Forecasting · documentation
↗ Open dashboard
Data & methodology

Forecast methodology

How forecasts are matched to markets, what the percentiles and predictability index mean, and why the settlement airport and its coordinates matter so much.

Settlement airports

Each Polymarket weather market resolves against the official reading at one specific airport weather station. The dashboard maps every tracked city to that settlement airport — its ICAO code, exact latitude/longitude, elevation, timezone, observation source and unit system. Forecasts are requested for those precise coordinates, so the model signal lines up with what will actually settle the market.

Why coordinates matter A market settles on one station. If the forecast is pulled for a slightly different point — a city centre instead of the airport, or the wrong field — the model number can be off by a degree or more, which is the difference between buckets. Keeping each city's coordinates pinned to its settlement station is what makes the comparison valid.

Two tiers of city

TierWhat works
Full pipeline (the 11 European settlement airports) Model forecast, exceedance percentiles, predictability, and the METAR observation chart — every panel is populated.
Market-only (most non-European cities) They appear as tabs with full Polymarket market data, but their per-city weather panels show “no settlement data” because no observation source is wired up.

See City & continent coverage for the full list.

Exceedance percentiles

Rather than a single point forecast, the model exposes a distribution as percentiles P5…P95. Each percentile gives the temperature at which that exceedance probability holds for the day. P50 is the median — the 50/50 value — and is also used to pick the consensus peak hour. Reading the spread between, say, P10 and P90 tells you how uncertain the day is.

In the detail table, lower percentiles (warm, upper tail) are tinted red and higher percentiles (cool, lower tail) are tinted blue, so the shape of the distribution is visible at a glance.

Predictability index

meteoblue publishes a daily predictability value from 0 to 100: the probability that the forecast for that day will not change materially as it approaches. The dashboard bands it to match meteoblue's own language:

TierRangeReading
high≥ 70%Forecast is settled; little expected movement.
medium40–69%Some model disagreement remains.
low< 40%Models still diverge; expect the forecast (and price) to move.

Units

Each airport carries a unit: European and several other cities are tracked in Celsius, United States airports in Fahrenheit, matching how the corresponding markets are quoted. Bucket parsing handles both — the integer in a label like 21°C or 65°F is the bucket value.

Observations vs settlement

The METAR “actual” readings are a live sanity check, not the settlement value. They can be sparse or delayed, and a market settles on its own designated official source. Use actuals to see how the day is developing, not to call the result.

Forecasting dashboard documentation · Meteo Bulgaria EOOD · 2026