Why we manually check automated classifiers’ bat identifications
A recent academic paper on automated classifiers (‘The use of automated identification of bat echolocation calls in acoustic monitoring: A cautionary note for a sound analysis‘) highlights the importance of a key part of our bat call identification method: manual checking of automated classifiers’ results.
All bats and their roosts are protected from harm and disturbance at all times by EU and UK law, and their foraging habitats also receive some protection within the planning system. Licensed professionals, such as ourselves, carry out a range of bat surveys at sites varying from large housing developments to the removal of individual trees or buildings. These surveys are a legal requirement and integral to planning permission submissions.
Spectrogram showing pipistrelle bat call
A useful tool to support surveys of bats is the use of automated classifiers. These analyse and attempt to recognise bat calls from recordings made by detectors such as SM2+s, and can provide an indication of the likely species. However, as the recent publication by Russo and Voight (2016) demonstrates, “no classifier has yet proven capable of providing correct classifications in 100% of cases or getting close enough to this ideal performance”.
Because of this, at Baker Consultants, our bat experts manually check each call following the use of automated classifiers to ensure accuracy. Automated classifiers remain an extremely useful tool when surveying bats, which are by their nature nocturnal, elusive and often cryptic, but, as this research shows, expert opinion is still needed to validate the results.