Use this URL to cite or link to this record in EThOS:
Title: Automated techniques for bat echolocation call analysis
Author: Scott, Christopher David
ISNI:       0000 0004 2740 7257
Awarding Body: University of Leeds
Current Institution: University of Leeds
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Access from Institution:
Acoustic bat detectors are an extraordinarily valuable tool in bat research as they enable researchers to listen in on the otherwise secretive world of bats, providing the means to nqn-invasively survey and monitor bats in their natural habitats. Technological advances facilitate unprecedented data collection, considerably expanding the scope of field studies. However, the burden of manual analysis, and difficulty in identifying some species reliably from their calls, hampers the development of systematic survey and long- term monitoring methods. We developed a series of algorithms for the automated analysis of bat detector recordings, used to detect and extract calls from continuous recordings, and measure temporal and spectral call variables. By hand-labelling the .location of calls in field recordings, we were able to evaluate the accuracy of the automated method at detecting calls. Comparison on the same dataset with two conventional bioacoustic signal detectors revealed our algorithm was more accurate and robust. Using machine learning (ML) classification algorithms that learn to identify calls following training using a reference library, we developed a fully automated species identification system. Evaluation of the system was carried out by cross-validation of our reference call library, containing recordings of >5000 calls from known British species, comparing classifier predictions to ground- truth labels. The ML approach outperformed conventional statistical analysis using discriminant function analysis (DFA). We applied our novel system to two field studies that highlight its utility. Firstly, monitoring multi- species bat activity at a remote cave system over a period of three months, analysing >20,000 audio files to investigate temporal patterns in activity. Secondly, separating acoustically cryptic Myotis species from data collected in the Lake District National Park, to generate presence data for species distribution modelling, facilitatinq the creation of species-specific habitat suitability maps projected over the entire Park (ea, 3,300 km").
Supervisor: Altringham, John Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available