A hierarchical machine learning framework for the analysis of large scale animal movement data

Area of Science:

• Movement ecology
• Animal behavior
• Statistical modeling

Background:

• Telemetry data revolutionized movement ecology, enabling fine-scale animal tracking.
• Existing models struggle with parameter uncertainty, intermittent data, and large datasets.
• Need for scalable and robust methods in movement analysis.

Purpose of the Study:

• Develop a novel approach for movement modeling using multilevel Gaussian processes.
• Infer continuous latent behavioral states underlying animal movement.
• Address challenges in analyzing large-scale telemetry data.

Main Methods:

• Implemented multilevel Gaussian processes for hierarchical movement modeling.
• Utilized trajectory segmentation for efficient likelihood approximation.
• Employed gradient-based Markov chain Monte Carlo for posterior inference.

Main Results:

• Gaussian process models offer flexible and powerful analysis of movement trajectories.
• Detected multiscale patterns and trends in animal movement data.
• Inference is accelerated using GPU-enabled machine learning libraries.

Conclusions:

• Multilevel Gaussian processes provide efficient inference for large movement datasets.
• Models allow fitting complex, flexible analyses of animal behavior.
• Applications include migration route analysis, activity pattern detection, and movement change identification.