Numbers of bats detected by AutoID in recordings using (near) continual recording was tested with various sample rates. These recordings were also manually IDed for verification.
- Result – Number of bats detected using sample rate of 192 and 256 kHz is broadly similar, while 384kHz records approx. 20-35% less bats.
- Outcome – Use 256kHz.
- As we are in a Lesser Horseshoe (LHS) area we standardised on 256kHz.
- In areas without LHS, it would be sensible to use 192kHz to reduce filesize, and possibly benefit from higher quality recordings due to oversampling within the AudioMoth.
- Some people have suggested using 394kHz to allow identification of Myotis spp. However rather than detect the start frequency of Myotis call, this is just likely to detect the frequency at which the AM mic becomes sensitive enough to detect the call. A external mic. may be the best option for 394kHz
We followed our usual testing process with a bank of 8 Audiomoths with 1.4.3 firmware, all set with near identical settings apart from sample rate, and two different setting for Hi-Pass. We tested several sessions of one night, using near continual recording of 10 second long files from sunset to midnight.
Sample rates of 196, 256 and 394kHz were tested using high gain, and hi-pass filters of 30 and 50 kHz.
Below are the results for the number of bats AutoIDed using BatClassify.
Note: See this page for comparison of BatClassify AutoID to manual identification.
|AudioMoth Configuration||PipPip AutoIDed||PipPyg AutoIDed||RhiFer AutoIDed||RhiHip AutoIDed|
|192kHz sample rate, 30kHz Hi-Pass||63||118||0||0|
|192kHz sample rate, 50kHz Hi-Pass||65||124||0||0|
|256kHz sample rate, 30kHz Hi-Pass||61||132||0||0|
|256kHz sample rate, 50kHz Hi-Pass||60||95||1||0|
|384kHz sample rate, 30kHz Hi-Pass||58||86||0||0|
|384kHz sample rate, 50kHz Hi-Pass||43||58||1||4|
As the results show most variance in 50kHz Hi pass these were selected for manula verification. In the table below are the results of manual identification from sunset to midnight for the 3 AudioMoths using 50kHz Hi-Pass filter.
|AudioMoth Configuration||Big bat (NSL) Manually IDed||PipPip Manually IDed||PipPyg Manually IDed||RhiFer Manually IDed||RhiHip Manually IDed|
|192kHz sample rate||17||67||100||0||0|
|256kHz sample rate||18||60||95||0||2|
|384kHz sample rate||14||43||58||0||2|
An example of the RhiHip recording is shown below, click to enlarge.
Full details of all configuration testing results are in the data tab of this spreadsheet.
Discussion of Results
A brief look at the manual ID results shows the results of 192kHz and 2656kHz are broadly similar, within the usual variance of detector and location, with the obvious exception of 192kHz not detecting the RhiHip at 112kHz.
However the 384Khz do detect far fewer bats – 22% less “Big Bat”, 28% less PipPip, 39% less PipPyg, and the same RhiHip.
It is easy to explain why 192 kHz detects more bats as:
“The lower sampling rates (including 192kHz) benefit from oversampling. In this case combiningAdrain Bicker – Ultrasound-triggered Firmware for the AudioMoth (1.0.8) – 4/8/19
two 384kHz samples to produce a cleaner, more accurate 192kHz recording”
However this does not explain why in real world bat testing why 256kHz without oversampling, performed as well as 192kHz. This needs more testing
Due to the limited manual ID performed it is difficult to give exact summary figures across all results, however a conservative statement is “384kHz records (at least) 20-35% less bats”
Some people have suggested using 394kHz to allow identification of Myotis spp. However rather than detect the start frequency of Myotis call, this is just likely to detect the highest frequency at which the AM mic becomes sensitive enough to detect the call. The internal mic. in the AudioMoth is so far outside its designed frequency response at this high frequency there is bound to be variability between individual AudioMoths. A external mic. may be the best option for 394kHz
Repeat this testing:
- Over more nights, at different times of year
- In areas with myotis and long-eared species
More manual verification