Predicting seabird-vessel interactions from bird-borne data alone

GPS-only and environmental random-forest models predict vessel-attendance patterns for 273 Calonectris diomedea (Scopoli’s shearwaters), with AUC = 0.85 against AIS in hourly bins.

• Why it matters: Spatiotemporal alignment of seabird and vessel tracks can miss interactions when vessel tracking is incomplete or unavailable, limiting bycatch exposure assessment.
• What they did: The study used GPS, wet-dry, accelerometer, and pressure data to detect 779 attendance events, then trained random-forest classifiers with different sensor combinations and validated them on 17 radar-equipped shearwaters.
• The result: Training classification exceeded 90% for movement-only models but produced 42% false positives on new individuals; GPS+environment models generalized with ~70%+ classification and detected nearby vessels in 19 predicted events without AIS support.

AI summary of the abstract · may contain errors · not medical advice · read the paper.