Uncovering hidden behaviours: a GPS-Accelerometry approach to behaviour classification in storm-petrels

Abstract

Biologging has transformed ecological research, yet studying the at-sea behaviour of small seabirds remains challenging due to trade-offs between device size and data resolution. Miniature GPS devices (<1 g) have been deployed on storm-petrels but their coarse temporal resolution limits behavioural classification, particularly resting. Accelerometers offer detailed behavioural data but lack spatial context, and some species are too small to simultaneously carry both devices. We deployed GPS devices on breeding European storm-petrels, including a subset recently developed with integrated accelerometers. Using accelerometer-informed rest periods, we developed a semi-supervised three-state Hidden Markov Model to classify transiting, foraging, and resting across all GPS tracks, even those without accelerometer data. Detection of resting behaviour by the semi-supervised model improved substantially compared to an unsupervised GPS-only model which only identified 17% of true resting events. Resulting time-activity budgets closely matched accelerometer-only studies, validating behaviour classifications, while also providing spatial context. Foraging peaked in the early hours of the night, whereas resting was most frequent in the afternoon. This is the first application of a combined GPS-accelerometer approach to a small storm-petrel species. Our method improves behavioural classification in small seabirds, providing new insights into the spatial and temporal dynamics of their at-sea behaviour.