Animal-borne acoustic data alone can provide high accuracy classification of activity budgets

Abstract

Background Studies on animal behaviour often involve the quantification of the occurrence and duration of various activities. When direct observations are challenging (e.g., at night, in a burrow, at sea), animal-borne devices can be used to remotely record the movement and behaviour of an animal (e.g., changing body posture and movement, geographical position) and/or its immediate surrounding environment (e.g., wet or dry, pressure, temperature, light). Changes in these recorded variables are related to different activities undertaken by the animal. Here we explored the use of animal-borne acoustic recorders to automatically infer activities in seabirds. Results We deployed acoustic recorders on Cape gannets and analysed sound data from 10 foraging trips. The different activities (flying, floating on water and diving) were associated with clearly distinguishable acoustic features. We developed a method to automatically identify the activities of equipped individuals, exclusively from animal-borne acoustic data. A random subset of four foraging trips was manually labelled and used to train a classification algorithm (k-nearest neighbour model). The algorithm correctly classified activities with a global accuracy of 98.46%. The model was then used to automatically assess the activity budgets on the remaining non-labelled data, as an illustrative example. In addition, we conducted a systematic review of studies that have previously used data from animal-borne devices to automatically classify animal behaviour (n = 61 classifications from 54 articles). The majority of studies (82%) used accelerometers (alone or in combination with other sensors, such as gyroscopes or magnetometers) for classifying activities, and to a lesser extent GPS, acoustic recorders or pressure sensors, all potentially providing a good accuracy of classification (> 90%). Conclusion This article demonstrates that acoustic data alone can be used to reconstruct activity budgets with very good accuracy. In addition to the animal’s activity, acoustic devices record the environment of equipped animals (biophony, geophony, anthropophony) that can be essential to contextualise the behaviour of animals. They hence provide a valuable alternative to the set of tools available to assess animals’ behaviours and activities in the wild.

Semi-Supervised Machine Learning for Livestock Threat Classification Using GPS Data

“…They achieve a near-perfect accuracy of 99%. Thiebault et al [11] studied Cape gannets using acoustic data in a supervised setting to predict whether a bird is flying, floating on water or diving. They also achieve a near-perfect accuracy of 98%.…”

Modelling reindeer rut activity using on‐animal acoustic recorders and machine learning

“…Although not yet widespread, early studies show their potential in categorizing events and behaviours (Stowell et al, 2017;Wijers et al, 2018), estimating metabolic costs of sound production (Ilany et al, 2013) and analysing feeding behaviour (Lynch et al, 2013;Studd et al, 2019;Thiebault et al, 2021). However, to date, on-animal recorders have not been used to study the reproductive activity of terrestrial species.…”

“…On‐animal acoustic recorders, commonly used in marine studies, are now being adapted for terrestrial research, offering continuous audio capture of both intentional and environmental sounds from individuals (Ilany et al., 2013 ; Lynch et al., 2013 ; Stowell et al., 2017 ; Studd et al., 2021 ; Thiebault et al., 2021 ; Wijers et al., 2018 ). These devices, custom‐built due to the lack of commercial options for land use, tend to be bulky due to power requirements, difficult to recover and prone to damage by wearers or the environment, affecting the continuity of recordings (Ilany et al., 2013 ; Lynch et al., 2013 ; Stowell et al., 2017 ; Studd et al., 2021 ; Thiebault et al., 2021 ; Wijers et al., 2018 ). Although not yet widespread, early studies show their potential in categorizing events and behaviours (Stowell et al., 2017 ; Wijers et al., 2018 ), estimating metabolic costs of sound production (Ilany et al., 2013 ) and analysing feeding behaviour (Lynch et al., 2013 ; Studd et al., 2019 ; Thiebault et al., 2021 ).…”

“…These devices, custom‐built due to the lack of commercial options for land use, tend to be bulky due to power requirements, difficult to recover and prone to damage by wearers or the environment, affecting the continuity of recordings (Ilany et al., 2013 ; Lynch et al., 2013 ; Stowell et al., 2017 ; Studd et al., 2021 ; Thiebault et al., 2021 ; Wijers et al., 2018 ). Although not yet widespread, early studies show their potential in categorizing events and behaviours (Stowell et al., 2017 ; Wijers et al., 2018 ), estimating metabolic costs of sound production (Ilany et al., 2013 ) and analysing feeding behaviour (Lynch et al., 2013 ; Studd et al., 2019 ; Thiebault et al., 2021 ). However, to date, on‐animal recorders have not been used to study the reproductive activity of terrestrial species.…”

“…Processing and analysing the large volume of data produced by these applications remains a significant challenge for researchers. In recent years, collaborations with computer scientists have led to the application of machine learning to bioacoustics (Blumstein et al., 2011 ; Enari et al., 2019 ; Stowell et al., 2017 ; Studd et al., 2021 ; Thiebault et al., 2021 ; Wijers et al., 2018 ). Machine learning, especially convolutional neural networks (CNNs), streamlines data processing and feature identification, outperforming commercial vocalization recognition tools (Dufourq et al., 2022 ; Knight et al., 2017 ; Zhong et al., 2020 ).…”