2026–2027 Winter Forecast: Coldest Regions, Snow and What NOAA Actually Published

NOAA’s flagship U.S. Winter Outlook for 2026–27 is not out. The outlook itself has landed on the third Thursday of October for four years running — though last year the announcement of it did not — and the third Thursday of October 2026 is the 15th, which is also the date on the Climate Prediction Center’s own published release calendar.

What NOAA has published is a genuine forecast for this winter, and it is unusually loaded. The agency has an El Niño Advisory in force and its forecasters put a greater than 90% chance on the event reaching the “very strong” tier between October and December. If that happens, the 2026/27 winter will be shaped by one of the largest El Niño events in the record going back to 1950.

That is a real signal, and it is worth being precise about what it does and does not tell you. It tilts the odds across whole regions over three months. It does not tell you whether it will snow at your house in January — and NOAA does not forecast snowfall totals at all.

What is the 2026/2027 US winter forecast?

Milder from the Pacific Northwest across to the Northeast, much wetter in the Southeast, drier through the northern Rockies and Great Lakes — and, the part nobody reports, nowhere at all forecast to be colder than normal.

This is the 20 August 2026 outlook set, read region by region from the Climate Prediction Center’s own prognostic discussion:

The forecaster’s own wording is that El Niño teleconnections “ramp up through the fall 2026 and winter 2026-2027”, with below-normal precipitation over the Northwest and north-central states “juxtaposed with above-normal precipitation over the southern tier, with highest probabilities over parts of the Southeast in winter 2026-2027 when teleconnections are strongest”.

“Not stated” is a real answer, not a gap in our reading. Where CPC gives no tilt for a region and season, that is the agency declining to lean either way — usually because the models disagree. Parts of the Gulf States sit there for exactly that reason: a strong El Niño would normally push them drier, but the dynamical models split and the recent decadal trend runs the other way.

Which regions will be coldest this winter?

NOAA does not name any. Not one region, in any season of the outlook, is favoured for below-normal temperature.

That is worth stating plainly because “coldest regions” is one of the most-searched things about a winter forecast, and the honest answer this year is that the official outlook has no candidates. Every temperature signal CPC states points the same direction — above normal — with the firmest probabilities, above 50%, running from the Pacific Northwest across the Great Lakes to the Northeast.

This was checked against every cell rather than read off a map. The chart above is built by a script that searches all eleven regions across all six seasons in the outlook for any mention of below-normal temperature, and refuses to draw if it finds one. It found none.

What that does not mean. It does not mean nowhere will be cold. A seasonal outlook is a three-month average against a 1991–2020 baseline; it says nothing about individual cold snaps, and a winter that finishes above normal can still contain a brutal fortnight. If you want the coldest places rather than the coldest anomalies, that is climatology, not a forecast — the northern Plains and interior Alaska are the coldest parts of the country most winters regardless of what the outlook says.

What does the winter outlook mean where I live?

NOAA forecasts regions; you live in a city. So this page does the join itself — and the first thing that falls out of doing it is that the most quotable version of the answer is wrong.

Measured against today’s normals, the eight strongest El Niño winters since 1950 look like a clean north–south split: every northern metro warm, every southern one cold. Tampa comes out 1.9 °F below normal, Las Vegas 1.9 °F below, Miami and Phoenix 1.7 °F below.

That split is mostly an artefact. Five of those eight winters are older than 1999 and three are older than 1975, so measuring them against a 1991–2020 baseline charges them for half a century of warming. Ask the question the other way — was each winter warm or cold for its own time? — and the southern signal collapses while the northern one gets stronger.

Tampa goes from −1.89 °F to −0.54 °F, Las Vegas from −1.91 °F to −0.40 °F, Atlanta from −1.10 °F to +0.37 °F. There is no southern cold signal in the record once the trend is out of it. What survives is northern and it is emphatic: Minneapolis +5.3 °F, Chicago +4.7 °F, Detroit +4.7 °F, Kansas City +4.0 °F. Boston, Buffalo, Cleveland and Minneapolis were warmer than their own era in all eight; Detroit, Chicago and Kansas City in every winter their record covers.

Both of those numbers were computed twice, once against a 31-winter window centred on each winter and once against a straight-line fit through the station’s whole record, because a centred window is lopsided at the start of a record that begins in 1948. The two methods agree on the direction for 29 of the 30 metros. Atlanta is the exception, and Atlanta is the one city sitting on zero.

