Local ecology changes foraging behaviour in Northern gannets: a machine-learning approach to determining the differences between colony foraging metrics
Ashley Bennison, Mark Jessopp, Stuart Bearhop, Steve Votier, Ewan Wakefield, Keith Hamer
Seabird researchers often identify key foraging areas using movement metrics within GPS tracking data. However, rarely are there any data from immersion loggers to inform or validate these predictions. We tracked Northern gannets (Morus bassanus) from Great Saltee (Ireland), Bass Rock (Scotland), and Grassholm (Wales) using a combination of GPS and time depth recorders (TDR). Machine-learning techniques were employed to build generalized boosted regression models of foraging dives based on movement patterns of birds in each colony. Models had good predictive power of dive events, with step length and turning angle accounting for 28.15% +/-0.964 and 2.56% +/-0.307 of model power in all colonies respectively. However, tortuosity values varied in importance across colonies with Celtic Sea colonies relying on values calculated over shorter distances than those in the North Sea. The predictive power of models built on data from one colony dropped considerably (average loss of >95% of predictive power) when applied to other colonies. That our models use the same metrics with differing levels of importance implies differences in behavior associated with foraging in different regions. This is discussed in terms of the local ecologies of the Celtic Sea and North Sea.