A Comparison of Bat Calls Recorded by Two Acoustic Monitors

Passive acoustic monitoring with autonomous recording units (ARUs; also called monitors) is a popular method that wildlife biologists use to collect presence/absence, relative activity, and behavioral data (Gibb et al. 2019; Sugai et al. 2019). Passive monitoring with ARUs is particularly useful for bat research given the difficulties of surveying for nocturnal species that emit inaudible vocalizations (Gibb et al. 2019). Further, ARUs can be deployed to sample multiple locations at once (Gibb et al. 2019), can detect rare species with greater success than active techniques (De Bondi et al. 2010; Diggins et al. 2016; but see Rojas et al. 2019), and are typically less expensive to implement than direct methods, such as live trapping (Diggins et al. 2016). The information gained from ARUs can help identify habitat characteristics that promote bat activity and species diversity (Bender et al. 2015; Caldwell et al. 2019; Austin et al. 2020), but most monitors that record ultrasound exceed US$800 per unit, and resources to monitor wildlife populations are often limited. Thus, access to low-cost ARUs that record ultrasound could help make large-scale passive monitoring projects for bats more feasible and reduce the barrier to entry for scientists or community members who want to participate in bat studies.

The simple-to-use, open-source AudioMoth (Open Acoustic Devices, UK) is a small (58 × 48 × 15 mm) and inexpensive (∼US$50) full-spectrum ARU with a built-in omnidirectional microphone that can record sounds up to 192 kHz (Hill et al. 2019). Researchers and hobbyists have purchased >22,000 AudioMoths since 2017 (GroupGets, LLC 2021) and used these monitors to study cicadas (superfamily Cicadoidea; Hill et al. 2018), gunshots (Hill et al. 2018), gray wolves Canis lupus (Barber-Meyer et al. 2020), and anurans (order Anura; Lapp et al. 2021). Researchers also used AudioMoths to investigate bat species occurrence (Katunzi et al. 2021), bat call plasticity (Montauban et al. 2021), and cave bat emergences (Revilla-Martín et al. 2020). The abovementioned studies used automated classification software (e.g., Tadarida, Kaleidoscope Pro) as a component of their acoustic analysis but provided few details regarding call quality. Further, when we investigated whether we could use AudioMoths for our own bat research, we could not find any peer-reviewed studies that compared classification results obtained using files recorded by this monitor type with any other ARUs that record ultrasound.

Variation in ARU performance can influence the results of bat research, inventory, and monitoring programs and the degree to which studies are comparable (Adams et al. 2012; Kaiser and O'Keefe 2015). Thus, understanding potential variation across ARUs is important and should be considered when researchers select a monitor for their projects. Our goal was to compare the number of bat call files, number of echolocation pulses, and number of species recorded by AudioMoths and full-spectrum Song Meter SM4BAT-FS monitors with omnidirectional SMM-U2 microphones (hereafter SM4BATs; Wildlife Acoustics, Inc., Maynard, MA) and identified using automated classification software (i.e., SonoBat; SonoBat, Arcata, CA). We predicted that SonoBat would identify a similar number of species from call files recorded by AudioMoths and SM4BATs. However, during a pilot study, we observed that AudioMoths recorded more background noise than our SM4BATs. As such, we predicted that SonoBat would identify fewer bat call files and echolocation pulses from recordings made by AudioMoths than from recordings made by SM4BATs.

We conducted our research at the Catahoula and Winn Ranger Districts in the Kisatchie National Forest located in Louisiana. We deployed pairs of AudioMoths and SM4BATs at 18 sites 10–140 ha in size representing six forest types (i.e., 3 sites each composed of clearcuts conducted within 5 y of our research, thinned loblolly pine Pinus taeda, loblolly pine with group selection harvests, loblolly or longleaf pine P. palustris stands managed for red-cockaded woodpeckers Leuconotopicus borealis, unmanaged loblolly pine >40 y old, and bottomland hardwood forests >40 y old). See Kunberger and Long (2022) for a detailed description of each forest type. At each site, we positioned two acoustic sampling points >200 m from other acoustic sampling points and >100 m from edge habitat (excluding group selection harvest sites, where we positioned acoustic sampling points at the edge of the group cut, and one study site managed for red-cockaded woodpeckers, where we positioned acoustic sampling points ∼80 m from edges due to area constraints). All 36 acoustic sampling points had minimal vegetation directly above the microphones.

