Part of the Vermont Center for Ecosystem Studies Vernal Pool Monitoring Project includes audio recordings to document the first calls of frogs at the pools. This year they started using AudioMoths to make the recordings. AudioMoths are open-source data loggers with a low-power sleep mode, real-time clock, microSD card slot, MEMS microphone, and circuitry to support audio capture. AudioMoths can be configured to automatically save audio recordings on a custom schedule. The AudioMoth software and hardware are well thought out and the audio quality is very good (see audio file below). The AudioMoths were deployed at about 50 Vermont vernal pools in weather-tight cases (Figure 1) and recorded for a few short sessions every night for several weeks in the spring.
Collecting these audio data is a big project but not nearly as big as the analysis. It is not reasonable to listen to all of the recordings in real time (50+ locations, for ~100 days per year, at 3 or 4 times per day, 10 minutes per session, for multiple years) so the plan is to train a computer to do it.
The microphone and circuitry on an AudioMoth are capable of recording ultrasonic sounds in the range of frequencies used by the bats that live in Vermont. So this summer some of us redeployed AudioMoths configured to record up to 125 kHz (humans can hear only up to about 20 kHz). We put AudioMoths on opposite sides of vernal pool MLS619 on Snake Mountian (Bridport). The two AudioMoths were about 40 feet apart and facing each other over the pool (Figure 2). They both recorded during 10 minute sessions at 9:00 PM, 12:00 AM, 3:00 AM, and 5:00 AM, and were deployed for two nights (August 15 and 16). Bat calls were recorded during each of the eight sessions (Figure 3).
The configuration of the AudioMoths for this deployment made recordings only when the sound was loud enough to cross an “amplitude threshold.” So for each 10 minute recording session, data were saved to the microSD card only when sound was sufficiently loud. This can prevent the recording of long audio files of silence or background noise. Audio files were broken up into segments of less than 30 seconds and then we counted how many segments had bat calls.
The two AudioMoths recorded more or less the same number of bat calls during this deployment (brown bars in Figure 3). That is a good result suggesting that the devices reliably record what is happening near the pool. The two nights were different from each other for reasons known only to the bats. The number of triggering audio events that were not bats differed between the two AudioMoths (gray bars in Figure 3). I assume this was because it was noisier near the east side AudioMoth due to insects, other animals, foliage, or who knows what was creeping around in the dark.
There were 240 (west side) or 290 (east side) individual files of audio events produced from the data saved on these two nights. I looked at each one by dragging them into the free version of SonoBat. This program makes a graphic display (sonogram) of each sound and bat calls are easy to visually distinguish from other sounds with a little practice. The frequency of the bat calls was always above about 40 kHz and the bats are therefore assumed to be either little brown bats (Myotis lucifugus) or Indiana bats (M. sodalis). Although both of these bats are on Vermont’s endangered species list, they have promising populations in the Champlain Valley. Other bat species have similar calls but are not very common in this area. No bat calls with major frequencies below 40 kHz (e.g., big brown bat, silver-haired bat) were observed.
Above: Screen capture from the SonoBat program displaying and “playing” a sonogram of two Myotis bat calls from vernal pool MLS619 and then two calls of big brown bats (from my backyard) for comparison. These recordings were made with an AudioMoth. It is difficult to distinguish the calls of little brown bats from calls of Indiana bats (both are in the genus Myotis). It is easy to distinguish the calls of these two bats from calls of big brown bats or silver-haired bats which are also common in this area. Bat calls do not sound like this (to us or to bats). This is what you get when you lower the frequency (pitch) and alter the speed and other attributes of the ultrasonic recording. A cursor at the bottom of the screen is animated but difficult to see.
This short observation suggests that there was a lot of bat activity close to this vernal pool. A follow-up observation would be to put one AudioMoth by the pool again and put the other one 300 feet away from the pool. Repeating that often enough would address the question of whether bats are more likely to be found near vernal pools than elsewhere in the forest.
There is also lots of additional work to do finding a good configuration for the AudioMoth for learning about bats. Instead of using the amplitude threshold to trigger recording, making a continuous recording and then counting the bat calls might produce more easily interpreted results. I did some comparisions in my backyard in Salisbury (Figure 4) and confirmed that amplitude triggering can save space on the microSD card and produce fewer files to inspect, but sometimes it doesn’t.
In one comparison continuous recording produced a lot more data. To quantifiy this, I broke the continuous recordings into 30 second pieces and counted how many had bat calls. As expected, most of them had no bat calls. During the same night, the other AudioMoth was recording only when it heard some sound. This required one third as much storage space and there were one third as many file segments to manually inspect. Both approaches produced similar, but not identical, census results for bat activity (Figure 5).
In some situations, using amplitude triggering can backfire. If sounds other than bat calls trigger the AudioMoth, the result can be similar to continuous recording. I used amplitude threshold triggering on two nights: one windy and one calm. The AudioMoths were in a woodland corridor along a stream (Figure 6). On the windy night the entire 15 minute session was recorded because wind noise triggered the AudioMoths repeatedly (Figure 7). Most of the recordings made on the windy night had no bat calls but had to be inspected to determine that.
Many configuration options allow AudioMoths to capture recordings of different sounds in different situations. Finding the best configuration for recording bats at vernal pools might take some additional testing. If an amplitude trigger is used, you must select which amplitude to use as the threshold and the amount of time to continue recording after the sound level drops. I have tried amplitudes between 3% and 20% and found that either 5% or 10% works, but results depend on how close the bats are and how much non-bat noise there is.
AudioMoths can be configured to listen to only part of the frequency spectrum, so sounds below 20 kHz (which are never bats) can be ignored by setting a low or bandpass filter. I tried this but not systematically enough to learn how much it helps. I also don’t know how abrupt the cutoff is e.g., when you select 20 kHz as the low end.
The sample rate can be selected and the high frequency calls of bats require a sample rate of 250 kHz or 384 kHz. I think 250 kHz is high enough for Vermont bats so that is what I used (384 kHz makes larger files that fill up the microSD card faster).
The audio gain determines how sensitive the microphone is and a level must be selected. I used the medium setting but in some conditions a different setting might be better.
Before the bats start to depart for their hibernacula we should deploy the AudioMoths again near a vernal pool to see what we can learn.
