Using non-continuous accelerometry to identify cryptic nesting events of Galapagos giant tortoises DOI Creative Commons

Edward F. Donovan,

Stephen Blake, Sharon L. Deem

et al.

Animal Biotelemetry, Journal Year: 2024, Volume and Issue: 12(1)

Published: Nov. 11, 2024

Triaxial accelerometers have revolutionized wildlife research by providing an unprecedented understanding of the behavior free-living animals. Machine learning is often applied to acceleration data classify diverse animal behaviors across taxa. However, high frequency, continuous collection typically favored for behavioral classification studies generates very large sets, which may inhibit remote acquisition and make storage challenging. Coarse-frequency sampling or non-continuous bursts reduce these problems. To analyze such data, a suite variables that summarize key features interest can be generated. These then used in numerous approaches, accommodating variation methods regimes. We demonstrate potential accelerometer identify long-duration employ machine nesting critically endangered eastern Santa Cruz giant tortoise (Chelonoidis donfaustoi). field validated 112 events from 21 tortoises. derived summary statistics based on accelerometry (e.g., overall dynamic body acceleration, metrics comparing before after probable event) them as inputs Random Forest Boosted Regression Tree algorithms. Our models produced harmonic mean precision sensitivity (F1-score) 0.91. tested generality our model found performs well when both novel individuals years. The most important variable accurately classifying sequences was proportion above activity threshold followed average value bursts. results feasibility efficacy using prolonged, biologically relevant wildlife. By do not require sampling, this approach facilitates long-term monitoring behavior. Similar methodology has inform priority questions ecology conservation, predicting responses climate change identifying critical habitats, with applications species behaviors.

Language: Английский

A benchmark for computational analysis of animal behavior, using animal-borne tags DOI Creative Commons
Benjamin Hoffman,

Maddie Cusimano,

Vittorio Baglione

et al.

Movement Ecology, Journal Year: 2024, Volume and Issue: 12(1)

Published: Dec. 18, 2024

Language: Английский

Citations

3
Classification of sex-dependent specific behaviours by tri-axial acceleration in the tegu lizard Salvator merianae DOI

Ane Guadalupe‐Silva,

Lucas A. Zena, Livia Saccani Hervas

et al.

Comparative Biochemistry and Physiology Part A Molecular & Integrative Physiology, Journal Year: 2024, Volume and Issue: 298, P. 111744 - 111744

Published: Sept. 16, 2024

Language: Английский

Citations

2
A tri‐axial acceleration‐based behaviour template for translocated birds: the case of the Asian houbara bustard DOI Creative Commons
Kareemah Chopra, Rory P. Wilson, Emily L. C. Shepard

et al.

Wildlife Biology, Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 2, 2024

Understanding the behaviours and time budgets of translocated animals post‐release has potential to improve rearing release protocols, therefore survival rate. Otididae (bustards) inhabit open landscapes across Middle East Asia, are highly mobile on ground have similar lifestyles body plans. The Asian houbara Chlamydotis macqueenii is a bustard conservation concern inhabiting Central Asia frequently reared in captivity for population management. We deployed tri‐axial accelerometers 20 captive houbaras two seasons catalogue basic behaviours, provide template applicable other species examine seasonal differences behaviour. created Boolean algorithms define following using raw acceleration data derived metrics: stationary, eating/drinking locomotion. used video recordings cross‐validate algorithms, yielding recalls from 95 97%, precisions between 97 98%. Houbaras spent significantly more ‘stationary' less ‘locomotion' summer (June) compared spring (March). Simple proved useful identifying several be species, wild post‐release. Keywords: accelerometer, animal behaviour, bustard, breeding, translocation

Language: Английский

Citations

1
Automatic identification of the endangered Hawksbill sea turtle behavior using deep learning and cross-species transfer DOI Creative Commons
Lorène Jeantet,

Kukhanya Zondo,

Cyrielle Delvenne

et al.

Journal of Experimental Biology, Journal Year: 2024, Volume and Issue: 227(24)

Published: Nov. 18, 2024

ABSTRACT The accelerometer, an onboard sensor, enables remote monitoring of animal posture and movement, allowing researchers to deduce behaviors. Despite the automated analysis capabilities provided by deep learning, data scarcity remains a challenge in ecology. We explored transfer learning classify behaviors from acceleration critically endangered hawksbill sea turtles (Eretmochelys imbricata). Transfer reuses model trained on one task large dataset solve related task. applied this method using green (Chelonia mydas) adapted it identify such as swimming, resting feeding. also compared with human activity data. results showed 8% 4% F1-score improvement turtle datasets, respectively. allows adapt existing models their study species, leveraging expanding use accelerometers for wildlife monitoring.

Language: Английский

Citations

1
Moving towards more holistic validation of machine learning-based approaches in ecology and evolution DOI Creative Commons
Charlotte Christensen, André C. Ferreira,

Wismer Cherono

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 21, 2024

Abstract Machine-learning (ML) is revolutionizing the study of ecology and evolution, but performance models (and their evaluation) dependent on quality training validation data. Currently, we have standard metrics for evaluating model (e.g., precision, recall, F1), these to some extent overlook ultimate aim addressing specific research question which will be applied. As improving has diminishing returns, particularly when data inherently noisy, biologists are often faced with conundrum investing more time in maximising at expense doing actual research. This leads question: how much noise can accept our ML models? Here, start by describing an under-reported source that cause underestimate true performance. Specifically, ambiguity between categories or mistakes labelling produces hard ceilings limit metric scores. common error biological systems means many could performing better than suggest. Next, argue show imperfect (e.g. low F1 scores) still useable. We first propose a simulation framework evaluate robustness hypothesis testing. Second, determine utility supplementing existing ‘biological validations’ involve applying unlabelled different ecological contexts anticipate outcome. Together, simulations case effects sizes expected patterns detected even relatively 60-70%). In so, provide roadmap approaches tailored evolutionary biology.

Language: Английский

Citations

0
Using non-continuous accelerometry to identify cryptic nesting events of Galapagos giant tortoises DOI Creative Commons

Edward F. Donovan,

Stephen Blake, Sharon L. Deem

et al.

Animal Biotelemetry, Journal Year: 2024, Volume and Issue: 12(1)

Published: Nov. 11, 2024

Triaxial accelerometers have revolutionized wildlife research by providing an unprecedented understanding of the behavior free-living animals. Machine learning is often applied to acceleration data classify diverse animal behaviors across taxa. However, high frequency, continuous collection typically favored for behavioral classification studies generates very large sets, which may inhibit remote acquisition and make storage challenging. Coarse-frequency sampling or non-continuous bursts reduce these problems. To analyze such data, a suite variables that summarize key features interest can be generated. These then used in numerous approaches, accommodating variation methods regimes. We demonstrate potential accelerometer identify long-duration employ machine nesting critically endangered eastern Santa Cruz giant tortoise (Chelonoidis donfaustoi). field validated 112 events from 21 tortoises. derived summary statistics based on accelerometry (e.g., overall dynamic body acceleration, metrics comparing before after probable event) them as inputs Random Forest Boosted Regression Tree algorithms. Our models produced harmonic mean precision sensitivity (F1-score) 0.91. tested generality our model found performs well when both novel individuals years. The most important variable accurately classifying sequences was proportion above activity threshold followed average value bursts. results feasibility efficacy using prolonged, biologically relevant wildlife. By do not require sampling, this approach facilitates long-term monitoring behavior. Similar methodology has inform priority questions ecology conservation, predicting responses climate change identifying critical habitats, with applications species behaviors.

Language: Английский

Citations

0