Bat capture, handling, and tag attachment were carried out in accordance with guidelines of American Society of Mammologists33 under permit from the California Department of Fish and Wildlife (# SC-002911). Experimental methods were approved by the Institutional Animal Care and Use Committee of the USDA Forest Service (IACUC 2017-014). We captured bats using 2.6-m high mist nets in a triple-high configuration. We measured forearm length and mass and determined species, age, sex, and reproductive status for each captured individual.
We used Vesper Pipistrelle on-board audio-recording devices with an accelerometer (ASD Tech, Haifa Israel) to quantify bat movement throughout the duration of attachment. We used the smallest possible battery (0.5 g) which was sufficient to allow a 3-h recording period on the first night and up to a 4-h recording period on the second night. Tags were programmed to record for 10 s once every 3 min from 23:00 to 02:00 on the first night and for 10 s once every minute from 19:00 to 23:00 on the second night. We recovered tags from bats tagged between September 28th and October 7th. Sunset was at 19:02 on September 29th and 18:47 on October 8th. Unfortunately, the timing mechanism on the tags malfunctioned some of the time, causing only some of the recordings to have synchronous audio and accelerometer data (See Results).
We attached Holohil LB-2X VHF transmitter (0.27 g) to the audio tags so we could locate the device once it detached from the bats. We coated the entire tag package (except the microphone opening) with liquid silicone followed by a latex sleeve covering to provide protection from the environment. The total tag package had a mass of 2.9 g which represented 10.6–12.5% of the mass of the bat. Several studies conducted in flight tents and in the field have shown no adverse consequences of payloads up to 15% for short duration deployments16,34. The diversity of natural behaviors that we observed, including prey pursuit, conspecific interaction, and extended flight over multiple nights indicates that hoary bats are capable fliers with this payload, however we cannot rule out the possibility that tags altered the behaviors that were observed.
We attached tags to the posterior dorsum of bats using latex surgical adhesive (Torbot Liquid Bonding Cement, Torbot Group Inc. Cranston, Rhode Island). We used the minimum quantity of adhesive that we estimated would be necessary for tags to remain affixed to bats for 2 nights. We recovered tags by using ground- and aircraft-based VHF telemetry to determine the general location of the shed tag, followed by homing in on the VHF signal using ground-based telemetry. Final recovery of tags was achieved using visual searches of the ground.
Microphone calibration
We calibrated on-board microphones to determine the minimum sound pressure level (SPL) at which we could reliably detect micro calls. We did this by broadcasting a series of micro calls from an Avisoft (Glienicke/Nordbahn, Germany) Scanspeak ultrasonic speaker to the on-board tags. The series of micro calls consisted of a single high-quality micro call that was broadcast 30 times with each successive call being 3 dB lower in SPL. The absolute intensity of the broadcast was calibrated by broadcasting the same signal to a G.R.A.S (Holte, Denmark) 40DP 1/8″ microphone, which itself was calibrated with a G.R.A.S 42AB sound calibrator. For both the calibration of the sound playback and the broadcasts to the on-board microphone, the microphones were placed 10 cm from the speaker. We repeated this procedure three times for each of three microphones that had been recovered from the bats and determined the SPL of the lowest amplitude micro call that could be detected on all nine broadcasts. This SPL was used as the minimum detectable level at which our microphones could detect micro calls.
Data processing
Determining whether bats are flying
We used custom MATLAB (Natick, MA) scripts to analyze ultrasound and accelerometer recordings. We first determined whether bats were in flight for each recording. Unfortunately, we were only able to record simultaneous accelerometer and acoustic data for 364 out of 2241 recordings. For these recordings, we independently classified each file as flight or no flight using only the accelerometer data and only the audio data. Accelerometer recordings showed clear and prominent wingbeat oscillations in the dorsoventral, or Z-axis (Fig. S2A). One observer used a custom program (AccelVis) to visualize and manually classify all accelerometer files. We also quantified the magnitude of wingbeat oscillations by measuring the root-mean-square magnitude of signals after applying a high-pass filter of 4 Hz (Bats used wingbeat frequencies of approximately 8 Hz).
A different observer classified all audio recordings as flight or no flight based on the presence or absence of low-frequency wind noise generated by the relative motion of the bats flying through the air (Fig. S2). The Individuals conducting the audio and acceleration analyses were blind to one another’s data. As with the accelerometer data, we analyzed all files both qualitatively and quantitatively. For the qualitative analysis, a user visualized files using a custom program (AudioBrowser; available with all data files as supplementary data) and noted presence or absence of low-frequency wind noise. We also quantified this wind noise by measuring the RMS magnitude of signals after applying a 1-Hz low pass filter. This resulted in a distinct bimodal distribution of low frequency magnitudes that corresponded to no wind and wind conditions with the two peaks being separated by approximately 30 dB. A small number of files (< 10%) had values between the two peaks because of abrupt noise bursts from an unknown origin. This noise had a distinct spectral-temporal profile and all files containing this noise were classified manually.
