Polymarket Weather Markets API — Daily Temperature Odds

Get every Polymarket daily temperature market ("Highest temperature in NYC on September 24?", "Lowest temperature in London on …") as clean, flat JSON — bracket by bracket: live bid/ask from the order book, implied probability (raw and normalized), volume, liquidity, results — plus what nobody else gives you:

• The exact settlement source for every city: station (ICAO), agency, unit, precision, hourly-only and fallback rules. Parsed from each market's own rules text, for all 51 cities.
• A live METAR running high/low at that exact station, and a per-bracket status: dead, leading, alive, final_proxy_win or final_proxy_loss.
• Settled history by date range, with the winning bracket's bounds.

It uses Polymarket's public market-data API and aviationweather.gov directly. No login, no API key, no browser. The default run takes a few seconds and costs about $0.08.

Things most people get wrong about these markets (all checked on 2026-09-24)

• Polymarket now settles 48 of 51 cities on NOAA station data, not Wunderground. The source is the weather.gov/wrh/timeseries page, using hourly data. Wunderground is only the fallback if NOAA has no data by 11:59 PM ET the next day. Only Jinan and Taipei still settle on Wunderground.

• Hong Kong settles on the Hong Kong Observatory's 0.1 °C reading, and the labels truncate. 32.9 °C falls in "32°C". This matched 62 of 62 settled HK events. The bracket_high_exclusive field encodes it.

• Kalshi and Polymarket use different stations in the same city:

  City	Polymarket	Kalshi
  NYC	LaGuardia (KLGA)	Central Park (KNYC)
  Chicago	O'Hare (KORD)	Midway (KMDW)
  Denver	Buckley (KBKF)	DEN (KDEN)
  Dallas	Love Field (KDAL)	DFW (KDFW)

• The METAR running extreme landed in the winning bracket 194/194 times (Sep 22–23, 48 stations, °C and °F). It is still a proxy, not the official value.

• Polymarket's displayed prices are not probabilities. On an empty or one-sided book the displayed price is roughly the midpoint. In NYC's Sep 26 low market, "72°F or higher" had no bid at all and an ask of 0.46, yet displayed 0.275. On Sep 24, 535 open brackets with no bid and no trade displayed 0.10 or more, and the displayed prices of the 254 live events summed to anywhere from 0.82 to 4.71. This actor never reports such a midpoint as a probability. implied_probability is null for those brackets, and event_favorite_bracket and implied_probability_normalized use real quotes only.

What you get

One dataset with three kinds of rows. record_type is always the first field.

record_type	One row per	Charged event
market	temperature bracket (≈11 per city-day-metric)	market-record
station_observation	settlement station × local day (METAR running high/low)	station-observation
notice	explanation (unknown city, no events, stopped at limit…)	free

