station climatology
Telegram alerts @tempterminal --:--:-- UTC

A climatology terminal for temperature markets

Live reports from the 48 airports Polymarket's temperature markets settle on, with live odds for each range, seven forecast models scored at each station, decoded TAFs, a live satellite cloud map and five years of hourly climatology.

What you get for a station

A live map around the station

The airport among the volunteer weather stations near it, each with its latest reading. Satellite cloud on every station page, and weather radar on the 11 US ones, play as a short loop, so you can see which way the weather is moving. Low cloud and fog show only faintly. See it on New York or London.

Live odds for each range

The chance of each market range, starting from our forecast (all seven models combined, using what 20 months of their misses at these stations taught it), then narrowed with every report the station files. Checked on 22 past days before going live: when it said 30%, it happened about 30% of the time. The market's price sits beside it.

The climb to peak

At 06:00 a station has most of its warming ahead of it; by 16:00 it usually does not. We hold the hourly distribution for every day of the year, so the page can say how much further this station typically climbs from the hour it is in now — and how often the peak is already behind it.

Which model to trust here

Seven forecast models scored against what the station actually recorded, per city. The best model is not the same everywhere: in a typical city the best and worst are 1.2°C apart, and 33 of 53 markets show a gap over a degree.

The aerodrome forecast

The station's own TAF, decoded into plain language, including the forecast maximum where the issuing service publishes one.

Where the numbers come from

Every figure traces to a public source, and the page says which. Station observations and climatology are Iowa Environmental Mesonet ASOS archives, which are public domain. Forecasts are Open-Meteo. Aerodrome forecasts are NOAA. The map's radar is the National Weather Service's NEXRAD, via the Iowa Environmental Mesonet; its cloud is from NOAA's GOES and Japan's Himawari satellites, via NASA GIBS, and EUMETSAT's Meteosat. 53 temperature markets at 48 airport stations. 5 settle on the day's low, the rest on its high, and each low market shares a station with its high twin. Every one has enough scored history to rank forecast models. Rebuilt every thirty minutes.

Sources and method

stationsall 53

Daily maximum — or the minimum, for the 5 markets badged MIN. Shown in °F where the market settles in °F, °C elsewhere.
— — rebuilt every 30 min
AmsterdamEHAM 19:25
now5°C
forecast9°C
AnkaraLTAC 21:25
now8°C
forecast15°C
AtlantaKATL 13:25
now55°F
forecast64°F
AustinKAUS 12:25
now69°F
forecast75°F
BeijingZBAA 02:25
now-4°C
forecast6°C
Buenos AiresSAEZ 15:25
now25°C
forecast28°C
BusanRKPK 03:25
now1°C
forecast8°C
Cape TownFACT 20:25
now17°C
forecast22°C
ChengduZUUU 02:25
now12°C
forecast17°C
ChicagoKORD 12:25
now67°F
forecast67°F
ChongqingZUCK 02:25
now14°C
forecast16°C
DallasKDAL 12:25
now70°F
forecast75°F
DenverKDEN 11:25
now41°F
forecast47°F
GuangzhouZGGG 02:25
now14°C
forecast21°C
HelsinkiEFHK 20:25
now1°C
forecast4°C
HoustonKHOU 12:25
now76°F
forecast78°F
IstanbulLTFM 21:25
now12°C
forecast16°C
JeddahOEJN 21:25
now29°C
forecast34°C
KarachiOPKC 23:25
now21°C
forecast32°C
Kuala LumpurWMKK 02:25
now27°C
forecast33°C
LondonEGLC 18:25
now8°C
forecast9°C
LondonMINEGLC 18:25
now8°C
forecast low3°C
Los AngelesKLAX 10:25
now61°F
forecast69°F
LucknowVILK 23:55
now20°C
forecast27°C
MadridLEMD 19:25
now10°C
forecast14°C
ManilaRPLL 02:25
now27°C
forecast32°C
Mexico CityMMMX 12:25
now21°C
forecast24°C
MiamiKMIA 13:25
now79°F
forecast83°F
MilanLIMC 19:25
now9°C
forecast16°C
MoscowUUWW 21:25
now0°C
forecast5°C
MunichEDDM 19:25
now3°C
forecast8°C
New YorkKLGA 13:25
now60°F
forecast64°F
Panama CityMPMG 13:25
now33°C
forecast32°C
ParisLFPB 19:25
now9°C
forecast11°C
ParisMINLFPB 19:25
now9°C
forecast low7°C
QingdaoZSQD 02:25
now-3°C
forecast13°C
San FranciscoKSFO 10:25
now48°F
forecast57°F
SeattleKSEA 10:25
now38°F
forecast49°F
SeoulRKSI 03:25
now-1°C
forecast5°C
SeoulMINRKSI 03:25
now-1°C
forecast low-3°C
ShanghaiZSPD 02:25
now9°C
forecast16°C
ShanghaiMINZSPD 02:25
now9°C
forecast low7°C
ShenzhenZGSZ 02:25
now18°C
forecast22°C
SingaporeWSSS 02:25
now26°C
forecast31°C
São PauloSBGR 15:25
now31°C
forecast33°C
TaipeiRCSS 02:25
now16°C
forecast24°C
Tel AvivLLBG 20:25
now21°C
forecast27°C
TokyoRJTT 03:25
now8°C
forecast14°C
TokyoMINRJTT 03:25
now8°C
forecast low9°C
TorontoCYYZ 13:25
now10°C
forecast13°C
WarsawEPWA 19:25
now2°C
forecast4°C
WellingtonNZWN 07:25
now14°C
forecast21°C
WuhanZHHH 02:25
now11°C
forecast17°C

