High duty cycle echolocation and prey detection by bats

The timing of production of echolocation calls separates bats using laryngeal echolocation into two categories: those signalling at high duty cycles (HDCs) and those signalling at low duty cycles (LDCs) (Fenton, 1999). The duty cycle of a periodic sound is defined as the proportion of time spent emitting signals in a given period of time. LDC echolocators separate pulse and echo in time and, to avoid self-deafening, do not simultaneously broadcast and receive. The search phase calls of LDC echolocators are typically short, separated by relatively long periods of silence, and consist of frequency-modulated sweeps of varying bandwidth. In contrast, bats using HDC echolocation avoid self-deafening by separating pulse and echo in frequency. The calls of HDC echolocators are long, separated by short periods of silence, and are typically dominated by a constant frequency component and begin and/or end with a short-frequency modulated sweep.

Physiologically, HDC echolocators have an acoustic fovea because their auditory systems are tuned to very narrow ranges of frequencies (Schuller and Pollak, 1979; Schnitzler et al., 1983; Schnitzler, 1987; Rübsamen et al., 1989; Kober and Schnitzler, 1990; Neuweiler, 1990). This narrow-band sensitivity enables these bats to readily detect moving prey as spectral variation around the carrier frequency. Flutter detection allows HDC bats to distinguish moving (usually referred to as fluttering because of the movement of prey wings) targets from stationary objects in the background (von der Emde and Schnitzler, 1986; Schnitzler, 1987; Neuweiler, 1990; Roverud et al., 1991). Furthermore, HDC bats ignore stationary or very slow fluttering prey (Goldman and Henson, 1977; Schnitzler and Henson, 1980; Bell and Fenton, 1984; Sum and Menne, 1988). Some LDC bats can detect fluttering insect-like targets in the laboratory (Kober and Schnitzler, 1990; Roverud et al., 1991; Moss and Zagaeski, 1994; Grossetête and Moss, 1998) but their ability to do this under natural conditions has not been tested. In captivity, the performance of LDC bats in detecting flutter is usually inferior (Roverud et al., 1991) or, at best, comparable to that of HDC bats (Sum and Menne, 1988).

LDC echolocation is considered to be ancestral in bats whereas HDC echolocation is thought to be derived (Fenton et al., 1995; Simmons and Geisler, 1998; Schnitzler et al., 2004; Eick et al., 2005; Jones and Teeling, 2006). This interpretation implies that HDC echolocation offers an advantage over LDC echolocation, resulting in its having been selected for. Improved flutter detection by HDC bats relative to syntopic LDC bats might provide part of the advantage. However, the role of competition in bat assemblages remains unclear. Some authors suggest that competition might structure bat communities and influence patterns of echolocation (Schoeman and Jacobs, 2008), but several disagree (Arita, 1997; Stevens and Willig, 1999; Jiang et al., 2008; Meyer and Kalko, 2008).

Compared with the echolocation calls of LDC bats, HDC bats often use echolocation calls that are higher in frequency and dominated by a single frequency (Jones, 1999; Jacobs et al., 2007). Emphasis on a single frequency could make bats' echolocation calls more conspicuous to insects with bat-detecting ears (Yager, 1999), so hearing-based defences of prey could influence foraging success of echolocating bats (Novick, 1977; Fenton and Fullard, 1979; Jacobs et al., 2008; Surlykke and Kalko, 2008). The allotonic frequency hypothesis (Novick, 1977; Fenton and Fullard, 1979; Schoeman and Jacobs, 2003) proposes that relatively low and high calling frequencies in bats result from strong selection pressure to become less audible to tympanate prey. The allotonic frequency hypothesis is commonly invoked to explain the tendency for HDC bats using high-frequency echolocation calls to eat more moths than LDC bats (Jones, 1992). Maximum sensitivity of insect ears is typically 20-60 kHz (Fullard, 1987; Yager, 1999), but many African moths are sensitive to frequencies up to 110 kHz (Jacobs et al., 2008). This raises the question of why there is a prevalence of moths in HDC bats' diet, considering that many moths can still hear them.

