Tracks estimated from archival tags deployed on juvenile SBT were used in determining their habitat preferences
in the Great Australian Bight (see Estimating habitat preferences).
A number of methods exist for estimating location from archival tag data; the method used in this study was developed
at CSIRO using the "twilight likelihood" (TL) method described in Basson et al. (2016).
It can be summarized as follows:
- The light data from each twilight event is used to estimate the likelihood of
having observed that light data given any location on the globe.
- The estimated likelihood surfaces get input to a grid-based Hidden Markov Model (HMM), where the location
of a fish (on a 1° x 1° grid) at each twilight is the hidden state to be estimated and the probability of moving
between states (i.e. grid cells) is assumed to be a random walk.
- Observational data which may be informative in determining location can be included in the HMM
in the form of a likelihood. The HMMs used here included surface temperature and depth data recorded on the tags
compared against global SST and bathymetry data.
- The output of the HMM is the posterior probability of a fish being in each possible grid cell on the globe
(excluding any cells that correspond to land masses) during each twilight event.
- A "most probable track" is estimated by taking a weighted average of all the grid latitudes and longitudes using
the posterior probabilities at each twilight as the weights.
Example tracks
- Release date and end date of the track, and length of fish at release are given in the title.
- Click on any map to open a full-size image in another window.