Do you hear what I hear? Implications of detector selection for acoustic monitoring of bats


Correspondence author. E-mail:


  1. The probability of detecting the echolocation calls of bats is affected by the strength of the signal as well as the directionality and frequency response of the acoustic detectors. Regardless of the research question, it is important to quantify variation in recording system performance and its impacts on bat detection results. The purpose of this study was to compare the detection of echolocation calls among five commonly used bat detectors: AnaBat SD2 (Titley Scientific), Avisoft UltraSoundGate 116 CM16/CMPA (Avisoft Bioacoustics), Batcorder 2·0 (ecoObs), Batlogger (Elekon AG) and Song Meter SM2BAT (Wildlife Acoustics).
  2. We used playback of synthetic calls to optimize detection settings for each system. We then played synthetic signals at four frequencies (25, 55, 85 and 115 kHz) at 5-m intervals (5–40 m) and three angles (0°, 45°, 90°) from the detectors. Finally, we recorded free-flying bats (Lasiurus cinereus), comparing the number of calls detected by each detector.
  3. Detection was most affected by the frequency dominating the signal and the distance from the source. The effect of angle was less apparent. In the synthetic signal experiment, Avisoft and Batlogger outperformed other detectors, while Batcorder and Song Meter performed similarly. Batlogger performed better than the other detectors at angles off-centre (45° and 90°). AnaBat detected the fewest signals and none at 85 kHz or 115 kHz. Avisoft detected the most signals. In the free-flying bat experiment, Batlogger recorded 93% of calls relative to Avisoft, while AnaBat, Batcorder and Song Meter recorded 40–50% of the calls detected by Avisoft.
  4. Numerous factors contribute to variation in data sets from acoustic monitoring; our results demonstrate that choice of detector plays a role in this variation. Differences among detectors make it difficult to compare data sets obtained with different systems. Therefore, the choice of detector should be taken into account in designing studies and considering bat activity levels among studies using different detectors.


Echolocation provides a window through which the behaviour and ecology of bats can be evaluated. Specifically, calls used by echolocating bats can be conspicuous to bat detectors, permitting biologists to distinguish among species by their calls and to identify foraging activity. Bat detectors, instruments sensitive to the acoustic frequencies dominating bat calls, have been extensively used in a range of bat studies, from those investigating echolocation behaviour, to others documenting patterns of distribution and activity levels. By 2012, the variety of commercially available bat detectors offered a spectrum of features at a range of prices (e.g. weatherproofing, temperature sensors and storage options; Table 1) but key features, such as microphone quality, sampling rate and recording technology will determine the ability to detect bats. Many published articles have used data from bat detectors to address questions about the echolocation behaviour of bats, as well as their patterns of activity and habitat use (e.g. Gillam 2007; Collins & Jones 2009; Müller et al. 2012).

Table 1. A comparison of the features associated with commercially available ultrasonic bat detector systems
System and manufacturer featureAnaBat SD2 titley ElectronicsAvisoft UltraSoundGate 116 w CM16/CMPA Avisoft BioacousticsBatcorder 2·0 ecoObsBatlogger Elekon AGSong Meter SM2BAT 192 kHz Wildlife Acoustics
  1. Pricing from company:

  2. a

    website or

  3. b


  4. c

    Price converted to USD

Recording technologyZero-crossing16-bit, full-spectrum16-bit, full-spectrum16-bit, full-spectrum16-bit, full-spectrum
Sampling rateN/A500 kHz500 kHz312·5 kHz192 kHz (384 kHz available)
Sound file typeAnaBat.wav.raw.wav & .xml.wav & .wac
Storage typeCompact Flash (CF)External through computerSDHCSDHCSDHC × 4
Storage capacity128 GBunlimited16 GB32 GB128 GB
Battery4 AA batteriesRuns off computerNiMH 6V 2700 mA, rechargeableLIB 3·7V 4600 mAh, rechargeable4 D batteries
Microphone typeCondenserCondenserElectretElectretElectret
Omnidirectional microphone?NoNoYesYesYes
Recording schedule?Yes, through CF ReaderYes, through Avisoft-RECORDER softwareYesYesYes
Post-process toolsAnaLookAvisoft-SASLab ProNoneBatExploreBatch noise scrubber, zero-cross converter, Wac2Wav converter
Weatherproof enclosure?NoNoYesWith StrongBoxYes
Weatherproof microphone?NoNoNoNoYes
GPS?Can connect externallyNoNoYesOptional
Temperature sensor?Internal temperatureNoNoExternal temperatureInternal and external
Price (USD)$2200b$5999a$3273ac$2035ac$999a