The two columns are different kinds of claim and should not be blurred. “NOAA tilt” is a forecast for this winter, taken from the 20 August outlook. “Warmer in” is history: how many of the eight strongest El Niño winters ran above their own era at that station’s first-order weather station. Eight is a small sample and the counts are printed rather than averaged away so you can see how small.

The region column is our join, not NOAA’s. CPC’s discussion names regions in prose — “the Great Lakes”, “the Northeast” — and publishes no city lookup, so each metro was assigned by hand. Seven sit near a seam and are marked with an asterisk: Washington and Baltimore (Mid-Atlantic, which the discussion does not name separately), Pittsburgh (on the Ohio Valley seam), Minneapolis (between the northern Plains and the Great Lakes), Denver (between the northern Plains and the Southwest), Salt Lake City (between the Southwest and the northern Rockies) and Houston (CPC splits the southern tier by longitude, not by state). Where a metro is marked, treat its tilt as approximate.

A station is also not a metro. These are single first-order stations — Central Park for New York, O’Hare for Chicago, Reagan National for Washington, which is physically in Virginia — and snowfall in particular varies sharply across one metro area.

When does the first freeze usually arrive?

Denver’s median first freeze is 8 October, Minneapolis 18 October, Chicago 28 October, New York 20 November, Dallas 24 November, Tampa 12 January — and Los Angeles and Miami have no median first freeze at all.

The width of each bar matters more than the dot in the middle. Minneapolis has a 30-day window between its early and late tenths; New Orleans has 61 days and Houston 55. In the north the first freeze is a date; in the south it is a season, and the practical planning question — when to drain a hose, lift tender plants, wrap a pipe — has a much softer answer there.

This is climatology, not a forecast for autumn 2026. These are NCEI’s 1991–2020 normals: the date by which a 32 °F low had occurred in 10%, 50% and 90% of those thirty years. Nothing here says anything about when the first freeze will arrive this year, and NOAA does not forecast that either.

Is it El Niño or La Niña this winter?

El Niño, and a fast-strengthening one. The Pacific flipped from a weak La Niña to a strong El Niño in about six months.

Chart: NOAA’s Oceanic Niño Index, the three-month running mean of Niño-3.4 sea surface temperature anomalies. Only completed seasons are plotted.

⚠️ That ranking is the ONI’s. On the RONI — the index CPC uses for its “historic” threshold — the same three events come out in a different order, and the current one sits lower. See below .

This matters because a lot of coverage still circulating online describes the coming winter as a La Niña winter. It is not. Here is the index, season by season through 2026:

June–August has now been published, and it is +1.80 — the strongest reading of the event so far, and a jump of 0.41 on the previous season. It appears on all three NOAA products: PSL’s ONI file, CPC’s oni.ascii.txt, and CPC’s ERSSTv6 table. The −99.90 placeholder this page described has moved on to the July–September cell.

⚠️ Every earlier value in that table was revised down in the same update, by 0.01 to 0.03. This page previously printed −0.37, +0.13, +0.98 and +1.40; the current figures are −0.39, +0.11, +0.95 and +1.39. That is a routine reanalysis rather than a correction of anyone’s error, but it is why a figure you noted a week ago may not match.

✅ And the version footnote this page carried no longer applies. It explained a 0.01 gap between NOAA PSL’s ERSSTv6 file (+1.40) and CPC’s table (+1.39) for May–July. Both now read +1.39, and the two agree to two decimal places on every 2026 season. A gap does persist deep in the historical record — PSL gives −1.53 for DJF 1950 against CPC’s −1.32 — but that is not simply a version difference, so the old explanation has been dropped rather than restated.

NOAA’s 9 July diagnostic discussion put the latest weekly Niño-3.4 reading at +1.2 °C, with the far-eastern Niño-1+2 region at +2.7 °C, and described a downwelling Kelvin wave deepening the thermocline. The SST consolidation forecast reaches just under +2.5 °C by late autumn.

The probabilities NOAA attached in that July discussion: a 97% chance El Niño persists through early spring 2027, an 81% chance it is very strong during October to December, and 75% for November to January.

Those have since firmed twice. The 13 August discussion put a very strong event at greater than 90% for Northern Hemisphere autumn and winter, and added a figure the July one did not carry: a 69% chance that October to December produces a historic event exceeding every El Niño back to 1950. The 10 September discussion holds the first figure and raises the second to 75%.

The threshold is a RONI, and that changes the arithmetic

🔴 CPC states the historic-event threshold as “+2.5 °C or more for a 3-month RONI value” — not ONI. Almost every article you will read about this winter quotes the ONI, currently +1.80, next to a +2.5 threshold and leaves you to do the subtraction. That subtraction is wrong, because the two numbers are not the same index.

The Relative Oceanic Niño Index subtracts the tropical-mean sea surface temperature anomaly from the Niño-3.4 anomaly. In a tropical ocean that has warmed overall, it runs lower. Here is 2026 on both, straight from CPC’s two files:

So the distance to “historic” is 1.14 °C, not 0.70 °C. Quoting the ONI makes the gap look 39% smaller than it is.

✅ And there is a reason CPC phrases it in RONI rather than ONI — the ONI version of the sentence would be false. On the ONI, 2015-16 already peaked at +2.59 and two of its seasons cleared +2.5. A “+2.5 ONI” event would not exceed every El Niño back to 1950; it would be the second one. On the RONI, nothing in the record since 1950 has ever reached +2.5 — the highest is +2.40, in 1982-83.

RONI also reorders the podium, which is the clearest illustration of what it is doing:

The three famous super-El Niños swap places depending on which index you use. That is not a quirk — it is the point of the relative index, and it is why the 75% figure is stated against it.

Where the warmth actually is

August’s index readings pull apart, and the spread is the useful part.

Every index rose over the month except Niño-4, which fell to +0.1 °C. CPC describes sea surface temperature anomalies exceeding +3.0 °C in the eastern equatorial Pacific, a subsurface index that held steady through August, and widespread subsurface anomalies in excess of +10.0 °C at depth.

This is an east-based event. That is worth holding on to when reading the analog winters below: the composite maps for El Niño are built mostly from east-based events, so the resemblance is closer than it would be for an event centred near the Date Line.

What did past strong El Niño winters actually do?

All eight of the strongest El Niño winters since 1950 were wetter than the US average. Six of the eight were warmer — and two were among the coldest fifteen winters in the record.

Chart: NOAA nClimDiv contiguous-US December-February average temperature. The marked winters are those whose Oceanic Niño Index peaked at +1.6 °C or higher.

We pulled NOAA’s national temperature and precipitation series for every winter since 1950 and looked at what the strong El Niño years actually delivered, rather than at what the composite maps say they should. The two right-hand columns are that winter’s rank out of 77, warmest and wettest first:

There is an interactive version of this table. Our US winter outlook explorer lets you move the ONI threshold and watch the hit rate change — at +1.6 every one of the eight was wetter than average, but drop the bar to +1.0 and it falls to 13 of 16. It also carries NOAA’s regional outlook for this winter, kept separate because that is a forecast and this table is not.

Two things fall out of that table.

The precipitation signal is the reliable one. Every single strong El Niño winter came in above the 1950-2026 average of 6.80 inches — although 1965-66 only just, by six hundredths of an inch. Two of them, 1997-98 and 1982-83, are the second and third wettest winters in the entire record.

The temperature signal is much weaker than its reputation. Five of the six warm ones are from 1982-83 onwards, the exception being 1957-58; the two cold ones are both from before 1975. That pattern is entangled with the warming trend running through the whole series, so “El Niño means a mild winter” is doing less work than it appears to. 1972-73 had a stronger El Niño than 1982-83 and was 4.3 °F colder.

For context on what you are comparing against, last winter — a weak La Niña — was the second warmest DJF on record at 37.07 °F and the fourth driest at 5.02 inches. The summer that followed it produced the July heat wave that ran across most of the eastern United States and an extreme heat watch across the Mid-South , so the ground the coming winter starts from is warm and, in much of the west, dry.

Which part of winter is most likely to turn — December, January or February?

NOAA’s own confidence rises as the winter goes on, and the history runs the other way: the eight analog winters put their biggest national anomaly in December and their smallest in January.

Detrended and averaged across the eight, December ran 2.15 °F above its own era and 0.45 inches wetter, and was above normal in six of the eight. February was next, at +1.70 °F in five of eight. January was a coin flip: +0.62 °F, above normal in four of the eight and below in the other four. That is the opposite of the intuition that a winter pattern needs a month or two to bed in.

