radar and release devices called dropwindsondes, which take similar measurements at various altitudes on their way down to the ocean.
NOAA’s P-3 Orion planes and the 53rd Weather Reconnaissance Squadron’s WC-130J planes fly through the eyes of storms. NOAA’s Gulfstream IV jet takes similar measurements from above hurricanes and thousands of square kilometers around them, also releasing dropwindsondes along the way. These planes gather information about the environment in which storms form. A 2025 study showed that hurricane forecasts that use data from the Gulfstream IV are 24% more accurate than forecasts based only on satellite imagery and ground observations.
The NOAA P-3 Hurricane Hunter aircraft captured this image from within the eye of Hurricane Katrina on 28 August 2005, 1 day before the storm made landfall. Credit:
NOAA,
Public Domain
Hurricane Hunters’ tactics have changed little since Katrina, but Brown said that in the past decade or so, more Hurricane Hunter data have been incorporated into models and have contributed to down-to-Earth forecasting.
Sundararaman “Gopal” Gopalakrishnan, senior meteorologist with NOAA’s Atlantic Oceanographic and Meteorological Laboratory’s (AOML) Hurricane Research Division, emphasized that Hurricane Hunter data have been “pivotal” for improving both the initial conditions of models and the forecasting of future storms.
With Hurricane Hunters, “you get direct, inner-core structure of the storm,” he said.
Hurricane Hunters are responsible for many of the improvements in hurricane intensity forecasting over the past 10–15 years, said Ryan Torn, an atmospheric and environmental scientist at the University at Albany and an author of the recent study about Gulfstream IVs. One part of this improvement, he explained, is that NOAA began flying Hurricane Hunters not just for the largest storms but for weaker and smaller ones as well, allowing scientists to compare what factors differentiate the different types.
“We now have a very comprehensive observation dataset that’s come from years of flying Hurricane Hunters into storms,” he said. These datasets, he added, make it possible to test how accurately a model is predicting wind, temperature, precipitation, and humidity.
In 2021, NOAA scientists also began deploying uncrewed saildrones in the Caribbean Sea and western Atlantic to measure changes in momentum at the sea surface. The drones are designed to fill observational gaps between floats and buoys on the sea surface and Hurricane Hunters above.
Modeling Track and Intensity
From the 1980s to the early 2000s, researchers were focused on improving their ability to forecast the path of a hurricane, not necessarily what that hurricane might look like when it made landfall, Gopalakrishnan explained.
Brown said a storm’s track is easier to forecast than its intensity because a hurricane generally moves “like a cork in the stream,” influenced by large-scale weather features like fronts, which are more straightforward to identify. Intensity forecasting, on the other hand, requires a more granular look at factors ranging from wind speed and air moisture to water temperature and wind shear.
Storms like 2005’s Katrina and Rita “showed the importance of [tracking a storm’s] intensity, especially rapid intensification.”
Gopalakrishnan said storms like 2005’s Katrina and Rita “showed the importance of [tracking a storm’s] intensity, especially rapid intensification.”
Without intensity forecasting, Gopalakrishnan said, some of the most destructive storms might appear “innocuous” not long before they wreak havoc on coastlines and lives. “Early in the evening, nobody knows about it,” he explained. “And then, early in the morning, you see a category 3 appear from nowhere.”
Gopalakrishnan came to AOML in 2007 to set up both the Hurricane Modeling Group and NOAA’s Hurricane Forecast Improvement Project. He had begun working on what is now known as the Hurricane Weather Research Forecast model (HWRF) in 2002 in his role at NOAA’s Environmental Modeling Center. With the formation of the hurricane modeling group in 2007, scientists decided to focus on using HWRF to forecast intensity changes.
HWRF used a technique called moving nests to model the path of a storm in higher resolution than surrounding areas. Gopalakrishnan compared a nest to using a magnifying glass focused on the path of a storm. Though a model might simulate a large area to provide plenty of context for a storm’s environment, capturing most of an area in lower resolution and the storm path itself in higher resolution can save computing power.
By 2014, Gopalakrishnan said, the model’s tracking and intensity forecasting capabilities had improved 25% since 2007. The model’s resolution also upgraded from 9 square kilometers in 2007 to 1.5 square kilometers by the time it was retired in 2023.
