Framework for Long-term Accelerometer Monitoring of Activity in Small and Secretive Terrestrial Spec
Desantis, Dominic L.
dldesantis@miners.utep.edu
Department of Biological Sciences
Department of Mathematical Sciences
University of Texas at El Paso
El Paso, TX USA
Mata-Silva, Vicente
Johnson, Jerry D.
Department of Biological Sciences
University of Texas at El Paso
El Paso, TX USA
Wagler, Amy E.
Department of Mathematical Sciences
University of Texas at El Paso
El Paso, TX USA
A series of movement steps or decisions made by an animal is an important reflection of the interactions between internal state and external conditions. For many small, cryptic, and secretive animals, an understanding of the various causes and consequences of these fine-scale behavioral decisions is often precluded by methodological limitations. Among the many recently developed animal-borne datalogger technologies, the use of miniature accelerometer dataloggers for remote and continuous recording of animal activity is becoming increasingly common. However, accelerometer applications remain largely biased toward large-bodied species due to body size limitations with smaller animals. We took advantage of ongoing miniaturization and advancement of devices and associated computational techniques to develop a robust and flexible framework for long-term accelerometer monitoring in small and secretive terrestrial species in natural settings. We internally implanted radio transmitters and tri-axial accelerometers in rattlesnakes (Crotalus atrox) for field data collection periods ranging from one to 289 days. During these field deployments, we conducted frequent field-validation observations of behavior to train and validate supervised learning models (Random Forest (RF) and GLMNET). GLMNET model recall was 95%, and RF was 99% for classifying periods of coarse behavioral mode (inactivity vs. activity) in rattlesnakes. We applied validated models to full acceleration datasets for automatized activity classification to establish and visualize long-term, fine-scale activity budgets. We demonstrate the utility of this data collection, processing, and analysis pipeline for placing behavior in an ecological context by examining timescale-dependent environmental drivers of activity decisions in this cryptic pitviper.