Acoustic sampling is a common, powerful technique for monitoring the activity of echolocating bats. Bat detectors are widely used by researchers, including those working for government agencies, environmental consulting firms and academics. Behavioural, presence/absence and relative abundance data are commonly collected with these devices. The results of research relying on bat detectors inform our understanding of bat ecology and behaviour and are frequently used to guide important wildlife management decisions (US Fish & Wildlife 2011). Acoustic monitoring is non-intrusive and capable of recording large quantities of data. However, the specific combination of hardware may affect the quality, precision and quantity of data collected.

Variation in microphone sensitivity and detection algorithms can produce data sets that differ among detectors. Both Downes (1982) and Fenton (2000) noted significant differences in detection sensitivity among brands of narrowband and broadband acoustic detectors. This variation has the potential to affect acoustic monitoring studies and their conclusions; whether the focus is curiosity-driven research or environmental assessments where low bat activity is assumed to equal low numbers of bats and therefore low risk (US Fish & Wildlife 2011).

The acoustic nature of bats is highly variable (e.g. frequency, intensity, etc.), which can influence detectability by even the ‘best’ detectors. Bats using low-intensity echolocation calls dominated by higher frequencies are less detectable than those using high-intensity calls dominated by lower frequencies. Higher-frequency sounds attenuate more quickly and will be detected less frequently than higher intensity, lower-frequency calls (Lawrence & Simmons 1982), resulting in under-representation of these species in acoustic surveys (Murray et al. 2007). Detection bias will be further compounded by the sensitivity and frequency response of the bat detector. Different systems vary in their performance over the range of biologically relevant frequencies. If the microphone has lower sensitivity to high frequencies, the bias caused by atmospheric attenuation will be further exaggerated. The consequences of detection bias will depend on the community being studied; the frequencies of bat calls range from ~8 to >200 kHz, and this range varies with a given bat community. Researchers must consider the community in question when choosing the most appropriate bat-detecting system for their research (Limpens & McCracken 2004).

Microphones with lower sensitivity will detect bats at shorter distances relative to more sensitive microphones. Detectors with shorter detection ranges will sample a smaller airspace and thus have a lower probability of detecting any bats present. Also, not all detectors are equal in their directionality and the orientation of the detector in relation to the bat affects detection (Britzke et al. 2010). When all other factors are equal, detectors with omnidirectional microphones will have a better chance of detecting a bat, compared with more directional microphones. However, a less directional microphone will be less sensitive, giving it a smaller detection range (Limpens & McCracken 2004). The smaller the microphone, the more omnidirectional it will be.

Three levels of variation can confound data acquired with bat detectors. First is the variation associated with the movement of sound through air. Second is that intrinsic to the instruments. Third is variation in echolocation behaviour and call design among bats. Whether the focus of a study is echolocation behaviour or documenting patterns of habitat use, it is important to distinguish between factors two and three. We presented synthetic acoustic signals and echolocation calls of free-flying bats in the wild to compare ultrasonic call detection by five commercially available bat detectors. Our goal was to provide data about relative bat detector performance and bat echolocation behaviour.

With an increasing number of commercially available bat detectors, it is important to address variation in the technologies. A fundamental factor of any methodology is addressing the capabilities and limitations of the equipment being used. It is vital to be aware of the differences that may result from the use of different equipment even when the same sampling method is employed. To date, no study has examined the differences in the detection efficacy among direct high-speed bat detector models.