We set out to test the hypothesis that HDC echolocation confers advantages over LDC echolocation because of improved detection of fluttering targets. We predicted that HDC echolocation would generate more detectable echoes from fluttering targets than LDC echolocation. We also predicted that bats using HDC echolocation would be better able to detect fluttering prey than syntopic bats using LDC echolocation. Finally, we assessed the hypothesis that prey detection is a primary selective force influencing the echolocation features of bats. To test our hypotheses, we used two complementary experimental approaches. First, we created synthetic echolocation calls representing a wide selection of both HDC and LDC signals. We used these calls to ensonify a fluttering target and measured the influence of call variables on echo characteristics. Second, we exposed artificial fluttering prey to free-living syntopic HDC and LDC bats to assess bats' responses to these targets. We used sound and video recordings to monitor the bats' behaviour. From the field experiments we also assessed the relationship between natural call parameters and assumptions about the detection ability of various species.

We digitally synthesised calls using MATLAB version 7.5 (The MathWorks Inc., Natick, MA, USA) and saved them as sound files (.wav format). We created a variety of calls (N=278) with a wide range of call designs, based on the echolocation calls of living bats (Table 1). Some synthetic calls resembled typical LDC calls (short pulse duration, large bandwidth), others HDC calls (long pulses, constant and high frequency). The rest were a combination of intermediate features. All calls had a constant sweep rate.

We constructed a custom fluttering target by attaching a piece of masking tape to a thin metal rod (1 mm diameter, 40 cm length) connected to a battery-powered 12 V DC motor (NexxTech 2730255, Orbyx Electronics, Concord, Ontario, Canada). We constructed a ‘large’ (1.2×2.2 cm wing area) target rotating at 70 Hz and a ‘small’ target (0.7×1.7 cm wing area) rotating at 95 Hz because smaller insects usually have faster wing-beat frequencies (Kober and Schnitzler, 1990). The rotation of the ‘wing’ produces an acoustic echo similar to that of a fluttering insect (Bell and Fenton, 1984; von der Emde and Schnitzler, 1986; Sum and Menne, 1988; Grossetête and Moss, 1998).

We played back simulated echolocation calls with a ScanSpeak ultrasonic loudspeaker (frequency range of 1-120 kHz; Avisoft Bioacoustics, Berlin, Germany) aimed at the fluttering target. Calls were digitised at a sampling rate of 250 kHz and a resolution of 16 bits with an Avisoft UltraSoundGate 116 and Recorder (version 2.9) software (Avisoft Bioacoustics), and stored as .wav files. Analog signals were generated by a digital-to-analog converter board (DAQCard 6062E, National Instruments, Austin, TX, USA) and amplified using an ultrasonic amplifier (Avisoft Bioacoustics) before delivery to the loudspeaker.

To detect echoes from the target, we mounted a condenser microphone (CM16, Avisoft Bioacoustics) 0.1 m from the speaker. To attenuate direct transmission of the sound, we put a plastic sheet and a piece of acoustic foam 0.02 m thick between the microphone and the speaker, which were 0.4 m from the target. Recording took place in a 45 m² room with a 2.9 m high ceiling.

We conducted most of our field-based experiments in June and July 2008 at two sites in Yangminshan National Park, Taipei County, Taiwan (25°09′N, 121°31′E), and three sites in Hsin Chu County, Taiwan (24°36′N, 121°07′E). All sites were in forested areas where bats foraged, based on previous observations by researchers from the Institute of Ecology and Evolutionary Biology at the National Taiwan University (Taipei, Taiwan). Yangminshan Park has six species of LDC bats and three species of HDC bats, all insectivorous (Lee and Huang, 2007). We also performed experiments from 23 to 27 April 2009 at Lamanai, Orange Walk County, Belize (17°45.848′N, 88°39.128′W), where over 40 species of bats have been reported, 21 of them insectivorous bats (20 LDC species; one HDC species, Pteronotus parnellii) (Fenton et al., 2001). We set up the apparatus close to flight paths used by bats foraging in the forest understory. We changed sites every night, though all sites were less than a 30 min walk apart.

