Change-point models for identifying behavioral transitions in wild animals.

Area of Science:

• Ecology
• Animal Behavior
• Statistical Modeling

Background:

• Direct observation of animal behavior is challenging, costly, and time-consuming.
• Hidden Markov Models (HMMs) infer behavior from movement but assume multiple state transitions.
• Single behavioral events like parturition or migration initiation are not well-suited for traditional HMMs.

Purpose of the Study:

• To develop and evaluate novel change-point models for identifying single, significant shifts in animal movement behavior.
• To assess the models' ability to detect transitions using simulated and real-world animal movement data.
• To apply these models to understand parturition behavior in ungulates.

Main Methods:

• Developed two change-point models: one location-based and one movement metric-based.
• Conducted simulation studies to test model performance with varying data and shift magnitudes.
• Applied the models to GPS and vaginal implant transmitter data from Pennsylvania deer and elk.

Main Results:

• Change-point estimation was successful with at least 12 hours of post-change data and a 15-minute fix interval in simulations.
• Model performance varied between deer and elk, indicating species- and individual-specific behavioral differences.
• The models demonstrated the ability to identify the timing of a single behavioral change when underlying assumptions were met.

Conclusions:

• Change-point models offer a viable alternative to HMMs for detecting singular behavioral events in animals.
• These models can be applied to various behaviors beyond parturition, provided they involve a distinct change in movement dynamics.
• Future research should consider individual variation and species-specific behaviors when applying these models.