Aerodrome Movement Monitoring Using ADS-B Data: A Case Study at Lommis Airfield

Abstract

This study presents and validates a machine learning pipeline that transforms raw Automatic Dependent Surveillance-Broadcast (ADS-B) data into structured aerodrome movement reports, addressing regulatory needs for continuous monitoring of aircraft operations at small, non-towered airfields. The approach automatically identifies aerodrome-specific flight events, particularly repetitive traffic circuits, which constitute a significant portion of General Aviation traffic at such airfields. Using ADS-B data observed at Lommis Airfield, a representative regional airfield in Switzerland, we filtered and preprocessed raw flight trajectories and segmented those meeting the filtering criteria into aerodrome circuit candidates. We then formulated circuit detection as a supervised binary classification problem and compared five machine learning approaches: Logistic Regression, Random Forest, unidirectional and bidirectional Long Short-Term Memory (LSTM) networks, and a 1D Convolutional Neural Network (CNN). Each traffic circuit candidate was characterised by eight engineered features capturing kinematic and flight-phase information. The 1D CNN model achieved 99.15% accuracy, outperforming rule-based heuristics by 25.5 percentage points in recall, while simpler models (Logistic Regression, Random Forest) reached comparable performance with higher interpretability and efficiency. End-to-end validation of the proposed pipeline over a three-month period yielded 67.6% overall detection coverage (438 of 648 flights), limited primarily by ADS-B data availability rather than model performance. The validated pipeline demonstrates the potential for a scalable path toward automated, data-driven movement reporting, with full end-to-end validation conducted at a single airfield and additional cross-airfield evidence shown at the model level.