Materials and methods

We simultaneously deployed five direct high-speed bat detectors for recording both synthetic playback and free-flying bats: AnaBat SD2 (Titley Scientific, Ballina, NSW, Australia), Avisoft UltraSoundGate 116 CM16/CMPA (Avisoft Bioacoustics, Berlin, Germany), Batcorder 2·0 (ecoObs, Nümberg, Germany), Batlogger (Elekon AG, Luzern, Switzerland) and Song Meter SM2BAT (Wildlife Acoustics Inc, Concord, MA, USA). There are a several other commercially available detectors that we were unable to include in this study, for example D500X and D1000X (Pettersson Elektronik, Uppsala, Sweden) and AR125 (Binary Acoustic Technology, Tuscon, AZ, USA). During all trials, microphones were within 10 cm of each other, on a parallel plane. Microphone order and position were rearranged randomly for each trial to change microphone position, but maintain consistent microphone spacing. We avoided variation by recording with only one detector of each model and recording with all detectors at the same time.

Optimizing detector recording settings

We used playback of synthetic signals to optimize detection settings for each system. Our synthetic signal file was 1478 ms in duration and consisted of 20, 57 ms long, constant frequency (CF) signals, five signals at each of four frequencies: 25, 55, 85 and 115 kHz. For playback, we used a laptop running Avisoft-RECORDER-NiDAQmx software connected to an ultrasonic playback interface with an integrated D/A power amplifier (UltraSoundGate Player 116). The interface was connected to an UltraSoundGate Dynamic Speaker ScanSpeak (hardware and software: Avisoft Bioacoustics, Berlin, Germany), which we did not calibrate. When possible, we recorded with all combinations of setting configurations for each detector. When combinations were prohibitively large (>100), we recorded in intervals spanning the full range of configurations. For each configuration, we played synthetic signals 5 m from each device. We analysed each recording visually to find the optimum settings for recording conditions. In cases where multiple configurations were equal, we chose the settings closest to the default settings for the detector. These settings were used for the remainder of our experiments (Table 2).

Table 2. Detector settings used in this study
AnaBat SD2Avisoft UltraSoundGate 116Batcorder 2·0BatloggerSong meter SM2BAT
  1. See each respective detector manual (available online) for the setting description.

Gain: 7

Data Div - 16

Gain: 7Trigger: permanent


Sampling rate: 500 kHz

Format: 16 bit

Buffer: 0·050

No. Buffers: 4

Critical frequency: 14 kHz

Threshold: −36 dB

Post trigger: 800 ms

Quality: 40


minCrest: 5

minRMS: 2

minPeak: 5

HighPass: 6

Sampling rate: 192 kHz

Compression: WAC0

Gain: 36 dB

Dig HPF: fs/16

Dig LPF: Off

Trigger Level: 15 SNR

Trigger Win Right: 1 s

Div Ratio: 16

Synthetic call playback

We played the synthetic CF signals three times at 5-m intervals (5–40 m) and three angles (0°, 45°, 90°) in an open field. This resulted in 15 calls of each frequency played at each distance and angle (24 combinations). We used the automated detection feature (Table 3) of callViewer18 (Skowronski & Fenton 2008), to count the number of calls detected by each system and manually inspected each recording to ensure that there were no false positives. CallViewer is a custom echolocation sound analysis program written with MatLab software (The MathWorks, Nadick, MA, USA). Because AnaBat file formats are not compatible with callViewer software, we visually inspected these recordings in AnaLook (Version 3·8; Titley Electronics). We used general linear models to analyse the number of signals detected (considering each frequency separately) with angle, detector, distance and all two-way interactions. To compare among detectors, we generated pairwise comparisons of the estimated marginal means, controlling for the effect of distance and angle. We used a similar approach to compare the effect among the three angles. We estimated the detection range by modelling the probability of detection of each signal frequency at each angle by all detectors with a logistic regression in PASW18 (SPSS Inc., Chicago, IL, USA). From the fitted logistic regression, we determined the distance corresponding to a detection probability of 0·50 as our estimate of detection range (i.e. beyond this distance there is less than a 50% chance that the signal would be detected).

