Validating hidden Markov models for seabird behavioural inference