To estimate the surfacing metrics (mean IBI, bout duration and number of surfacings in a bout), Monte Carlo methods were used to simulate individual time series (follows) of dive types based on the transition probability matrix obtained from the Markov model.
The methods are the same as those described in Christiansen et al. (in press).
1000 simulations were run. First, an empty vector of dive typeswas created in R, with each empty value representing a sampling unit to which a dive type and duration (i.e. IBI), were randomly assigned. The initial dive type was arbitrarily categorised as a surface dive.
The next dive type was then randomly chosen based on the transition probability matrix obtained from the Markov chain model. This procedure was repeated for the entire vector.
To account for the heterogeneity in duration of dive types (i.e. the variation in IBI) a duration was assigned to each dive type by randomly selecting with replacement an IBI from the “distribution” of IBIs obtained fromthe raw data.
Each dive assigned as a surface dive was given a random IBI from the “distribution” of IBIs classified as surface dives, while each dive assigned as a deep dive was given a random IBI from the “distribution” of IBIs classified as deep dives.
After allocating dive types, and durations of dive types, the first 100 dives in the time series were removed as a burn-in period so that each simulation began with a randomly chosen dive type.
The time series was then cut at an upper limit of 7 h, which represents the time between the earliest (07:50:50) and latest (15:18:55) recordings in a day, rounded to the nearest hour, to avoid extrapolation.
From the resulting time series, the mean IBI, bout duration and number of surfacings per bout were estimated. This was done for each simulation, so that a density distribution around each surfacing metric was obtained.