We presented the same artificial fluttering targets in the field that we had used to generate echoes in the laboratory. In the field, we mounted the targets so they could flutter and move (Fig. 1). The prey moved back and forth (along the plane of rail) at ∼0.7 m s^-1. We remotely controlled the motors responsible for fluttering rate and movement from a distance of 4 m.

We varied prey wing size and wingbeat frequency, as well as the linear motion of the prey. We tested three wing sizes (0.58, 0.72 and 2.89 cm²), four wingbeat rates [0 (control), 20, 50 and 80 Hz] and whether the prey moved laterally (moving) or not (stationary control), resulting in a total of 24 possible combinations. This variation should reflect a diversity of prey types ranging from a small fast-flying insect like a fly to a large moth with slower wing beats.

We monitored approach and attack behaviour by bats during a 4 h period beginning at dusk. Bats were presented with various combinations of wingbeat flutter rate, prey size and motion. During 2 min recording trials we changed the wingbeat speed or motion every 20 s. Between each recording trial, we changed the target size. The order of presentation of different combinations of target characteristics was randomly determined prior to the experiment. We recorded bat calls with an ultrasound microphone (CM16, Avisoft Bioacoustics) connected to a laptop computer running Recorder software version 2.9 (Avisoft Bioacoustics). Sounds were digitised at a sampling rate of 250 kHz and a resolution of 8 bits, and stored as .wav files. We filmed bats as they approached the target using a video camera equipped with ‘night vision’ (DCR-SR46, Sony, Tokyo, Japan) and an infrared light source (IRLamp6, Wildlife Engineering, Tucson, AZ, USA). We recorded at 30 frames sec^-1. The field of view captured by the camera measured ∼160×120 cm, with the target at the centre of the screen. The ultrasonic microphone and video camera were synchronised while running side-by-side 4 m from the target and 1.5 m above the ground.

We counted the number of bat passes in our acoustic recordings. We defined a bat pass as the sequence of echolocation calls produced by a bat as it passed through the airspace sampled by the recording microphone (Fenton et al., 1998). We determined the following call variables using BatSound Pro version 3.31b (Petterson Electronik AB, Uppsala, Sweden): call duration (ms), peak spectral frequency (kHz), bandwidth (kHz), sweep rate (bandwidth/pulse duration; kHz ms^-1) and duty cycle [pulse duration/(pulse duration + interpulse interval); %]. We assumed that detection of prey would result in an observable change in both the bat's echolocation and flight behaviour. We assigned each acoustic pass to either: (1) approach calls coupled with a bat approaching the fluttering target on the video or (2) no approach when the bat did not alter its flight path and calling pattern. We used the video recordings to validate the categorisation of passes. We counted approaches to the prey when the bat changed its trajectory to face and then fly towards the target.

We conducted a principal component analysis (PCA) on the data collected in the synthetic call experiment. This reduced the dimensionality of the synthetic call data and the potential for interaction effects associated with highly intercorrelated call variables. We then plotted the PC values against recorded target strength. We used regression analysis, and examined different relationships (linear, quadratic, inverse, logistic and logarithmic) to determine which best fit the data to identify significant relationships. Prior to these analyses, we transformed the data by adding 10 units to the PCA results, shifting the entire distribution by a positive constant, to accommodate functions that cannot handle negative values. We compared the coefficient of determination (R² values) to decide which regression model was the best fit. We compared the echo target strengths of the small fast-fluttering targets and large slow-fluttering targets using a two-sample t-test.