todaywhat is unusual right now

Every station against its own normal for this hour, most unusual first. 25 of 48 are more than 3°C from normal. Stations still climbing come first — those are the days not yet decided. Rebuilt every thirty minutes, and it moves as the world turns.

Open to everyone while this is new. From 1 February 2027 the board below the first five stations becomes part of Pro, alongside a watchlist and a morning brief on the stations you pick; the top five, and every station page, stay free. Hear about it first.

Still climbing25 stations — today is undecided

stationnow vs normal unusual left to go

Peak already set23 stations — today's extreme is in

stationnow vs normalunusual status

“vs normal” is the current reading minus this station's median for this hour across five years, in the unit the market settles in. “Unusual” divides that by the middle-half spread of the same station-hour, so a degree counts for more where a station is steady than where it swings; stations report whole degrees, so that spread has a floor and the figure is shown as a bound past 4σ. A gap of 3°C or more calls the direction of the day's extreme 98.9% of the time when measured at 10:00–13:00 local with the peak still ahead (n=76,096, against an 88.3% base rate). That is persistence, not a forecast edge — it says what today is doing, not that anything is mispriced.

forecast modelsaccuracy by station

53 of 53 markets and 7 models, each model's early-morning forecast scored against what the station recorded over the last 30 days, with its usual lean taken out. Misses are in °C everywhere, including markets shown in °F. The 7-model blend, the site's forecast where it runs, combines all seven; its miss is its plain miss, with nothing taken out.
league table typical miss across all stations lower is better
modelaccuracy typical missused atscored at
7-model blend0.77°C4545
DWD ICON0.81°C453
ECMWF IFS0.95°C253
UK Met Office1.00°C053
NOAA GFS1.02°C253
Environment Canada GEM1.10°C053
JMA GSM1.13°C053
CMA GRAPES1.16°C053
the forecast used at each market the 7-model blend, or the model that has missed least lately
marketused here its typical missworst modelwithin a degree
Amsterdam7-model blend0.57°C1.16°C87%Ankara7-model blend0.88°C1.52°C77%Atlanta7-model blend0.66°C1.18°C50%Austin7-model blend0.85°C1.88°C43%Beijing7-model blend1.01°C1.44°C57%Buenos Aires7-model blend1.18°C1.83°C50%Busan7-model blend1.01°C1.68°C57%Cape Town7-model blend0.78°C1.63°C70%Chengdu7-model blend0.94°C1.29°C63%Chicago7-model blend0.64°C1.10°C53%Chongqing7-model blend0.75°C1.26°C70%Dallas7-model blend0.74°C1.30°C50%DenverDWD ICON0.87°C1.91°C37%Guangzhou7-model blend0.85°C1.20°C63%Helsinki7-model blend0.46°C0.64°C90%Houston7-model blend0.88°C1.73°C47%Istanbul7-model blend0.75°C1.75°C80%Jeddah7-model blend0.70°C1.07°C77%KarachiNOAA GFS0.50°C0.86°C82%Kuala Lumpur7-model blend0.86°C1.66°C70%London7-model blend0.48°C1.03°C90%LondonMINECMWF IFS0.53°C1.10°C93%Los Angeles7-model blend0.54°C1.03°C60%Lucknow7-model blend0.67°C0.95°C80%Madrid7-model blend0.77°C1.42°C67%Manila7-model blend0.68°C1.00°C83%Mexico City7-model blend0.80°C1.03°C60%Miami7-model