Table 3. Automated detection parameter settings used for call analysis in callViewer18
Minimum link length10
Window length (ms)0·3
Frame rate (fps)10 000
Chunk size (s)1
Minimum energy (dB)14
Echo filter threshold (dB)10
UPPER cut-off freq. (kHz)Inf
LOWER cut-off freq. (kHz)15
Window typeBlackman
Delta size (±frames)1

Recording free-flying bats

Free-flying bats produce complex, frequency-modulated calls that vary in intensity in contrast to the simple, constant frequency signals we used for the synthetic playback experiment. To introduce the variability that is present in natural settings, we recorded free-flying bats. We deployed the detectors for 2 h per night on three separate nights in a suburban area in London, Ontario and Canada. The Avisoft system detected more bat echolocation calls than any of the other detectors so we used the data from it as a baseline. We chose 26 easily identifiable passes (minimum seven consecutive calls), from hoary bats (Lasiurus cinereus) and counted the number of calls in each pass. We manually counted all calls recorded regardless of call quality or completeness. We used callViewer18 to analyse the full-spectrum system calls and AnaLook to analyse calls from AnaBat. We calculated the proportion of calls detected per pass relative to Avisoft, arcsine square root transformed the data and compared detector performance with anova and Tukey's post hoc test in PASW18 after finding no effect of recording night.


Synthetic call playback

Overall, Avisoft detected the most signals (1067 signals, 25% of all signals presented), and AnaBat detected the fewest (240 signals, 5% of all presented). Avisoft was the only system that detected the 115 kHz signal and only at 5 m (Fig. 1a). AnaBat did not detect CF signals at 85 kHz and 115 kHz (Fig. 1e). The other detectors only recorded 85 kHz signals at 5 m, except Avisoft which recorded these signals at 10 m (Fig. 1). All systems detected the 55 kHz signals, but detection range varied from 7 to 16 m at 0° (Fig. 2). Song Meter did not detect 115 kHz signals because the frequency is outside of this model's detection capabilities. A model with a higher sampling frequency is available and would likely have detected higher-frequency signals.

Figure 1.

Mean number of calls detected by each bat detector system at four frequencies at each distance and angle during the synthetic playback experiment. There were 15 calls played for each frequency/distance/angle combination.

Figure 2.

Distance of 50% probability of detection calculated with a logistic regression for each frequency at 0° by each bat detector system during the synthetic playback experiment. Patterns were similar for all detectors at 45° and 90°, but with lower overall probability of detection.

The number of signals detected at 25 kHz varied significantly among detectors (F4,348 = 21·32, P < 0·001; Fig. 3), except Batcorder and Song Meter. AnaBat recorded the fewest 25 kHz CF signals. There were also differences among detectors in the number of 55 kHz signals detected (F4,346 = 22·74, P < 0·001; Fig. 3); Avisoft recorded more than Song Meter and AnaBat, while Batcorder recorded more than AnaBat. Batlogger recorded significantly more signals than any other detector for at 25 and 55 kHz.

Figure 3.

Performance varied among detectors with a strong effect of frequency. Call detection (arcsine square root transformed number of calls) ± SE by call frequency evaluated at a distance of 22·5 m. Detectors with the same letter superscript were not significantly different from each other within each frequency.

There was a significant interaction between detector and distance for both 25 and 55 kHz signals (F4,348 = 9·42, P < 0·001; F4,346 = 13·63, P < 0·001; Fig. 1). For 25 kHz, Batcorder and Song Meter detections reflected a greater rate of attenuation with distance than AnaBat, Avisoft and Batlogger. For 55 kHz, AnaBat had the greatest rate of attenuation with distance and Batlogger had the lowest (Fig. 1).

