Answers are grounded in this paper and cite their sources.
Radio meteor observations produce large numbers of spectrograms whose automatic analysis is challenging because meteor echoes vary strongly in morphology and contrast and may overlap with aircraft reflections or other signals. In the Radio Meteor Zoo project, volunteers identify meteor echoes by drawing bounding boxes around their spectrogram signatures, which we refer to as annotations in this paper. We investigate whether multiple independent volunteer annotations can be aggregated into reliable meteor detections for scientific use. We benchmark five aggregation methods: the original pixel-consensus procedure, DBSCAN, HDBSCAN, graph-based clustering, and equal-weight box fusion. The input dataset of 2,199 BRAMS (Belgian RAdio Meteor Stations) spectrograms was divided into a development set (used for method selection) and a held-out test set of 556 spectrograms containing 5,302 expert-marked meteors. In the held-out subset, equal-weight box fusion provided the best overall event recovery, whereas the pixel consensus baseline produced the most conservative catalogue. DBSCAN and graph-based clustering produced identical detections under the selected settings and tighter bounding boxes than box fusion, while HDBSCAN recovered fewer events. Low-contrast events were less likely to be recovered. Moreover, many detections with limited bounding-box overlap still corresponded to the same meteor, highlighting differences in annotated extent rather than event identity. These results show that citizen-science annotations can yield reliable meteor catalogues, with the preferred aggregation method depending on whether completeness or accurate localisation is more important for the intended scientific application.
No takes yet. Share an insight, caveat, or question.
Calders et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: