Forecasts vs METAR truth · where each source goes wrong
Error and bias maps
How to use this page · tap any ? for a full explanation
- Pick a model and an element (wind, temperature, ceiling, fog...). Each dot on the map is an airport, coloured by how that model has verified there against the METARs.
- Start with bias: which way the forecast leans (wind too light, ceilings too high). For yes/no events like fog, frequency bias says whether it is forecast too often or too rarely.
- Narrow it down by lead time, season or weather regime. Many errors only show in one kind of weather, such as clear calm nights or chinook wind.
- Click an airport on the map or in the table for its full numbers.
- Trust the clear ones: faded dots are within noise; empty rings have too few cases to judge.
New here? Start with what this page is for, then bias and error scores. Words are explained in the glossary at the bottom.
Each dot is an airport, coloured by how a forecast source has verified there against the METARs. Pick a model, an element and a score; narrow it to a lead time, a season or a weather regime to find patterns, such as where a model forecasts wind too light, ceilings too high, or ends fog too early. Every forecast is scored as it was shown on this site at the time. Airports with fewer than 30 cases for the selection are drawn as empty rings rather than as a number that is mostly noise.
Map
Airports
How it is worked out
Confidence: forecasts at one airport on the same day share their errors, so the number of verified hours overstates how much evidence there is. For bias, the page counts about two independent cases per day with data (fewer when there are fewer hours) and calls a bias clear when it is at least twice its uncertainty, possible when it is one to two times, and within noise below that; within-noise airports are drawn faded.
Glossary: words used on this page
See the site glossary.