Development of an automated method of detecting stereotyped feeding events in multisensor data from tagged rorqual whales

• Published in: Ecology and Evolution
• Main Authors: Allen, Ann N., Goldbogen, Jeremy A., Friedlaender, Ari S., Calambokidis, John
• Other Authors: Office of Naval Research
• Format: Article in Journal/Newspaper
• Language: English
• Published: Wiley 2016
• Subjects: Rorqual; Balaenoptera physalus
• Online Access: http://dx.doi.org/10.1002/ece3.2386; https://api.wiley.com/onlinelibrary/tdm/v1/articles/10.1002%2Fece3.2386; https://onlinelibrary.wiley.com/doi/pdf/10.1002/ece3.2386; https://onlinelibrary.wiley.com/doi/full-xml/10.1002/ece3.2386; http://api.wiley.com/onlinelibrary/chorus/v1/articles/10.1002%2Fece3.2386

Abstract The introduction of animal‐borne, multisensor tags has opened up many opportunities for ecological research, making previously inaccessible species and behaviors observable. The advancement of tag technology and the increasingly widespread use of bio‐logging tags are leading to large volumes of sometimes extremely detailed data. With the increasing quantity and duration of tag deployments, a set of tools needs to be developed to aid in facilitating and standardizing the analysis of movement sensor data. Here, we developed an observation‐based decision tree method to detect feeding events in data from multisensor movement tags attached to fin whales (Balaenoptera physalus ). Fin whales exhibit an energetically costly and kinematically complex foraging behavior called lunge feeding, an intermittent ram filtration mechanism. Using this automated system, we identified feeding lunges in 19 fin whales tagged with multisensor tags, during a total of over 100 h of continuously sampled data. Using movement sensor and hydrophone data, the automated lunge detector correctly identified an average of 92.8% of all lunges, with a false‐positive rate of 9.5%. The strong performance of our automated feeding detector demonstrates an effective, straightforward method of activity identification in animal‐borne movement tag data. Our method employs a detection algorithm that utilizes a hierarchy of simple thresholds based on knowledge of observed features of feeding behavior, a technique that is readily modifiable to fit a variety of species and behaviors. Using automated methods to detect behavioral events in tag records will significantly decrease data analysis time and aid in standardizing analysis methods, crucial objectives with the rapidly increasing quantity and variety of on‐animal tag data. Furthermore, our results have implications for next‐generation tag design, especially long‐term tags that can be outfitted with on‐board processing algorithms that automatically detect kinematic events and transmit ethograms via ...