AeroScope provides researchers with real-time and historical ADS-B flight data through multiple export formats and a comprehensive API. Designed for academic rigour and reproducibility.
The AeroScope ADS-B Anomaly Benchmark (v1) is a free, citable dataset for evaluating ADS-B anomaly and spoofing detectors. It pairs real airborne traffic captured live from crowdsourced receiver feeds (multi-region, deduplicated by ICAO 24-bit address) with synthetically injected attacks following the standard taxonomy in the ADS-B security literature — altitude tampering, velocity tampering, ghost kinematics, and integrity-field downgrade (methodology after Habler & Shabtai, Computers & Security, 2018).
Each record carries 38 documented columns: raw ADS-B fields (position, kinematics), the signal-integrity fields that matter for spoofing detection (NIC, NACp, NACv, SIL), derived self-consistency features (geometric-vs-barometric altitude difference, ground-speed-vs-Mach residual, barometric-vs-geometric vertical-rate difference, track-vs-heading difference), and the anomaly score from our deployed scikit-learn IsolationForest baseline. That baseline's measured performance at training time is ROC-AUC ≈ 0.87 with roughly a 50% detection rate at a strict ~2% false-positive rate (threshold calibrated on held-out real traffic).
Download: CSV JSONL 🤗 Hugging Face ↗ Dataset card & column reference
Also published on Hugging Face Datasets with a load_dataset() loader. A Zenodo DOI is being minted.
Licence: CC-BY 4.0. Please cite as “AeroScope ADS-B Anomaly Benchmark v1, aeroscope.live”. Honest limitations: synthetic injection is a proxy for real attacks; coverage is biased toward busy terminal areas; enhanced-surveillance fields (IAS/TAS/Mach/heading) are broadcast by only a subset of aircraft, so those columns are intentionally sparse. Full details in the dataset card.
AeroScope runs six independent, torch-free detectors — River (HalfSpaceTrees), scikit-learn IsolationForest, a pykalman innovation test, OpenAP envelope checks, stumpy matrix profiles and PyOD — and flags an aircraft only when at least two agree. Researchers can study how consensus voting trades sensitivity against false positives on real traffic, and benchmark unsupervised detectors against the published baseline. There is no reinforcement learning; an earlier RL prototype was retired.
Detectors are stress-tested by injecting synthetic attacks into real traffic — altitude tampering, velocity tampering, ghost kinematics and integrity-field downgrade — following the standard ADS-B security taxonomy. Labels are used only to compute ROC-AUC and detection at a fixed false-positive budget, not to train any self-play agent. The captured records, injection labels and deployed-model scores are published so the evaluation can be reproduced. Method →
AeroScope builds per-aircraft behavioral baselines using DBSCAN clustering for route patterns and Isolation Forest for point anomaly detection. Researchers studying anomaly detection can access baseline data showing normal speed/altitude/route distributions and deviation scores. This provides a labeled dataset of behavioral anomalies in real-world aviation traffic — useful for benchmarking unsupervised learning methods.
Each broadcast is checked against its own integrity fields (NIC/NACp/NACv/SIL), a Kalman normalised-innovation test, and self-consistency residuals — geometric-vs-barometric altitude, ground-speed-vs-Mach and track-vs-heading. This gives researchers a quantitative, reproducible surface for studying ADS-B spoofing and message integrity on real-world traffic. Integrity model →
AeroScope fuses 60+ public ADS-B feeds plus a satellite-ADS-B layer, de-duplicated by ICAO 24-bit address. Because independent receiver networks often see the same airframe, the overlap is a natural cross-validation surface — researchers can study coverage geometry, inter-network agreement, and how reconciling sources affects position quality and integrity. Fusion →
Study aircraft movement patterns including holding patterns, approach procedures, departure routes, and traffic flow. AeroScope's pattern detection algorithm identifies orbits, racetracks, loiter behavior, and grid searches automatically, providing a pre-classified dataset for pattern research.
Analyze traffic density, altitude layer distribution, temporal patterns, and sector loading. AeroScope's airspace complexity scoring provides a quantitative framework for studying congestion, capacity, and safety margins in different airspace volumes.
Study ADS-B signal characteristics including reception rates, position accuracy, message types, and spoofing detection. AeroScope's DO-260B 7-check validation and Kalman NIS test provide a foundation for cybersecurity research in aviation surveillance systems. The 9-dimension signal integrity score offers a quantitative metric for signal quality research.
Ideal for spreadsheet analysis and statistical software (R, SPSS, Stata). Each row represents one aircraft at one point in time, with columns for ICAO hex, callsign, latitude, longitude, altitude, speed, heading, vertical rate, squawk, aircraft type, threat score, and all analysis flags.
GET /api/export/csv
Accept: text/csv
Authorization: Bearer <token>Structured data format ideal for programmatic analysis in Python, JavaScript, and other languages. Includes nested objects for analysis results (threat scoring breakdown, pattern classification details, drone detection rationale).
GET /api/export/json
Accept: application/json
Authorization: Bearer <token>Geographic data format compatible with GIS tools (QGIS, ArcGIS, Mapbox, Leaflet). Each aircraft is a GeoJSON Feature with Point geometry and properties containing all telemetry and analysis data. Ideal for spatial analysis and mapping projects.
GET /api/export/geojson
Accept: application/geo+json
Authorization: Bearer <token>For automated data collection, use the REST API with over 50 endpoints. Common research workflows include:
/api/aircraft at regular intervals (e.g., every 12 seconds) and store results for longitudinal analysis/ws for continuous real-time data without polling overheadSee the full API documentation for endpoint details, authentication, and rate limits.
When using AeroScope data in academic publications, please cite as follows:
@misc{aeroscope2026,
title={AeroScope Real-Time ADS-B Flight Tracking Platform},
author={{AeroScope}},
year={2026},
url={https://aeroscope.live},
note={Accessed: [date]}
}If your research uses specific analysis features (threat scoring, pattern detection, drone detection), please mention the specific module version in your methodology section for reproducibility.