We deployed the monitors during two breeding (June–August 2020 and 2021) and two nonbreeding (January–February 2020 and December–February 2021) periods (Harvey et al. 2011), although, due to logistical constraints, we did not monitor all 36 acoustic sampling points during each sampling period. The climate of our study area is subtropical and has an average annual rainfall of 150 cm, and daily temperatures range from ∼5°C in January to ∼35°C in July (National Weather Service 2021). During the 2020 nonbreeding period, the average dry bulb temperature was 9°C, and the average relative humidity was 83% (National Oceanic and Atmospheric Administration 2022). During the 2020 breeding period, the average dry bulb temperature was 25°C, and the average relative humidity was 79% (National Oceanic and Atmospheric Administration 2022). During the 2021 nonbreeding period, the average dry bulb temperature was 4°C, and the average relative humidity was 83% (National Oceanic and Atmospheric Administration 2022). Finally, during the 2021 breeding period, the average dry bulb temperature was 26°C, and the average relative humidity was 78% (National Oceanic and Atmospheric Administration 2022).

We deployed SM4BAT microphones and AudioMoths ∼15 cm apart on a 2-m-tall polyvinyl chloride pole at the acoustic sampling points with both microphones facing up (Wildlife Acoustics, Inc. 2022). Our study area is wet all year, and weatherproof cases were unavailable for AudioMoths when we started our research. Thus, we began collecting data in 2020 with AudioMoths housed in disposable plastic bags, as recommended by Open Acoustic Devices (Open Acoustic Devices 2021). We lost several units to water damage (both rain and condensation) and placed the AudioMoths in plastic containers with the microphone hole covered with clear tape, as recommended by other researchers that used these monitors. We again lost several units to water damage and could not deploy them without enclosures; so, for the remainder of our study, we housed our AudioMoths in plastic bags designed to protect smartphones, which protected the circuit board and built-in microphone from water ingress.

We calibrated the AudioMoths to record at a 256-kHz sample rate with medium gain and, in accordance with firmware settings for V. 1.2.2 and previous research using AudioMoths to record bat activity (Katunzi et al. 2021), we programmed the monitors to record cyclically at 10-s intervals followed by 5-s pauses, during which they stored data files to micro-SD cards located inside each device. We programmed our SM4BATs with the default settings, which include triggering at a minimum detected frequency of 16 kHz for a minimum recording length of 1.5 ms and recording at a 256-kHz sample rate with a 12-dB gain. We programmed all ARUs to turn on 30 min before sunset and to turn off 30 min after sunrise.

For both AudioMoth and SM4BAT data, we used the SonoBat (V. 4.4.1) batch file scrubber to exclude files without bat calls (e.g., noise) and used SonoBat's Southeastern southeast classifier (i.e., reference library of bat calls for species expected to occur in our region) with default values (acceptable call quality of 0.60 and decision threshold of 0.90) to automatically identify files to species. The bat species represented in the Southeastern southeast classifier include southeastern myotis Myotis austroriparius, tricolored bats Perimyotis subflavus, evening bats Nycticeius humeralis, eastern red bats Lasiurus borealis, big brown bats Eptesicus fuscus, silver-haired bats Lasionycteris noctivagans, Rafinesque's big-eared bats Corynorhinus rafinesquii, northern yellow bats Dasypterus intermedius, Mexican free-tailed bats Tadarida brasiliensis, and hoary bats Aeorestes cinereus. SonoBat classified files as high-frequency (typically >35 kHz) or low-frequency (typically <35 kHz) bat activity, which is helpful in estimating the activity of narrow and edge space aerial foragers (i.e., high-frequency calls) and open space aerial foragers (i.e., low-frequency calls; Data S1, Supplemental Material; Denzinger and Schnitzler 2013). Hereafter, we refer to files with bat echolocation as call files because we assumed that each file contained one full bat echolocation sequence. Last, SonoBat identified the number of echolocation pulses in each call file, and we defined an echolocation pulse as a single echolocation call (Data S1, Supplemental Material). We used an automatic classifier without manual verification of species identifications because the automatic classification process is more reproducible than manual identification (Nocera et al. 2019) and because determining species occurrence was not the goal of our study.

