Figure 6. PBD dataset UMAP ordination at ~1 min resolution. Samples grouped by hydrophone deployment location (left). Samples grouped by sound source (right). All other samples are shown in grey.
Samples labelled as seismic airgun noise and ship noise separated and occupied two distinct portions of the UMAP ordination plot (Fig. 6). A small number of samples from the two sources overlapped, indicating ship transits occurring during seismic exploration. However, we could not observe a clear distinction between presences and absences of the sources.
Lastly, we inspected how UMAP ordinated the ~1 s acoustic features labelled by their chance of containing a humpback whale vocalization (Fig. 7). Detections per hour peaked at 1:00 and 13:00 and 15:00 for the Burin samples, while the Red Island samples showed a single distinct peak at 12:00 (Appendix S2.1). The ~1 s resolution UMAP ordination showed a concentration of humpback whale detection scores (> 0.8) towards the right end of the plot, with samples densely aggregated along the second UMAP dimension. However, and similarly to the anthropogenic noise sources, we could not observe a clear separation between presences and absences.