Heaviside's Dolphins (Cephalorhynchus Heavisidii)

Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

Heaviside’s dolphins (Cephalorhynchus rspb.royalsocietypublishing.org heavisidii) relax acoustic crypsis to increase communication range

Morgan J. Martin1, Tess Gridley2, Simon H. Elwen1 and Frants H. Jensen3,4 Research 1Mammal Research Institute, Department of Zoology and Entomology, University of Pretoria, c/o Sea Search Cite this article: Martin MJ, Gridley T, Elwen Research and Conservation NPC, 4 Bath Rd, Cape Town 7945, South Africa 2Centre for Statistics in Ecology, Environment and Conservation, Department of Statistical Sciences, University of SH, Jensen FH. 2018 Heaviside’s dolphins Cape Town, c/o Sea Search Research and Conservation NPC, 4 Bath Rd, Cape Town 7945, South Africa (Cephalorhynchus heavisidii) relax acoustic 3Aarhus Institute of Advanced Studies, Aarhus University, 8000 Aarhus C, Denmark crypsis to increase communication range. 4Biology Department, Woods Hole Oceanographic Institution, 266 Woods Hole Rd, Woods Hole, MA 02543, USA Proc. R. Soc. B 285: 20181178. MJM, 0000-0002-3556-6632; TG, 0000-0001-6167-9270; SHE, 0000-0002-7467-6121; http://dx.doi.org/10.1098/rspb.2018.1178 FHJ, 0000-0001-8776-3606

The costs of predation may exert significant pressure on the mode of communication used by an animal, and many species balance the benefits of Received: 29 May 2018 communication (e.g. mate attraction) against the potential risk of predation. Accepted: 26 June 2018 Four groups of toothed whales have independently evolved narrowband high-frequency (NBHF) echolocation signals. These signals help NBHF species avoid predation through acoustic crypsis by echolocating and communicating at frequencies inaudible to predators such as mammal-eating killer whales. Heaviside’s dolphins (Cephalorhynchus heavisidii) are thought Subject Category: to exclusively produce NBHF echolocation clicks with a centroid frequency Behaviour around 125 kHz and little to no energy below 100 kHz. To test this, we recorded wild Heaviside’s dolphins in a sheltered bay in Namibia. We Subject Areas: demonstrate that Heaviside’s dolphins produce a second type of click with lower frequency and broader bandwidth in a frequency range that is audible behaviour, evolution, ecology to killer whales. These clicks are used in burst-pulses and occasional click series but not foraging buzzes. We evaluate three different hypotheses and Keywords: conclude that the most likely benefit of these clicks is to decrease transmission acoustic crypsis, active space, communication, directivity and increase conspecific communication range. The expected echolocation, Heaviside’s dolphin, narrowband increase in active space depends on background noise but ranges from 2.5 high-frequency clicks (Wenz Sea State 6) to 5 times (Wenz Sea State 1) the active space of NBHF signals. This dual click strategy therefore allows these social dolphins to maintain acoustic crypsis during navigation and foraging, and to selectively relax their crypsis to facilitate communication with conspecifics. Author for correspondence: Morgan J. Martin e-mail: [email protected] 1. Introduction Social animals inevitably need to balance effective communication with conspecifics against the costs associated with communication, including eavesdropping and potential detection by predators and prey [1]. Trade-offs to decrease predator detection often involve shifting communication to periods or locations with lowered predation risk [2], but such acoustic avoidance can be costly if the social or ecological functions of communication are not fulfilled [3]. Alternatively, animals may use quiet, low-amplitude or high-frequency signals with short detection ranges for social interactions [4], which can be difficult for predators to locate [5]. In the aquatic environment, where light diminishes quickly, cetaceans (whales, dolphins and porpoises) rely on sound as the primary medium for Electronic supplementary material is available orientation, foraging and communication [6]. In water, sound travels faster online at https://dx.doi.org/10.6084/m9. and attenuates less than in air [7], increasing the necessity of balancing communication with the associated risk of distant eavesdroppers. Mammal-eating figshare.c.4154456. killer whales (Orcinus orca) have been shown to fall silent as they hunt so as

not to alert their acoustically sensitive prey [8]. Antipredator however, group sizes tend to be slightly larger, with more 2 strategies that decrease the risk of passive detection by preda- socializing activity than described for other NBHF species rspb.royalsocietypublishing.org tors have potentially large benefits because echolocation used [32]. Heaviside’s dolphins have only been reported to pro- by all toothed whales puts them at heightened risk of detec- duce NBHF clicks with little to no energy below 100 kHz, tion by eavesdroppers [9]. For example, Blainville’s beaked like other NBHF species [33]. Here we present evidence whales (Mesoplodon densirostris) only produce sound at that Heaviside’s dolphins produce lower-frequency broad- depth and remain silent within several hundred metres of band signals, despite residing in an area with killer whale the surface, and this has been proposed to represent a strat- predation risk. We show that burst-pulses are generally com- egy to reduce risk of detection by killer whales, which tend posed of these lower-frequency broadband signals, and thus not to dive deeper than a few tens of metres [10]. Addition- present evidence of a NBHF species with a dual click type ally, delphinids [11] and seals [12] seem to suppress vocal strategy. We discuss three possible theories to explain how activity in the presence of killer whales. the production of these lower-frequency broadband signals B Soc. R. Proc. Toothed whales are grouped into four acoustic categories may help this species compensate for the socio-ecological by the type of biosonar pulses they emit [13,14]. While most trade-offs imposed by communicating with NBHF signals. delphinids produce broadband, extremely short biosonar We use an acoustic model to show that a major advantage clicks, 13 species from four separate clades (Kogiidae, Phocoenidae, of communicating using lower-frequency clicks is that trans- 285 Pontoporiidae and 6 delphinid species from the genera Cepha- mission directivity is lower and active space is larger over a lorhynchus and Lagenorhynchus) have evolved a narrowband, wide range of noise levels, thus facilitating social interactions 20181178 : high-frequency (NBHF) click type [15,16] with energy almost over a greater area. exclusively above 100 kHz [17]. These four independent cases of convergent evolution have spurred several hypotheses regarding the evolution of NBHF signals [16]. Some 2. Material and methods authors have argued that NBHF signals exploit a natural low noise window occurring at frequencies above 100 kHz Twenty-five hours of acoustic recordings of Heaviside’s dolphins were collected in Shearwater Bay, Namibia (2268 370 S, 158 050 E), to favour detection in an otherwise noisy environment [18]. over 12 days during April and May 2016. Recordings were Other authors propose that the evolution of NBHF signals made by deploying two hydrophones (SoundTrap 300 HF; and the concurrent loss of producing lower-frequency whis- Ocean Instruments, New Zealand) mounted 1 m apart and tles is evidence for an ‘acoustic crypsis’ strategy [15,19] suspended 1.5 m below an ocean kayak. Only data from a where NBHF species have shifted their acoustic signals to single hydrophone were analysed for this study. Sound was frequencies above the hearing limit of killer whales which digitized at a sampling rate of 576 kHz with a 16-bit resolution cuts off around 100 kHz [20]. The ‘acoustic crypsis’ hypo- (sensitivity: 2171 dB re 1V mPa21, flat frequency response: thesis has become a commonly accepted explanation for 400 Hz–150 kHz + 3 dB). Behaviour and group size information the evolution of NBHF signals [14,21,22]. were collected concurrently with sound recordings (see electronic This cryptic biosonar strategy has had consequences for supplementary material). A land-based observer team stationed communication and social behaviour in NBHF species. at a vantage point (20 m elevation) monitored the presence of cetaceans within the bay. Many broadband delphinid species produce a wide variety of communication signals [23,24] including low-frequency calls and whistles that can travel several kilometres under- (a) Acoustic data extraction water [25,26] and are easily distinguished from foraging Recordings made within a visually estimated 50 m range of dol- sounds. By contrast, NBHF species seem to have lost the abil- phins were selected for analysis. Acoustic signals produced by ity to whistle [15] and communication is therefore limited to Heaviside’s dolphins were identified through visual inspection clicks. Both harbour porpoises (P. phocoena) [21,27] and Hec- of a spectrogram display in Adobe AUDITION CC (Adobe Systems tor’sdolphins(C. hectori) [28] are NBHF species which have Inc.). Heaviside’s dolphin NBHF echolocation clicks have been been shown to communicate acoustically with short, iso- previously described [33], and only a subset were selected for lated burst-pulses during social and aggressive encounters. analysis. We defined three functional groups of signals based on signal context and interclick intervals (ICI, calculated as the However, there are socio-ecological drawbacks for species time between subsequent clicks [9]). Click trains were defined constrained to producing NBHF signals for both echolocation as series of clicks with ICI exceeding 10 ms. Such click trains and communication. First, the signal repertoire and thus com- are likely to be echolocation signals produced by the animals. munication complexity [29] are limited, potentially reducing A subset of click trains were composed of lower-frequency, options for resolving and differentiating social interactions broader-bandwidth signals than previously described [33], and with sound. Second, communicating with signals that are we therefore divided click trains into NBHF click trains and also used for echolocation and foraging may increase signal broadband click trains by inspecting spectrograms (figure 1). ambiguity for a receiver [30] which then needs to differentiate Foraging buzzes are used during prey capture by echolocating communication from foraging signals. Finally, as NBHF clicks animals [34,35], including NBHF species [36]. These were defined are highly directional and attenuate rapidly with distance due as click series with ICIs less than 10 ms, which were preceded by to high-frequency-dependent absorption [22], the detection a slower click train. Since buzzes occurred at the end of a click train, we defined the start of a buzz as the point when the ICI range for nearby conspecifics is typically short (less than first decreased below 10 ms and the end of the buzz as the 1 km) and dependent on the relative orientation of the source point where the click train ended or where the ICI increased to and the receiver [21]. greater than 10 ms. Finally, we defined burst-pulse signals as Heaviside’s dolphins (Cephalorhynchus heavisidii) are small discrete, isolated series of high repetition rate clicks that began, (less than 1.7 m) delphinids endemic to the west coast of persisted and generally ended with interclick intervals less than southern Africa. They are typically found in shallow coastal 10 ms following Lammers et al. [37]. Burst-pulses are commonly waters to approximately 100 m depth [31] in small groups; considered to have an intra-specific communicative function Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

(a)((i) NBHF click train b) (i) foraging buzz 3 60 40 rspb.royalsocietypublishing.org

40 20 ICI (ms) ICI (ms) 20 0 (ii) (ii) 250 250

200 200

150 150

100 100

frequency (kHz) 50 frequency (kHz) 50 rc .Sc B Soc. R. Proc.

0 0.2 0.4 0.6 0.8 1.0 1.2 0 0.2 0.4 0.6 0.8 1.0 1.2 time (s) time (s)

(iii) (iv) 0 (iii) (iv) 0 1.0 1.0 285 0.5 –10 0.5 –10 0 –20 0 –20 20181178 :

amplitude –0.5 amplitude –0.5 power (dB) power –30 (dB) power –30 –1.0 –1.0 –40 –40 –100 0 100 200 0 100 200 300 –100 0 100 200 0 100 200 300 time (ms) frequency (kHz) time (ms) frequency (kHz)

(c)((i) BB click train d) (i) burst-pulse call 150 1.6

100 1.4 ICI (ms) ICI (ms) 50 1.2 (ii) (ii) 250 250

200 200

150 150

100 100

frequency (kHz) 50 frequency (kHz) 50

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.80 0.2 0.4 0.6 0.8 1.0 time (s) time (s)

(iii) (iv) 0 (iii) (iv) 0 1.0 1.0 0.5 –10 0.5 –10 0 0 –20 –20 –0.5 amplitude –0.5 amplitude power (dB) power –30 (dB) power –30 –1.0 –1.0 –40 –40 –100 0100 200 0 100 200 300 –100 0 100 200 0 100 200 300 time (ms) frequency (kHz) time (ms) frequency (kHz)

Figure 1. Examples of Heaviside’s dolphin pulsed signal types. (a) Narrowband high-frequency (NBHF) click train, (b) foraging buzz, (c) broadband (BB) click train and (d) burst-pulse call. For each signal, panels represent (i) the interclick intervals throughout the signal, (ii) the spectrogram of the signal (512 pt FFT, Hamming window, 50% overlap), (iii) the normalized waveform (solid line) and envelope (dashed line) of a single click extracted from the pulsed signal shown in panel (ii) and (iv) the normalized power spectrum of the extracted click (512 pt rectangular window, 576 kHz sampling rate).

[24,37,38], including in NBHF species [21]. Only distinguishable, inspected and manually corrected for missed detections. To high-quality pulsed signals measuring more than 10 dB above compare signals with highly variable numbers of clicks, we the background noise measured immediately before the signal finally calculated the 5th, 50th and 95th percentile ICI across were selected for further analysis. each click series. To quantify temporal and spectral differences of component clicks, we extracted the highest amplitude click from each click (b) Acoustic feature extraction series following the methods for on-axis click analysis [39,40]. To quantify temporal differences in repetition rate across signals, While these signals were recorded from an unknown aspect, we used a click detection algorithm developed in MATLAB 2013B the minute difference in the waveform and spectrum of NBHF (MathWorks, USA). We first filtered the input signal with a six- clicks across varying off-axis angles [41] means that spectral par- pole Butterworth bandpass filter (20–275 kHz), calculated the ameters are likely reasonably close to on-axis signals. Individual signal envelope, and extracted peaks in the envelope that were signals were filtered in MATLAB with a four-pole Butterworth separated by more than 0.5 ms. Click detections were visually bandpass filter between 20 and 275 kHz. Individual click Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

Table 1. Biosonar parameters of pulsed signal types produced by Heaviside’s dolphins. The median and 5th–95th percentile values are reported for each 4 parameter. ICI5th, ICIMED and ICI95th represent 5th, median (50th) and 95th percentile interclick intervals, respectively. FP, peak frequency; FC, centroid frequency; rspb.royalsocietypublishing.org BW3dB, 23 dB bandwidth; BW10dB, 210 dB bandwidth; BWRMS, root mean square bandwidth; QRMS,FC/BWRMS; Dur10dB, 210 dB click duration.

NBHF click train BB click train buzz burst-pulse

n 5 33 n 5 28 n 5 40 n 5 58

source parameters median (5–95%) median (5–95%) median (5–95%) median (5–95%)

a B Soc. R. Proc. ICI5th (ms) 23.5 (14.9–41.2) 24.8 (7.8–78.9) 6.0 (2.1–9.9) 1.5 (1.2–1.9) a ICIMED (ms) 28.9 (22.3–55.4) 28.8 (11.7–110.1) 7.2 (3.0–11.2) 1.6 (1.3–2.2) a ICI95th (ms) 46.1 (29.4–104.8) 40.9 (17.4–215.8) 10.0 (5.0–13.0) 1.7 (1.4–3.2) F (kHz)b 127.1 (121.5–136.6) 113.6 (78.4–141.3) 123.8 (115.8–137.3) 112.5 (90.0–133.1) P 285 b

FC (kHz) 131.3 (125.3–136.9) 110.8 (87.2–146.8) 132.4 (124.9–143.3) 119.5 (94.4–149.0) 20181178 : b BW3dB (kHz) 15.8 (9.5–22.7) 21.4 (4.9–79.1) 12.4 (3.3–23.7) 16.3 (3.2–62.2) b BW10dB (kHz) 31.5 (22.7–69.8) 79.9 (38.1–142.5) 37.1 (21.2–86.1) 75.4 (31.0–137.9) b BWRMS (kHz) 12.8 (8.2–23.2) 27.5 (17.1–38.6) 18.3 (10.2–31.6) 26.6 (18.1–38.7) b QRMS 10.2 (5.9–15.6) 4.1 (2.8–7.1) 7.2 (4.3–12.5) 4.4 (3.0–6.8) b Dur10dB (mm) 63.9 (50.6–85.1) 37.0 (16.6–50.7) 71.1 (42.6–129.0) 41.1 (21.1–82.3) aParameters measured across a click series. bParameters measured for an individual click. power spectra were calculated with a 512-point 50% Tukey distribution of on-axis source levels from Heaviside’s dolphins window centred on the peak envelope of each click. Spectral [33]) to examine how varying noise conditions and output and temporal click parameters were calculated according to levels affect the relative change in active space between the two methods for measuring on-axis click parameters [9,42]. signal types. The full model and sensitivity analysis are described in electronic supplementary material. (c) Statistical analysis of signal discrimination Signal parameters, including spectral and temporal click par- 3. Results ameters as well as interclick intervals, were compared across Acoustic data were collected during recording sessions with signal categories using a non-parametric Kruskal–Wallis test Heaviside’s dolphins during which foraging, resting, socializing, and subsequent Dunn’s post-hoc tests for pairwise comparisons in R v. 3.4.2 [43,44]. We then used a random forest classifier [45] interacting with the kayak and travelling behaviours were to measure prediction accuracy as a function of buzz and burst- observed. No other cetacean species were sighted visually pulse signal categories using either ICI parameters (5th, 50th and or detected acoustically during recording sessions. A total 95th ICI percentiles for each click series), spectral and temporal of 90 broadband click trains, 706 buzzes and 954 burst- individual click parameters, or all signal parameters combined pulses and a subset of 33 NBHF click trains were indexed to test the potential benefit of spectral differences in decreasing from recordings made when Heaviside’s dolphins were signal ambiguity. The random forest classifier was built in within 50 m of the kayak. MATLAB 2017b using a ‘bagged trees’ ensemble classifier with Broadband click trains and burst-pulse signals were 30 learners [45]. Prediction accuracy was measured using 5-fold composed of clicks with lower frequency and broader band- cross-validation to prevent overfitting. To measure consistency width (figure 1) compared to typical NBHF signals (table 1). in prediction accuracy, a classifier was trained 100 times and Q-ratios (centroid frequency/RMS bandwidth) are an indi- prediction accuracy measured for each iteration. cator of click type, and generally burst-pulse signals and broadband click trains had Q-ratios less than 5, whereas (d) Acoustical modelling of detection range NBHF click trains and buzz signals had Q-ratios greater than To test the potential benefit for communication, we modelled the 7 (table 1; electronic supplementary material, figure S1A). detection range for typical NBHF clicks and for lower-frequency Initially, buzz and burst-pulse signals were visually differ- clicks extracted from burst-pulses. We first filtered the input signal entiated by the presence or absence of a preceding click train with a six-pole Butterworth bandpass filter (10–150 kHz), and we as burst-pulses occur as isolated signals. The measured used a piston model [46] to estimate changes in transmission signal parameters confirmed there were significant differences beam and empirical measurements of hearing sensitivity of a har- in both ICI parameters and spectral parameters between these bour porpoise [47] to estimate changes in directional hearing. We two signal types (figure 2; see electronic supplementary modelled the detection range (m) for a noise-limited scenario with Wenz Sea State 2 noise levels,and we accounted forchanges in trans- material for a full comparison of different signal types). mission loss due to lower frequency-specific absorption. A separate Based on these findings, a random forest classification algor- sensitivity analysis was conducted across a 25 dB variation in wind- ithm was implemented to evaluate importance of different generated ambient noise (reflecting calm sea conditions to storms) parameters and test if discrimination of communication sig- and a 25 dB variation in signal source levels (reflecting the full nals (burst-pulses) from feeding signals (buzzes) benefits Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

(a) 5 16 rspb.royalsocietypublishing.org 14 foraging buzz burst-pulse 12

10 RMS

Q 8

4 B Soc. R. Proc.

2 80 90 100 110 120 130 140 150 160 centroid frequency (kHz)

(b) 285

1.8 20181178 :

1.6

1.4 RMS 10 1.2 log bandwidth (kHz) 1.0

0.8 0 0.2 0.4 0.6 0.8 1.0 1.2

log10 median ICI (ms)

(c) importance of acoustic parameters for (d) discriminating of foraging buzz from discrimination (ranked) burst-pulse 100

ICIMED

ICI95th 90

ICI5th

FC 80

QRMS

Dur10dB 70

FP classification accuracy (%) 60 BWRMS

BW10dB 50 0 0.02 0.04 0.06 0.08 all ICI spectral feature importance acoustic parameters for classification Figure 2. Signal parameters and discrimination of buzz and burst-pulse signal types. (a) Q-ratio (centroid frequency/RMS bandwidth) as a function of centroid frequency. (b) Log-transformed RMS bandwidth as a function of log-transformed median ICI. (c) Relative feature importance of acoustic signal parameters for classification accuracy. (d) Random forest classification accuracy following three scenarios: discrimination using all signal parameters (All); discrimination using interclick intervals (ICI); or discrimination using spectral and temporal click parameters (spectral). For plots (c,d), values are reported as mean (+s.d.) for 100 independently trained random forest models. Both feeding buzzes and burst-pulse calls can be accurately classified by interclick intervals without including frequency, bandwidth or other individual click parameters. (Online version in colour.) from spectral differences. The random forest classifier demon- accuracy using all available parameters (figure 2d). Classifi- strated that ICI parameters were most important for accurate cation accuracy decreased only marginally (95% prediction classification of buzz and burst-pulse signal categories accuracy) when only interclick interval parameters were (figure 2c). Signal categories could be predicted with 97% included in the model, whereas a larger drop in accuracy Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

(a) change in beamwidth (c) change in active space 6 0 rspb.royalsocietypublishing.org –5 500 80 90 100 110 120 130 450 –10 Fc (kHz)

–15 400

–20 350

SL : –25 300 RMS 174 dB (max)

relative intensityrelative (dB) –30 250 166 dB (+1 s.d.) 161 dB (mean) 156 dB (–1 s.d.) –35 200

potential change in active spacepotential change in active (%) Wenz Wenz Wenz Wenz 149 dB (min) rc .Sc B Soc. R. Proc. –40 150 SS1 SS2 SS4 SS6 0 10203040 50 60 70 80 90 50 55 60 65 70 75 80 m 2 –1 angle (°) NL1kHz¢ dB re 1 Pa Hz

(b) change in detection range

(i)receiver aspect 0° (ii)receiver aspect 45° (iii) receiver aspect 90° 285

(head-on) 60° source angle 60° source angle (broadside) 60° source angle 20181178 : 800 m 800 m 800 m 30° 30° 30° 600 m 600 m 600 m 400 m 400 m 400 m

0° 0° 0°

–30° –30° –30° BP BP BP NBHF –60° NBHF –60° NBHF –60° Figure 3. Switching to lower-frequency burst-pulse signals increases beamwidth and active space. (a) Transmission beam modelled for 10 standard NBHF clicks and 9 burst-pulse clicks of varying frequency using a circular piston model. (b) Detection range modelled for a typical NBHF signal (blue solid line) and a lower-frequency (80 kHz) burst-pulse signal (green dashed line) under Wenz Sea State (SS) two noise conditions and with 161 dB RMS source level, with receivers oriented (i) towards the source, (ii) at a 458 angle to the source or (iii) at a 908 angle to the source. (c) Relative active space for burst-pulse signals compared to the active space of NBHF signals. Note that while detection range will depend on specific model parameters, the qualitative relationship between the detection range of NBHF and burst-pulse signals is consistent under a wide range of noise levels (including Wenz SS1 to Wenz SS6 wind-generated noise, with thermal noise constant) and source levels (covering full range of source levels measured for Heaviside’s dolphins in [33]). was seen when only spectral and temporal click parameters 4. Discussion were included in the model (86% prediction accuracy). The effect of signal type on beamwidth was two-fold: Members of the genus Cephalorhynchus are thought to have first, the sidelobes seen in NBHF signals were suppressed evolved the exclusive use of NBHF biosonar signals to because of the broader bandwidth of burst-pulse signals; become acoustically cryptic, thereby reducing predation risk second, the transmission directivity was lower and conse- by killer whales [15]. This has consequences for the evolution quently sound intensity away from the centre of the sound and function of communication signals within the genus, beam was higher (figure 3a). The detection range for NBHF because acoustic communication is thought to be limited to clicks and burst-pulse signals was modelled for a typical taking place through click series [21,28]. Here, we show 130 kHz NBHF signal and for a burst-pulse signal with a cen- that Heaviside’s dolphins produce a second click type that troid frequency of 80 kHz. While detection range depends on is distinct from normal NBHF clicks by having a lower- the modelled noise levels as well as source and receiver geo- frequency content and broader bandwidth which circumvents metry, the estimated detection range was consistently greater some of the limitations of communicating with NBHF clicks. for burst-pulse signals at all estimated source and receiver Heaviside’s dolphins produce these lower-frequency broad- angle combinations (figure 3b). The potential gain in active band signals occasionally in the form of slow click trains space depended on noise level but was relatively unaffected but predominately in the form of burst-pulses, presumably by large changes in sound source level (figure 3c). At wind- used for communication [21,24,27,28,37,38]. generated noise levels corresponding to Wenz Sea State 1 Communication with burst-pulses is normally achieved (approximately 4–6 knots of wind), the active space of a using clicks that are nearly indistinguishable from echoloca- burst-pulse signal would be around five times greater than tion clicks in delphinids [28,37] and phocoenids [21,48], the active space of a NBHF click (figure 3c). At an estimated apart from low-frequency pulsed signals such as bottlenose wind-generated noise level corresponding to Wenz Sea State dolphin (Tursiops sp.) pops [49] or jaw claps [50]. However, 6 (approximately 28–47 knots of wind), the active space in Heaviside’s dolphins, clicks comprising most burst-pulses would be approximately 2.5 times greater than for a NBHF appear to be a modified and clearly distinguishable version click (figure 3c). (86% classification success based only on spectral differences: Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

figure 2d). Most of the burst-pulses analysed (63%) contained primarily thermal noise, increasing wind-generated ambient 7 energy beginning at approximately 50 kHz, which is an noise decreases the potential gain in active space, but rspb.royalsocietypublishing.org octave lower than signals reported for other NBHF species active space remains higher for burst-pulse signals across [21,22,28,51]. Consequently, most of the recorded broadband the entire range of modelled noise levels from Wenz Sea signals are well within the hearing limit of killer whales State 1 through Wenz Sea State 6 conditions (figure 3c). (upper limit at approximately 100 kHz) [20]. This makes Furthermore, the change in active space may be greater if these signals risky to produce, especially in Namibia where animals simultaneously change transmission aperture killer whales are known to occur and predate on cetaceans through manipulations of air sacs or soft tissue structures, [52], including Heaviside’s dolphins in the study area (J.-P. such as suggested for echolocating delphinids [46] or harbour Roux 2016, personal communication). porpoises emitting foraging buzzes [55]. Thus, the most likely One explanation for the use of lower-frequency broadband reason for Heaviside’s dolphins to use risky, lower-frequency signals could be to reduce signal ambiguity by allowing con- broadband signals is to circumvent the restrictions in com- B Soc. R. Proc. specifics to differentiate communication signals from foraging municating with a short-range, highly directional NBHF buzzes. We addressed this theory by using a cross-validated signal imposed by shifting their biosonar above the hearing random forest classification algorithm with feature vectors range of killer whales. The estimated increase in active containing only ICI parameters, only spectral and temporal space achieved by the lower-frequency broadband signals is 285 click parameters, or containing all parameters combined. still far less than could be achieved by using whistles [26], Both burst-pulses and foraging buzzes were accurately classi- thus this secondary click type represents a compromise 20181178 : fied (95% accuracy) by interclick intervals without including between remaining acoustically cryptic (especially when spectral and temporal click parameters, so these do not foraging) and possessing the ability to communicate over a seem to be necessary for accurate discrimination of burst- greater range when necessary. pulses from foraging buzzes. Rather, it seems likely that It is possible that other NBHF species may take advantage ICIs by themselves may allow animals to identify communi- of selectively increasing their active space. Neonatal phocoe- cation signals and it will be interesting to see if that is the nids have been reported to produce pulsed signals with a case for other NBHF species as well. strong low-frequency (approx. 1–3 kHz) content just after A second, similar explanation for the use of lower- birth and begin to exclusively produce NBHF clicks between frequency broadband signals is to increase signal complexity four [56] and 20 [57] days postnatal. It is not yet understood in the repertoire, thus allowing for encoding a greater variety if this is related to morphological changes or learned call of messages. Repertoire complexity could be augmented behaviour. Regardless, calves’ ability to produce lower- either by producing non-NBHF communication signals at frequency signals with greater active space may be useful repetition rates that are also used for foraging signals, or for mother–offspring cohesion during the first days of life. by composing communication signals with different click Additionally, sporadic broadband clicks and low-frequency types. However, we see only little evidence for either of (4–16 kHz) whistle sounds have been recorded in the pres- these explanations: burst-pulses were composed predomi- ence of mother and calf pairs of Commerson’s dolphins nantly of lower-frequency clicks, with no evidence of (C. commersonii) [58]. Thus, we should not unequivocally dis- burst-pulses composed of different click types, and with rep- miss the possibility of finding lower-frequency communication etition rates consistently higher than for other signal types signals in species that are considered acoustically cryptic such as click trains or foraging buzzes. However, the lower- NBHF species. frequency cut-off did vary between burst-pulses, and it is unclear how much of this is due to off-axis distortion Ethics. This research was conducted by the ‘Namibian Dolphin Project’ [46,53] or could be used to encode information. with permission from the Namibian Ministry of Fisheries and Marine Finally, a third possible explanation for the use of these Resources and with ethics 252 clearance from the University of Pretoria signals is that the lower frequency helps to increase the detec- Animal Use and Care Committee (Reference: ec061-09 AUCC). tion range and thus favours signal detection for nearby Data accessibility. The datasets supporting this article have been conspecifics. High-frequency signals suffer from increased uploaded as part of the electronic supplementary material and in sound absorption as they propagate through water, and Dryad Digital Repository (doi:10.5061/dryad.64048p0) [59]. Authors’ contributions. thus attenuate faster than lower frequencies [7]. By reducing M.J.M., T.G. and S.H.E. conceived and designed the experiments. M.J.M. performed the experiments and collected the predominant frequency, signals will suffer less frequency- data. M.J.M. and F.H.J analysed the data. M.J.M., T.G., S.H.E. and dependent absorption and thus travel farther underwater F.H.J. contributed materials and analysis tools. F.H.J. contributed to [51]. At the same time, both transmission directivity and acoustic modelling. M.J.M., T.G., S.H.E. and F.H.J. wrote the paper receiving directivity will be lower (figure 3a), and thus and approved final submission. energy will be more equally distributed around the vocaliz- Funding. This research was supported by a Fulbright U.S. Research ing animal [47,54]. The modelled detection ranges of NBHF Fellowship, the National Geographic Society’s Emerging Explorers Grant in conjunction with the Waitt Foundation (38115) and the and burst-pulse signals support this hypothesis and show University of Pretoria’s Zoology Department. T.G. was funded by that significant improvements in detection range are poss- the Claude Leon Foundation, and S.H.E. was funded by the South ible by switching to lower-frequency burst-pulse signals, African National Research Foundation. F.H.J. acknowledges funding especially for receivers that are oriented away from or from the Office of Naval Research (N00014-1410410) and an AIAS- located outside the centre of the sound beam (figure 3b). COFUND fellowship from Aarhus Institute of Advanced Studies. Competing interests. The relative change in active space is driven mostly by the We declare we have no competing interests. Acknowledgements. change in sound radiation and partly by a lower sound The authors gratefully acknowledge Heiko Metzger, Dr J.-P. Roux, Jeff Slater, Melissa Nel, Robert Sparg, the leaders of the absorption and thus is relatively independent of the actual SNAK Acoustic Communication Course, Aarhus University, and source level and the absolute detection range of the animal four anonymous reviewers for providing comments and helpful (figure 3c). Since the noise at NBHF signal frequencies is feedback to improve this manuscript. Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

References 8 rspb.royalsocietypublishing.org 1. Bradbury J, Vehrencamp S. 1998 Signal design 16. Morisaka T. 2012 Evolution of communication 30. Wiley, RH. 2013 Communication as a transfer of rules. In Principles of animal communication, sounds in odontocetes: a review. Int. J. Comp. information: measurement, mechanism, and pp. 571–618. Sunderland, MA: Sinauer Psychol. 25, 1–20. meaning. In Animal communication theory: Associates. 17. Au WWL. 1997 Echolocation in dolphins with a information and influence (ed. U Stegmann), 2. Curio E. 1976 The ethology of predation. Berlin, dolphin-bat comparison. Bioacoustics 8, 137–162. pp. 113–129. Cambridge, UK: Cambridge University Germany: Springer-Verlag. (doi:10.1080/09524622.1997.9753357) Press. 3. Ruxton GD. 2009 Non-visual crypsis: a review of the 18. Møhl B, Andersen S. 1973 Echolocation: high- 31. Elwen S, Mey¨er MA, Best PB, Kotze PGH, Thornton empirical evidence for camouflage to senses other frequency component in the click of the harbour M, Swanson S. 2006 Range and movements of than vision. Phil. Trans. R. Soc. B 364, 549–557. porpoise (Phocoena ph.). J. Acoust. Soc. Am. 54, female Heaviside’s dolphins (Cephalorhynchus rc .Sc B Soc. R. Proc. (doi:10.1098/rstb.2008.0228) 1368–1372. (doi:10.1121/1.1914435) heavisidii), as determined by satellite-linked 4. Wilson DR, Hare JF. 2006 The adaptive utility of 19. Andersen S, Amundin M. 1976 Possible predator- telemetry. J. Mammal. 87, 866–877. (doi:10.1644/ Richardson’s ground squirrel (Spermophilus related adaption of sound production and hearing 05-MAMM-A-307R2.1) richardsonii) short-range ultrasonic alarm signals. in the harbour porpoise (Phocoena phocoena). 32. Behrmann CA. 2011 Occurrence and group

Can. J. Zool. 84, 1322–1330. (doi:10.1139/Z06-120) Aquat. Mamm. 4, 56–58. dynamics of Heaviside’s dolphins (Cephalorhynchus 285

5. Jones KJ, Hill WL. 2001 Auditory perception of 20. Szymanski MD, Bain DE, Kiehl K, Pennington S, heavisidii) in Table Bay, Western Cape, South Africa. 20181178 : hawks and owls for passerine alarm calls. Ethology Wong S, Henry KR. 1999 Killer whale (Orcinus orca) MSc thesis, University of Pretoria, Pretoria, 107, 717–726. (doi:10.1046/j.1439-0310.2001. hearing: auditory brainstem response and South Africa. 00698.x) behavioral audiograms. J. Acoust. Soc. Am. 106, 33. Morisaka T, Karczmarski L, Akamatsu T, Sakai M, 6. Tyack P. 1998 Acoustic communication under the sea. 1134–1141. (doi:10.1121/1.427121) Dawson S, Thornton M. 2011 Echolocation signals of In Animal acoustic communication (eds SL Hopp, MJ 21. Clausen KT, Wahlberg M, Beedholm K, DeRuiter Heaviside’s dolphins (Cephalorhynchus heavisidii). Owren, CS Evans), pp. 163–220. Berlin, Germany: SL, Madsen PT. 2010 Click communication in J. Acoust. Soc. Am. 129, 449–457. (doi:10.1121/1. Springer. (doi:10.1007/978-3-642-76220-8_6) harbour porpoises Phocoena phocoena. 3519401) 7. Clay CS, Medwin H. 1977 Acoustical oceanography: Bioacoustics 20, 1–28. (doi:10.1080/09524622. 34. Griffin DR, Webster FA, Michael CR. 1960 The principles and applications. New York, NY: 2011.9753630) echolocation of flying insects by bats. Anim. Behav. Wiley-Interscience. 22. Kyhn LA, Tougaard J, Beedholm K, Jensen FH, Ashe 8, 141–154. (doi:10.1016/0003-3472(60)90022-1) 8. Deecke VB, Ford JKB, Slater PJB. 2005 The vocal E, Williams R, Madsen PT. 2013 Clicking in a killer 35. Miller LA, Pristed J, Møshl B, Surlykke A. 1995 The behaviour of mammal-eating killer whales: whale habitat: narrow-band, high-frequency click-sounds of narwhals (Monodon monoceros)in communicating with costly calls. Anim. Behav. 69, biosonar clicks of harbour porpoise (Phocoena Inglefield Bay, Northwest Greenland. Mar. Mamm. 395–405. (doi:10.1016/j.anbehav.2004.04.014) phocoena) and Dall’s porpoise (Phocoenoides dalli). Sci. 11, 491–502. (doi:10.1111/j.1748-7692.1995. 9. Au W. 1993 The sonar of dolphins. New York, NY: PLoS ONE 8, e63763. (doi:10.1371/journal.pone. tb00672.x) Springer-Verlag. (doi:10.1007/978-1-4612-4356-4) 0063763) 36. Wisniewska DM, Johnson M, Teilmann J, Rojano- 10. Aguilar de Soto N, Madsen PT, Tyack P, Arranz P, 23. Tyack P. 1986 Population biology, social behavior Don˜ate L, Shearer J, Sveegaard S, Miller LA, Siebert Marrero J, Fais A, Revelli E, Johnson M. 2012 No and communication in whales and dolphins. Trends U, Madsen PT. 2016 Ultra-high foraging rates of shallow talk: cryptic strategy in the vocal Ecol. Evol. 1, 144–150. (doi.org/10.1016/0169- harbor porpoises make them vulnerable to communication of Blainville’s beaked whales. Mar. 5347(86)90042-X) anthropogenic disturbance. Curr. Biol. 26, Mamm. Sci. 28, 75–92. (doi:10.1111/j.1748-7692. 24. Herzing DL. 1996 Vocalizations and associated 1441–1446. (doi:10.1016/j.cub.2016.03.069) 2011.00495.x) underwater behavior of free-ranging Atlantic 37. Lammers MO, Au WWL, Aubauer R, Nachtigall PE. 11. Rankin S, Archer F, Barlow J. 2013 Vocal spotted dolphins, Stenella frontalis, and bottlenose 2004 A comparative analysis of the pulsed activity of tropical dolphins is inhibited by the dolphins, Tursiops truncatus. Aquat. Mamm. 22, emissions of free-ranging Hawaiian spinner dolphins presence of killer whales, Orcinus orca. Mar. 61–79. (Stenella longirostris). In Echolocation in bats and Mamm. Sci. 29, 679–690. (doi:10.1111/j.1748- 25. Miller PJ. 2006 Diversity in sound pressure levels dolphins (eds JA Thomas, CF Moss, M Vater), 7692.2012.00613.x) and estimated active space of resident killer whale pp. 414–419. Chicago, IL: The University of 12. Thomas JA, Ferm LM, Kuechle VB. 1989 Silence as vocalizations. J. Comp. Physiol. A 192, 449. (doi:10. Chicago Press. an antipredation strategy by weddell seals. San 1007/s00359-005-0085-2) 38. Blomqvist C, Amundin M. 2004 High-frequency Diego, CA: Naval Ocean Systems Center. 26. Jensen F, Beedholm K, Wahlberg M, Bejder L, burst-pulse sounds in agonistic/aggressive 13. Wahlberg M, Beedholm K, Heerfordt A, Møhl B. Madsen PT. 2012 Estimated communication range interactions in bottlenose dolphins, Tursiops 2011 Characteristics of biosonar signals from the and energetic cost of bottlenose dolphin whistles truncatus.InEcholocation in bats and dolphins northern bottlenose whale, Hyperoodon ampullatus. in a tropical habitat. J. Acoust. Soc. Am. 131, (eds JA Thomas, CF Moss, M Vater), pp. 425–431. J. Acoust. Soc. Am. 130, 3077–3084. (doi:10.1121/ 582–592. (doi:10.1121/1.3662067) Chicago, IL: The University of Chicago Press. 1.3641434) 27. Amundin M. 1991 Sound production in odontocetes 39. Madsen PT, Wahlberg M. 2007 Recording and 14. Fenton BM, Jensen FH, Kalko EK, Tyack PL. 2014 with emphasis on the harbour porpoise Phocoena quantification of ultrasonic echolocation clicks from Sonar signals of bats and toothed whales. In phocoena. PhD dissertation, Stockholm, Sweden. free-ranging toothed whales. Deep Sea Res. 1 Biosonar (eds A Surlykke, P Nachtigall, R Fay, 28. Dawson SM. 1991 Clicks and communication: the Oceanogr. Res. Pap. 54, 1421–1444. (doi:10.1016/ A Popper), pp. 11–59. New York, NY: Springer. behavioral and social contexts of Hector’s dolphin j.dsr.2007.04.020) (doi:10.1007/978-1-4614-9146-0_2) vocalizations. Ethology 88, 265–276. (doi:10.1111/ 40. Jensen FH, Rocco A, Mansur RM, Smith BD, Janik 15. Morisaka T, Connor RC. 2007 Predation by killer j.1439-0310.1991.tb00281.x) VM, Madsen PT. 2013 Clicking in shallow rivers: whales (Orcinus orca) and the evolution of whistle 29. Fischer J, Wadewitz P, Hammerschmidt K. 2016 short-range echolocation of Irrawaddy and Ganges loss and narrow-band high frequency clicks in Structural variability and communicative complexity river dolphins in a shallow, acoustically complex odontocetes. J. Evol. Biol. 20, 1439–1458. (doi:10. in acoustic communication. Anim. Behav. 134, habitat. PLoS ONE 8, e59284. (doi:10.1371/journal. 1111/j.1420-9101.2007.01336.x) 229–237. (doi:10.1016/j.anbehav.2016.06.012) pone.0059284) Downloaded from http://rspb.royalsocietypublishing.org/ on July 18, 2018

41. Au WWL, Kastelein RA, Rippe T, Schooneman NM. 48. Soerensen PM, Wisniewska DM, Jensen FH, 54. Jakobsen L, Ratcliffe JM, Surlykke A. 2013 9 1999 Transmission beam pattern and echolocation Johnson M, Teilmann J, Madsen PT. 2018 Convergent acoustic field of view in echolocating rspb.royalsocietypublishing.org signals of a harbor porpoise (Phocoena phocoena). Click communication in wild harbour porpoises bats. Nature 493, 93–96. (doi:10.1038/ J. Acoust. Soc. Am. 106, 3699–3705. (doi:doi.org/ (Phocoena phocoena). Sci. Rep. UK 8, 9702. nature11664) 10.1121/1.428221) 49. Connor RC, Smolker RA. 1996 ‘Pop’ goes 55. Wisniewska DM, Ratcliffe JM, Beedholm K, Christensen 42. Madsen PT, Kerr I, Payne R. 2004 Source parameter the dolphin: a vocalization male bottlenose CB, Johnson M, Koblitz JC, Wahlberg M, Madsen PT. estimates of echolocation clicks from wild pygmy dolphins produce during consortships. 2015 Range-dependent flexibility in the acoustic field killer whales (Feresa attenuata). J. Acoust. Soc. Am. Behaviour 133, 643–662. (doi:10.1163/ of view of echolocating porpoises (Phocoena 116, 1909–1912. (doi:10.1121/1.1788726) 156853996X00404) phocoena). Elife 4, e05651. (doi:10.7554/eLife.05651) 43. Fox J, Weisberg S. 2011 An R companion to applied 50. Caldwell MC, Haugen RM, Caldwell DK. 1962 High- 56. Delgado-Garcia L. 2016 Acoustic development and regression, 2nd edn. Thousand Oaks, CA: Sage energy sound associated with fright in the dolphin. behaviour of odontocete calves. PhD thesis,

Publications. Science 138, 907–908. (doi:10.1126/science.138. University of Southern Denmark, Odense, Denmark. B Soc. R. Proc. 44. Ogle D. 2017 FSA: Fisheries Stock Analysis. R 3543.907) 57. Li S, Wang K, Wang D, Dong S, Akamatsu T. 2008 package version 0.8.16. 51. Madsen PT, Carder D, Bedholm K, Ridgway S. 2005 Simultaneous production of low-and high-frequency 45. Breiman L. 2001 Random forests. Mach. Learn. 45, Porpoise clicks from a sperm whale nose: sounds by neonatal finless porpoises. J. Acoust. Soc. 5–32. (doi:10.1023/A:1010933404324) Convergent evolution of 130 kHz pulses in toothed Am. 124, 716–718. (doi:10.1121/1.2945152) 46. Jensen FH, Wahlberg M, Beedholm K, Johnson M, whale sonars? Bioacoustics 15, 195–206. (doi:10. 58. Reyes Reyes MV, Tossenberger VP, In˜iguez MA, 285 de Soto NA, Madsen PT. 2015 Single-click beam 1080/09524622.2005.9753547) Hildebrand JA, Melco´n ML. 2016 Communication 20181178 : patterns suggest dynamic changes to the field of 52. Best PB, Mey¨er MA, Lockyer C. 2010 Killer whales in sounds of Commerson’s dolphins (Cephalorhynchus view of echolocating Atlantic spotted dolphins South African waters: a review of their biology. commersonii) and contextual use of vocalizations. (Stenella frontalis) in the wild. J. Exp. Biol. 218, Afr. J. Mar. Sci. 32, 171–186. (doi:10.2989/ Mar. Mamm. Sci. 32, 1219–1233. (doi:10.1111/ 1314–1324. (doi:10.1242/jeb.116285) 1814232x.2010.501544) mms.12321) 47. Kastelein RA, Janssen M, Verboom WC, de Haan D. 53. Wahlberg M et al. 2011 Source parameters of 59. Martin MJ, Gridley T, Elwen SH, Jensen FH. 2018 2005 Receiving beam patterns in the horizontal echolocation clicks from wild bottlenose dolphins Data from: Heaviside’s dolphins (Cephalorhynchus plane of a harbor porpoise (Phocoena phocoena). (Tursiops aduncus and Tursiops truncatus). heavisidii) relax acoustic crypsis to increase J. Acoust. Soc. Am. 118, 1172–1179. (doi:10.1121/ J. Acoust. Soc. Am. 130, 2263–2274. (doi:10. communication range. Dryad Digital Repository. 1.1945565) 1121/1.3624822) (doi:10.5061/dryad.64048p0) Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B

Heaviside’s dolphins (Cephalorhynchus heavisidii) relax acoustic crypsis to increase communication range Morgan J. Martin1, Tess Gridley2, Simon H. Elwen1 and Frants H. Jensen3,4

1Mammal Research Institute, Department of Zoology and Entomology, University of Pretoria. C/o Sea Search Research and Conservation NPC, 4 Bath Rd, Cape Town 7945, South Africa 2 Centre for Statistics in Ecology, Environment and Conservation, Department of Statistical Sciences, University of Cape Town. C/o Sea Search Research and Conservation NPC, 4 Bath Rd, Cape Town 7945, South Africa. 3Aarhus Institute of Advanced Studies, Aarhus University, Aarhus 8000, Denmark 4Biology Department, Woods Hole Oceanographic Institution, 266 Woods Hole Rd, Woods Hole, MA 02543, USA Contact: [email protected]

Field site and data collection The study area, Shearwater Bay, located near Lüderitz in southern Namibia (-26° 37' S, 15° 05' E), is a small bay (6.5 km2) which consists of shallow water with a maximum depth of 12 m. Data analysed in this study were collected from wild Heaviside’s dolphins located in Shearwater Bay during April and May 2016. Underwater acoustic recordings of Heaviside’s dolphin vocalisations were made under calm weather conditions (Beaufort sea state ≤ 2) using two high frequency recording hydrophones (SoundTrap 300 HF; www.oceaninstruments.co.nz). The hydrophones were mounted 1 m apart and suspended 1.5 m below a 4.2 m fiberglass ocean kayak. Sound was digitised at a sampling rate of 576 kHz with a 16-bit resolution, and settings were configured to include high gain (+12 dB) and a high pass filter (400 Hz), effective sensitivity: -171 dB re 1 V/µPa, flat frequency response: 400 Hz – 150 kHz ± 3 dB. A built in anti-aliasing filter exists at 150 kHz. Recordings were stored as compressed 30-min SUD files on the SoundTraps. The kayak and hydrophone array were deployed when Heaviside’s dolphins were observed from shore and weather conditions permitted. When an individual or group of dolphins was sighted, the observer on board the kayak would attempt to approach with minimal disturbance. A group was defined as two or more dolphins in close proximity (< 50 m radius), generally carrying out the same activity. Behaviour and focal group information were collected concurrently with sound recordings using a Dictaphone. A visual survey group-follow with incident sampling protocol [1, 2] was used to record surface behaviour along with group size, group composition (presence or absence of calves), group spacing, and estimated distance from the hydrophone array. Definitions

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B of behavioural states and events were adapted from [3, 4]. A secondary visual survey method was implemented from shore (20 m elevation) using two observers with walkie talkies and a Sony Handycam camcorder to assist the kayak-based observer to locate and maintain focal groups, monitor other Heaviside’s dolphin groups present in the bay, provide information on behaviour and to assess inter-observer reliability [5]. Statistical analyses All high-quality measured signals visually classified into the four proposed categories were evaluated to examine the ability to quantitatively distinguish pulsed signal types. Signal parameters were compared across signal categories using non-parametric Kruskal-Wallis tests and subsequent Dunn’s post-hoc tests for pairwise comparisons in R version 3.4.2 [6, 7] (Suppl. Table 1). Further in R, all high-quality signals were evaluated with a principal component analysis (PCA) as it is robust to correlated variables. The PCA was used to identify the most influential parameters for signal classification. Nine parameter variables were included in the PCA: 5th, median (50th) and 95th percentile interclick intervals (ICI), peak frequency, centroid frequency, -10 dB bandwidth, RMS bandwidth, Q-ratio, and -10 dB click duration (Suppl. Fig. 1). All values were log- transformed prior to the analysis. The Kaiser criterion was used to identify the number of principal components to retain and was determined by eigenvalues > 1 (Suppl. Table 2). We then used a Random Forest classifier [8] to measure prediction accuracy as a function of buzz and burst-pulse signal categories using either interclick intervals (5th, 50th and 95th percentiles for each signal), spectral and temporal click parameters (peak frequency, centroid frequency, -10 dB bandwidth, RMS bandwidth, Q-ratio, and -10 dB click duration), or all parameters combined as features to test the potential benefit of spectral differences in decreasing signal ambiguity in the repertoire. The Random Forest classifier was built in MATLAB 2017b using a ‘bagged trees’ ensemble classifier with 30 learners. Prediction accuracy was measured using 5-fold cross-validation to prevent overfitting. To measure consistency in prediction accuracy, the classifier was trained 100 times and prediction accuracy was measured for each iteration.

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B

Supplementary Table 1. Dunn’s post-hoc tests of measured parameters across signal categories. All parameters were log-transformed before statistical analysis. Note that initially click trains were differentiated visually from buzzes and burst-pulses using click rates with interclick intervals exceeding 10 ms. A subset of click trains were composed of lower-frequency, broader bandwidth signals than previously described [9], and we therefore divided click trains into NBHF click trains and broadband click trains by inspecting spectrograms. Initially, buzz and burst-pulse signals were visually differentiated by the presence or absence of a preceding click train as burst-pulses occur as isolated signals.

ICI5th ICIMED ICI95th FP FC BW10dB BWRMS QRMS Dur10dB Signal Comparison p p p p p p p p p NBHF Train : BB Train 0.9695 0.8520 0.6452 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 NBHF Train : Buzz < 0.0001 < 0.0001 < 0.0001 0.1466 0.7288 0.3543 0.0062 0.0189 0.4201 NBHF Train : Burst-pulse < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 Buzz : Burst-pulse < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 < 0.0001 Buzz : BB Train < 0.0001 < 0.0001 < 0.0001 0.0029 < 0.0001 < 0.0001 0.0002 < 0.0001 < 0.0001 Burst-pulse : BB Train < 0.0001 < 0.0001 < 0.0001 0.7380 0.4669 0.5611 0.9823 0.6735 0.0459

α = 0.05, p-values below this threshold are shown in boldface th Abbreviations: NBHF Train = narrowband high-frequency click train; BB Train = broadband click train; ICI5th, ICIMED and ICI95th = 5 , th th median (50 ) and 95 percentile interclick intervals (ms); FP = peak frequency (kHz); FC = centroid frequency (kHz); BW10dB = -10 dB bandwidth (kHz); BWRMS = root mean square bandwidth (kHz); QRMS = FC/BWRMS; Dur10dB = -10 dB click duration (µs)

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B

Supplementary Table 2. PCA output of the nine measured parameter variables from 159 signals. All parameter values were log-transformed prior to the PCA. Parameter abbreviations: ICI5th, th th th ICIMED and ICI95th = 5 , median (50 ) and 95 percentile interclick intervals (ms); FP = peak frequency (kHz); FC = centroid frequency (kHz); BW10dB = -10 dB bandwidth (kHz); BWRMS = root mean square bandwidth (kHz); QRMS = FC/BWRMS; Dur10dB = -10 dB click duration (µs)

PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9 Importance of Components Standard Deviation 2.099 1.545 1.088 0.794 0.461 0.383 0.172 0.058 0.010 Prop. of Variance 0.490 0.265 0.132 0.070 0.024 0.016 0.003 0.000 0.000 Cumulative Prop. 0.490 0.755 0.886 0.956 0.980 0.996 1.000 1.000 1.000

Loadings with Rotation = (9 x 9) ICI5th -0.326 -0.460 0.104 -0.037 0.063 0.050 -0.602 -0.548 -0.005

ICIMED -0.329 -0.459 0.116 -0.056 0.045 0.031 -0.166 0.797 0.002

ICI95th -0.328 -0.451 0.126 -0.058 -0.009 -0.010 0.777 -0.255 0.003

FP -0.254 0.332 0.514 0.231 0.704 -0.107 0.026 -0.002 -0.001

FC -0.226 0.306 0.619 -0.160 -0.585 0.252 -0.035 -0.002 0.201

BW10dB 0.369 -0.218 0.398 0.000 -0.173 -0.791 -0.048 -0.001 0.001

BWRMS 0.405 -0.139 0.328 -0.361 0.144 0.359 0.018 -0.002 -0.656

QRMS -0.426 0.210 -0.124 0.280 -0.292 -0.255 -0.019 0.007 -0.728

Duration10dB -0.284 0.244 -0.179 -0.840 0.141 -0.320 -0.017 -0.008 0.000

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B

Suppl. Fig. 1: Signal parameters and discrimination of signal types. A: Q-ratio (centroid frequency / RMS bandwidth) as a function of centroid frequency. B: Log-transformed RMS bandwidth as a function of log-transformed median ICI. C: Principal component analysis of signal types including nine parameter variables. Each data point represents one measured pulsed signal. PC 1 primarily represents RMS bandwidth and Q-ratio parameters. PC 2 represents click rate parameters (5th, 50th and 95th percentile interclick intervals).

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B

Acoustical modelling of detection range and impact on active space To investigate how click type affects conspecific detection range and active space, we built an acoustic model of detection range under a noise-limited scenario using the passive sonar equation [10] and assuming successful detection when received sound energy exceeded masking noise energy integrated across auditory bandwidth:

Eq. 1: 푅퐿 = 푆퐿 − 푇퐿 > 푁퐿 Here, RL is the received echo level, SL is the source level measured in energy flux density, TL is the transmission loss between source and receiver, NL is the masking noise level; all in decibels. Since toothed whales have directional sound emission and directional hearing, we modelled detection range explicitly as a function of the outgoing source angle 휃푆 and the incoming receiver aspect 휃푅. Directional sound emission was modelled through a transmission gain (TG), the difference between off-axis apparent source level and the on-axis source level, with values always negative. Directional hearing was modelled through an auditory gain (AG), the difference between off-axis hearing sensitivity and on-axis hearing sensitivity, with values always negative.

Eq. 2: Successful detection when: 푆퐿 + 푇퐺(휃푆) − 푇퐿 + 퐴퐺(휃푅) > 푁퐿 Source level: On-axis source level for Heaviside’s dolphin NBHF clicks has been measured to 161±5 dB [min 149, max 174] re. 1 µPa RMS for a -10 dB duration of 74 µs [9]. To reflect the temporal integration of the auditory system, we corrected these source levels for a temporal integration time of 264 µs for a bottlenose dolphin [11] by adding 10푙표푔10(74휇푠⁄264휇푠). Directional sound emission: Off-axis apparent source level was modelled using a circular, symmetric piston which has frequently been used to approximate the sonar beam of toothed whales [12]. Transmission beams were calculated for 10 NBHF clicks and 9 burst-pulse clicks using a piston size of 6.4 cm diameter and a waveform filtered with a 10 kHz - 150 kHz 6-pole Butterworth bandpass filter (Fig. 3A). This resulted in a directivity index (DIT) of 24 dB for NBHF clicks, similar to that of other NBHF species [13-15] and decreasing to 20 dB for burst-pulse clicks. For the rest of the paper, we used a model burst-pulse click with a centroid frequency of 80 kHz1 to calculate the possible change in detection range. We assumed that animals were energy limited and that a change in directivity would therefore lead to a lower on-axis source level, so on-axis SL for burst-pulse clicks was set 4 dB lower than for NBHF clicks. Transmission loss: We estimated transmission loss as the combination of spherical spreading loss and frequency dependent absorption, so that TL = 20 log10(R) + αR. Here, R was the range to the target (m), and the absorption coefficient α was calculated using the centroid frequency of each

1 Note that click parameters reported in manuscript are for signals filtered with a wider bandwidth Butterworth filter (20 kHz – 275 kHz), and centroid frequency measurements here are therefore similar but not directly comparable.

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B click type [16], resulting in an absorption coefficient of 0.40 dB/m for a 128 kHz NBHF click and 0.22 dB/m for an 80 kHz burst-pulse click. Directional hearing: No data were available for hearing directivity at the exact frequencies required. Instead, we used auditory sensitivity measurements as a function of angle reported for a harbour porpoise (Phocoena phocoena) at 64 kHz and 100 kHz [17], which represents a similar shift in frequency (a little over half an octave) as the difference between NBHF and burst-pulse clicks. While this study found a slightly asymmetric receiving beam, we simplified this by taking the mean acoustic sensitivity between the left and right side, and then interpolated across values using a piecewise cubic interpolation. This resulted in a receiver directivity index (DIR) of 8.8 dB (100 kHz) and 5.1 dB (64 kHz; Suppl. Fig. 2).

Suppl. Fig. 2: Aspect dependent hearing sensitivity (implemented here as auditory gain AG) based on measurements from a harbour porpoise [17].

Masking noise level: The masking noise energy was estimated as the spectral noise level N0 (in dB re 1 µPa2Hz-1, i.e. noise intensity per Hz bandwidth) integrated over the auditory filter bandwidth of the animal and suppressed by the auditory directivity of the animal. Since we did not have reliable estimates of auditory filter bandwidth for clicks, we assumed a simple 1/3rd octave bandwidth similar to terrestrial mammals:

Eq. 3: 푁퐿 = 푁0(퐹푐) + 10푙표푔10(0.23 ∗ 퐹퐶) − 퐷퐼푅 The spectral noise level was estimated as Wenz Sea State 2 deep-water noise levels (approximately 58 dB at 1 kHz, with a gradual decrease of 17 dB per decade increase in frequency), plus the addition of thermal noise (generally at frequencies above 100 kHz) (both in dB re. 1 µPa2Hz-1):

퐹 Eq. 4: 푁 (퐹 ) = 푁 (1 푘퐻푧) − 17푙표푔 ( 푐 ) − 75 + 20푙표푔 (퐹 ) 0 푐 0 10 1 푘퐻푧 10 푐 ----(wind generated noise)------(thermal noise)---

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B

Equations from http://www.usna.edu/Users/physics/ejtuchol/documents/SP411/Chapter11.pdf Detection range: We solved equation 2 numerically in MATLAB 2013b to find the maximum detection range where the received level exceeded the masking noise level. The detection range was calculated as a function of source angle (θS) and receiver aspect (θR), for both a NBHF click and an 80 kHz burst-pulse click (Suppl. Fig. 3).

Suppl. Fig. 3: Modelled conspecific detection range for a Heaviside’s dolphin NBHF click and an 80 kHz burst-pulse click, as a function of both source angle and receiver aspect.

Active space: To calculate active space, we assumed a 2D habitat with conspecifics located at the water surface. We then calculated the total detection area A (m2) by integrating detection range R as a function of source angle from 0 to 180 degrees and assuming rotational symmetry:

휋 Eq. 5: 퐴 = 2 푅(휃 ) sin 푑휃 ∫0 푠 푠

Since detection range depends both on source angle θS and receiver aspect θR, we assumed an equal probability of receiver aspect and used the mean detection range as a function of receiver aspect, so only source angle appears in equation 5.

Sensitivity analysis: The two most important parameters for detection range are source level and noise level. We therefore conducted a sensitivity analysis to measure the change in total detection area (ABP/ANBHF) across a wide range of possible source levels and noise levels. We varied wind- 2 -1 generated noise [N0(1kHz)] from 50 to 75 dB re. 1 µPa Hz while keeping thermal noise constant, thus increasingly favouring NBHF signals that are primarily limited by thermal noise. We used 5 different source levels reflecting the full range of NBHF source levels measured empirically from Heavisides dolphins [9]. For all simulations, the modelled active space for burst pulse signals was at least twice as large, and for quieter (Sea State 1 or Sea State 2 conditions) as high as 4 to 5 times as large as for NBHF signals.

Martin et al. 2018 Supplementary Methods Proceedings of the Royal Society B

References 1. Altmann J. Observational study of behaviour: sampling methods. Behaviour. 1974. 49: 227- 67. 2. Mann J. Behavioural sampling methods for cetaceans: A review and critique. Mar. Mamm. Sci. 1999. 15: 102-22. 3. Herzing DL. Vocalizations and associated underwater behavior of free-ranging Atlantic spotted dolphins, Stenella frontalis, and bottlenose dolphins, Tursiops truncatus. Aquat. Mamm. 1996. 22: 61-79. 4. Henderson EE, Hildebrand JA, Smith MH, Falcone EA. The behavioral context of common dolphin (Delphinus sp.) vocalizations. Mar. Mamm. Sci. 2011. 28: 439-60. (doi:10.1111/j.1748-7692.2011.00498.x) 5. Kaufman AB, Rosenthal R. Can you believe my eyes? The importance of interobserver reliability statistics in observations of animal behaviour. Anim. Behav. 2009. 78: 1487-91. 6. Fox J, Weisberg S. An R Companion to Applied Regression, 2nd Ed. Thousand Oaks, California: Sage Publications. 2011. 7. Ogle D. FSA: Fisheries Stock Analysis. R package version 0.8.16. 2017. 8. Breiman L. Random Forests. Machine learning. 2001. 45: 5-32. (doi:10.1023/A:1010933404324) 9. Morisaka T, Karczmarski L, Akamatsu T, Sakai M, Dawson S, Thornton M. Echolocation signals of Heaviside's dolphins (Cephalorhynchus heavisidii). J. Acoust. Soc. Am. 2011. 129: 449-57. (doi:10.1121/1.3519401) 10. Au W. The Sonar of Dolphins. New York: Springer-Verlag. 1993. (doi:10.1007/978-1-4612- 4356-4) 11. Au WW, Moore PW, Pawloski DA. Detection of complex echoes in noise by an echolocating dolphin. J. Acoust. Soc. Am. 1988. 83: 662-8. 12. Jensen FH, Wahlberg M, Beedholm K, Johnson M, de Soto NA, Madsen PT. Single-click beam patterns suggest dynamic changes to the field of view of echolocating Atlantic spotted dolphins (Stenella frontalis) in the wild. J. Exp. Biol. 2015. 218: 1314-24. (doi:10.1242/jeb.116285) 13. Kyhn LA, Tougaard J, Jensen F, Wahlberg M, Stone G, Yoshinaga A, et al. Feeding at a high pitch: source parameters of narrow band, high-frequency clicks from echolocating off- shore hourglass dolphins and coastal Hector's dolphins. J. Acoust. Soc. Am. 2009. 125: 1783-91. (doi:10.1121/1.3075600) 14. Kyhn LA, Jensen FH, Beedholm K, Tougaard J, Hansen M, Madsen PT. Echolocation in sympatric Peale's dolphins (Lagenorhynchus australis) and Commerson's dolphins (Cephalorhynchus commersonii) producing narrow-band high-frequency clicks. J. Exp. Biol. 2010. 213: 1940-9. (doi:10.1242/jeb.042440) 15. Kyhn LA, Tougaard J, Beedholm K, Jensen FH, Ashe E, Williams R, et al. Clicking in a killer whale habitat: narrow-band, high-frequency biosonar clicks of harbour porpoise (Phocoena phocoena) and Dall's porpoise (Phocoenoides dalli). PLOS ONE. 2013. 8: e63763. (doi:10.1371/journal.pone.0063763) 16. Kinsler LE, Frey AR, Coppens AB, Sanders JV. Fundamentals of Acoustics, 4th Ed. Wiley- VCH. 1999. p. 560. 17. Kastelein RA, Janssen M, Verboom WC, de Haan D. Receiving beam patterns in the horizontal plane of a harbor porpoise (Phocoena phocoena). J. Acoust. Soc. Am. 2005. 118: 1172-9. (doi:10.1121/1.1945565)

Martin et al. 2018 Supplementary Appendix Proceedings of the Royal Society B

Heaviside’s dolphins (Cephalorhynchus heavisidii) relax acoustic crypsis to increase communication range Morgan J. Martin1, Tess Gridley2, Simon H. Elwen1 and Frants H. Jensen3,4

Appendix S1: Measured parameters from 159 Heaviside’s dolphin pulsed signals grouped by th th th signal type. Parameter abbreviations: ICI5th, ICIMED and ICI95th = 5 , median (50 ) and 95 percentile interclick intervals (ms); FP = peak frequency (kHz); FC = centroid frequency (kHz); BWRMS = root mean square bandwidth (kHz); BW10dB = -10 dB bandwidth (kHz); QRMS = FC/BWRMS; Dur10dB = -10 dB click duration (µs)

Signal Type ICI5th ICI50th ICI95th FP FC BWRMS BW10dB QRMS Dur10dB NBHF Click Train 24.4 27.9 33.0 121.5 130.5 12.8 27.0 10.2 60.8 NBHF Click Train 30.4 50.2 98.6 130.5 132.9 11.8 32.6 11.2 84.6 NBHF Click Train 25.2 29.0 65.2 131.6 132.3 10.6 25.9 12.4 61.3 NBHF Click Train 23.5 31.9 48.3 130.5 131.0 13.9 24.8 9.5 61.3 NBHF Click Train 24.1 28.9 49.5 121.5 124.1 8.5 19.1 14.5 70.3 NBHF Click Train 31.4 38.6 60.1 127.1 134.1 19.9 58.5 6.8 74.1 NBHF Click Train 15.5 29.8 58.5 127.1 131.3 15.8 57.4 8.3 85.9 NBHF Click Train 21.3 29.0 42.5 128.3 127.6 9.6 21.4 13.3 77.8 NBHF Click Train 25.6 30.1 36.7 136.1 132.4 9.7 31.5 13.7 53.7 NBHF Click Train 27.1 34.5 63.8 122.6 136.6 20.4 61.9 6.7 72.2 NBHF Click Train 21.0 25.7 29.0 127.1 129.4 10.5 25.9 12.4 54.9 NBHF Click Train 33.9 53.8 141.2 127.1 128.8 10.4 31.5 12.4 51.7 NBHF Click Train 42.8 67.2 114.1 123.8 126.0 7.1 23.6 17.8 63.9 NBHF Click Train 22.4 28.1 34.3 128.3 130.1 8.5 23.6 15.3 69.6 NBHF Click Train 14.0 22.3 67.0 122.6 135.0 26.7 93.4 5.1 39.8 NBHF Click Train 26.6 33.8 44.4 128.3 123.2 7.7 27.0 15.9 59.4 NBHF Click Train 17.3 23.5 61.5 137.3 131.3 10.4 31.5 12.6 60.1

Martin et al. 2018 Supplementary Appendix Proceedings of the Royal Society B

NBHF Click Train 40.1 57.0 87.5 122.6 139.6 27.8 69.8 5.0 74.7 NBHF Click Train 24.1 28.7 34.3 127.1 131.3 13.6 33.8 9.7 51.2 NBHF Click Train 10.6 13.5 21.9 122.6 130.2 14.5 37.1 9.0 49.8 NBHF Click Train 25.2 27.7 31.9 135.0 130.7 10.0 23.6 13.1 72.1 NBHF Click Train 26.1 35.6 73.4 122.6 131.9 14.2 23.6 9.3 70.3 NBHF Click Train 22.9 27.9 42.6 130.5 130.5 9.0 28.1 14.5 65.5 NBHF Click Train 20.9 25.2 36.9 138.4 136.0 16.3 33.8 8.3 51.0 NBHF Click Train 17.6 26.6 42.7 121.5 134.0 20.8 69.8 6.5 51.2 NBHF Click Train 23.2 28.7 46.1 136.1 130.5 14.0 28.1 9.3 56.8 NBHF Click Train 18.7 25.0 49.7 127.1 127.2 11.1 23.6 11.5 64.9 NBHF Click Train 21.2 25.2 42.3 130.5 131.7 15.9 33.8 8.3 54.3 NBHF Click Train 19.8 30.4 49.3 121.5 132.4 15.4 32.6 8.6 63.7 NBHF Click Train 17.1 22.2 29.7 136.1 131.6 11.5 31.5 11.4 67.9 NBHF Click Train 18.1 27.3 37.7 131.6 137.3 20.9 69.8 6.6 77.6 NBHF Click Train 27.3 29.9 41.4 121.5 133.8 20.3 46.1 6.6 65.3 NBHF Click Train 47.9 54.3 61.2 122.6 130.3 11.7 30.4 11.1 91.5 Buzz 6.4 7.0 10.9 137.3 136.6 14.5 30.4 9.4 55.2 Buzz 9.0 9.5 10.1 130.5 132.6 13.2 43.9 10.1 55.9 Buzz 4.2 5.4 7.7 121.5 143.2 31.4 77.6 4.6 39.8 Buzz 2.6 4.4 7.9 123.8 132.3 23.1 40.5 5.7 62.0 Buzz 2.1 2.3 4.6 122.6 130.2 12.7 36.0 10.3 76.6 Buzz 11.2 13.5 15.0 118.1 150.5 39.9 113.6 3.8 54.2 Buzz 5.6 6.6 10.4 122.6 134.0 25.6 51.8 5.2 95.5 Buzz 2.9 3.3 4.4 128.3 132.6 11.3 38.3 11.8 42.7 Buzz 6.7 10.2 11.0 137.3 134.9 17.4 47.3 7.8 54.9 Buzz 4.9 6.3 8.2 126.0 128.0 9.9 18.0 12.9 83.5 Buzz 8.8 9.0 10.3 135.0 138.6 17.2 39.4 8.0 84.6 Buzz 6.5 7.9 9.3 120.4 138.8 25.4 85.5 5.5 71.2 Buzz 6.7 7.7 11.4 135.0 146.1 16.8 50.6 8.7 51.0 Buzz 7.2 7.4 9.2 130.5 135.9 21.8 69.8 6.2 74.3 Buzz 9.9 11.1 13.0 115.9 125.3 21.6 24.8 5.8 145.3 Buzz 6.7 7.8 9.1 111.4 125.4 25.4 33.8 4.9 135.9 Buzz 6.1 7.3 9.0 121.5 125.0 15.1 10.1 8.3 128.7 Buzz 6.2 6.6 9.5 115.9 130.7 25.0 49.5 5.2 79.3 Buzz 8.1 8.7 9.9 124.9 129.0 14.8 21.4 8.7 121.2 Buzz 3.8 4.8 12.9 124.9 126.9 14.8 24.8 8.6 83.7 Buzz 5.9 8.9 12.4 118.1 127.9 18.1 21.4 7.1 79.5 Buzz 5.8 6.7 10.5 121.5 128.5 10.3 27.0 12.5 55.0 Buzz 9.4 10.6 12.3 121.5 124.0 12.9 27.0 9.6 54.9 Buzz 4.3 4.7 6.6 124.9 130.4 19.5 25.9 6.7 62.0 Buzz 8.0 8.3 8.7 120.4 108.5 23.3 68.6 4.7 30.4 Buzz 10.7 11.2 12.6 126.0 131.5 16.3 33.8 8.1 66.5 Buzz 5.7 7.0 8.2 128.3 128.0 13.4 24.8 9.6 78.1 Buzz 4.4 4.5 5.0 113.6 136.8 32.1 97.9 4.3 86.3