The forecast side of the question moves the other way, because CPC’s overlapping three-month seasons get more confident later:

In CPC’s own words the Southeast’s probabilities “exceed 60 percent … by NDJ and enhanced probabilities persist through JFM”, while the California signal runs “from DJF through FMA, peaking in coverage during JFM”. The Northeast is the clearest reversal on the page: modestly dry in the autumn, then “the tilt toward above normal precipitation extends along almost the entire East Coast from NDJ through FMA”.

Neither of those is a monthly forecast, and NOAA does not publish one at this range. CPC issues exactly one monthly outlook at a time: the current one covers September 2026: it went out on 20 August and was updated on 31 August. The next monthly outlook, for October, comes with the 17 September set. The long-lead set that runs into winter is thirteen overlapping three-month seasons with no individual month among them, so the first outlook naming December on its own is due in mid-November. The month-by-month numbers above are a composite of eight past winters — history, with a sample of eight.

Will it snow more or less this winter?

More in the southern mountains and the southern tier, less across a much larger northern area — and NOAA’s own phrase for El Niño is “the great snowfall suppressor over most of North America”.

The short version: above-average snow shows up in California, Nevada, Utah, Colorado, New Mexico and Arizona, on the high plains and in the Mid-Atlantic and southern Appalachians; below-average snow shows up from the Pacific Northwest through the northern Rockies and Great Lakes to interior New England and Alaska. Wetter is not the same as snowier, because the southern tier is wetter partly by being milder.

The snow half of the record is where the city-level numbers get emphatic, and unlike the temperature signal it survives the trend test without shrinking:

Detroit, Chicago, Indianapolis and Kansas City did not have a single above-normal December-February snowfall in any strong El Niño winter their records cover — 0 of 7, 0 of 6, 0 of 8 and 0 of 6. Cleveland and Pittsburgh managed one each out of eight. In inches the deficits are large: Buffalo averaged 17 inches below its own era, Cleveland 15, Detroit 14, Indianapolis 13, Boston 12.

Around Washington and Baltimore the sign flips — an average of about two inches more — but each was snowier in only four of eight, which is what no signal looks like. Denver is the same: +0.7 inches, five of eight.

We have given snow its own page, because the region-by-region picture and the reasons a stronger event does not mean more snow need more room than a section: 2026-27 US El Niño snow predictions .

What could actually disrupt daily life this winter?

Not snowfall totals. The things that close a road, empty a fridge or kill someone are ice, brief violent squalls and long cold snaps, and none of them show up in a seasonal outlook.

This section is background on the hazards, not a forecast of any of them. A seasonal outlook cannot tell you whether there will be an ice storm; what it can do is tell you which failure modes are worth being ready for where you live.

Ice is the one to plan for in a mild, wet winter. The National Weather Service treats a quarter of an inch on an elevated surface as the benchmark for significant accumulation, and its own directive says why: “Significant accumulations of ice pull down trees and utility lines, resulting in loss of power and communication.” The warning threshold itself is not a national number — it is set locally, and NWS explicitly allows “both higher thresholds for regions that are accustomed to ice events and lower thresholds for areas where lesser amounts can cause major problems”. State College’s office, for instance, publishes 0.25 or 0.50 inch depending on where you are in its area.

Snow squalls are the most dangerous winter product almost nobody knows about. They are short — NWS puts them at 30 to 60 minutes against a snowstorm’s hours or days — and the accumulation is “typically an inch or less”. The danger is not the snow. It is that “the sudden white-out conditions combined with falling temperatures produce icy roads in just a few minutes”, on a road that was merely wet a moment earlier. The Snow Squall Warning is modelled on a tornado warning: valid 30 to 60 minutes, aimed at a small area, and since the 2023-24 season only those tagged “Significant” set off a phone alert. NWS’s advice is blunter than usual — “There truly is no safe place on the highway during a snow squall” — and it warns specifically against braking hard, because of the chain-reaction risk.

The cold products changed name and a lot of coverage has not caught up. From 1 October 2024 the Wind Chill Warning became the Extreme Cold Warning, the Wind Chill Watch became the Extreme Cold Watch, and the Wind Chill Advisory became the Cold Weather Advisory. NWS’s reason was that the old name taught the wrong lesson: the change was meant “to clarify that cold can be dangerous with or without wind”. The current warning triggers on air temperature or wind chill against locally set criteria — in central Pennsylvania, −20 °F south of I-80 and −25 °F north of it.

