Everything you wanted to know about how SkyMeter ingests, processes, and replays flights.
SkyMeter turns raw flight broadcasts into something you can actually watch. We pull every ADS-B trace we can get our hands on, run it through our processing pipeline, attach the weather that was happening at the time, and rebuild the flight as a 3D replay you can scrub through.
If FlightAware tells you "AAL123 went from JFK to LAX," SkyMeter tells you the pitch attitude on rotation, the headwind on final, where the touchdown happened relative to the threshold, and whether it was a stable approach.
Flights come from the global volunteer ADS-B receiver network — thousands of ground stations on every continent that pick up the position broadcasts modern aircraft transmit several times per second. We aggregate the raw traces into our archive, then enrich them with airport, aircraft, weather, and runway data from FAA / OurAirports / NOAA / ECMWF.
Aircraft type and tail-number details come from the FAA registry plus crowd-sourced ICAO24 mappings. Airline data comes from OpenFlights and our own correction layer for callsigns the public databases miss.
Yes. Every replay you see on SkyMeter is a real flight that actually happened — no simulation, no demo data. Position, altitude, ground speed, and squawk code come straight from the aircraft's ADS-B transponder. We compute attitude (pitch, roll, yaw) and indicated airspeed by combining position + time with weather, but we don't invent flights.
No. SkyMeter is a historical replay platform. Most flights show up within 24-48 hours of completion, after our pipeline has had time to ingest the full trace, match it against schedules, attach weather, detect runway events, and build the proto file. If you need live tracking, FlightAware or FlightRadar24 are better tools — we do the part where you go back and study what happened.
ADS-B is line-of-sight, which means the aircraft has to be inside the reception cone of at least one ground receiver to be heard. Coverage is dense over land, near airports, and near anywhere ADS-B hobbyists tend to live. It's much thinner over oceans, polar regions, big chunks of central Africa, parts of Russia, and most of Antarctica.
When a flight crosses one of those gaps, the trace genuinely goes silent — we aren't hiding the data, we just don't have it. Our pipeline interpolates over short gaps (a minute or two) so the replay still feels continuous, but for longer stretches we'd rather end the flight cleanly than fabricate a position.
If a flight stops over land in a region we'd expect to have coverage, the most common cause is the transponder being switched off after landing on a taxiway without a visible runway — pilots vary on when exactly they kill the squawk.
Yes — and not just at the surface. Every flight is enriched with a hybrid weather model that combines surface METAR observations from the airports near the route with upper-air data from NOAA's GFS forecast and ECMWF's ERA5 reanalysis. Our weather servers refresh METARs continuously and re-bake the wind cube on a daily cadence, so by the time you watch a replay the wind triangle has already been solved at every sample.
Practically, that's why the cockpit panel can show you Indicated Airspeed and Mach (computed from ground speed + true wind + air density) instead of just GPS ground speed, and why the runway dashboards can break down crosswind and headwind components per landing.
Honest answer: very good for what it is, but not a substitute for an AHRS. We don't have a real attitude sensor on the aircraft — we reconstruct attitude from position derivatives, computed airspeed, and aerodynamic models specific to the aircraft type.
We've validated the output against real Garmin AHRS exports from light aircraft that share the same flight: pitch correlates within about a degree on average, roll within about r=0.88 across a full flight, and the model picks up pilot-induced events (stall break, side-slip, base-to-final overbank) cleanly. It's good enough to make the replay look right and to drive incident detection; it's not good enough to certify against an FDR.
Each detector is its own algorithm. Go-arounds use the MDPI 2020 Method-5 paper: a short descent below ~500 ft AGL within a few miles of the runway, followed by a positive climb — that's the canonical signature. Unstable approach uses the stable-approach gates from the Flight Safety Foundation (target speed, descent rate, configuration) sampled at 1,000 ft and 500 ft AGL.
Stalls and overspeeds are detected from computed IAS vs. type-specific Vs, Vmo, and Mmo. Rejected takeoffs come from speed-then-stop signatures on departure runways. Runway excursions look for centerline deviation past the threshold threshold lights. Every flag is a heuristic — they're meant to surface interesting flights for a human to look at, not to certify what actually happened.
Most small airports and helipads don't publish their own METAR — only the larger fields run an automated weather observing system. When the airport you're looking at doesn't broadcast one, our cache borrows from the closest medium or large airport that does, so you still get a reasonable read on conditions. The badge on the weather card tells you which station and how far away — typically within 10-20 miles, which is close enough that surface wind and pressure are usually a good match.
We pull from a different ADS-B aggregator and we run a different processing pipeline, so small differences in landing time, distance, or max altitude are expected. We also tend to be more aggressive about splitting traces into separate flights (e.g. a touch-and-go at an enroute airport), and we recompute things like routing, landing runway, and taxi distance from the trace rather than trusting the FAA flight plan.
If something looks really wrong (wrong airport, wrong aircraft type), please tell us — usually it's a stale registration record we can fix.
Free covers the most-recent 7 days of flights with full replay, weather, and airport / runway dashboards. Pro extends that to a rolling 365 days, unlocks bulk download (FDR for X-Plane & MSFS, raw .smr proto, and CSV exports), and adds advanced filters and the API. See /pricing for the current tiers.
Yes. Every flight has a Download button on its replay page that exports:
X-Plane 12/Output/FDR and replay it inside the sim with full attitude.Yes — Pro users get a REST API with the same data the dashboards use: flight search, route summaries, airport stats, runway events, and proto download URLs. Documentation lives at /api.
Ground Speed (GS) is what GPS measures — how fast the aircraft is moving over the dirt. It's the only "true" speed in the cockpit panel; the others are computed.
True Airspeed (TAS) is GS minus the wind component along the aircraft's heading. Indicated Airspeed (IAS) is what the pilot actually flies the airplane on — TAS corrected backward for air density (basically, "what the pitot tube would read"). At cruise altitude, IAS can be 160-200 kt slower than TAS. Mach is the speed of sound ratio, which matters above ~Mach 0.7 where compressibility starts to bite.
Our pipeline computes IAS and Mach from GS + wind + temperature on every sample, which is why the EFIS tape on the replay reads more like a real cockpit than a ground-track readout.
.smr is the SkyMeter Replay format — a Protocol Buffers file that
contains the full enriched flight: every position sample, computed attitude,
weather snapshot, runway events, and flight-phase classification. It's typically
100-500 KB per flight, gzip-compressed.
We treat the underlying ADS-B trace as an ODbL Derivative Database, which means the .smr download is free and attributed (you can find the source notice inside the file). The convenience formats (FDR, CSV, KML) are paid because they require conversion and additional metadata.
We try to answer everything ourselves before asking you to write in. If you've got a flight that looks wrong, an aircraft type that's mis-identified, or a feature you'd love to see, drop us a note at contact@skymeter.io. We read every message.