Martin et al. 2018 Supplementary Appendix Proceedings of the Royal Society B

Buzz 8.0 9.8 11.3 122.6 133.7 24.9 32.6 5.4 66.2 Buzz 2.1 2.9 5.5 119.3 138.1 24.9 67.5 5.6 52.3 Buzz 4.3 5.9 8.2 119.3 133.2 21.5 57.4 6.2 80.2 Buzz 5.3 7.4 11.3 126.0 135.8 18.9 25.9 7.2 102.1 Buzz 3.4 6.1 12.9 138.4 134.0 21.0 40.5 6.4 71.0 Buzz 1.8 3.0 5.7 122.6 136.0 23.7 52.9 5.7 68.9 Buzz 3.6 6.6 11.3 123.8 131.5 18.5 27.0 7.1 117.7 Buzz 8.0 10.1 11.7 124.9 131.0 12.3 47.3 10.6 71.7 Buzz 3.3 6.2 7.7 128.3 129.4 13.6 30.4 9.5 55.2 Buzz 8.5 8.9 10.7 130.5 134.7 15.8 28.1 8.6 71.5 Buzz 7.0 7.6 12.4 122.6 132.6 31.6 83.3 4.2 57.8 Buzz 3.3 3.5 8.4 137.3 131.5 9.2 28.1 14.3 56.8 Burst-pulse 1.0 1.1 1.1 122.6 169.9 39.1 132.8 4.3 29.5 Burst-pulse 1.8 2.1 2.7 120.4 132.5 34.4 110.3 3.8 35.4 Burst-pulse 1.3 1.6 11.2 99.0 122.3 36.3 126.0 3.4 32.8 Burst-pulse 1.4 1.8 2.1 119.3 127.1 11.3 31.5 11.3 75.5 Burst-pulse 1.4 1.5 1.6 108.0 119.5 23.7 78.8 5.1 57.3 Burst-pulse 1.4 1.4 1.6 124.9 148.6 36.6 88.9 4.1 60.6 Burst-pulse 1.6 1.6 1.8 97.9 120.9 40.1 133.9 3.0 81.9 Burst-pulse 1.6 1.7 1.8 126.0 131.6 29.9 56.3 4.4 57.8 Burst-pulse 5.4 6.7 7.7 122.6 129.2 18.3 28.1 7.1 58.3 Burst-pulse 1.6 1.7 1.8 145.1 121.4 24.4 81.0 5.0 29.0 Burst-pulse 1.4 1.6 1.7 136.1 122.0 30.4 115.9 4.0 26.4 Burst-pulse 1.5 1.5 1.5 126.0 125.6 24.0 31.5 5.2 63.2 Burst-pulse 1.5 1.6 1.6 91.1 95.2 24.7 47.3 3.9 47.6 Burst-pulse 1.6 1.6 1.7 124.9 124.1 19.9 56.3 6.2 34.7 Burst-pulse 1.3 1.3 1.3 112.5 123.6 20.5 66.4 6.0 53.0 Burst-pulse 1.3 1.3 1.6 135.0 135.5 14.3 14.6 9.5 96.9 Burst-pulse 1.6 1.6 1.8 120.4 112.7 16.7 56.3 6.7 40.8 Burst-pulse 1.4 1.4 1.6 120.4 123.9 21.0 60.8 5.9 33.7 Burst-pulse 1.6 1.6 1.7 105.8 111.5 18.4 45.0 6.0 71.0 Burst-pulse 1.5 1.6 1.7 109.1 110.5 24.7 63.0 4.5 55.4 Burst-pulse 1.4 1.4 1.4 115.9 109.4 36.0 68.6 3.0 36.6 Burst-pulse 1.5 1.6 1.9 104.6 105.3 23.3 40.5 4.5 35.1 Burst-pulse 1.3 1.5 1.6 132.8 121.4 32.3 132.8 3.8 32.3 Burst-pulse 2.7 2.9 4.8 109.1 122.9 34.3 91.1 3.6 57.5 Burst-pulse 1.4 1.4 1.5 106.9 151.5 40.9 129.4 3.7 15.5 Burst-pulse 1.4 1.4 1.5 91.1 105.2 28.6 79.9 3.7 40.1 Burst-pulse 1.6 1.6 1.9 111.4 106.8 27.9 61.9 3.8 31.1 Burst-pulse 1.1 1.2 1.7 105.8 121.7 36.3 112.5 3.4 19.6 Burst-pulse 1.6 1.6 1.9 113.6 105.2 32.8 141.8 3.2 24.7 Burst-pulse 1.6 1.7 2.1 112.5 105.8 29.6 137.3 3.6 29.2 Burst-pulse 1.2 1.3 1.5 102.4 102.7 22.9 51.8 4.5 30.0 Burst-pulse 1.4 1.4 1.6 110.3 105.0 19.1 49.5 5.5 29.5

Martin et al. 2018 Supplementary Appendix Proceedings of the Royal Society B

Burst-pulse 1.4 1.4 1.7 118.1 113.5 21.2 57.4 5.3 41.5 Burst-pulse 1.5 1.6 1.8 131.6 117.8 37.8 126 3.1 38.4 Burst-pulse 1.4 1.5 1.5 126.0 115.0 26.2 77.6 4.4 27.6 Burst-pulse 1.5 1.6 1.7 127.1 122.1 20.5 74.3 6.0 19.3 Burst-pulse 1.6 1.6 1.8 105.8 121.4 38.7 118.1 3.1 33.0 Burst-pulse 1.5 1.6 1.7 83.3 89.8 36.6 154.1 2.5 22.7 Burst-pulse 1.3 1.6 1.8 108.0 104.0 28.3 72.0 3.7 67.0 Burst-pulse 1.5 1.5 1.5 74.3 89.2 28.3 84.4 3.2 25.4 Burst-pulse 1.4 1.4 1.5 115.9 115.9 26.9 94.5 4.3 41.3 Burst-pulse 1.2 1.2 1.3 104.6 103.7 22.8 34.9 4.6 79.9 Burst-pulse 1.6 1.6 1.7 103.5 107.6 24.1 76.5 4.5 40.3 Burst-pulse 1.4 1.5 1.6 94.5 95.4 31.1 78.8 3.1 21.4 Burst-pulse 1.6 1.6 1.7 120.4 119.5 18.6 37.1 6.4 44.3 Burst-pulse 1.6 1.7 1.9 111.4 133.3 38.3 124.9 3.5 49.0 Burst-pulse 1.5 1.6 1.6 124.9 128.6 28.7 95.6 4.5 53.7 Burst-pulse 1.4 1.6 1.6 112.5 122.1 35.7 78.8 3.4 46.5 Burst-pulse 1.4 1.6 1.7 103.5 112.8 21.1 52.9 5.3 56.8 Burst-pulse 1.5 1.5 1.6 101.3 116.6 38.5 147.4 3.0 33.7 Burst-pulse 1.6 1.6 1.7 91.1 104.3 38.3 123.8 2.7 43.2 Burst-pulse 1.3 1.5 1.7 130.5 131.8 24.4 25.9 5.4 62.5 Burst-pulse 1.3 1.6 1.7 112.5 113.5 22.7 55.1 5.0 84.2 Burst-pulse 1.4 1.4 1.5 122.6 118.9 23.8 37.1 5.0 49.7 Burst-pulse 1.8 1.9 2.0 74.3 84.2 24.7 73.1 3.4 61.6 Burst-pulse 1.4 1.5 1.5 117.0 124.7 26.1 43.9 4.8 62.0 Burst-pulse 1.8 1.8 2.0 119.3 126.9 25.6 74.25 5.0 95.5 Burst-pulse 1.9 2.3 2.9 126.0 151.2 36.0 102.4 4.2 33.0 BB Click Train 15.4 24.0 34.1 113.6 110.7 37.9 142.9 2.9 27.4 BB Click Train 8.1 13.7 17.8 180.0 172.3 39.1 141.8 4.4 22.1 BB Click Train 13.8 16.0 20.1 130.5 129.2 18.7 23.6 6.9 50.9 BB Click Train 15.0 18.7 27.1 113.6 155.3 41.9 156.4 3.7 27.4 BB Click Train 42.9 53.4 87.3 79.9 101.3 25.5 50.6 4.0 43.4 BB Click Train 16.5 20.5 23.4 142.9 108.3 37.9 129.4 2.9 25.7 BB Click Train 26.2 27.9 30.7 109.1 114.5 21.5 51.8 5.3 31.1 BB Click Train 19.8 25.6 41.2 82.1 84.0 21.0 67.5 4.0 20.7 BB Click Train 64.3 83.8 133.3 54.0 89.8 29.9 118.1 3.0 19.4 BB Click Train 104.0 118.1 254.7 91.1 86.6 29.4 78.8 2.9 34.6 BB Click Train 32.9 79.2 194.4 77.6 88.4 34.2 129.4 2.6 41.2 BB Click Train 57.4 125.7 227.3 82.1 104.0 33.2 108.0 3.1 41.3 BB Click Train 23.6 26.9 28.6 119.3 117.4 16.3 33.8 7.2 46.2 BB Click Train 33.0 43.4 127.2 136.1 129.2 14.9 55.1 8.7 32.3 BB Click Train 7.7 10.7 17.1 112.5 110.9 28.0 81.0 4.0 36.1 BB Click Train 36.5 39.2 48.4 119.3 101.0 23.7 69.8 4.3 69.4 BB Click Train 20.0 21.3 23.1 133.9 118.6 21.2 50.6 5.6 40.5 BB Click Train 52.7 70.1 135.2 121.5 130.5 27.0 81.0 4.8 37.9

Martin et al. 2018 Supplementary Appendix Proceedings of the Royal Society B

BB Click Train 15.6 21.6 75.9 138.4 128.1 27.4 113.6 4.7 46.2 BB Click Train 22.6 31.4 57.6 100.1 91.7 30.4 87.8 3.0 22.6 BB Click Train 86.8 95.1 118.8 108.0 105.5 24.5 72.0 4.3 43.6 BB Click Train 30.2 36.0 47.4 94.5 104.1 30.0 61.9 3.5 33.3 BB Click Train 26.1 29.6 39.5 118.1 131.0 24.5 46.1 5.3 50.4 BB Click Train 30.2 34.7 40.5 79.9 101.3 25.6 50.6 4.0 43.6 BB Click Train 18.8 22.6 24.4 122.6 120.4 19.9 78.8 6.0 46.7 BB Click Train 33.4 41.8 47.0 120.4 129.2 27.9 101.3 4.6 45.1 BB Click Train 3.6 4.1 4.6 92.3 88.8 31.9 117.0 2.8 15.1 BB Click Train 21.3 22.3 24.6 135.0 122.3 27.5 101.3 4.4 13.9