Commercial aircraft report temperature, wind and sometimes humidity as they climb and descend, giving vertical profiles much like a radiosonde ascent. NOAA's MADIS is the usual way to get them — but its decoder is currently losing most reports from outside the United States.
Measured over one six-hour window on 5 August 2026, the same day from both sources:
| Region | PREPBUFR | MADIS | × |
|---|---|---|---|
| North America | 162,203 | 161,154 | 1.0 |
| Europe | 53,916 | 6,866 | 7.9 |
| Asia | 19,709 | 4,843 | 4.1 |
| South America | 7,072 | 905 | 7.8 |
| Africa | 6,880 | 1,631 | 4.2 |
| Oceania | 5,889 | 1,496 | 3.9 |
North America agrees to within one percent while everywhere else runs four to eight times short. That pattern is a formatting problem, not a sampling or licensing difference. NOAA has acknowledged it and expects a fix by the end of 2026.
A few airframes report humidity that swings by a factor of a hundred between readings taken seconds apart. That is a sensor fault, not atmospheric structure, and it draws a sawtooth dewpoint trace. Those profiles are marked with an exclamation mark on their tab and a caution below the plot.
The trace is shown as measured, never smoothed. A median filter would produce a clean curve that no instrument recorded, and a forecaster can discount an obviously bad trace far more safely than an invented smooth one. Temperature and wind on these profiles are unaffected.
Any profile can be laid over a GFS or ECMWF IFS sounding for the same place and hour. These are the forecasts that were current at the time, not a reanalysis — the guidance a forecaster would actually have had. Useful for judging whether an observation your service never received would have changed the picture.
Model soundings reach back to mid‑2024; the aircraft archive goes back to 2015, so the comparison is unavailable on older case studies. GFS carries 19 pressure levels here, ECMWF 13.
It reads the same observations from a different processing chain: NCEP PREPBUFR, assembled from GTS reports independently of MADIS, so a message MADIS rejects can still survive. Individual reports are grouped by aircraft and split into ascents and descents, then plotted on a Skew-T.
Two honest limits. Profiles are reconstructed here, not supplied ready-made as MADIS supplies them — reports are grouped by aircraft and broken at gaps longer than 15 minutes, which will occasionally merge a quick turnaround or split a slow climb. And humidity is sparse: about 3% of reports carry it, almost all over North America.
Aircraft observations: NCEP ADP Global Upper Air and Surface Weather Observations (PREPBUFR), NSF NCAR GDEX dataset d337000, used under CC BY 4.0. Unlike the MADIS real-time feed these files carry no 48-hour restriction. Model soundings from Open-Meteo (CC BY 4.0), served as GFS (NOAA) and ECMWF IFS. Map tiles © Esri. Airport list from OurAirports (public domain). Skew-T rendering shared with the AMDAR Archive.
This site exists to cover a gap. When MADIS is fixed it should stop being necessary.