On the roads, the widely quoted figure is the wrong one. FHWA’s Snow and Ice page says 24% of weather-related crashes happen on snowy, slushy or icy pavement and that over 1,300 people are killed on it annually — and attaches no date and no source to either number. FHWA’s own dated statistics, five-year averages for 2019–2023 built from NHTSA data, say something different and smaller: snow or slushy pavement is 186,076 crashes and 496 deaths a year (16% of weather-related crashes), icy pavement 156,164 crashes and 521 deaths (13%). We use the dated pair. Winter’s real signature on traffic is delay rather than death: arterial speeds fall 30–40% on snowy or slushy pavement.

Power is the thing that turns a storm into a week. Climate Central’s analysis of two decades of utility filings puts winter weather behind 23% of major weather-related US outages between 2000 and 2023 — 398 events — behind severe weather at 58%. That is Climate Central’s own re-analysis, not a federal statistic; EIA’s reliability reporting does not break outages down by weather type. What EIA does report is that the average US customer lost 11 hours of electricity in 2024, nearly twice the previous decade’s average, with major events accounting for 80% of those hours.

For scale on the human cost, NWS’s own hazard statistics put the 10-year averages at 35 US deaths a year from cold and 27 from winter weather, against 279 from heat. Those counts swing hard: cold killed 106 in 2021 and 11 in 2023. NWS itself names the CDC, not NWS, as the official source for weather-related cause of death.

Does El Niño raise Florida’s winter tornado risk?

Yes, and it is the one part of the country where that is true. We counted it out of the Storm Prediction Center’s own database: 19 Florida tornadoes per winter in the eight strongest El Niño winters, against 7 in every other winter since 1950-51.

The same holds for the stronger ones: 10.6 rated (E)F1 or above per winter against 3.9. Because tornado reports rise through the record as detection improves, we also expressed Florida as a share of the national December-February count, which cancels a nationwide reporting trend — the gap narrows but survives, 18.1% against 11.7%, and 15.4% against 9.3% for F1 and above. Four of the eight winters land in the top decile of all 75, where chance would put 0.8 of them.

NWS says the same thing from the physics. Its Tampa Bay office finds that since 1969-70 “Strong El Niño events have more than double the number of severe weather events … than the other ENSO categories”, because “with the jet stream further south and stronger we tend to see a higher number of severe weather events”. Its Tallahassee office puts the Florida average at “20 tornado events per year” in a strong El Niño. Its Melbourne office is the one that put the causal limit best: “El Niño does not cause violent tornadoes in Florida, but it enables them.”

The spread is wide and the average hides it. 1965-66 produced no Florida winter tornadoes at all. 1997-98 produced 46, including the outbreak of 22-23 February 1998 that killed 42 people and injured more than 260 — the deadliest in Florida’s history, and one NWS’s service assessment ties to “a stronger-than-normal subtropical jet stream” at 140 knots.

This does not generalise to the rest of the country, and that is the part coverage usually gets wrong. The Storm Prediction Center’s own study of the record found that during El Niño winters US tornado outbreaks tend to be weaker and less frequent nationally, with the activity relocating rather than increasing — “mainly limited to the Gulf Coast states, including central Florida”. Of the fifty states, Florida is the only one whose winter tornadoes are “markedly stronger” with “longer tracks” in El Niño winters. Mississippi, Alabama and Louisiana all come in below their expectation.

Will heating bills be higher this winter?

The dollar figure is not published yet — EIA’s Winter Fuels Outlook accompanies the October Short-Term Energy Outlook, scheduled for 6 October 2026 — but the weather input already is, and it points slightly mild: EIA’s September outlook expects 3,392 population-weighted heating degree days this winter, about 2% fewer than the 2016-2025 average.

Heating degree days are the mechanism by which a weather forecast becomes a bill. EIA measures them against a 65 °F base, weights them by population, and assumes heating consumption “varies in proportion to heating degree days”. Its current month-by-month expectation for 2026-27, against the ten-year average:

October and March both run slightly above their ten-year averages; January is furthest below, by 5%. ⚠️ This page previously said October was the only month above, which was an artefact of its own rounding: the March average is 525.47, which rounds to 525 rather than the 526 printed here before, so what looked like an exact tie is in fact 525.79 against 525.47. The difference is a third of a degree day and changes nothing about the winter — but “the one month” was wrong, so it is fixed.

