Using approximate Bayesian inference for a "steps and turns" continuous-time random walk observed at regular time...

Area of Science:

• Ecology
• Movement Ecology
• Statistical Modeling

Background:

• Animal movement is complex, influenced by factors across various spatial and temporal scales.
• Existing models often use discrete-time formulations, creating a mismatch with naturally continuous animal movement.
• Observing animal movement at fixed intervals complicates analysis.

Purpose of the Study:

• To develop and evaluate a continuous-time state-space model for animal movement analysis.
• To investigate the use of Approximate Bayesian Computation (ABC) for inferring movement parameters from this model.
• To assess how the discrepancy between observation and movement decision time scales affects model performance.

Main Methods:

• A continuous-time state-space model was employed where movement decisions (steps, turns) occur at any time.
• Approximate Bayesian Computation (ABC) was used for model inference due to an intractable likelihood function.
• A simulation study explored the model's performance across different observation and movement time scales.

Main Results:

• Model parameters were successfully recovered when the observation time scale was moderately close to the movement decision time scale.
• Accurate estimates were achieved when the observation scale was up to five times that of the movement decision scale.
• The model was successfully applied to high-resolution sheep trajectory data.

Conclusions:

• Continuous-time state-space models coupled with ABC offer a robust framework for analyzing animal movement.
• The temporal scale of data collection is crucial and should align with the animal's movement decision-making process.
• This approach provides an intuitive way to link observation scales with movement dynamics.