On Thursday, May 21, NOAA will hold a news conference at its Aircraft Operations Center in Lakeland, Florida, at 11:00 a.m. EDT to announce its official 2026 Atlantic Hurricane Season Outlook, the agency’s annual prediction of how active the coming storm season will be. Early forecasts from competing agencies, including Colorado State University and Tropical Storm Risk, are already calling for a somewhat below-normal season, with developing El Niño conditions expected to increase wind shear across the tropical Atlantic and suppress storm formation. But before the media takes NOAA’s numbers at face value and the inevitable “busier-than-normal season” headlines roll in, it’s worth asking: how good is NOAA actually at this? A look at 26 years of NOAA’s May seasonal outlooks versus what actually happened reveals a more nuanced picture; the agency hits its own stated forecast range roughly 69% of the time on named storms and hurricanes, just shy of its self-declared 70% confidence target, with a notable tendency to miss badly in the most extreme seasons, and a historical bias toward under-forecasting activity when the Atlantic is running hot.
Overall hit rate: NOAA’s May outlooks land within their stated range for named storms roughly 17 of 25 years; about 68%, just shy of their own 70% confidence target. The hurricane count accuracy is similar. That said, NOAA aims for a range (not a point forecast), so some “hits” are easier than others in wide-range years.
The big misses:
- 2005 was the most extreme failure — NOAA forecast 12–15 named storms; 28 formed, including a record 15 hurricanes. Katrina and Wilma made it a historically catastrophic season.
- 2020 was another blowout — the season blew past the forecast (30 named storms vs. a predicted 13–19), exhausting the alphabetical list for the second time ever.
- 2006 went the other direction — NOAA forecast an active season (13–16 storms) following 2005’s record activity, but El Niño suppressed the season to just 10 named storms and 5 hurricanes.
- 2013 was also a notable underperformance year — the forecast called for 13–20 storms but only 14 formed, and just 2 became hurricanes (the lowest since 1982).
Systematic bias: Research on NOAA’s seasonal outlooks found that forecasts for named storms have tended to run low on average, and that May outlooks have sometimes not verified within the outlook ranges at the target rate of 70%. In other words, NOAA has historically been more likely to under-forecast activity than over-forecast it, which matters given the trend of record-breaking seasons since 2017.
The direction call: Even when the exact numbers miss, NOAA generally gets the character of the season right, calling above-normal, near-normal, or below-normal correctly most of the time. That directional skill has improved over the 25-year period.
Year-by-year verdict — did actual named storms and hurricanes fall within NOAA’s forecast range?
For 2025, NOAA forecasted a 60% chance of an above-normal season with 13–19 named storms and 6–10 hurricanes — a somewhat more conservative range than 2024’s blockbuster outlook. 2025 is highlighted with a ★. It was a “hit” year for NOAA because the season produced 13 named storms and five hurricanes, falling within the predicted ranges for named storms, hurricanes, and major hurricanes. Source: National Oceanic and Atmospheric Administration
What made 2025 interesting was the story behind the numbers: despite a below-average number of named storms and hurricanes, the season had an above-normal accumulated cyclone energy rating of 130.8 units, and three Category 5 hurricanes formed; the second most of any year on record. So NOAA got the count right, but the intensity distribution was extreme. Tropical storms and hurricanes during the 2025 season were 50% more challenging to predict compared to average.
We’ll soon learn what they think about 2026.
Maybe they should just stick to weather.
Technically, and it is a bit of a stretch, this is weather forecasting, just not daily.
NOAA’s charter is oceans and atmosphere, so this is not outside their sandbox.
A possible reason that seasonal differences are running low could be to bolster the narrative of increasing storms. If the actual are constantly higher than predictions then the weather (and climate by extension) look worse.
It looks no better than a dice roll.
With 2 dice.
It looks like they could just have said 5 to 9 hurricanes each year and been right about two times out of three.
How about monkeys with darts
If they don’t know what causes rapid acceleration, they don’t know what causes formation.
If you ask them, they know.
If course that applies to any field with “experts.”
They have some of the clues identified and under study. But the search for the golden “control knob” misleads them down deep, dark rabbit holes.
I thought CO2 was the golden control knob 😉
If I say the term ‘Mandajulian oscillation’ three times fast I turn into a flawless forecaster.
People who believe in predictive weather models often find themselves living outside of reality.
Hourly is good, constant data updates for the locale. Beyond that, they might as well use tarot cards than “predictive” models.
I think the weather forecast accuracy for rain actually occurring where I live, drops below 50% 48 hours out. Why they still put out a 10 day forecast is beyond me.
NOAA wants to get some degree of legitimacy back? Stop naming every halfassed collection of clouds over water as “hurricanes”, stop naming every weather pattern, period.
+1,000,000
It’s fun looking at the SF evolution. It started based on landfall damage assessments to estimate wind speeds. Now, it is a hodgepodge and one, if one were so inclined, could chose one over another or just average them.