Overall, there was an effect of angle for both 25 and 55 kHz signals (F2,348 = 24·92, P < 0·001; F2,346 = 21·06, P < 0·001; Fig. 1); the number of signals detected declined as the angle increased. The effect of angle was the same among all detectors (P > 0·05). There was no interaction between angle and distance for 25 kHz signals (P > 0·05), but there was an interaction for 55 kHz signals (F2,346 = 12·62, P < 0·001). For 55 kHz signals, there was no difference between 0° and 45°, but these two angles had a greater rate of decline in number of signals over distance than 90°.

Recording free-flying bats

Batlogger recorded significantly more hoary bat echolocation calls (relative to Avisoft) than any other system (F3, 100 = 45·26, P < 0·001; Fig. 4), while AnaBat, Batcorder and Song Meter did not differ significantly from each other. Only AnaBat and Batcorder failed to detect all 26 passes; both of these systems did not record any calls from two passes. One of the 26 passes included a feeding buzz that was recorded by all of the detectors. Avisoft, Batcorder, Batlogger and Song Meter recorded more calls (23–25 calls) in the feeding buzz than AnaBat (11 calls).

Figure 4.

Mean number of calls ± SE per pass relative to Avisoft for each bat detector from recordings of free-flying Lasiurus cinereus on three nights. Batlogger detected more calls than any of the other systems (detectors with the same letter superscript were not significantly different from each other).


Our results demonstrate that there is significant variation in detection efficacy among commercially available bat detectors. The differences in the detection abilities of these microphones, particularly in relation to differing frequency sensitivity, illustrate the hazards of comparing data collected by different detecting systems. Our results show that detection of different frequencies varied among detector systems and was affected by the distance and angle of the signal from the detector. Avisoft and Batlogger detected more of the highest frequency signals we tested than the other detectors, but as expected, these signals were detected at much shorter ranges. Detection distance for the 55 kHz synthetic signals (detected by all systems) is particularly relevant because this frequency is in the range of most species of bats that occur in temperate regions. In Hawaii, where only one species of bat occurs (L. cinereus semotus), any of the systems we used would suffice, although each would provide quite a different view of bat activity. In Newfoundland, where two species occur (Myotis lucifugus, Myotis septentrionalis) any of the systems we tested would suffice for M. lucifugus (echolocation call frequency of most energy ~40 kHz, maximum frequency ~81 kHz), but only some would accurately document activity by M. septentrionalis, which uses calls dominated by higher frequencies (frequency of most energy ~60 kHz, maximum frequency ~126 kHz; Faure, Fullard & Dawson 1993; Ratcliffe & Dawson 2003). In Newfoundland, some systems would be better than others. In other parts of the world, some bat species use echolocation calls dominated by frequencies >85 kHz. For these bat communities, the detection distance of the 85 kHz synthetic signals in our study is important to consider. Monitoring the activity of vespertilionid bats in the subfamilies Kerivoulinae and Murininae would be difficult with any of the systems we tested, because these species produce high frequency (80–200 kHz), frequency-modulated sweeps.

Variation in detection distance among detectors has important practical implications. For many studies, it is particularly important to understand the volume of airspace being sampled, such as when interpreting the results of pre-construction acoustic surveys conducted at potential wind energy facility sites where high bat mortality is a concern (Kunz et al. 2007). On modern wind turbines, the lower edge of the blade swept area is ~20 m above-ground (Barclay, Baerwald & Gruver 2007). Our data demonstrate detection ranges of 7–16 m, and therefore, none of the ground-based microphone systems we tested can detect bats flying in the area swept by the blades of wind turbines. Even a detector placed on the nacelle of a turbine (in the centre of the blade swept area) would sample no more than one-third of the area swept by 50-m-long blades (Kunz et al. 2007).