For data from the field experiment, we used Pearson's chi-square test to determine whether approach toward the fluttering target was paired with echolocation type, location or bat species. We used Fisher's exact test when it was more appropriate (2×2 contingency table and small samples). We compared HDC bats with LDC bats, LDC bats from Taiwan with LDC bats from Belize, and further compared taxa within LDC and HDC bats. To assess whether the number of approaches was influenced by prey characteristics (wing size, wingbeat frequency and prey movement), we performed Pearson's chi-square tests. We performed these analyses separately for LDC and HDC bats. To examine call variables separately, we conducted binary logistic regressions considering a binary response variable (‘approach’ and ‘no approach’) for each call variable. We only used data from Taiwan for logistic regression analysis because of the large sample size and we wanted to avoid bias stemming from differences in sampling effort and geography.

The strength of echoes from larger and more slowly fluttering targets was stronger than that from smaller and faster fluttering targets (t=4.47, d.f.=136, P<0.001).

Following a PCA on call variables, we chose the first three PCs that accounted for >75% of the variation, defining biologically relevant groups of variables (see supplementary material Tables S1 and S2). The relationship between spectral variables (PC1) and target strength was best described by a quadratic equation (F_2,273=105.30, P<0.001, R²=0.44; Fig. 2A). Weaker target echoes were significantly associated with increasing call frequency, but only the decreasing portion of the curve fitted within a realistic range of signalling frequencies. Bandwidth, sweep rate variables (PC2) and target strength were significantly related (F_2,273=40.22, P<0.001; Fig. 2B), with a best-fit quadratic equation (R²=0.23). The weakest target strength occurred in the mid range of the distribution. Finally, temporal acoustic variables (PC3) and target strength were also significantly related (F_2,273=62.66, P<0.001; Fig. 2C). A quadratic function had the best fit (R²=0.32), with all values in the increasing portion of the curve, indicating a stronger target strength with increasing duration and duty cycle.

During 23 nights of sampling in Taiwan, we recorded 2727 passes and echolocation calls from three species of HDC bats (Rhinolophus monoceros, Hipposideros armiger and R. formosae). We were unable to identify all LDC bats to species level because of the similarity of echolocation calls among species and intraspecific variability in call features. Almost all approaches to the target (442 of 446) were by HDC echolocators. The mean approach rate was 18.6% (N=2382 passes) for HDC bats and 1.2% for LDC bats (N=345). The four LDC bats that approached fluttering targets were vespertilionids in the subfamily Murininae and/or Kerivoulinae. These bats have distinctive echolocation calls characterised by short durations (∼1 ms), high sweep rates (21-26 kHz ms^-1) and broad bandwidths (>60 kHz).

In Belize, we recorded five passes of the HDC bat P. parnellii and 370 passes of LDC species over five sampling nights. The relative percentage of HDC individuals over the total number of passes we recorded (1.3%) is less than the percentage of HDC bats caught by Fenton et al. in the same location (3.5%) (Fenton et al., 2001). Among the LDC bats recorded, 134 were members of the genus Pteronotus (Pteronotus personatus and P. davyi). Three of the LDC Pteronotus spp. passes included an approach to the fluttering target, yielding an approach rate of 2.2%. We saw no approaches from other LDC species (vespertilionids, emballonurids and molossids) in Belize.

