Statistical modelling of collective animal movement: with an application to reindeer movement in northern Sweden

The ways in which animals move are a complex phenomena, from small scale interactions to
larger migratory movement. Internal and external stimuli govern a variety of behavioural
patterns whose processes are vital for species survival. Analysing these movement and
behavioural processes can have significant applications for conservation and management.
Although there are many statistical tools readily available for investigating animal movement,
they are largely directed towards individual-level cases and do not consider the group
movement present in collective species such as ungulates.
This thesis aims to redress the shortcomings of statistical literature by providing a modelling
framework for collective animal movement in continuous time. Our modelling approach
builds upon general themes of group movement originally put forward by Langrock et al.
(2014), where each individual in the group is at times attracted to an unobserved leading point.
However, the behaviour of each individual can switch between ‘following the group’ and
‘moving independently’, modelled as an Ornstein Uhlenbeck process and Brownian motion
respectively. The movement of the leading point is also modelled as an Ornstein-Uhlenbeck
process or, if we forgo the leader’s drift term, as Brownian motion. An inhomogeneous
Kalman filter Markov chain Monte Carlo algorithm is used to estimate the diffusion and
switching parameters and the behavioural states of each individual at a given time point.
We assess the model’s performance in a variety of simulated settings before providing
a real world application using the location data of semi-domesticated reindeer (rangifer
tarandus). We extend this methodology by allowing switching to depend explicitly on
covariate information. We define a general auxiliary model for the inclusion of covariate
data which accounts for a wide range of environmental heterogeneity. We give a simulated
illustration where the animals switch behaviour sinusoidally depending on the time of day.
Then, we revisit the reindeer application by including covariate data on insect harassment,
which is thought to influence reindeer movement.