🔴 These numbers are not a forecast of this winter, and EIA says so in the footnote under its own table. The source note on Table 9c reads: “Current month based on forecasts by the NOAA Climate Prediction Center. Remaining months based on the 30-year trend.” September was the current month when this issue closed. Everything from October onward is trend extrapolation, not a prediction.

That is worth stating plainly because the table looks like a winter forecast and is widely reported as one. It is the answer to “what would a normal-ish winter on the current climate trend cost to heat”, which is a useful question and a different one.

✅ This also explains something we got wrong. This page previously warned that the table was “about to be superseded” by the September STEO on 9 September. The September issue came out and did not move a single one of these numbers. We pulled the monthly series from both releases’ own base-case workbooks and compared them cell by cell: every month from October 2026 through March 2027 is identical to two decimal places. The only cells that changed are May to September 2026 — the months that turned from forecast into observation. A figure built from a 30-year trend does not go stale monthly, and our warning that it would was wrong. 3,392 is now the September figure.

What lands on 6 October is the money. The Winter Fuels Outlook covers the four main US heating fuels — natural gas, electricity, propane and heating oil — and forecasts average household expenditures for November through March, for the country and, depending on the fuel, for up to four census regions. It is not a single number: EIA publishes the base case alongside two side cases in which every region is 10% colder or 10% warmer in heating degree days, and notes that nine of the previous ten winters fell inside that band. The figures are nominal, on an assumed 2% inflation, and EIA warns against reading them across fuels as a switching comparison.

Two cautions about that date. It is the scheduled date of the October STEO, which EIA’s own release schedule annotates “Winter Fuels Outlook” — not a separately announced date for the supplement. Last year the schedule said 7 October 2025 and the supplement appeared on 15 October; the year before, it came out with the STEO on the day. And no 2026-27 edition exists yet: the URL the 2026 article would live at returns 404, as do two invented siblings on the same host, so nothing has been quietly pre-staged.

The one piece of background already on the record is on the supply side, and it is EIA’s own — but it was revised down in September and the language around it was toned down with it.

⚠️ The word “record” is gone, and so is the comparison to 2016. The September issue drops both, along with the administrator’s quote about inventories providing “a cushion” for winter heating. It says instead that inventories are “set to be relatively high heading into the winter heating season” at 5% above the previous five-year average and 1% above October 2025. The volume moved by 0.4%; the framing moved considerably more, and a page still quoting “record” would now be quoting a withdrawn description.

September also breaks it out by region, which is the part that matters for heating. Entering the withdrawal season, EIA puts the Mountain region 21% above its five-year average, the Pacific 10%, the Midwest 6% and South Central 4% — while the East is about level with it, having started the injection season 11% below after a cold winter drew stocks down. The national surplus is real, and it is not evenly distributed: the region carrying the smallest one is the populous East.

This page will not turn any of that into a number on your bill. Storage volumes and degree days are physical quantities EIA has published; what a household pays depends on retail rates, the state you are in and the weather that actually happens. The expenditure forecast is EIA’s to make, and it makes it on 6 October.

Why do NOAA and the almanacs disagree?

Less than the headlines suggest — and the sharpest disagreement is over a corridor where NOAA has declined to forecast anything at all.

Three publications are involved and two of them are constantly mistaken for each other. The Old Farmer’s Almanac (almanac.com, founded 1792) published its 2026-27 winter forecast on 27 August 2026. The Farmers’ Almanac (farmersalmanac.com, founded 1818) published its own on 6 August 2026. They are separate businesses. NOAA is the third.

Rows: how the forecast is expressed; the normals it is measured against; how far ahead it is made; whether it is revised afterwards; whether it gives snowfall; the stated method; and how accuracy is claimed.

In their own words: the Old Farmer’s Almanac’s forecasts rest on “a unique formula that combines solar science, climatology, and meteorology”, descended from a method its 1792 founder built around sunspots and which it still calls “secret” and “proprietary”. The Farmers’ Almanac’s formula “incorporates … sunspot activity, tidal action of the Moon, the position of the planets, and more”, is attributed to a prognosticator whose identity is a brand secret, and is fixed a long way ahead: “the Almanac’s forecasts are calculated two years in advance. Once the new edition is printed, the editors never go back to change or update its forecasts.”

Start with what they agree on. All three describe a broadly mild winter driven by El Niño. The Old Farmer’s Almanac’s own headline is “Mostly mild, with a colder stretch from central New England through the Appalachians and Ohio Valley, plus Hawaii”, and it names the same driver NOAA does — “A strong El Niño has returned and is expected to continue through winter” — while adding one NOAA does not use: “Solar Cycle 25 is past its peak, and solar activity is waning.” The Farmers’ Almanac likewise puts El Niño behind “wetter-than-normal conditions across much of the southern United States”, and for the north forecasts temperatures that “should average near normal overall” with cold arriving in bursts rather than as a cold season.

Now the real disagreement, which is narrower than it sounds. The Almanac’s cold corridor runs central New England → Appalachians → Ohio Valley. NOAA’s 20 August outlook favours above-normal temperature at better than 50% “from the Pacific Northwest to the Northeast” — a direct conflict at the New England end. But over the Ohio and Tennessee Valleys, NOAA’s temperature cell for this winter is not stated: the agency declines to lean. So the place the Almanac is most confident about is the place NOAA has explicitly refused to forecast.

It is also worth noticing that the Almanac’s own Northeast region is split rather than uniformly cold — “Above normal in the north; below normal in the south” — so “the Almanac says New England will be cold” is not quite what the Almanac says.

They are not measured against different yardsticks. Both use 1991–2020 normals; the Almanac moved to that base with its 2023 edition, the same period CPC adopted in May 2021. The difference is not the baseline. It is the form of the claim.

NOAA publishes a probability of a three-month average landing in the warmest, middle or coldest third of the 1991–2020 distribution, and says “equal chances” when it has no lean. An almanac publishes a regional statement, in advance, and does not revise it. A probability and a statement cannot be scored against each other without first agreeing what counts as a hit — which is exactly the ground the accuracy argument is fought on. The Old Farmer’s Almanac’s own definition is narrow and it says so: 80% “means that, on average, The Old Farmer’s Almanac correctly predicts the direction of seasonal temperature and precipitation departures from normal about 80% of the time. It does not mean that every daily temperature or precipitation event is predicted correctly.” It grades itself against one representative city per region, and scored last winter at 88.9%.

The practical answer for a reader. If you want to know how likely a warm winter is where you live, the outlook is the document with a number attached to it. If you want a seasonal snowfall figure, only the almanacs will give you one, and NOAA’s reason for not giving you one is worth reading first: “The outlook does not project seasonal snowfall accumulations as snow forecasts are generally not predictable more than a week in advance.”

When does NOAA’s winter forecast actually come out?

Two different things come out on two different schedules, and 2026 is the year they collide on the same day: 15 October.

There are two clocks here and confusing them is the whole problem. The routine long-lead outlook set — the one every figure above comes from — is reissued on the third Thursday of every month, and has carried a December–January–February forecast since the spring. NOAA’s branded Winter Outlook, the one that gets a press release and a map in the news, is an annual event.

The dates line up in a way worth noticing: the routine reissue in October 2026 falls on 15 October, and that is also the third Thursday NOAA has put its Winter Outlook out on for four years running. So the flagship and the routine release are due the same day — assuming nothing gets in the way of the announcement, which is exactly what happened last year.

The one everybody means is not out yet, and the date to put in the diary is Thursday 15 October 2026.

There are two different NOAA products with a claim on the phrase “winter forecast”, and almost every argument about this topic is really two people talking about different documents.

The flagship U.S. Winter Outlook is the one with the press release and the three coloured maps of the country — temperature, precipitation and drought, for December through February. It is issued once a year, in October, and it is what a news story means when it says NOAA’s winter forecast is out.

The long-lead seasonal outlooks are the monthly workhorse. The Climate Prediction Center reissues a rolling set of them every month, each covering the next twelve overlapping three-month seasons — which means a December-February 2026/27 forecast has already been in the set for months, and simply gets reissued at a shorter lead each time. The current one was issued on 20 August. It is a real NOAA forecast for this winter. The regional tilt set out above is read from that 20 August set, region by region — it replaced the 16 July version this page previously quoted.

Why 15 October, and how firm is that?

