Searching for separation between frequency-modulated calls of blue and fin whales.

Reliable differentiation of similar acoustic signals remains a major challenge for passive acoustic monitoring (PAM) of sympatric marine mammal species. Blue (Balaenoptera musculus) and fin whales (Balaenoptera physalus) produce frequency-modulated (FM) downsweeping calls with overlapping frequency ranges and durations, making species attribution difficult. This study combines long-term PAM datasets with animal-borne tag recordings to investigate how these FM calls can be reliably distinguished by species. Tag recordings with confirmed species identity showed substantial overlap in most acoustic features, but the parameter Duration 90% offered partial separation: with most fin whale FM calls < 0.6 s, and blue whale FM calls > 0.7 s. Applied to high-quality calls from moored data with signal-to-noise ratios above 12 dB, this approach may provide a conservative, rule-based means of classification. A more robust approach was achieved by combining deep learning feature extraction with non-linear dimensionality reduction. Integrating confirmed ground-truth tag recordings as a reference space enabled previously unseen FM calls from long-term PAM recordings to be attributed by acoustic similarity to known calls while maintaining an explicit measure of uncertainty by leaving acoustically ambiguous calls unassigned. Together, these approaches provide a practical framework for more reliable call attribution, enhancing long-term PAM-based monitoring of blue and fin whales.