blend0.53°C1.03°C67%Milan7-model blend0.92°C1.43°C60%Moscow7-model blend0.73°C1.65°C67%Munich7-model blend0.93°C1.28°C57%New York7-model blend0.44°C0.97°C63%Panama CityDWD ICON0.60°C1.40°C73%Paris7-model blend0.74°C1.30°C80%ParisMINDWD ICON0.79°C1.45°C77%Qingdao7-model blend0.98°C1.27°C63%San Francisco7-model blend0.74°C1.38°C33%Seattle7-model blend0.92°C1.26°C33%Seoul7-model blend0.69°C1.32°C73%SeoulMINDWD ICON0.68°C1.87°C70%Shanghai7-model blend0.46°C0.93°C90%ShanghaiMINECMWF IFS0.62°C1.18°C83%Shenzhen7-model blend0.70°C1.25°C73%Singapore7-model blend0.65°C1.30°C73%São Paulo7-model blend1.23°C1.98°C50%Taipei7-model blend0.78°C1.26°C73%Tel Aviv7-model blend0.80°C1.04°C70%Tokyo7-model blend0.96°C1.55°C63%TokyoMINNOAA GFS1.01°C2.74°C43%Toronto7-model blend0.79°C1.75°C67%Warsaw7-model blend0.83°C1.17°C60%Wellington7-model blend0.63°C1.00°C87%Wuhan7-model blend0.94°C1.65°C57%

Scored against the reading the market settles on, not against the midpoint of a winning range: a midpoint rounds every miss to the nearest degree and flatters every model equally. Each station page shows all seven models and which way each one leans there.

how to reada station

one worked example, using whatever the weather is doing right now

Every station page answers one question in four steps: how hot will it actually get today. Here is São Paulo, read the way the tabs are meant to be used. This example is rebuilt with the rest of the site, so the numbers below are the same ones on São Paulo's own page.

1. Today — what is happening now

It is 31°C there now, about 4.0° warmer than this station usually is at 15:00. This station usually reaches its maximum around 13:00 local, so a reading taken before then still has climbing left to do.

2. Forecasts — whose number to believe

Seven models forecast this station and they disagree. Each is scored against what this station actually recorded, and the site's forecast, the 7-model blend, combines all seven, using what 20 months of their forecasts and misses at these stations taught it. At São Paulo it misses by about 1.2° on a typical day — its call for today, 33°C, is the one the page leads with.

3. Climatology — what this station normally does

Five years of hourly observations say what a day like this usually looks like here: how much further it climbs from now, when it peaks, and how far from normal it lands under a given sky or wind.

The cloud regime is the same idea applied to the sky: across five years, overcast days here land -5.8° against that normal and scattered days +1.7°. Read it as an indicator rather than a cause — a rare sky usually arrives with the weather system that made it rare.

4. Markets — what is priced, beside what has happened

The market currently favours 31°C at 91%. Across the 60 days near this date in the record, 1 landed there — 2%.

The one thing worth understanding

Those two numbers are not a disagreement to be settled, and the gap is not a mistake by either side. Climatology is a base rate: it knows nothing whatsoever about today. It describes every November day this station has recorded, warm and cold alike. The market is pricing this day, with today's forecast in hand.

So when the two are far apart, it is usually telling you that today is not a typical day — which is exactly when the base rate is least useful as a prediction and most useful as a yardstick. Here the market is well above the historical rate, which is what an unusually warm spell looks like from the outside: the record has rarely been this hot near this date, and today is forecast to be.

The site shows the base rate and the price side by side and stops there. It does not tell you which is right, and nothing here is advice.

Now try one

Pick any station and read it in that order — now, forecast, normal, priced. Open São Paulo, or browse all 53.

sourcesand method

what each number is built from, and what it will not tell you

Observations

Station observations and all climatology come from the Iowa Environmental Mesonet ASOS archive, which republishes METAR reports from national weather services. IEM places its archive in the public domain. Five years of hourly observations per station underpin the climb-to-peak distributions, the day-of-year normals and the wind regimes.

Day-of-year figures are smoothed over a ±7 day window, because a single calendar day gives roughly five samples in five years — too few to quote. The window is why a panel can show a mean that sits close to its own quartiles.

Forecasts

Model forecasts come from Open-Meteo (CC-BY-4.0), covering ECMWF IFS, GFS, ICON, UKMO, GEM, JMA and CMA. Aerodrome forecasts are the TAFs published by NOAA Aviation Weather Center, a United States Government work.

Radar

US station pages show the last hour of the US National Weather Service's NEXRAD radar composite, a frame every five minutes, as map tiles from the Iowa Environmental Mesonet (Iowa State University). Radar for the other stations would need a paid source, so it is not shown yet.

Every station page can also show clouds: infrared images from the weather satellite that sees it best, a frame every 10 or 15 minutes. We redraw them as white cloud: the colder the cloud top, the whiter.

How models are scored

Each model's first forecast after the station's local midnight is compared with the reading the market settles on: the day's highest (or lowest) routine report, to a tenth of a degree. We do not score against the midpoint of a winning market bucket: a one-degree bucket rounds every error to the nearest degree and its outer buckets are unbounded, which flatters accurate and inaccurate models alike. A market needs 7 scored days before it appears.

Our forecast and the range odds

The forecast high on each station page is the 7-model blend: all seven models combined by a machine-learned model (gradient boosting) trained on 20 months of their forecasts against what these stations recorded. Before it went live it was tested on 34 days it had never seen: a typical miss of 0.78°C, against 0.95°C for the best single model at each station. The daily-low markets, and stations it has no history for, still use that single model.

The chance of each range starts from that forecast and narrows with every report the station files, using how much it usually still climbs after each hour. Replayed on past days, ranges it gave about 30% happened about 30% of the time.

What this does not do

It does not beat the market. In our checks the market's favourite range came true more often than ours, so when the two disagree, the market has usually been closer. The forecast, the odds and the climatology (what a station has done historically, given the hour and the day of year) are estimates, and on any individual day the weather is free to ignore them.

Temperatures are shown in the unit each market settles in — Fahrenheit for United States stations, Celsius elsewhere. Climbs and anomalies are differences, so they carry no offset when converted.

Markets

Some pages compare a market's own prices with this station's history. Those bucket probabilities come from Polymarket temperature markets and are shown as published, refreshed with each rebuild. A market column may sum to slightly more or less than 100% — prices are quoted independently per bucket and rounded for display. Nothing here is trading advice.

What settles these markets

Polymarket settles each daily temperature market on the airport station's own reports as published on aviationweather.gov (since 23 October 2026; Wunderground before that, and now only a fallback). The figure is the highest reading of the station's local calendar day — the lowest, on a lowest-temperature market — across every report not labelled SPECI, half-hourly reports included. Readings are taken to the tenth of a degree where the report carries it and rounded to the nearest whole degree, with .5 going up; US markets convert those tenths to °F first. Hong Kong is the exception: it settles on the Hong Kong Observatory. The “so far” figures on each city page follow the same rules, from the same reports.

