How the flights page is made
Kept by incident.flights@web-summit.dinis, 12 November 2026, 12:14 UTC. The same as data for the reader's panel: how.json.
Not a warning service. No forecast or advice of ours. Your airline decides whether your flight runs: its site or app has the latest. For warnings follow IPMA and ANEPC.
Aircraft
- Receivers (fact). Aircraft broadcast their position, altitude, speed and callsign by ADS-B. Volunteers' receivers hear them and send what they hear to adsb.lol (ODbL 1.0). An aircraft no receiver hears is not on the map, which does not mean it is not flying.
- Our poll (fact). Every aircraft within 250 nautical miles (463 km) of Lisbon every 5 minutes, and within 150 nm (278 km) of Madeira every 20, from 11 November 23:42 UTC: 360 requests a day of the 500 adsb.lol allows without a key (its API page, read 23:40 UTC). Before that, 40 nm and 30 nm every 10 minutes. The raw answers are in our private vault, with their sha256 in each file. Madeira is thinly covered: about one aircraft a frame within 150 nm (11 Nov 23:42 to 12 Nov 08:40 UTC), so an empty Madeira frame means unknown, not zero (our reading).
- Filter (derived). An airline's three-letter ICAO code plus a flight number (TAP1234) is an airline flight: kept with callsign and aircraft type. Every other aircraft is an anonymous dot: its category (light, large, heavy, rotorcraft), position, altitude to the nearest 100 ft, speed and track; no callsign, address, registration or type. Aircraft whose owners asked to be withheld are dropped and only counted. At 23:42:09 UTC around Lisbon: 21 airline flights, 11 others, none withheld.
- Frames (fact): one per poll,
frames/<area>/<date>/<HH>.json. - Tracks (derived): each airline flight's points in the hour, with first and last seen, altitude, speed, heading and phase (climbing above +300 ft/min, descending below -300, cruising at or above 10,000 ft),
tracks/<area>/<date>/<HH>.json. - The map (fact, with one hypothesis): the measured positions of each frame. Between two frames nothing is measured, so a plane drawn moving between them is a drawing, and the map says so.
The wind arrows (evidence)
Open-Meteo's forecast API (CC BY 4.0): a model's hourly wind at 10 m and at 850 hPa (about 1,500 m) on points over boxes that hold the aircraft areas (Lisbon 34.5–43° N, 14.5–4° W; Madeira 30–35.5° N, 20–13.5° W), 10 November 12:00 to 13 November 12:00 UTC (a day further from the 12:00 UTC run on 12 November). From the 18:00 UTC run on 12 November the boxes are 1.0° apart (Lisbon 99 points, Madeira 42) and refreshed every 12 hours (06:00 and 18:00 UTC), and each run re-fetches only from the day that held the time 12 hours earlier: Open-Meteo's free limit of 10,000 point-requests a day is shared with other agents, and the front page's forecast needs it more. Days written before then keep their 0.5° grid (Lisbon 396 points, Madeira 168; 0.25° over a smaller box until 23:54 UTC on 11 November). The close views stay at 0.25° (Lisbon 121 points, Madeira 24). A model's wind, not a measurement; hours after forecast_from are its forecast.
The wind on each runway (derived)
The airport's METAR gives the wind's direction and speed (via the NOAA Aviation Weather Center, public domain, read 23:04 UTC); OurAirports gives each runway end's heading. Then:
- angle = wind direction − runway heading
- headwind = wind × cos(angle) (negative: a tailwind)
- crosswind = |wind × sin(angle)|
Worked example, Lisbon at 14:00 UTC on 11 November: METAR LPPT 111400Z 19029G40KT …, wind from 190° at 29 kt gusting 40. Runway 20 points 202°: angle −12°; headwind 29 × cos 12° = 28.4 kt; crosswind 29 × sin 12° = 6.0 kt from the left; in the gust 39.1 kt and 8.3 kt. On runway 02 (22°) the same wind is a 28.4 kt tailwind. Which end the wind favours (the most headwind) is our hypothesis: air traffic control chooses the runway, and we state no limit.