Key market fields

Field	Example	Notes
city, metric, event_date	NYC, high, 2026-09-24	event_date is the station-local calendar day. city is stable across history (e.g. January's "Seoul" events are output as Seoul (Incheon), like today's); city_key is the short key
bracket_label	68-69°F	Polymarket's label
bracket_kind	range	range, exact, or_below, or_above
bracket_low, bracket_high	68, 69	inclusive bounds as labelled; null = open end
bracket_high_exclusive	70	the bracket wins iff bracket_low ≤ official value < bracket_high_exclusive
unit, settlement_precision	F, 1	C/F; 0.1 only for Hong Kong (HKO)
yes_bid, yes_ask, bid_size, ask_size	0.99, 0.992, 10, 208.39	best levels from the live CLOB book (USD per share)
mid, spread, last_trade_price	0.991, 0.002, 0.98	missing bid counts as 0, missing ask as 1. last_trade_price is the last YES-side trade (null if the YES token never traded)
implied_probability	0.991	mid if spread ≤ 0.10. On a wider book: the last YES trade clamped into [bid, ask], else the best bid (the only firm price). null when there is no bid and no YES trade on a wide or empty book, so an empty-book midpoint is never reported. Always within [bid, ask]
implied_probability_normalized	0.980703	implied ÷ event_implied_sum, only when that sum is 0.8–1.25 (a rescale of at most 25%). null on thin events (most of the probability unpriced) and on stale ones
price_flag	ok	ok, wide_spread, no_bid, no_book, resolved
price_source	clob_orderbook	or gamma_cached (catalog price, up to 5 min old)
status, result	open, null	open / closed_awaiting_resolution / resolved; yes/no when resolved
winning_bracket_label, settled_value_low, settled_value_high_exclusive	70-71°F, 70, 72	settled events: the winner's label and bounds, repeated on every row. Polymarket publishes no temperature, only the winner.
obs_extreme, obs_extreme_time, obs_count	68, 2026-09-24T19:51:00Z, 21	METAR running max (high) / min (low) so far, in settlement units
proxy_bracket_status	leading	dead / leading / alive, then final_proxy_win / final_proxy_loss once the local day is over
station_id, station_icao, station_name	KLGA, KLGA, LaGuardia Airport Station	HKO has no ICAO code
settlement_agency, settlement_source_url	NOAA, https://www.weather.gov/wrh/timeseries?site=klga	NOAA, Wunderground or Hong Kong Observatory
settles_on_hourly_obs_only, fallback_source	true, Wunderground	from the rules text
volume_usd, volume_24h_usd, liquidity_usd	9388.96, 5953.49, 440.58	per bracket; event_* versions per event
event_price_sum, event_implied_sum	1.0045, 1.0105	per event: sum of Polymarket's displayed yes_price / sum of implied_probability. A large gap usually means the displayed prices include midpoints of empty or one-sided books
event_favorite_bracket	68-69°F	the bracket with the highest implied_probability, if it is unique and at least 0.10; else null (thin book or a tie)
taker_fee_rate, tick_size, min_order_size	0.05, 0.001, 5
ids and links	event_id, market_id, condition_id, yes_token_id, no_token_id, event_url, market_url	ready for Polymarket's CLOB API

Input

Every field is optional. An empty input runs NYC live markets (high and low), with the order book and METAR proxy. That is about 66 bracket rows in a few seconds, for about $0.08.

Field	Type	Default	What it does
cities	list of strings	["NYC"]	Names (London, Seoul, Hong Kong), aliases (New York, LA, SF, HK, KL, CDMX, TLV), station ids (KLGA, EGLC) or series slugs. Case, accents and extra spaces are ignored, as are suffixes like "Chicago, IL". "all" selects every city.
metric	both | high | low	both	Daily highest or lowest temperature markets
mode	live | settled | both	live	live = open markets, usually today plus 1–2 days. settled = closed markets with results.
lookbackDays	integer 1–120	7	Settled/both mode only, when no dates are given
dateFrom / dateTo	string	—	Event-date filter: 2026-09-25, today, tomorrow, yesterday, -3d
includeObservation	boolean	true	Adds the METAR proxy fields and one station_observation row per station-day
liveOrderBook	boolean	true	Takes prices from the live CLOB book. Off = catalog prices, up to 5 min old.
maxRecords	integer	5000	Cap on charged market rows (1–100,000; 0 = no cap)

Unknown cities never fail a run. They are skipped, and you get one free notice row listing every valid city. Values that pass the input schema but cannot be used, such as a dateFrom of "next week" or a lookbackDays of 500, fall back to defaults or are clamped, with a warning. Dates are kept within 2020–2100. In settled mode dateTo never goes past tomorrow, and a window longer than 120 days keeps its latest 120 days.

Example inputs

{ "cities": ["all"] }

{ "cities": ["London", "Tokyo", "Chicago"], "metric": "high" }

{ "cities": ["NYC"], "mode": "settled", "dateFrom": "2026-09-01", "dateTo": "2026-09-23" }

Output example (real run, 2026-09-24 21:04 UTC)

This is the NYC high on Sep 24, at 5 PM ET. The METAR running max at LaGuardia had reached 68 °F, so every bracket below 68 is dead. 68-69°F is leading, and the book agrees: bid 0.99, ask 0.992, so it is also event_favorite_bracket. The event's implied values sum to 1.0105, inside the 0.8–1.25 band, so they are normalized (0.991 becomes 0.980703).