The likelihood of approach to prey was related to echolocation call type (χ²=135.5, d.f.=1, P<0.001; Figs 3, 4). Hipposiderids and rhinolophids differed significantly in their approach rate (χ²=302.6, d.f.=1, P<0.001), with rhinolophids approaching more often than hipposiderids (30.3 versus 2.2%, respectively). The approach rate for the LDC bats in Taiwan did not differ significantly from that for LDC bats in Belize (Fisher's exact test, P=0.72). We further examined the data from Belize, comparing LDC Pteronotus spp. with all other LDC bats, and found no relationship between species and approach rate (Fisher's exact test, P=0.08). LDC bats exhibited a low approach rate to fluttering targets that was not influenced by wingbeat frequency (χ²=1.16, d.f.=2, P=0.56; Fig. 3A), wing size (χ²=2.12, d.f.=2, P=0.35) or the presence of lateral movement (Fisher's exact test, P=0.46; Fig. 4). However, the approach rate of HDC bats was significantly influenced by wingbeat frequency (χ²=20.24, d.f.=2, P<0.001; Fig. 3) and prey size (χ²=52.77, d.f.=2, P<0.001; Fig. 3B) but not by lateral prey movements (Fisher's exact test, P=0.18).

We performed a multiple binary regression with call characteristics as predictor variables and echo detection (yes/no) as the dependant variable. We conducted the regression using pulse duration, peak frequency and bandwidth to avoid multicollinearity. An increase in approach rate was related to an increase in pulse duration (W=25.74, P<0.001; Fig. 4A) and peak frequency (W=130.99, P<0.001; Fig. 4B), but not bandwidth (W=0.61, P=0.44; Fig. 4C). The midpoint of the logistic curve for peak frequency was 120 kHz, corresponding to an approach rate of 0.5.

Our data demonstrate that, in forest understory habitats, where hunting echoloctaing bats receive multiple echoes from objects other than insect prey, HDC bats approached fluttering targets significantly more often than LDC bats. Our data confirm that HDC echolocation is better than LDC echolocation for detecting fluttering prey in cluttered habitat (von der Emde and Schnitzler, 1986; Schnitzler, 1987; Neuweiler, 1990; Roverud et al., 1991). Our results also indicate that rhinolophids may be more efficient at detecting fluttering targets than hipposiderids. In previous experiments, rhinolophid bats showed higher sensitivity to echoes containing flutter information compared with hipposiderids, indicating that they were more effective at detecting flutter (von der Emde and Schnitzler, 1986; Roverud et al., 1991).

The prevalence of moths in the diet of rhinolophid bats cannot entirely be explained by the allotonic frequency hypothesis (Jacobs et al., 2008; Fullard et al., 2008). The diets of rhinolophids and hipposiderids may reflect their superior performance in flutter detection than their LDC counterparts. Our data suggest that, among HDC echolocators, rhinolophids are more effective flutter detectors than hipposiderids, a prediction that is testable in the field.

Size of target and rate of flutter influenced approach rate by HDC bats, but our data set did not allow us to determine whether this reflects differences in detection ability and/or prey preference. For both wingbeat frequency and prey size, targets in the mid range were most often approached. Our experiments using synthetic calls showed that larger prey reflected stronger echoes than smaller prey, supporting Waters et al. (Waters et al., 1995), who used LDC calls and dried immobile insects. The most common HDC bats recorded in Taiwan were the small R. monoceros (36-40 mm forearm length) and increased approaches to medium-sized prey could reflect a compromise between bats' ability to detect and then catch and handle prey of different sizes.

Bats in the subfamilies Murininae and Kerivoulinae occasionally approached the fluttering targets in the forest understory. This response may be a function of their distinct echolocation calls and behaviour (Kingston et al., 1999). Their range of target detection is likely less than that of HDC bats because of the low-intensity and high-frequency echolocation calls (Murininae calls sweep from 180 to 43 kHz and Kerivoulinae calls sweep from 152 to 86 kHz) (Griffin, 1971; Kingston et al., 1999). In Belize, some LDC bats use a very different strategy. The two LDC Pteronotus spp. we recorded, and other species in this genus, emit relatively long calls for LDC bats, with some narrowband elements, and P. personatus accomplishes partial Doppler shift compensation (Smotherman and Guillén-Servent, 2008). However, their reaction to the fluttering prey did not differ from that of other LDC bats in the present study.