We conducted our statistical analyses in R (V. 4.1.1; R Core Team 2021) with packages AICcmodavg (V. 2.3-1; Mazerolle 2020), ggplot2 (V. 3.3.5; Wickham 2016), MASS (V. 7.3-58; Venables and Ripley 2002), and ggforce (V. 0.3.3; Pedersen 2021). We first used a Kruskal–Wallis test (Kruskal and Wallis 1952) with α = 0.05 to determine if the three enclosures we used for the AudioMoths had an influence on the number of call files identified to species and species richness during the nonbreeding period. We also summarized the average duration of SM4BAT call files during each sampling period given differences in the settings that were available for each ARU at the time of our study (i.e., SM4BATs recorded a minimum of 1.5 ms after they were triggered, and AudioMoths recorded cyclically at 10-s intervals followed by 5-s pauses). We used generalized linear modeling with negative binomial distributions and log-link functions to examine our response variables of interest, including total identified call files, total call files, call files with high-frequency bat calls, call files with low-frequency bat calls, echolocation pulses, and species richness. Our explanatory variables included forest type, monitor type, and period (i.e., nonbreeding or breeding). We included forest type as an explanatory variable in our analyses because stand characteristics (e.g., understory vegetation structure) can influence the detectability of bat echolocation calls (Limpens and McCracken 2004; Britzke et al. 2013; O'Keefe et al. 2014).

Our final candidate set for each response variable included a null model, main effects models for each explanatory variable, and additive models representing all combinations of two explanatory variables. We used Akaike's information criteria corrected for small sample sizes (AIC_c) to rank each model and examined relative support for each model using ΔAIC_c and model weights (Sugiura 1978; Anderson 2008). We considered models with a ΔAIC_c value of <2 equally plausible. We examined parameter estimates and their 95% confidence intervals for the best fit models and considered a 95% confidence interval that included 0 as uninformative (Burnham and Anderson 2002; Grueber et al. 2011). Last, we visualized the predicted values for each response variable from the best fit models and their 95% confidence intervals.

We collected data on 237 nights (26 nights during the 2020 nonbreeding period, 75 nights during the 2020 breeding period, 54 nights during the 2021 nonbreeding period, and 82 nights during the 2021 breeding period). SonoBat did not detect bat calls from files recorded during 61 nights for AudioMoths and 24 nights for SM4BATs. SonoBat did not detect bat calls from files recorded during eight nights from either ARU, so we excluded these eight nights from our analyses. We found no significant differences in the number of identified call files (H₂ = 0.67, P = 0.72) or species richness (H₂ = 1.10, P = 0.58) identified by SonoBat across the three enclosures we used for our AudioMoths during the nonbreeding period, so we pooled recordings from all of the enclosures for our subsequent analyses. We found that the mean duration of bat call files from SM4BAT recordings was approximately 10 s; however, we found considerable variation in the duration of individual call files (Table S1, Supplemental Material).

The best fit model for total identified call files, call files with high-frequency bat calls, echolocation pulses, and species richness was monitor type + period (Table 1). The predicted values for these response variables were two to four times higher for SM4BATs than for AudioMoths during both the breeding and nonbreeding periods (Figure 1; Table S2, Supplemental Material). The best fit model for total call files and call files with low-frequency bat calls was forest type + period (Table 1). The predicted values for total call files and call files with low-frequency bat calls were eight and ten times higher for the breeding period than for the nonbreeding period when we averaged the predicted values across forest types (Figure 1; Table S2, Supplemental Material). Compared with other forest types, bottomland hardwood forests had higher predicted total call files and call files with low-frequency bat calls during both the breeding and nonbreeding periods (Figure 1; Table S2, Supplemental Material).

Many factors can influence detectability of bat calls by ARUs, including the acoustic properties of the species' calls (e.g., frequency and intensity), the foraging behavior of the species or individuals, habitat conditions, and microphone sensitivities (i.e., the distance at which a microphone can record a subject; Limpens and McCracken 2004; Britzke et al. 2013). In particular, ARUs with less sensitive microphones record fewer detections of bats with high- or low-frequency calls (Downes 1982; Fenton et al. 2001; Adams et al. 2012) because they detect sounds at shorter distances from the microphone (Britzke et al. 2013; Kaiser and O'Keefe 2015). Further, the distance that bat calls travel can vary greatly due to environmental conditions (Limpens and McCracken 2004; Britzke et al. 2013). For example, high-frequency sound travels shorter distances in hot and humid conditions, meaning that microphones must be hypersensitive to detect high-frequency bat calls in these environments (Petterson 2004; Goerlitz 2018). We conducted our study in Louisiana, which has high temperatures and humidity throughout the year (National Oceanic and Atmospheric Administration 2022), suggesting that ARUs deployed in Louisiana should be highly sensitive to detect high-frequency bat species at far distances from the microphone (Petterson 2004). Our results indicate that SM4BATs record more high-frequency call files than AudioMoths, which may be attributed to differences in microphone sensitivity between the two ARUs.