What counts to the people affected is landfall wind strength, duration over land, rainfall, and sometimes those girls go swimming again before strutting on a new beach.
The orange and pink are very close in hue and difficult to distinguish, even for non-colorblinds.
NO idea what you are talking about. Those colors don’t appear in either the graph or the table.
If you are seeing those colors, then your monitor is out of whack.
I had to scroll up and look. lol
We should not trust either their predictions, nor the results. Since 2017 they have been consistently over rating Hurricane and Tropical storm intensity as in wind speeds. They do this with far more overblown winds out at sea and reduce this fudging of the numbers as storms come ashore, because too many actual anemometers exist to debunk their fudging on land.
It is like a baseball game where the umpire is part of the pitching team. So of course the umpire (the so called measured data) will attempt to favor the pitching team’s case (the prediction, and the global warming propaganda)
There’s a limit to the tricks employed while producing plausible results- after they run out everything goes flat in trend at the overestimated level.
The above graph shows the models (dots) match the data fairly well without even knowing that the data is tropical systems. The models do miss the peaks of named storms, but otherwise fit the data. However, I think one could just as easily say each year that there will be 14 named storms and 8 hurricanes and draw such with a straight line and be just as accurate as any model.
“Approximately 1,200 to 1,300 degrees in atmospheric science and meteorology are awarded annually in the USA”
Not free, They gotta do something.
“Predictions are hard, especially about the future”. Yogi Berra I think, maybe paraphrased.
The 2025 season occurred under the highest CO2 levels in the modern era.
A significant portion of the ACE came from three massive Cat 5 hurricanes. Two of those were fish storms.
Can we have some more of that please?
(Unfortunately, Cat 5 Melissa did kill 95 people.)
If my job performance had been this bad I’d have had to find another line of work a long time ago, because I’d have been canned.
Why can’t they just admit that weather is so variable that SWAGs are the best they can do and are almost as worthless as a bucket of warm…oh, never mind…
> Those colors don’t appear in either the graph or the table.
I can see pink and orange on my Asus laptop with an external 16″ monitor…but only by copying the image and dropping it into my preferred image-editor IrfanView …then selecting a rectangular portion … tapping it, to see it expand into full-screen…voila…the lines are green solid, blue dotted, pink solid, and orange dotted.
Confusingly, the key to the colours show squares, but the data points are circles. The squares also show as pink and orange when expanded, as described above.
Thanks to Hans E. for indirectly making me Register. and make my first comment, to correct a perceived injustice ! 🙂
i comes before h?
Anthony:
The graph suggests a slight increase in number of named storms over the 26 years, but
nothing like how the climate alarmists wail about year after year. Note this ~70% prediction is for something that begins in one month [Atlantic: June- November]].
Call me when they can predict the timing & strength of El Nino/La Nina 2-3 years ahead of time. Lol
NOAA implemented their latest, major operational high-performance computing (HPC) system in June 2022. The system tripled NOAA computing capacity by addition of two, identical-twin supercomputers, named “Dogwood” and “Cactus”, each operating at 12 petaflops.
For reference, a petaflop is one thousand trillion, or one quadrillion, floating-point operations per second . . . a high-end consumer gaming computer with top-tier hardware can achieve something like 0.1 to 0.2 petaflops in 32-bit floating-point performance.
The two supercomputers, part of the Weather and Climate Operational Supercomputing System (WCOSS), were delivered under a contract with General Dynamics Information Technology valued at approximately $150 million for initial cost to build, house, and support the machines and having potential total contract ceiling of $505 million over 10 years for operational support costs and further upgrades.
In August 2023, the supercomputers were upgraded to operate at a speed of 14.5 petaflops each, at a cost of $100 million in federal funding. This investment supported the installation of the “Rhea” supercomputer, adding 8 petaflops to NOAA’s capacity, and bringing total capacity from 35 to 43 petaflops.
In September 2024, NOAA announced a further $99 million upgrade for “Rhea”, adding 22 petaflops to bring combined WCROSS speed to a total of 65 petaflops.
In addition, NOAA is integrating a separate HPC system “Ursa” that will bring the grand total of R&D HPC to approximately 73 petaflops. (ref: https://www.noaa.gov/news-release/biden-harris-administration-invests-100-million-in-new-high-performance-computer-system )
However, the hit-or-miss statistics in the above article, comparing NOAA predictions pre-2022 to those post-2022, show that NOAA would have been fine just using a $1,000 (or less expensive) laptop computer running an Excel spreadsheet.
/sarc
If I guess the average ± the standard deviation, I too will bat .600 like those guys with the supercomputers. They should be doing much better than that, yet they are not.
Here is table with the CSU and TSR forecasts along with NOAA’s. This includes their early season forecast as well as their midseason updates.
https://www.aoml.noaa.gov/hrd/hurdat/SeasonalVerification.html