Limitations of using surrogates for behaviour classification of accelerometer data: refining methods using random...

Area of Science:

• Animal behavior analysis
• Biologging technology
• Ecological research

Background:

• Biologging devices offer insights into animal movement and behavior but often require direct observation for accurate data interpretation.
• Captive or domestic surrogates are commonly used to calibrate biologging data for elusive species, yet their utility is debated.

Purpose of the Study:

• To assess the validity of using captive conspecifics and domestic surrogate species for calibrating animal behavior classification from biologging data.
• To refine methods for accurate behavior classification in both captive and free-living animals.

Main Methods:

• Employed tri-axial accelerometers and magnetometers alongside direct behavioral observations.
• Developed random forest models to predict animal behaviors and terrain-specific locomotion.
• Utilized captive Alpine ibex (Capra ibex) and domestic pygmy goats (Capra aegagrus hircus) as study subjects.

Main Results:

• Achieved high accuracy (>98%) in behavioral classification for captive Alpine ibex and pygmy goats.
• Model performance significantly decreased (56.1% ± 11%) when classifying behavior of individuals not included in the training data.
• Using domestic pygmy goat data to predict Alpine ibex behavior yielded insufficient accuracy (<55%) for all behaviors.

Conclusions:

• Biologging device calibration requires training data from similar conspecifics observed in environments reflecting natural terrain.
• Domestication-induced morphological differences and environmental terrain significantly impact the accuracy of surrogate-based behavior classification.
• Accurate behavior prediction necessitates training models with data from individuals within the target species and appropriate environmental conditions.