State-switching continuous-time correlated random walks

Théo Michelot, Paul G. Blackwell

Research output: Contribution to journalArticlepeer-review

17 Citations (Scopus)


1.Continuous‐time models have been developed to capture features of animal movement across temporal scales. In particular, one popular model is the continuous‐time correlated random walk, in which the velocity of an animal is formulated as an Ornstein‐Uhlenbeck process, to capture the autocorrelation in the speed and direction of its movement. In telemetry analyses, discrete‐time state‐switching models (such as hidden Markov models) have been increasingly popular to identify behavioural phases from animal tracking data.

2.We propose a multistate formulation of the continuous‐time correlated random walk, with an underlying Markov process used as a proxy for the animal's behavioural state process. We present a Markov chain Monte Carlo algorithm to carry out Bayesian inference for this multistate continuous‐time model.

3.Posterior samples of the hidden state sequence, of the state transition rates, and of the state‐dependent movement parameters can be obtained. We investigate the performance of the method in a simulation study, and we illustrate its use in a case study of grey seal (Halichoerus grypus) tracking data.

4.The method we present makes use of the state‐space model formulation of the continuous‐time correlated random walk, and can accommodate irregular sampling frequency and measurement error. It will facilitate the use of continuous‐time models to estimate movement characteristics and infer behavioural states from animal telemetry data.

Original languageEnglish
JournalMethods in Ecology and Evolution
VolumeEarly View
Early online date14 Feb 2019
Publication statusE-pub ahead of print - 14 Feb 2019


  • Animal movement
  • Continuous time
  • Multistate model
  • Ornstein–Uhlenbeck process
  • Random walk
  • State-space model


Dive into the research topics of 'State-switching continuous-time correlated random walks'. Together they form a unique fingerprint.

Cite this