Issue 056 - Meteorology - Poisson probability
How unusual is an Atlantic season with no hurricanes by mid-September?
The 2026 Atlantic hurricane season has been exceptionally quiet. As of mid-September, no Atlantic hurricanes have formed, despite this period normally being near the seasonal peak. Five weaker tropical storms have formed so far.
The problem
Estimate the probability that a typical Atlantic hurricane season would reach September 15 without producing a single hurricane.
You will need to estimate how many hurricanes normally occur during an entire Atlantic season, what fraction of annual hurricane activity normally occurs by mid-September, and how independently hurricane formations can reasonably be treated for a rough probability estimate.
Then convert your estimated probability into an intuitive frequency, such as about one season in every ___ years.
Because Fermi problems target an order of magnitude, I normally use no more than two significant digits and write most calculations in scientific notation; the Fermi reference explains both conventions.
Before checking sources
Matt's first pass
I really suck with these types of probability estimates, so I wanted to cover this one so I would learn how to approach it. Any answers I give are straight up guesses.
My intuitions are that there are about 2 hurricanes per hurricane season, on average, and they last 3 to 4 days per storm. I also assumed hurricane season stretches from about June to the end of September, or about 120 days.
So I would guess that most seasons, we would experience about 6 to 8 storm days in a 123 day season.
expected hurricane-days
~= 2 hurricanes/season x 3 to 4 days/hurricane
~= 6 to 8 hurricane-days/seasonThat means each day of the season there is about a 6.5% chance that it would have stormed on that day.
daily hurricane-day chance
~= 8 storm-days / 123 days
~= 6.5%We are about 107 days into the season this year with 0 storm days, and we would have expected about 5 to 6 storm days in an average season by this time, with about 90% confidence.
So does that mean 10% of the time we would not expect to see any storms by this time, or once in a decade? If my confidence was 95%, would that mean we would not expect to see any storms by this date about once every 20 years?
Calibration Score
Matt's Calibration Score: 25 / 100
Higher is better: earn points for accurate pegs, sound models, correct math, and a result close to the sourced answer. The image shows percent full of it: 100 minus the Calibration Score.
Pegs: 0/30. The average Atlantic season has about 7 hurricanes, not 2, and the formal season runs June through November, not June through September.
Model: 0/30. Hurricane-days are not the same as hurricane formations, and the probability needs an expected-event model rather than a confidence guess.
Math: 5/10. The basic ratios were internally plausible, but the key probability conversion was missing.
Result: 20/30. A once-per-decade guess is within an order of magnitude of the simple recalibrated answer, though it understates how rare this looks in the satellite era.
Grounding facts
NOAA's 1991-2020 Atlantic hurricane climatology gives an average season about 14 named storms, 7 hurricanes, and 3 major hurricanes. The formal Atlantic season runs from June 1 through November 30.
AP reported that, as of mid-September 2026, the Atlantic season had produced only five weak tropical storms and no hurricanes, the slowest start since the satellite era began. AP also reported that El Nino-related wind shear, dry air, and sinking air helped suppress development.
A local scan of the NHC HURDAT2 Atlantic database through 2025 shows that, from 1966 through 2025, the Atlantic averaged about 3.8 hurricanes by September 15. In that same satellite-era span, zero of 60 seasons reached September 15 without at least one hurricane.
After checking sources
Check and recalibrate
For this problem, the simplest useful model is a Poisson model: if the expected number of hurricane formations by September 15 is lambda, then the probability of seeing exactly zero hurricanes by then is:
P(0 hurricanes) = e^-lambdaThe key is choosing lambda. NOAA says the average Atlantic season has about 7 hurricanes. By September 15, the satellite-era HURDAT2 average is about 3.8 hurricanes. Round that to 4 as the memory-friendly peg.
expected hurricanes by Sept. 15
~= 4
P(0 hurricanes by Sept. 15)
~= e^-4
~= 1.8 x 10^-2
~= 2%That is about:
frequency
~= 1 / 0.02
~= 50 seasonsSo the clean Fermi answer is that a typical-season model puts this outcome around 2%, or roughly one season in 50. If you use lambda = 3 instead of 4, you get e^-3 ~= 5%, or about one in 20. If you use lambda = 4.5, you get about 1%, or one in 100. The stable answer is not "every few years"; it is more like once every few decades.
The historical check is even sharper. In the satellite era before 2026, 0 of 60 seasons reached September 15 without a hurricane in the HURDAT2 record. That does not mean the true probability is literally zero, but it tells us the event is at the edge of the observed modern record. Looking farther back to 1851-2025, five seasons had no hurricane by September 15, including 1907 and 1914 with no recorded hurricanes at all, but pre-satellite records are less complete.
The best plain-English conclusion: no Atlantic hurricanes by September 15 is unusual at the one-in-several-decades scale under ordinary season assumptions, and record-setting in the modern satellite-era record. The Poisson model is useful for the back-of-the-napkin probability; the historical record reminds us that real seasons are not independent coin flips because El Nino and other climate patterns can push the whole basin toward quiet or active conditions.
Post-check reflection
Matt's reflection
Ok, I knew this one was going to sting. Technically I was within an order of magnitude of the correct answer, but that is only due to luck. My intuition pegs about the period of time the hurricane season lasts and how many hurricanes we might experience in a normal year were both off, and I had no reliable model for calculating the relevant probabilities.
This has come up before, so moving forward I need to take seriously the Poisson distribution and memorize the pegs for at least the first few expected events.
Recommended memory peg
For Atlantic hurricane-season probability problems, remember: a typical season has about 7 hurricanes total, and by September 15 the satellite-era average is about 4 hurricanes. For Poisson zero-event checks, P(0) = e^-lambda, so lambda ~= 3 gives 5%, lambda ~= 4 gives 2%, and lambda ~= 5 gives 0.7%.
Reader results
Bars show how submitted estimates sort into the answer choices from the gut-check prompt.