{
  "record_type": "market", "platform": "polymarket",
  "city": "NYC", "city_key": "nyc", "metric": "high", "event_date": "2026-09-24",
  "bracket_label": "68-69°F", "bracket_kind": "range",
  "bracket_low": 68.0, "bracket_high": 69.0, "bracket_high_exclusive": 70.0, "unit": "F",
  "status": "open", "accepting_orders": true,
  "yes_bid": 0.99, "yes_ask": 0.992, "bid_size": 10.0, "ask_size": 208.39,
  "mid": 0.991, "spread": 0.002, "last_trade_price": 0.98, "yes_price": 0.985,
  "implied_probability": 0.991, "implied_probability_normalized": 0.980703,
  "price_flag": "ok", "price_source": "clob_orderbook", "price_as_of": "2026-09-24T21:04:01Z",
  "result": null, "winning_bracket_label": null, "settled_value_low": null, "settled_value_high_exclusive": null,
  "obs_extreme": 68, "obs_extreme_time": "2026-09-24T19:51:00Z", "obs_count": 21,
  "obs_last_time": "2026-09-24T20:51:00Z", "proxy_bracket_status": "leading",
  "proxy_note": "METAR proxy from aviationweather.gov, not the official settlement value",
  "local_day_over": false, "timezone": "America/New_York",
  "station_id": "KLGA", "station_icao": "KLGA", "station_name": "LaGuardia Airport Station",
  "settlement_agency": "NOAA", "settlement_source_url": "https://www.weather.gov/wrh/timeseries?site=klga",
  "settlement_precision": 1, "settles_on_hourly_obs_only": true, "fallback_source": "Wunderground",
  "volume_usd": 9388.96, "volume_24h_usd": 5953.49, "liquidity_usd": 440.58,
  "event_volume_usd": 50281.73, "event_volume_24h_usd": 36726.56, "event_liquidity_usd": 67370.16,
  "event_open_interest_usd": 20866.33, "event_price_sum": 1.0045, "event_implied_sum": 1.0105,
  "event_favorite_bracket": "68-69°F",
  "tick_size": 0.001, "min_order_size": 5.0, "taker_fee_rate": 0.05, "bracket_index": 6,
  "end_date": "2026-09-24T12:00:00Z", "closed_time": null, "uma_resolution_status": null,
  "event_id": "1062921", "event_slug": "highest-temperature-in-nyc-on-september-24-2026",
  "event_title": "Highest temperature in NYC on September 24?",
  "market_id": "4834071", "market_slug": "highest-temperature-in-nyc-on-september-24-2026-68-69f",
  "question": "Will the highest temperature in New York City be between 68-69°F on September 24?",
  "condition_id": "0x7f415d40308196b3249e80e624753e1c64813c0810a5f1a2d1385f2f527d979b",
  "yes_token_id": "44328198862325790762178720785097949971157428241633354692885220915372009707360",
  "no_token_id": "64933037534463726121263305277451737728012588069625629357647321937985468868801",
  "neg_risk_market_id": "0x9ff4391a0a8fd7bd1d96480f1091f7882b3a9cffe1571323640268511a621900",
  "series_slug": "nyc-daily-weather",
  "event_url": "https://polymarket.com/event/highest-temperature-in-nyc-on-september-24-2026",
  "market_url": "https://polymarket.com/event/highest-temperature-in-nyc-on-september-24-2026/highest-temperature-in-nyc-on-september-24-2026-68-69f",
  "scraped_at": "2026-09-24T21:04:04Z"
}

The matching station row from the same run:

{
  "record_type": "station_observation", "station_id": "KLGA", "station_icao": "KLGA",
  "city": "NYC", "event_date": "2026-09-24", "timezone": "America/New_York", "unit": "F",
  "running_high": 68, "running_low": 54, "running_high_c": 20.0, "running_low_c": 12.2,
  "running_high_time": "2026-09-24T19:51:00Z", "running_low_time": "2026-09-24T07:43:00Z",
  "obs_count": 21, "first_obs_time": "2026-09-24T04:51:00Z", "last_obs_time": "2026-09-24T20:51:00Z",
  "local_day_over": false,
  "last_raw_metar": "METAR KLGA 242051Z 04013G27KT 10SM FEW040 SCT170 BKN250 20/06 A3029 RMK AO2 PK WND 06027/2048 SLP257 T02000061 56024",
  "source": "aviationweather.gov METAR",
  "proxy_note": "METAR proxy from aviationweather.gov, not the official settlement value",
  "scraped_at": "2026-09-24T21:04:04Z"
}

A thin market from the same run shows why implied_probability is sometimes null. This is the NYC low on Sep 26, two days out:

bracket_label	yes_bid	yes_ask	yes_price (Polymarket's display)	implied_probability
53°F or below	null	0.45	0.245	null (no bid, no trade)
58-59°F	0.27	0.40	0.315	0.36 (last YES trade)
60-61°F	0.21	0.25	0.23	0.23 (mid)
62-63°F	0.05	0.21	0.16	0.05 (bid: wide book, no trade)
72°F or higher	null	0.46	0.275	null (no bid, no trade)

Polymarket's displayed prices for that event sum to 1.715 (event_price_sum). The implied values sum to 1.05 (event_implied_sum). The favourite is 58-59°F, not one of the two unbid tails.

Pricing

Pay per event. You pay only for rows you actually receive.

Event	Price	When
actor-start	$0.005	once per run, only when at least one charged row is returned. If your maximum cost per run cannot cover the start plus one row, nothing is charged.
market-record	$0.001	per bracket row
station-observation	$0.002	per station-day METAR running high/low row
notice rows	free	unknown city, no events, limits reached, partial results

Typical bills, measured on 2026-09-24:

Run	Rows	Cost
Default (NYC, live, high and low, with proxy)	66 brackets + 1 station row	≈ $0.073
All cities, live, with proxy	≈ 2,800–3,200 brackets + ≈ 50–56 station rows	≈ $2.90–3.32
All cities, settled, last 7 days	≈ 8,200 brackets (raise maxRecords)	≈ $8.20
One city, settled, last 7 days	154 brackets	≈ $0.16

Use maxRecords and the run's maximum cost setting to cap spend. When a limit is hit, the run stops cleanly with what it has, plus a free notice. Bracket rows are bought first, so a tight budget never pays for a station row whose brackets it could not deliver.

How settlement works (FAQ)

What exactly decides the winning bracket? The official value is the day's highest (or lowest) reading at the station named in each market's rules. For 48 cities that is NOAA's weather.gov/wrh/timeseries?site=<ICAO> page, and US °F cities use the "Show Hourly Data" readings only. If NOAA has no data by 11:59 PM ET the next day, the Weather Underground daily observations table is used. With no data at all, the market resolves to the lowest bracket. Every row carries settlement_agency, settlement_source_url, settles_on_hourly_obs_only and fallback_source, parsed from that market's own rules. Nothing is hard-coded per city.

Why is Hong Kong different? Hong Kong settles on the Hong Kong Observatory's "Absolute Daily Max/Min" at 0.1 °C, and whole-degree labels truncate, so 32.9 °C is "32°C". Use bracket_low ≤ value < bracket_high_exclusive and it works for every city. HKO has no METAR station, so Hong Kong rows have no proxy.

How good is the METAR proxy? It takes the max (or min) of every METAR and SPECI report at the settlement station in the station-local calendar day, from aviationweather.gov. US °F values are converted as floor(°C × 1.8 + 32 + 0.5), the value NOAA displays. On 194 settled events on Sep 22–23 it fell in the winning bracket 194 times. Using every report is deliberate. NOAA's page for US °F cities lists hourly readings only, but on the 184 settled events of Sep 22–24 still inside the 72-hour window, all reports matched the winner 184 times and routine hourly METARs alone 183 times. The miss was Panama City's Sep 23 high, where a SPECI reading of 33 °C won. It is still a proxy: NOAA revisions or missing reports can differ. Every row says METAR proxy … not the official settlement value. The proxy is computed only for unresolved markets whose local day has started, within aviationweather.gov's 72-hour window. It is not available for Hong Kong (no ICAO) or Jinan (no METAR feed).

