Treed Gaussian processes for animal movement modeling.

Area of Science:

• Wildlife Ecology
• Computational Biology
• Statistical Modeling

Background:

• Telemetry data is crucial for wildlife ecology and management.
• Modeling animal movement is challenging due to temporal variations and nonstationarity.
• Existing models can be complex and difficult for practitioners to implement.

Purpose of the Study:

• To introduce and demonstrate treed Gaussian process (TGP) modeling for analyzing animal movement data.
• To showcase TGP's ability to automatically capture nonstationarity and abrupt transitions in movement.
• To enable the derivation of statistically comparable movement descriptors with uncertainty measures.

Main Methods:

• Application of treed Gaussian process (TGP) modeling, a Bayesian machine learning approach.
• Utilizing an existing R package for implementation via Markov chain Monte Carlo (MCMC).
• Deriving movement descriptors (e.g., distance traveled, residence times) from estimated trajectories.

Main Results:

• TGP modeling effectively captures nonstationarity and transitions in animal movement.
• The approach allows for automated modeling and provides uncertainty measures for movement descriptors.
• Case study on lesser prairie-chickens demonstrated TGP's utility in comparing movement patterns across individuals and populations.

Conclusions:

• TGP modeling offers a powerful, accessible tool for analyzing wildlife telemetry data.
• Combining machine learning and Bayesian inference facilitates the estimation of statistically comparable movement descriptors.
• This method enhances the application of telemetry data for wildlife management and ecological research.