Differences in the settings used to record bat calls may have also influenced our results. Song Meter SM4BAT-FS monitors are designed to increase the signal-to-noise ratio, meaning that they adjust to record clearer files as background noise increases or decreases throughout the night (Wildlife Acoustics, Inc. 2022). In addition, SM4BATs are designed to record when the monitor detects a bat call (i.e., triggered) rather than recording cyclically throughout the night (Wildlife Acoustics, Inc. 2022). By contrast, the AudioMoth firmware that was available when we started our study did not have signal-to-noise ratio adjustments, amplitude thresholds, or frequency thresholds that allowed for triggered recording. Open Acoustic Devices released a firmware update in 2020 that added these features (Open Acoustic Devices 2020), but we did not adopt it when it was released to avoid introducing additional variation to the current study and to ensure that our data could be compared to previous work. The firmware update might improve the clarity of recordings and, thus, minimize differences in calls recorded by AudioMoths and SM4BATs. Addressing this question was beyond the scope of our research, but we recommend additional studies to compare call files recorded by SM4BATs and other commercially available ARUs with call files recorded by AudioMoths that use the firmware update. We also urge caution when comparing data across time, as firmware updates may improve the quality of data collected by AudioMoths or other monitors but may simultaneously introduce variation that should be considered when scientists interpret their results.

It is also important to note that enclosure type could have influenced the patterns we observed. We ultimately housed our AudioMoths in waterproof smartphone bags during our field study because we experienced equipment failure due to water damage when we used the disposable plastic bags and plastic box enclosures recommended by Open Acoustics and other users. We did not find differences in the number of call files identified to species or species richness across the three enclosures we used to protect the AudioMoths. However, previous studies showed that enclosure design can influence the detectability of bats (Britzke et al. 2010; Kaiser and O'Keefe 2015), and all three enclosure types likely limited the amount of sound that reached the AudioMoth's internal microphone. Louisiana is wet throughout the year, so it was not feasible for us to deploy AudioMoths without enclosures at our study sites in the Kisatchie National Forest. However, in September and October 2022, we deployed both ARUs with and without enclosures locally so that we could bring the AudioMoths indoors when weather conditions would likely cause damage to the monitors. Given our small sample sizes, we did not include our results here, but, anecdotally, we found that the number of call files, echolocation pulses, and species recorded by the two monitors were more similar when the AudioMoths were deployed without enclosures (A.M. Long, Louisiana State University Agricultural Center, unpublished data; see Figures S1, S2, and S3, Supplemental Material, for a visual comparison of call quality among these treatments). In the short time since we concluded our study, Open Acoustics released a waterproof enclosure for AudioMoths that could limit water ingress, but, as discussed above, it should be tested to determine what effect the new enclosure type has on recordings.

No ARU will record 100% of the bat calls present in the environment, and the challenges we described above are expected when new technology becomes available. Based on our results, we recommend caution when comparing data collected by different ARUs, especially through time as firmware updates and new enclosures become available. In addition, we encourage additional research to examine variation in monitor performance across a wide range of environmental conditions. Differences in recording quality across ARUs can have a direct influence on the results of monitoring studies, particularly when researchers use automated classifiers to identify species presence in their study area. With declining bat populations worldwide, it is crucial that estimates of species occurrence and activity are accurate and reproducible to best manage for species of conservation concern. Our results suggest that SM4BATs record a greater number of call files, echolocation pulses, and species than AudioMoths, but AudioMoth recordings could be influenced by recording settings and enclosure design and may be improved with recent updates. We urge researchers using AudioMoths to report the firmware and all the settings they used, details on their enclosures, typical environmental conditions in their study area, and any issues they had during the recording or classification process. Such transparency could help researchers compare AudioMoth results across studies, leading to improved estimates of bat species occurrence and activity that can inform management for species of conservation concern.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Data S1. Comma separated value file of the numbers of call files, echolocation pulses, and species richness identified by SonoBat each night from recordings that we collected using AudioMoths and SM4BATs across forest types in the Kisatchie National Forest in central Louisiana during the nonbreeding (January–February 2020 and December 2020–February 2021) and breeding (June–August 2020 and 2021) periods. Data fields include monitor (AudioMoth monitor [AudioMoth] or Song Meter SM4BAT-FS [SM4BAT]), date (the night we sampled), enclosure (plastic box, disposable bag, phone bag, or not applicable [NA]), forest type (forest types include bottomland hardwood [BH], clearcut [CC], group selection harvest [GS], red-cockaded woodpecker habitat [RC], thinned [TH], and unmanaged loblolly [UL]), ID (unique identifier for each deployment), period (nonbreeding or breeding), year (2020 or 2021), pulses (number of echolocation pulses), TotalIDdCallFiles (call files identified to species), HighFCallFiles (call files with high-frequency bat calls), LowFCallFiles (call files with low-frequency bat calls), TotalCallFiles (total call files), SpeciesRichness (species richness), and exclude (Y for yes and N for no; we excluded all sampling occasions during which both Song Meter SM4BAT-FS monitors and AudioMoth monitors did not record bat echolocation).

