Mac Aodha et al., Towards a General Approach for Bat Echolocation Detection and Classification. Licensed CC BY-NC 4.0.
Known limitations:
- Trained on 17 UK species.
- Poor on feeding buzzes and social calls.
Method
Recording#
Recordings currently come from pippyg devices. The pippyg is a self build static detector capturing full spectrum audio at 384 kHz in multi-second clips when triggered.
Device clocks are set once at configuration and then count seconds, with no awareness of daylight saving. Times are therefore an offset from when the device was configured. Configuration is done in the field, right as the device is deployed.
Detection#
Every recording is put through 3 classifiers. BatDetect2 1.3.1 is run first and detects the species for each.
The same recordings then go through BSG-BAT 0.21 and Kaleidoscope Pro 5.9.2 (which is commercial software, so its results are imported from the outputs of running the tool directly). The results from these classifiers are used to agree or disagree with the first. If there's at least one agreement, the detection is considered confirmed.
Limitations and claims#
Species detection here is automated, and hasn't been verified by a qualified person. If you'd like to help verify, this would be very welcome and you can contact me to discuss this: [email protected]
Locations#
Locations are deliberately a little imprecise. Recording data (from the pippyg) contains precise locations (at least, where the device was first deployed), but those are not published. The location maps are designed to give a general idea of where species have been recorded.
Classifiers used#
Wildlife Acoustics. Licensed commercial.
Meramo et al., BSG-BATS: an open-source data annotation portal and classifier for European bat vocalizations, Methods in Ecology and Evolution. Licensed CC BY-4.0.
Known limitations:
- Trained on 21 European species, of which 13 occur in the UK. It cannot return Natterer's, whiskered, Brandt's, Bechstein's or greater mouse-eared bat.
- Can return species that are not UK residents, so here we assume that it misread.
Runs#
| Classifier | Run | Recordings | Detections | Threshold |
|---|---|---|---|---|
| BatDetect2 1.3.1 | batdetect2-1.3.1-20261004T2012 | 514 | 104586 | 0.1 |
| BatDetect2 1.3.1 | batdetect2-1.3.1-20261004T1551 | 141 | 11985 | 0.1 |
| BSG-BAT 0.21 | bsgbat-0.21-20261004T2052 | 514 | ||
| Kaleidoscope Pro 5.9.2 | run-20261004 | 141 | ||
| Kaleidoscope Pro 5.9.2 | run-20261004-warleigh | 373 |
BSG-BAT and Kaleidoscope Pro return one label per recording rather than per call, so they have no detection count or threshold to report.
Generated 06/10/2026 10:45 UTC.