Local Level Lab

Airborne measurements. Open methods.

Directly scraped flight geometry — 2026-10-06

Four public FlightAware track tables cover high-latitude, equatorial and north–south geometry, including Europe/Middle East to South Africa. The tables were retrieved without an account or browser session. Each flight departed on 2026-10-02 UTC.

Flight Geometry Coordinate rows Provider-estimated rows Recorded span
Icelandair FI680, SEA–KEF High latitude; reported positions reach 67.97°N 560 70 7h00m
Singapore Airlines SQ938, SIN–DPS Equatorial, 1.50°N to 8.81°S 270 5 2h27m
Lufthansa LH572, FRA–JNB Europe–South Africa, 50.00°N to 26.13°S 602 262 9h38m
Emirates EK761, DXB–JNB Middle East–South Africa, 25.24°N to 26.15°S 518 174 7h36m

Source track pages:

The CSV filenames follow the yyyy-mm-dd airline number origin destination convention. Each has a matching .provenance.json with the source URL, retrieved HTML/export hashes, retrieval time, extractor hash, page timezone and UTC observation bounds. All coordinate rows in the public tables were extracted; event annotations and duplicate rounded mobile-display values were omitted. "Complete table" does not mean uninterrupted measured coverage.

The pages explicitly display EDT. The extractor reconstructs the dated New York clock from each actual history link, verifies the dated timezone against the page label, and exports UTC. normalized-tracks.json.gz preserves source labels and estimate flags for each position, elapsed time and SI units. summary.json reports gaps, missing values and position-coverage windows.

Estimated/approximate points are kept for broad path inspection but excluded from observed coverage. They often lack speed and altitude. In particular, the African routes have large estimated portions and cannot establish continuously observed motion through those portions. Maximum sampling gaps are 430, 148, 148 and 174 seconds. With a 90-second gap limit, the tracks have 10, 0, 12 and 0 overlapping 75-minute windows with >=95% reported-position coverage. This screening threshold is provisional; zero qualifying windows does not erase the route's usefulness for coarse geometry. It does preclude pretending that the public track is dense receiver data. Windows are not independent flight samples or accepted scientific flights.

The raw HTML downloads remain temporary inputs. The factual CSVs and provenance are the durable source artifacts for later geometry work. Public-history links can expire or change. These tracks have no IMU, measured wind or mount-turn schedule. Sensor/nuisance processes must be simulated explicitly in any later trajectory study. The active 1,000-flight pilot is unchanged.

Extraction: analysis/tests/extract_flightaware_track.py; normalization: analysis/tests/import_flight_tracks.py. Seven lightweight checks cover table annotations, full coordinate precision, UTC day crossing, gaps, missing values and provider estimates.