In the laboratory, measures of echo strength indicated that the best bandwidths and sweep rates for detecting fluttering targets occurred at both ends of the distribution (either narrowband or very broadband), but we found no clear relationship in field experiments. Bandwidth, sweep rate and even pattern of frequency change over time were not directly related to detection ability. Previous experiments showed that bandwidth and sweep rate were related to angular localization, reduction of ranging error and feature extraction (Boonman et al., 2003; Schnitzler et al., 2003; Holderied et al., 2006; Boonman and Ostwald, 2007). Therefore, bandwidth is probably not a call characteristic shaped by selection for improved detection of prey. Reducing signal bandwidth is a way to achieve increased call intensity when total power is constant. If signal power is conserved, every 10× reduction in call bandwidth would increase power spectral density at the call resonant frequency by a factor of 10, or 10 dB (Oppenheim and Schafer, 1989). The obvious advantage of a narrowband echolocation signal is an increase in call energy with no overall increase of energy expenditure in signal production.

Field and laboratory data showed that longer pulse durations and a higher duty cycle translated into higher flutter detection performance because temporal variables are important to aerial-hawking bats in forested habitat. At higher duty cycles, it may be easier to detect, track and lock onto a fluttering target because the silent period between calls is dramatically reduced. Enhanced tracking of flying targets by HDC bats is a function of their approach to avoiding self-deafening (separating pulse and echo in frequency). Additionally, longer signals are better for encoding target movements because of relatively uninterrupted modulations in echo amplitude and frequency (Schnitzler et al., 2003). Although increased pulse length can be advantageous for LDC bats, the constraint imposed by pulse-echo overlap limits call duration.

Results from our synthetic call experiment matched field data for temporal but not spectral variables. In the field, the midpoint of approach rate calculated for peak frequency occurred at 120 kHz (Fig. 4B). Species with echolocation calls that have maximum energy at 120 kHz approached fluttering targets more often than those using lower frequencies with maximum energy. Higher frequencies might permit better flutter detection by HDC bats, but suffer increased attenuation (Lawrence and Simmons, 1982). Houston et al. (Houston et al., 2004) showed that lower frequencies (20-30 kHz) yielded weaker target strengths from smaller insects, because of Rayleigh scattering (Pye, 1993). Surlykke and Kalko reported that LDC bats calling at higher frequencies emitted stronger calls to compensate for increased atmospheric attenuation, resulting in similar ranges at which prey were detected (Surlykke and Kalko, 2008). Focusing energy at one frequency in the calls of HDC bats increases the operational range of their calls (Waters and Jones, 1995). Intraspecific variation in the frequencies dominating the echolocation calls of HDC bats may reflect some combination of environmental factors or interspecific competition. Furthermore, the acoustic communication hypothesis states that social interactions among rhinolophids impose selection pressure on peak frequency in these HDC bats (Möhres, 1967; Heller and von Helversen, 1989; Kingston et al., 2000) through intraspecific communication and identification, and avoidance of interspecific competition.

In conclusion, our results support the hypothesis that bats using HDC are more effective at detecting fluttering targets than LDC bats. We found no clear relationship between prey detection ability and frequency-dominating echolocation calls likely because many selective forces act on frequency. Prey detection is unlikely to influence the use of particular bandwidth or sweep rate. However, longer pulses and higher duty cycle improve the bats' ability to detect fluttering prey, and could have been a major selective force for an increased pulse duration and duty cycle in HDC bats.

We thank Mark Skowronski for technical support and comments. We also thank Dr Ling-Ling Lee and students from the Institute of Ecology and Evolutionary Biology at the National Taiwan University (Chun-Chia Huang, Ting-Ting Lin and Hsuan-Yu Wu) for outstanding logistical support, as well as the Lamanai Oupost Lodge in Belize for facilitating our work. John Ratcliffe, Liam McGuire, Paul Faure, Mark Brigham and two anonymous reviewers provided constructive comments of previous versions of the manuscript.