The BCT guidelines (4th edition, section 10.2) allow automatic identification and recognise that most systems use AI, but they do not accept raw output. Set an error threshold from a random sample checked by an experienced analyst, verify by hand the uncommon, rarely detected, poorly classified and oddly timed records, and report how you did it. This applies to Kaleidoscope, BTO Acoustic Pipeline, BioSonic and any other classifier. BCT does not certify any of them.
We make BioSonic. The method on this page works with any bat analysis software.
The four steps in section 10.2
- Classify. Know what the system was trained on, how it scores confidence and how it handles two species in one recording.
- Validate. An accomplished analyst checks a random selection to find the false positive rate for each species, and you set the threshold from it.
- Verify. Check by hand any species that is locally uncommon, detected infrequently or classified poorly, and any record at an unusual time of day or year.
- Calculate. Turn the validated records into the metrics you report. Raw output is unlikely to be fit to report.
Whatever the method, section 10.2.9 says a proportion of the data should always be verified for quality assurance. It does not set that proportion. An experienced bat acoustics analyst should be involved in every project.
How many static detector nights
| Habitat suitability for bats | Minimum in Table 8.3, for each location |
|---|---|
| Low | Five consecutive nights per season: spring, summer and autumn. Use judgement on whether activity surveys are needed. |
| Moderate or high | Five consecutive nights per month, April to October. |
Surveys need suitable weather, and April and October depend on weather and location. BCT notes that rarer and quiet calling species may need more nights.
Wording for the method section
Recordings were classified with [software and version]. For each species, [number] randomly selected records were verified by [name, experience] and a confidence threshold of [value] was set for a false positive rate of [value]. All records of [locally uncommon species] and records at unusual times were verified by hand, following Collins (2023), section 10.2.
Where BioSonic fits
BioSonic is an AI classifier with a confidence for each identification, up to seven species per recording and manual review by a team. It scored 98.7% F1 against 83.0% for Kaleidoscope Pro in the Wilder Sensing benchmark. A benchmark does not replace validation on your own project.
Common question
Is AI bat identification accepted for planning reports?
BCT sets out how to use auto ID, with validation, manual verification and an experienced analyst. Whether a report is accepted is for the planning authority and its ecologist to decide, so describe the method in full.
Sources
- Bat Surveys for Professional Ecologists, 4th edition (Collins, ed., Bat Conservation Trust, 2023, amended March 2024), sections 8.2.49 to 8.2.50, Table 8.3 and section 10.2.
- Guidelines for passive acoustic surveys of bats in woodland (Bat Conservation Trust), for woodland surveys, which differ.
Related: Cut the time spent checking calls · Auto ID for Myotis · Bat survey software for UK consultants