What does dead mean? For a high market the running max can only go up, so any bracket entirely below it can no longer win. For a low market the running min can only go down, so any bracket entirely above it is dead. leading means the bracket contains the running extreme right now. Once the local day is over, rows show final_proxy_win or final_proxy_loss until Polymarket resolves them, usually within hours.

How fresh are prices? With liveOrderBook on (the default), bid/ask/sizes come from Polymarket's live CLOB order book at run time (price_source = clob_orderbook), and last_trade_price from the CLOB's last YES-side trade. (The order book's own "last trade" field is the market's last fill on either outcome, so a NO fill at 0.99 would show up as 0.99; it is not used.) The catalog's own prices can be up to 5 minutes old (Cloudflare cache) and are used only as a fallback (gamma_cached).

How far back does history go? Polymarket's daily temperature markets go back to at least January 2026. The daily-temperature tag was applied inconsistently before March 2026. Named cities are fetched through their series and are not affected; all-city runs before March get a free partial_history notice. Some early events have 7 brackets instead of 11.

Which cities? All 51 cities as of September 2026: NYC, Chicago, Denver, Dallas, Atlanta, Miami, Austin, Houston, Seattle, Los Angeles, San Francisco, Toronto, Mexico City, Panama City, Sao Paulo, Buenos Aires, London, Paris, Amsterdam, Munich, Milan, Madrid, Warsaw, Helsinki, Moscow, Ankara, Istanbul, Tel Aviv, Jeddah, Karachi, Lucknow, Cape Town, Singapore, Kuala Lumpur, Manila, Beijing, Shanghai, Qingdao, Jinan, Zhengzhou, Wuhan, Chengdu, Chongqing, Guangzhou, Shenzhen, Hong Kong, Taipei, Seoul (Incheon), Busan, Tokyo, Wellington. The list is read live, so new cities work without an update.

Is this official? No. This is an unofficial tool, not affiliated with or endorsed by Polymarket. It reads Polymarket's documented public market-data API (no scraping of polymarket.com, no personal or wallet data). You are responsible for complying with Polymarket's Terms of Use, which restrict use of Polymarket data by capital-markets firms and market-data distributors without Polymarket's written consent. Weather data comes from aviationweather.gov (US public domain). Nothing here is investment advice.

For AI agents (Apify MCP server)

This actor is built to be called by agents.

• Inputs are forgiving. City names in any case, accent or spacing, aliases, station ids, unknown cities, a "city" key instead of "cities" and "date": "tomorrow" all work. Unknown values become warnings or a free notice row, never a failed run. The platform checks field types and allowed values before the run starts, so cities must be a list and metric / mode must use the listed lower-case values.
• Rows are flat, typed and self-describing. There are no nested objects and no long rules text. record_type tells rows apart, and proxy_note explains every empty proxy field. A null price field means "no real quote", never zero.
• Incomplete data is flagged in the dataset. If Polymarket's API fails partway, you still get the rows fetched so far plus a free partial_results notice. A run that could fetch nothing ends with a free upstream_unavailable (or upstream_format_changed) notice instead of an empty dataset.
• Small by default. Ask for one or two cities; "all" returns about 3,000 rows.
• Useful questions it answers: "Which NYC high brackets are already dead today?" (proxy_bracket_status = dead), "What's the market-implied distribution for London's low tomorrow?" (implied_probability_normalized; null when the book is too thin to say, so check event_implied_sum), "Which bracket won Tokyo's high on Sep 20?" (mode: settled, winning_bracket_label).

Related actors

• Kalshi Weather Markets Data — KXHIGH & KXLOW covers the same kind of daily temperature markets on Kalshi, with official settled temperatures. Watch the station differences above.
• METAR & NWS Weather Station Observations API gives raw and decoded METAR history for any station, including deep US history.