When we focus on detection of echolocation calls from free-flying bats, bat detectors fell into one of two performance groups. AnaBat, Batcorder and Song Meter did not differ significantly in the number of hoary bat echolocation calls detected. These bats produce high-intensity echolocation calls with a minimum frequency which is typically ~17 kHz (Obrist 1995). The minimum frequency of L. cinereus calls is lower than the lowest frequency of our synthetic calls. Consequently, our free-flying bat results represent a best-case scenario; we used only high-intensity, low-frequency calls and our sampling method, counting all calls regardless of quality, presented the most optimistic view of activity. In reality, many species are much less detectable and the quality of many recorded calls is too poor to be identified to species or counted as a bat call. Using automated detection algorithms with recording quality standards will provide more objective call counts when measuring activity. If we had looked at passes from any of the Ontario Myotis species (calls with a minimum frequency range of ~34–40 kHz; Thomas, Bell & Fenton 1987), it is likely that the results from our free-flying passes would have mirrored the results from our synthetic call trials.

Among the detectors we tested, AnaBat is unique in that it is the only detector to use zero-crossing analysis which may (Corben & Fellers 2001) or may not (Fenton 2000) provide an adequate picture of bat activity. Our data contribute to this discussion, demonstrating that AnaBat is capable of performing similarly to a full-spectrum detector (Fig. 4), but in most cases it detects fewer calls (Fig. 3). Therefore, we emphasize the importance of considering the research questions and local bat fauna. While our results from the synthetic call trials agree that full-spectrum detectors are more sensitive, our free-flying bat trial showed that there are circumstances where the differences are not substantial. Ultimately, the specific hypotheses and objectives of a study will dictate the suitability of various detectors (Limpens & McCracken 2004). No one recording system is ideal for all situations, and thus, it is the responsibility of the researcher (and the reader) to consider how the performance of the recording system will impact the results and conclusions of the study.

It is important to note that regardless of recording system, all microphones detect only a subset of the calls present in the environment (e.g. in our playback experiment the best system detected only 25% of the calls we played). However, our findings show that some subsets are significantly larger than others. This discrepancy is essential to remember when attempting to compare data sets collected with different detecting systems. Even when comparing multiple detectors of the same model, the microphones must be calibrated to ensure comparable performance (Larson & Hayes 2000). With an increasing number of threats to bat populations (e.g. wind turbines and white-nose syndrome), there may be a drive to develop more rigorous monitoring programmes with standardized protocols for bat surveys. Our results highlight the importance of considering the specific detector used, and the variation that may arise from different microphones.

As technology continues to evolve, the number of commercially available detectors will increase. As with the current proliferation in detectors on the market, many brands will persist (e.g. AnaBat and Avisoft) and new brands will emerge (e.g. Batlogger). In such a specialized market, there will probably be few dramatic changes in the technology; we would expect to see increases in microphone sensitivity, battery life and storage capacity, along with continued software upgrades to improve detection algorithms. With a high diversity of detectors, each with a wide range of settings and technical capabilities, it is now necessary to report not only the type of detector used, but also the settings chosen (e.g. Table 2) and as many hardware details as possible. The extent that detector-specific settings have on performance and accuracy between detectors of the same brand remains to be seen. Finally, it comes to the issue of comparability of results; different detectors will give different results, which must be taken into account.

Whether the bat-detecting system you are using hears the same signals as the one I am using depends upon the echolocation calls. There are numerous factors that contribute to variation in data sets from acoustic monitoring; our results demonstrate that the detector plays a role in this variation. Ultimately, it is crucial that differences in detector performance be considered when designing studies and comparing results from different detectors, whether among models included in our study, other extant models, or those yet to be invented. No detector is ideal for all research questions and methods, and conversely, not all detectors are appropriate for a given question or methodology.


We thank L. P. McGuire, E. E. Fraser, J. F. Miller and A. Cameron for their support and input for this study. Thanks to L. Lazure and M. D. Skowronski for creating the synthetic calls. D. Morningstar, M. Gumprich, M. K. Obrist, R. M. R. Barclay and E. F. Baerwald generously allowed us to use their equipment.