Withdrawal behaviour of the sea pen Pennatula aculeata (Pennatuloidea): a 10-month fine-scale study with implications for conservation monitoring

Abstract Sea pen fields function as critical biogenic structures that add three-dimensional complexity to soft-sediment habitats, making them priority targets for marine conservation and subsequent monitoring. However, the capacity of certain species to completely retract beneath the sediment introduces a significant detection bias into non-invasive visual surveys. To quantify these autogenic retraction dynamics, two high-resolution camera-equipped benthic landers were deployed 600 m apart within the Gully MPA (northwest Atlantic) from October 2022 to July 2023, capturing seafloor images at 30-min intervals. Image time-series from the primary site, dominated by the sea pen Pennatula aculeata , were analyzed using a combination of manual annotations (Lightroom, BIIGLE) and supervised machine learning pixel-segmentation (RootPainter). While individual P. aculeata colonies remained visible across 70–90% of the frames, distinct group retraction events occurred in approximately 1.5% of the imagery, during which at least 97% of the localized population simultaneously withdrew into the sediment. Random Forest modeling identified semilunar cycles (~14.7-day intervals) as a primary driver of these synchronized retractions, with behavioural periodicities heavily coupled to tidally driven forcing during the autumn. Conversely, winter and spring months exhibited high, sustained colony expansion, potentially mediated by seasonal pulses of high-quality organic matter confirmed by sediment trap biomarker fluxes. Density- and size-dependent cues and diel rhythms were not primary drivers. These findings demonstrate that optical detection probability varies significantly by season and lunar phase; standardizing visual censing frameworks to explicit temporal windows will mitigate non-random sampling absences and substantially enhance the statistical power of deep-sea conservation monitoring.