These airports are listed on other venues too, and they do not all settle the same way. Kalshi names The Weather Company rather than the station reading, and for London and Paris it is not even the same airport — its contracts settle on Heathrow and Charles de Gaulle, against the London City and Le Bourget stations used here. So treat this site as describing the Polymarket markets it is built from; elsewhere it is background at best, and for those two cities it does not describe the market at all.

Nearby personal weather stations

City maps show nearby personal weather stations alongside the official airport observation. They are volunteer-operated, unevenly sited and not the instrument any market settles on; they are there for spatial context, not for resolution.

What the counts mean

53 temperature markets at 48 airport stations. 5 settle on the day's low, the rest on its high, and each low market shares a station with its high twin. Every one has enough scored history to rank forecast models.

Attribution

Iowa Environmental Mesonet, Iowa State University · Open-Meteo (CC-BY-4.0) · NOAA / National Weather Service · NOAA GOES and Japan Meteorological Agency Himawari satellite images via NASA GIBS · contains modified EUMETSAT Meteosat data 2026 (CC-BY-4.0).

termsand privacy

what this is, what it is not, and what we keep

This is not financial advice

TempTerminal publishes weather observations, climatology and published forecasts. Some pages place a prediction market's own quoted prices beside what a station has historically done. That is a comparison, not a recommendation. Nothing here is financial, investment or trading advice, no outcome is predicted or guaranteed, and you should not make a financial decision on the basis of this site. Prediction markets can lose money and are not available or lawful everywhere.

A climatological base rate describes what has happened before, given the date and the hour. It knows nothing about today's weather. Where the market and the base rate disagree, the market usually holds information the base rate cannot.

Not affiliated

TempTerminal is an independent project. It is not affiliated with, endorsed by, or connected to Polymarket, the Iowa Environmental Mesonet, Iowa State University, NOAA, the National Weather Service, NASA, EUMETSAT, the Japan Meteorological Agency, Open-Meteo, any national meteorological service, or any airport or station operator. Their names appear only to say where a number came from.

Accuracy

Everything here is provided as is. Observations can be missing, late or wrong at source; a station can fail; an upstream service can go down. We show a source and a timestamp on the numbers so you can judge them, and we say when a reading is unavailable rather than showing a stale one silently. We do not warrant that any figure is correct, and we accept no liability for decisions made using them.

What we collect

The site is a static page. There are no accounts, no logins and no cookies set by us. We do not collect personal information, we do not build a profile of you, and we do not sell or share anything, because we do not have anything to sell or share.

The web server records ordinary request logs -- IP address, time, the page requested -- which is what any web server does and what is needed to keep it running and to spot abuse. Your browser also fetches fonts from Google Fonts, map tiles from OpenStreetMap and, on US station pages, radar images from the Iowa Environmental Mesonet and, if you switch clouds on, satellite images from NASA GIBS or EUMETSAT’s EUMETView. Those services will see your IP address as a result; their own privacy terms apply. Your browser also remembers which map layers you switched on or off, in its own local storage; that choice is never sent anywhere.

The one thing we count

We count how many browsers arrive for the first time and how many come back on a later day, and where they came from. That is the only measurement here, and it decides whether this project continues.

It works without identifying you. Your browser stores two dates in its own local storage — the day you first arrived and the last day it was counted — compares them with today, and sends one word, at most once a day: new or return. Alongside it we record which of four buckets your browser falls into — mobile, tablet, desktop or bot — so we know how much of the site is read on a phone and where to spend the work. The bucket comes from the User-Agent your browser already sends; it is not kept, only which of the four it matched. No version, no operating system, no model, and no identifier. With the same word we send where the visit came from: the name of the site that linked here (such as x.com or google.com, never the page you were on), or nothing if you typed the address, and the short tag some of our own links carry (such as ?src=x). What we keep is a handful of numbers per day. If your browser blocks local storage, nothing is sent at all.

There is no third-party analytics script, no advertising network and no profile of you anywhere in this.

Contact

Corrections about a station, a source or a number are genuinely welcome — they are how this gets more accurate. Write to hello@tempterminal.com. If a figure here looks wrong, say which station and which date and it can be checked against the raw observations.