Firm on the schedule, and a four-year pattern on the press release. Two separate things point at the same day.

The Climate Prediction Center publishes its release calendar in advance. Its own wording is that forecasts “are released on the third Thursday of the month”, generally around 8:30 AM Eastern. The four releases that carry us into winter, with the August release now past:

That is the schedule, published by NOAA, and 15 October is a fact on it rather than a guess.

What is a pattern rather than a promise is the flagship. NOAA has put its Winter Outlook out on that same October release day four years running, and last year it was the announcement, not the data, that came apart:

The first four are third Thursdays of October, which we computed rather than eyeballed.

The 2025 split is worth understanding, because it is the clearest evidence for the two-clocks argument on this page. The 43-day federal shutdown ran from 1 October to 12 November 2025, straight through the October release. The outlook itself still went out on the third Thursday — coverage that weekend described NOAA as having “quietly released it’s outlook for the upcoming winter season on Thursday, sans press conference since the government shutdown continues”, with the press conference “rescheduled to November”. NWS’s own local write-ups then dated the flagship to the announcement: “NOAA’s U.S. Winter Outlook released on November 21”.

NOAA has not announced the 2026–27 date. The honest form of the claim is: the seasonal outlook release is scheduled for 15 October, the flagship has accompanied that release every year we checked, and in one of those four years the public announcement of it arrived five weeks later than the data did.

So what is published right now?

*Not announced by NOAA. 15 October is the third Thursday and matches four years of precedent, and it is on CPC’s published release calendar for the routine set. **The 2025–26 outlook went out on 16 October during the federal shutdown; NOAA’s public announcement of it followed on 21 November.

The 20 August set is the current one: the Climate Prediction Center’s long-lead prognostic discussion for it is timestamped 8:30 AM EDT on Thursday 20 August 2026, and the regional tilt at the top of this page is now read from it region by region.

The 8 October date is not our inference either. The 10 September discussion ends by naming it: “The next ENSO Diagnostics Discussion is scheduled for 8 October 2026.”

The practical consequence for a reader in mid-September. Anything you see today headlined “NOAA’s winter forecast” is describing the monthly seasonal outlook, not the flagship — which is fine, because the monthly outlook is a genuine NOAA product covering this winter. What it is not is necessarily the newest. The version most stories are written from went out on 16 July; a fresher set landed on 20 August, and the next lands on 17 September.

Three things could still move the picture between now and October: whether the event peaks near +2.5 °C on the RONI or falls short, how the Madden-Julian Oscillation behaves through the winter, and whether the polar vortex holds. None of the three is settled yet, and the last two are essentially unpredictable at this range. The first one has further to travel than the widely-quoted ONI makes it look — see the index section above.

The bottom line

The 2026/27 US winter will be an El Niño winter, very likely a strong one, and NOAA gives it a real chance of ranking among the biggest events since 1950. The official tilt is wetter across the south with the Southeast most strongly favoured, drier across the Northwest and the northern tier, and milder over most of the country outside the southern states.

The historical record supports the wet part more firmly than the mild part. Every strong El Niño winter since 1950 was wetter than average; two of them were among the coldest in the record. If you want one sentence to carry through to December: El Niño reliably moves the storm track, and much less reliably moves the thermometer.

At the scale of a street rather than a region, three things in the record are firmer than the national average is. The northern tier really did run warm in these winters — Boston, Buffalo, Cleveland and Minneapolis in all eight. The Great Lakes and Ohio Valley really did lose snow, and Detroit, Chicago, Indianapolis and Kansas City did not have a single above-normal winter among them. And Florida really did get more winter tornadoes, by a factor of about two and a half. Everywhere else, and especially across the southern tier, the honest reading of eight winters is that they do not tell you much.

Sources

Checked 21 August 2026; re-checked 2 September 2026. Re-checked again 13 September 2026, when the September ENSO discussion and the September STEO were both read in full: the ENSO figures moved, the STEO’s winter degree days did not, and its storage figure and framing did. Both ONI tables still end at June–August and are unchanged. Every source added on 30 August 2026 was opened and read on that date. No affiliate links, and no payment was received for any link on this page.

Seasonal outlooks are probabilistic statements about three-month averages over large regions. They are not forecasts of individual storms, snowfall totals or conditions on any given date. For decisions that depend on actual weather, use your local National Weather Service forecast office.