swadhinbiswas/air-traffic

EU Air Traffic Lake

Live and scheduled European air traffic as versioned Parquet: raw Bronze intake windows from the collector, plus curated Silver snapshots refreshed by a scheduled pipeline.

Updates

• Bronze windows land continuously (positions every ~5 min, weather every 5 min, departures twice an hour, arrivals backfilled nightly).
• Each of the Silver snapshots below is replaced every 15 minutes by the lake pipeline.
• The Eurostat airport-traffic benchmark and dbt Gold marts are not stored here; see eu-air-traffic for the warehouse.

Contents

Silver snapshots (silver/<source>/data.parquet)

• silver/positions/data.parquet — Live aircraft position snapshots (deduped by airframe). Key columns: icao24, callsign, latitude, longitude, altitude, velocity.
• silver/flights/data.parquet — Completed movements from OpenSky plus AirLabs schedules. Key columns: flight_id, callsign, departure_icao, arrival_icao, status, delay_minutes.
• silver/weather/data.parquet — METAR observations with flight categories. Key columns: station_icao, timestamp, temperature_c, wind_speed_kt, flight_category.
• silver/weather_forecast/data.parquet — Open-Meteo hourly forecasts. Key columns: station_icao, timestamp, temperature_c, condition.
• silver/weather_taf/data.parquet — Raw terminal aerodrome forecasts. Key columns: station_icao, issue_time, valid_from, valid_to, raw_taf.
• silver/fuel/data.parquet — Daily jet-fuel prices by region. Key columns: date, region, price_per_litre, currency.
• silver/routes/data.parquet — EU route network, ICAO-mapped with great-circle distance. Key columns: airline, origin, destination, distance_km, stops, equipment.
• silver/airports/airports.parquet — Enriched EU airport reference. Key columns: ident, name, type, latitude_deg, longitude_deg, iata_code.
• silver/holidays/data.parquet — Public-holiday calendar by country. Key columns: country, date, name.
• silver/aircraft/data.parquet — Aircraft type reference (ICAO 8643 + specs). Key columns: type_icao, manufacturer, family, capacity, range_km.
• silver/notams/data.parquet — Notices to air missions. Key columns: notam_id, icao_location, notam_type, valid_from, valid_to.
• silver/emissions/data.parquet — Per-type hourly fuel and CO2 rates. Key columns: aircraft_type, fuel_burn_kg_per_hour, co2_kg_per_hour.

Bronze intake windows

• bronze/parquet/<source>/<source>_YYYY-MM-DDTHHMMSSffffffZ.parquet — immutable windows exactly as drained from Kafka.
• bronze/raw/<source>/<date>/*.jsonl — the same records before Parquet conversion.
• Windows accumulate: download a prefix such as bronze/parquet/flights/ for history.

Field reference

silver/positions/data.parquet

icao24, callsign, registration, aircraft_type, latitude, longitude, altitude, altitude_geom, velocity, heading, vertical_rate, mach, ias, tas, oat, wind_dir, wind_speed, squawk, emergency, category, on_ground, source, collected_at, fuel_burn_kg_per_hour, co2_kg_per_hour, aircraft_class, emitter_class, is_cargo, is_military, operator_name, operator_country, operator_category, type_name, manufacturer, airframe, wake_category

Note: Type-specific columns may be null (e.g. squawk only for transponders that send it).

silver/flights/data.parquet

flight_id, callsign, airline_icao, airline_name, departure_icao, departure_iata, arrival_icao, arrival_iata, scheduled_departure, scheduled_arrival, actual_departure, actual_arrival, status, delay_minutes, cancelled, source, collected_at, ingestion_date

Note: delay_minutes/actual_* are null when the source has no schedule (OpenSky movements).

silver/weather/data.parquet

station_icao, timestamp, temperature_c, humidity_pct, wind_speed_ms, visibility_m, condition, pressure_hpa, source, ingestion_date, name, latitude, longitude, dewpoint_c, wind_dir_deg, wind_speed_kt, gust_kt, visibility, altimeter_hpa, flight_category, cover, raw_metar, observed_at, collected_at, visibility_raw

Note: wind_dir_deg is stored as a string.

silver/weather_forecast/data.parquet

station_icao, timestamp, is_forecast, temperature_c, humidity_pct, precipitation_mm, wind_speed_ms, wind_direction_deg, weather_code, condition, wind_unit, source, collected_at, ingestion_date

silver/weather_taf/data.parquet

station_icao, issue_time, valid_from, valid_to, raw_taf, source, collected_at

silver/fuel/data.parquet

date, region, price_per_litre, currency, source, collected_at, ingestion_date, series_key

silver/routes/data.parquet

airline, origin, destination, stops, equipment, distance_km, source, collected_at, ingestion_date, _kind, id

silver/airports/airports.parquet

ident, name, type, latitude_deg, longitude_deg, elevation_ft, iso_country, municipality, iata_code, score, scheduled_service, ingestion_date

silver/holidays/data.parquet

country, date, name, source, collected_at, ingestion_date, _kind, id

silver/aircraft/data.parquet

type_iata, type_icao, manufacturer, family, engine, capacity, range_km, source, collected_at, ingestion_date, _kind, id

silver/notams/data.parquet

notam_id, icao_location, notam_type, message, qualification, valid_from, valid_to, source, collected_at, ingestion_date

silver/emissions/data.parquet

aircraft_type, fuel_burn_liters_per_hour, fuel_burn_kg_per_hour, co2_kg_per_hour, co2_tonnes_per_hour, emission_factor_kg_per_kg_fuel, fuel_density_kg_per_liter, source, collected_at, ingestion_date, _kind, id

Data sources

• Positions: ADS-B via adsb.lol, with airplanes.live and OpenSky fallbacks.
• Movements and delays: OpenSky (/flights/*) plus AirLabs schedules.
• Weather: METAR/TAF from aviationweather.gov, forecasts from Open-Meteo.
• Reference: airports, routes, aircraft, emissions and holidays built from bundled reference data.
• Official passengers: Eurostat avia_paoa (monthly, ~2 months behind), loaded as a benchmark layer.

Reproducibility

• Bronze files are immutable and timestamped; Silver files are snapshots of the latest state.
• Local paths mirror the repo: warehouse/bronze/<source>/…, warehouse/silver/<source>/….
• Regenerate locally with uv run python -m scripts.lake_sync pull-silver, then uv run pytest; see docs/huggingface-dataset.md for the scripts and cadence.

License

MIT — same license as the eu-air-traffic repository.

Citation

@misc{swadhinbiswas_air_traffic_lake,
  author = {Swadhin Biswas},
  title = {EU Air Traffic Lake},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/swadhinbiswas/air-traffic}
}

The software that produces this dataset is archived at 10.5281/zenodo.22790202 (v0.1.0).

Limitations

• OpenSky movements carry no schedule, so delay fields are null there; use AirLabs schedules or Gold delay marts for punctuality.
• Arrivals are backfilled nightly; same-day arrival coverage lags.
• METAR wind direction is a string in the source feed (wind_dir_deg), not numeric.
• TAF validity bounds are epoch integers, not ISO timestamps.
• This dataset mirrors the live pipeline: expect drifting schemas over months, pinned by Silver snapshots.