Predicting animal behaviour using deep learning: GPS data alone accurately predict diving in seabirds

Abstract

Abstract1. To prevent further global declines in biodiversity, identifying and understanding key habitats is crucial for successful conservation strategies. For example, globally, seabird populations are under threat and animal movement data can identify key at-sea areas and provide valuable information on the state of marine ecosystems. To date, in order to locate these areas, studies have used global positioning system (GPS) to record position and are sometimes combined with time-depth recorder (TDR) devices to identify diving activity associated with foraging, a crucial aspect of at-sea behaviour. However, the use of additional devices such as TDRs can be expensive, logistically difficult and may adversely affect the animal. Alternatively, behaviours may be resolved from measurements derived from the movement data alone. However, this behavioural analysis frequently lacks validation data for locations predicted as foraging (or other behaviours).2. Here, we address these issues using a combined GPS and TDR dataset from 108 individuals by training deep learning models to predict diving in European shags, common guillemots and razorbills. We validate our predictions using withheld data, producing quantitative assessment of predictive accuracy. The variables used to train these models are those recorded solely by the GPS device: variation in longitude and latitude, altitude and coverage ratio (proportion of possible fixes acquired within a set window of time).3. Different combinations of these variables were used to explore the qualities of different models, with the optimum models for all species predicting non-diving and diving behaviour correctly over 94% and 80% of the time, respectively. We also demonstrate the superior predictive ability of these supervised deep learning models over other commonly used behavioural prediction methods such as hidden Markov models.4. Mapping these predictions provides useful insights into the foraging activity of a range of seabird species, highlighting important at sea locations. These models have the potential to be used to analyse historic GPS datasets and further our understanding of how environmental changes have affected these seabirds over time. K E Y W O R D Scommon guillemot, European shag, foraging, machine learning, prediction, razorbill, time-depth recorder This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

State‐space modeling reveals habitat perception of a small terrestrial mammal in a fragmented landscape

“…Movement patterns provide a finer-scale understanding of how animals perceive their habitat (Browning et al, 2018). Movement patterns provide a finer-scale understanding of how animals perceive their habitat (Browning et al, 2018).…”

“…Management of animals threatened by habitat fragmentation often attempt to preserve or restore habitat by planting native vegetation, with the assumption that the resultant structure is suitable for the animals. Effective restoration therefore requires fine-scaled understanding of how animals respond to details of habitat (Allen & Singh, 2016;Browning et al, 2018;McClintock, London, Cameron, & Boveng, 2017). Effective restoration therefore requires fine-scaled understanding of how animals respond to details of habitat (Allen & Singh, 2016;Browning et al, 2018;McClintock, London, Cameron, & Boveng, 2017).…”

The Ornithodolite as a tool to quantify animal space use and habitat selection: a case study with birds diving in tidal waters

“…Therefore, VOD positioning error is probably comparable to what is generally accepted for GPS data, which is estimated to be in the range of 3-28 m (Frair et al 2010). However, while spatial error in tagging technology can lead to the misrepresentation of behaviors in a scale-dependent manner (Costa et al 2010;Browning et al 2017), animal locations can be coded according to behavior (as well as species, age and other factors that may be of interest) with the VOD. The downside of the VOD is that it has relatively intensive requirements when it comes to survey effort.…”

Can AI predict animal movements? Filling gaps in animal trajectories using inverse reinforcement learning

“…First, some methods assume specific movement properties and/or constraints regarding animal behavior between the observation records. These mechanistic models should include realistic movement properties and certain parametric assumptions (Royer and Lutcavage 2008), which depend on species and environmental conditions (Humphries et al 2010, de Jager et al 2011, Browning et al 2018. Other frequently assumed animal movement behaviors are that animals moving in fluid (i.e., water and air) follow curvilinear rather than linear paths due to flows, vortices, and turbulence (Tremblay et al 2006).…”

“…A recent network analysis has shown that trajectories have a higher predictability than expected; the location of a person can be predicted with 93% accuracy (Song et al 2010), indicating that machine learning, which infers knowledge from trajectory data, is a promising approach to the prediction of animal movements. The use of machine learning has become more common in ecological, behavioral, and environmental research over the last decade (Guilford et al 2009, Resheff et al 2014, Appelhans et al 2015, Valletta et al 2017, Weinstein 2017, Browning et al 2018), but reinforcement learning (RL; Fig. 3; Kaelbling et al 1996, Sutton andBarto 1998), which is a major machine learning paradigm along with supervised and unsupervised learning, has not been applied to animal behavioral data in the field to date.…”