More noise, fewer bats

After 12 nights of recording bat calls near an Indiana bat maternity roosting colony, we deployed the AudioMoths for a week at the vernal pool where we recorded bat calls in August. Instead of putting two AudioMoths at the vernal pool, we put one by the pool and two in the forest surrounding the pool. One of the forest AudioMoths recorded nothing (a battery was inserted backwards), so we got data from only one non-pool AudioMoth.

Figure 1. Vernal pool MLS619 (Bridport, VT) on September 23, 2021 at the end of a week of recording bat calls. There was no standing water in the pool but the liquid mud was 6 inches deep. This might be as low as the water level gets in the pool this year (it was dry and firm last October). There were a few inches of water in the pool at the beginning of the AudioMoth recording on September 16.

The non-pool AudioMoth site was 130 m away from the pool on a ridge 10 m higher than the pool. It was dry and rocky with oak, hickory, maple, and hop hornbeam trees. The presence or absence of a vernal pool is not the only ecological difference between the two sites, but it might be the most important one.

Figure 2. A shagbark hickory (Carya ovata) on ridge of Monkton quartzite near the vernal pool. These were scattered through this forest but none was very big. This was a large one about 10 inches diameter.

When we deployed the AudioMoths (September 16) we waited until dark and listened with the Echo Meter Touch 2 for bat calls. During the 70 minutes (6:30 PM to 7:40 PM) we listened near the vernal pool we heard no bats. When I retrieved the AudioMoths a week later I listened for 70 minutes near the AudioMoth on the high ridge and again heard no bats. I had hoped to learn whether it was more likely to hear little brown or Indiana bats in that forest but I learned nothing. Except that maybe there was not as much bat activity there as in August. The AudioMoth data for the week confirmed this (Figure 3).

Figure 3. Number of audio files recording bat calls during four recording sessions each night for seven nights at two locations adjacent to (upper graph) and 130 m away from (lower graph) vernal pool MLS619 (Bridport, VT). Each 70-minute recording session included five 10 minute listening periods separated by five minute sleep sessions.

Both AudioMoths were configured for four recording sessions per night for seven nights. At the vernal pool, 17 of the 28 recording sessions produced no recordings with bat calls and most of the sessions with bat calls had only 1 or 2 recordings with bat calls (Figure 3, upper graph). A month earlier, two AudioMoths by this vernal pool averaged more than 25 recordings with bat calls during each recording session during two nights (see Figure 3 here). The AudioMoth 130 m from the vernal pool recorded even fewer bats — only 4 of the 28 recording sessions recorded bats (Figure 3, lower graph).

There was not much water in the vernal pool during the September week of AudioMoth recording (see Figure 1). If bats are visiting vernal pools because of the standing water, that might explain why there was more bat activity in August than in September.

It is also possible that bat activity had slowed regionally between mid August and mid September, but that is not supported by our casual observations. On September 16 and September 23 we listened for bats along Mountain Road 0.25 miles from the vernal pool and there was much activity. Both evenings the Echo Meter Touch 2 told us that there were little brown bats, Indiana bats, big brown bats, and silver-haired bats nearby. On both nights there was rarely a minute without a bat audible by the Echo Meter Touch 2. These bat calls were heard from 8:00 PM to 8:30 PM soon after the end of the 70 minute listening periods near the AudioMoths when no bats were heard. So bats were still very active in the area, just not where the AudioMoths were.

There is another difference between the results of the August and September AudioMoth deployments. In August there were more than three times more recordings with bat calls than without bat calls. This made manual inspection of the files a reasonable task. In September there were 38 times more recordings without bats than with bats. In September there were far fewer bats and also a lot more non-bat noises triggering the AudioMoth to start recording (Figure 4). The two AudioMoths produced almost 2,000 files that I inspected to learn that only 52 of them included bat calls.

Figure 4. Number of audio files produced by the AudioMoths that were triggered by something other than a bat. These 1,971 files were produced when the special T.wav files saved by the AudioMoths were expanded into files less than 30 seconds each.

There are a few suspects we can blame for the difference between August and September. First, the amplitude threshold differed between the two deployments. In August I configured the AudioMoths to start recording only when the sound level was greater than 3328 (-26dB or 10%). In September I used a lower threshold of 1536 (-20dB or 5%). (The AudioMoth configuration utility allows the threshold to be set along three different scales, none of which is explained anywhere I could find.) So in September the AudioMoths were triggering recordings when there was a lower sound volume and were expected to make more recordings. I have used the 1536 threshold before and did not get so many false triggers, but that lower threshold probably contributed to the 1,971 recordings of noise.

Second, I added a horn to the AudioMoth cases to protect the delicate acoustic membrane covering the microphone hole (Figure 5). This was not intended to be an acoustic horn, it was designed to protect the membrane and direct rainwater away from it. It’s possible that the horn amplified sounds, but some tests suggested that it didn’t have much effect on sounds (Figure 6). I don’t think it is likely that it caused hundreds of false triggers.

Figure 5. A 3D printed case for an AudioMoth with a 3D printed horn (blue) added to protect the acoustic membrane over the microphone hole.
Figure 6. Sonogram of a single bat chirp recorded simultaneously by two AudioMoths. Both AudioMoths were in 3D printed cases and one (right) had a horn added over the microphone hole. This casual test suggested that the added horn had little acoustic impact. These images are screenshots from SonoBat, a program for visualizing, analyzing, and identifying bat calls. The call is probably from a little brown bat but I don’t have the bat call ID part of Sonobat because it costs $384/year. The annotations on the images and measurement data below are part of the analytical capabilities of the free version of Sonobat.

A third culprit is crickets. Maybe crickets were louder in September than in August and triggered hundreds of recordings. There were lots of crickets chirping in September. It also could have been wind or rain but many of the nights in question were calm and dry. I might never know why there were so many false (non-bat) triggers or why there were so few bat calls compared to a month earlier.

Producing hundreds of files with no bat calls uses up precious storage on the SD card and uses more battery power. It also requires more manual inspection if you don’t have an automated system for analyzing the files. For a one week deployment the storage space and battery life are not critical, but the prospect of 2,000 files that need to be opened and viewed manually to find out whether I learned anything about bats is dampening my enthusiasm for this project.

For the next deployment I will make at least one change to the AudioMoth configuration. I will apply a filter to the incoming sound so frequencies below 25kHz (i.e., sounds we can hear) are quieted some. This high pass filter is applied first and then the filtered sound is compared to the amplitude threshold to decide whether to start recording. That should help prevent the recording of some normal sounds with frequencies too low to be bats (like crickets). I don’t know yet whether I will use the lower or higher amplitude threshold. All three AudioMoths are recording now in the backyard with different configurations to help make that decision.

Here is the configuration file used for this seven day deployment in both AudioMoths.

Leave a Reply

Your email address will not be published.