We compared the classification of flight/no-flight from accelerometer and audio data (N = 364 files) and found 100% correspondence. Therefore, for audio files lacking synchronous accelerometer data, we used low-frequency wind noise to determine flight vs. no flight conditions.
Quantifying flight maneuvering
For each accelerometer recording, we quantified the magnitude of flight maneuvering (Fig. 3) using the lateral and anterior–posterior components of the accelerometer measurements. We first detrended each signal by subtracting the mean signal value. We next applied a 4 Hz low-pass filter to each signal to remove wingbeat frequency oscillations and highlight maneuvers lasting more than one wingbeat. Finally, we took the root-mean-square value of the filtered signals as the overall measure of flight maneuvering.
Classifying acoustic behaviors
We manually classified each 10-s acoustic recording as either silence, micro calls, high-intensity calls, feeding buzz or social interaction (Fig. 1). Recordings containing a feeding buzz or social interaction were classified as such (see below for how these events were identified). Otherwise, recordings were classified as silence, micro calls or high-intensity calls based on the call type that occupied the majority of the recording (> 5 s). This 5 s threshold is twice the longest pulse interval recorded for echolocation calls (Supplementary Information), and therefore represents a conservative threshold for identifying silent periods.
High-intensity calls could be identified by their consistently high signal levels. For recordings where no calls were initially detected, the observer made a second examination of the recording using a custom 55–90 kHz bandpass filter setting that highlights micro calls (Fig. 1D). A second observer also examined all files where either no calls or micro calls were detected by the first observer to confirm classification. Recordings were processed both by visualization of spectrograms and by listening to slowed-down recordings through headphones.
Hoary bat feeding buzzes have a characteristic pattern involving a rapid increase in calling rate, and progressively decreasing call intensity (Fig. 1B)35,36. In contrast, social interactions involve prolonged (often several seconds) high-intensity echolocation calls produced at a high rate (e.g., 50–100 Hz) with a second bat also producing echolocation calls at a relatively high calling rate14. Echolocation calls of “other” bats (which could be present in any of the recordings) could be distinguished from the calls of the bat with the tag because they were typically recorded at a much lower intensity levels that increased and decreased, presumably as the other bat approached and then withdrew from the focal bat and were temporally out of phase with calling rate of the tagged bat. Calls classified as “other bat” also had lower calling rates compared to social interactions.
Statistical analysis
Acoustic recordings were organized by individual bat (Table 1) and by time of night (Fig. 2). To determine if bats exhibited consistent differences in the use of high-intensity echolocation, we measured the proportion of recordings including high-intensity echolocation for each bat night. Initial analysis of the data indicated that bats produced high-intensity echolocation during either most or all of the recordings (96–100%, including feeding buzzes) or at a considerably lower rate (< 80%). We used a one-sided binomial probability test to determine the likelihood of all five bats with two nights of data using high (> 96%) or low (< 80%) rates of high-intensity echolocation on each of the two nights they were recorded.
We used a permutation test to determine whether feeding buzzes and social calls were produced during periods when bats were producing fewer or more high-intensity echolocation calls than expected by chance. For this test, we kept the timing of recordings constant, but randomized the acoustic classifications shown in Fig. 2. For each feeding buzz and social call, we identified all recordings that occurred within 15 min and determined the proportion of those recordings that included high-intensity echolocation calls. This was iterated 10,000 times to generate a random distribution of the percent of recordings within 15 min having high-intensity echolocation calls (See Fig. 2B for an example randomization). The actual percent of recordings within 15 min containing high-intensity echolocation calls was then compared to this distribution to determine a P value.
We used linear regression to determine if cloud cover or moon illumination was correlated with the proportion of time bats used high-intensity echolocation calls. For each recording night, we used moon calendars to look up percent moon illumination. We also confirmed that the moon was up during the recording period for each night. We downloaded weather data for the weather station closest to our release site (Fortuna Rohnerville Airport, KFOT) and measured the proportion of time during our recording period each night when the cloud cover was either “overcast” or “mostly cloudy” versus “clear” or “partly clear”. We then conducted a linear regression to determine if there was a relationship between the proportion of recordings with high-intensity calls and either our nightly measure of moon illumination or cloud cover. We did not attempt to conduct an analysis of calling behavior and cloud cover at a finer temporal resolution (such as hourly) to avoid temporal autocorrelation that might confound our analysis.
Finally, we compared our measures of flight maneuvering across recordings that were classified as different acoustic behaviors (Fig. 3D). Because of the highly skewed distribution, we used a non-parametric Kruskal–Wallis test with Tukey post-hoc comparisons to determine whether median levels of flight maneuvering differed between recordings classified as different acoustic behaviors.
Source: Ecology - nature.com