Since 2007, the National Hurricane Center’s official (OFCL) track forecast errors decreased between 30% and 50%, and intensity errors shrank by up to 55%. MAE = mean absolute error; VMAX = maximum sustained 10-meter winds. Credit: Alaka et al., 2024,
https://doi.org/10.1175/BAMS-D-23-0139.1
Over time, advances in how data are introduced into models meant that the better data researchers were receiving from satellites, radars, and Hurricane Hunters improved modeling abilities even further. Gopalakrishnan estimated that by 2020, his office could predict hurricane track and intensity with somewhere between 50% and 54% more accuracy than in 2007.
NOAA began transitioning operations to a new model known as the Hurricane Analysis and Forecast System (HAFS) in 2019, and HAFS became the National Hurricane Center’s operational forecasting model in 2023. HAFS, developed jointly by several NOAA offices, can more reliably forecast storms, in part by increasing the use of multiple nests—or multiple high-resolution areas in a model—to follow multiple storms at the same time. HAFS predicted the rapid intensification of Hurricanes Helene and Milton in 2024.
Just as they did with HWRF, scientists run multiple versions of HAFS each year: an operational model, used to inform the public, and a handful of experimental models to see which of them work the best. At the end of hurricane season, researchers examine which versions performed the best and begin combining elements to develop the next generation of the operational model. The team expects that as HAFS improves, it will lengthen the forecast from the 5 days offered by previous models.
“As a developer [in 2007], I would have been happy to even get 2 days forecast correctly,” Gopalakrishnan said. “And today, I’m aiming to get a 7-day forecast.”
NOAA’s budget plan for 2026 could throw a wrench into this progress, as it proposes eliminating all NOAA labs, including AOML.
The Role of Communication
An accurate hurricane forecast does little good if the information isn’t shared with the people who need it. And communication about hurricane forecasts has seen its own improvements in the past 2 decades. NWS has partnered with social scientists to learn how to craft the most effective messages for the public, something Beaman said has paid dividends.
Communication between the National Hurricane Center and local weather service offices can be done over video calls, rather than by phone as was once done. Sharing information visually can make these calls more straightforward and efficient. NWS began sending wireless emergency alerts directly to cell phones in 2012.
In 2017, the National Hurricane Center began issuing storm surge watches and warnings in addition to hurricane watches and warnings. Beaman said storm surge inundation graphics, which show which areas may experience flooding, may have contributed to a reduction in storm surge–related fatalities. In the 50-year period between 1963 and 2012, around 49% of storm fatalities were related to storm surge, but by 2022, that number was down to 11%.
“You take [the lack of visualization] back to Katrina in 2005, one of the greatest storm surge disasters our country has seen, we’re trying to express everything in words,” Beaman said. “There’s no way a human can properly articulate all the nuances of that.”
Efforts to create storm data visualization go beyond NOAA.
Carola and Hartmut Kaiser moved to Baton Rouge, La., just weeks before Hurricane Katrina made landfall. Hartmut, a computer scientist, and Carola, an information technology consultant with a cartography background, were both working at Louisiana State University. When the historic storm struck, Hartmut said they wondered, “What did we get ourselves into?”
Shortly after the storm, the Kaisers combined their expertise and began work on the Coastal Emergency Risks Assessment (CERA). The project, led by Carola, is an easy-to-use interface that creates visual representations of data, including storm path, wind speed, and water height, from the National Hurricane Center, the Advanced Circulation Model (ADCIRC), and other sources.
The Coastal Emergency Risks Assessment tool aims to help the public understand the potential timing and impacts of storm surge. Here, it shows a forecast cone for Hurricane Erin in August 2025, along with predicted maximum water height levels. Credit:
Coastal Emergency Risks Assessment
“We know of a lot of people who said, ‘Yes, thank you, [looking at CERA] caused me to evacuate.”
What started as an idea for how to make information more user-friendly for the public, emergency managers, and the research community grew quickly: Hundreds of thousands of people now use the tool during incoming storm events, Hartmut said. The Coast Guard often moves its ships to safe regions on the basis of CERA’s predictions, and the team frequently receives messages of thanks.
