Hobart

Seminar Abstract

Friday 22 May 2009, 11.30am (Tas time)
CSIRO Auditorium, Hobart

Martin Wæver Pedersen
Visiting Phd student
Technical University of Denmark
Lyngby

Tracking movements and estimating behaviour of fish using electronic tags

I’ll give a broad overview of the movement and behaviour estimation methods I have developed in my Ph.D. study so far and show some of the results we have obtained using these methods on archival data from Atlantic cod in the North Sea and from PSATs from southern bluefin tuna in Australian waters.

The methodology was developed with the purpose to analyse achival tag data from Atlantic cod in the North Sea, primarily to infer movements from recordings of depth and temperature. It is formulated in the statistical framework of state space models which often leads to estimation using some variant of the Kalman filter. This is however not feasible in this case as the problem is highly non-linear and therefore the fundamental assumptions of the Kalman filter are invalidated. Advanced alternatives like particle filters would in theory be applicable, but do have drawbacks when it comes to smoothing and parameter estimation in particular. Therefore we opted for an approach encompassing hidden Markov models which essentially discretises the state space i.e. the two-dimensional horizontal position space of the fish into a finite number of states. This comes with great computational requirements but does in turn deliver very informative results.

I will not go into the theoretical details of the method but rather give an overview of the general idea and thereafter present some examples of the results that it is capable of producing. The outcome of a successful analysis is the probability distribution of the position of the fish for the time spanned by the data. An outcome of this distribution is a set of positions in time i.e. a movement trajectory and therefore is the mode of the distribution the most probable track of the fish given the data. Marginal distributions of the joint posterior distribution can be viewed in succession resulting in a very illustrative animation and unknown model parameters related to the swim speed of the fish can be straightforwardly estimated by maximum likelihood from data sets of sufficient quality.

I’ll finish with showing how the model can be expanded to encompass distinct behaviour states of the animal such as a residing state and a migratory state and present the results we obtain from applying this model to the southern bluefin tuna PSAT data.

Seminar Recording

[back]

Location:
CSIRO = Marine Laboratories Auditorium, Castray Esplanade, Hobart

For further information, or to schedule a seminar, contact:
To schedule a seminar, contact:
Clothilde Langlais, (Oceanographic seminars) CSIRO Marine and Atmospheric Research (03) 6232 5399
Natalie Kelly, (Biology/Modelling seminars) CSIRO Marine and Atmospheric Research 0438 452 483
Jillian Enraght-Moony, (seminar administrator) CSIRO Marine and Atmospheric Research (03) 6232 5320
Communications Manager, Antarctic Climate and Ecosystems CRC (03) 6226 2265
Margaret Hazelwood, Institute of Antarctic and Southern Ocean Studies (IASOS) University of Tasmania (03) 6226 2971

Last updated 21/07/09