Available: https://doi.org/10.3996/JFWM-22-028.S1 (44 KB XLSX)

Table S1. The mean ± standard error (SE), minimum, and maximum values of Song Meter SM4BAT-FS monitor call file durations in seconds for bat call files that we recorded in the Kisatchie National Forest in central Louisiana during the nonbreeding (January–February 2020 and December–February 2021) and breeding (June–August 2020 and 2021) periods and identified using automated classification software (SonoBat) for our study comparing results we obtained using AudioMoth monitors and full-spectrum Song Meter SM4BAT-FS monitors. We set AudioMoth monitors to record cyclically at 10-s intervals, with 5-s pauses for the device to write the recording data to the SD card.

Available: https://doi.org/10.3996/JFWM-22-028.S2 (14 KB DOCX)

Table S2. Parameter estimates for variables in the best fit generalized linear models (ΔAIC_c < 2) for bat calls that we recorded in the Kisatchie National Forest in central Louisiana during the nonbreeding (January–February 2020 and December–February 2021) and breeding (June–August 2020 and 2021) periods and identified using automated classification software (SonoBat) for our study comparing results obtained using AudioMoth monitors and full-spectrum Song Meter SM4BAT-FS monitors.

Available: https://doi.org/10.3996/JFWM-22-028.S3 (17 KB DOCX)

Figure S1. A comparison of bat call quality among Song Meter SM4BAT-FS monitors and AudioMoth monitors both with and without smartphone bag enclosures for an exploratory analysis of call files that we recorded in Baton Rouge, Louisiana, over 12 nights from September to October 2022 and identified using the automated classification software SonoBat. These sequences occurred within 1 min of each other. SonoBat identified all of these sequences as a high-frequency bat, but SonoBat did not identify any of these sequences to the species level.

Available: https://doi.org/10.3996/JFWM-22-028.S4 (1.094 MB DOCX)

Figure S2. A comparison of bat call quality between a Song Meter SM4BAT-FS monitor and an AudioMoth monitor without smartphone bag enclosures for an exploratory analysis of call files that we recorded in Baton Rouge, Louisiana, over 12 nights from September to October 2022 and identified using the automated classification software SonoBat. These sequences occurred within 1 min of each other. SonoBat identified both of these sequences as an evening bat Nycticeius humeralis.

Available: https://doi.org/10.3996/JFWM-22-028.S5 (503 KB DOCX)

Figure S3. A comparison of bat call quality among Song Meter SM4BAT-FS monitors and AudioMoth monitors both with and without smartphone bag enclosures for an exploratory analysis of call files that we recorded in Baton Rouge, Louisiana, over 12 nights from September to October 2022 and identified using the automated classification software SonoBat. These sequences occurred within 1 min of each other. SonoBat identified all of these sequences as a Mexican free-tailed bat Tadarida brasiliensis.

Available: https://doi.org/10.3996/JFWM-22-028.S6 (792 KB DOCX)

We thank E. Smith from the U.S. Forest Service for her help in selecting field sites in the Kisatchie National Forest. We also thank C. Bergeron from the Louisiana State University AgCenter's Camp Grant Walker for providing housing during the project. Additionally, we acknowledge the lab members who helped collect these data, including N.M. Raginski, M.D. Gamble, E.L. Munch, S.M. Pettibone, N.M. Black, A.K. Tunstall, T.J. Price, R.A. Garcia, P.M.K. Oramous, C.E. Crawford, and A.C. Erwin. We thank the Associate Editor and anonymous reviewers for their comments, which greatly improved our manuscript. We received funding from the Gilbert Foundation's Graduate Research Assistantship awarded by Louisiana State University's School of Renewable Natural Resources, McIntire Stennis project number LAB94479 from the U.S. Department of Agriculture National Institute of Food and Agriculture, and the Louisiana Forestry Association's Bob Blackmon Graduate Scholarship. The undergraduates who assisted with our research received funding from the A. Wilbert's Sons Research Internship awarded by Louisiana State University's School of Renewable Natural Resources.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.