“We know of a lot of people who said, ‘Yes, thank you, [looking at CERA] caused me to evacuate,” Hartmut said. “And now my house is gone, and I don’t know what would have happened if I didn’t go.”
Looking Forward
Unlike hurricane season itself, the work of hurricane modelers has no end. When the season is over, teams such as Gopalakrishnan’s review the single operational and several experimental models that ran throughout the season, then work all year on building an upgraded operational model.
“It’s 365 days of model developments, testing, and evaluation,” he said.
NOAA scientists aren’t the only ones working to improve hurricane forecasting. For instance, researchers at the University of South Florida’s Ocean Circulation Lab (OCL) and the Florida Flood Hub created a storm surge forecast visualization tool based on the lab’s models. The West Florida Coastal Ocean Model, East Florida Coastal Ocean Model, and Tampa Bay Coastal Ocean Model were designed for the coastal ocean with a sufficiently high resolution to model small estuaries and shipping channels.
Though Yonggang Liu, a coastal oceanographer and director of OCL, cited examples of times his lab’s models have outperformed NOAA’s models, the tool is not used in operational NOAA forecasts. But it is publicly available on the OCL website (along with a disclaimer that the analyses and data are “research products under development”).
The Cyclone Global Navigation Satellite System (CYGNSS) is a NASA mission that pairs signals from existing GPS satellites with a specialized radar receiver to measure reflections off the ocean surface—a proxy for wind levels. The constellation of eight satellites can take measurements more frequently than GOES satellites, allowing for better measurement of rapid intensification, said Chris Ruf, a University of Michigan climate and space scientist and CYGNSS principal investigator.
It might seem that if a method or mission offers a way to more accurately forecast hurricanes, it should be promptly integrated into NOAA’s operational models. But Ruf explained NOAA’s hesitation to use data from university-led efforts: Because they are outside of NOAA’s control and could therefore lose funding or otherwise stop running, it’s too risky for NOAA to rely on such projects.
“CYGNSS is a one-off mission that was funded to go up there and do its thing, and then, when it deorbits, it’s over,” Ruf said. “They [at NWS] don’t want to invest a lot of time learning how to assimilate some new data source and then have the data disappear later. They want to have operational usage where they can trust that it’s going to be there later on.”
“These improvements cannot happen as a one-man army.”
Whatever office they’re in, it’s scientists who make the work of hurricane forecasting possible. Gopalakrishnan said that during Katrina, there were two or three people at NOAA associated with model development. He credits the modeling improvements made since then to the fact that, now, there’s a team of several dozen. And more advances may be on the horizon. For instance, NOAA expects a new Hurricane Hunter jet, a G550, to join the ranks by 2026.
However, some improvements are stalling. The Geostationary Extended Observations (GeoXO) satellite system is slated to begin expanding observations of GOES satellites in the early 2030s. But the 2026 U.S. budget proposal, which suggests slashing $209 million from NOAA’s efforts to procure weather satellites and infrastructure, specifically suggests a “rescope” of the GeoXO program
Hundreds of NOAA scientists have been laid off since January 2025, including Hurricane Hunter flight directors and researchers at AOML (though NWS received permission to rehire hundreds of meteorologists, hydrologists, and radar technicians, as well as hire for previously approved positions, in August).
In general, hurricane fatalities are decreasing: As of 2024, the 10-year average in the United States was 27, whereas the 30-year average was 51. But this decrease is not because storms are becoming less dangerous.
“Improved data assimilation, improved computing, improved physics, improved observations, and more importantly, the research team that I could bring together [were] pivotal” in enabling the past 2 decades of forecasting improvements, said Gopalakrishnan. “These improvements cannot happen as a one-man army. It’s a team.”
—Emily Dieckman (@emfurd.bsky.social), Associate Editor
Citation: Dieckman, E. (2025), How researchers have studied the where, when, and eye of hurricanes since Katrina,
Eos, 106, https://doi.org/10.1029/2025EO250320. Published on 28 August 2025.
Text © 2025. AGU.
CC BY-NC-ND 3.0Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.
