Characterization of the acoustic community of vocal fishes in the Azores

Introduction

Acoustic communication in fishes is widespread and occurs in different environmental and behavioural contexts (Zelick, Mann & Popper, 1999). More than 800 fish species are known to produce sounds (Fish & Mowbray, 1970; Rountree et al., 2006). These sounds are species-specific and are associated with courtship, spawning, parental care, feeding, aggressive or territorial behaviours, and can be effective indicators of fish species richness and diversity in biological processes (Fine, Winn & Olla, 1977; Amorim et al., 2008; Lobel, Kaatz & Rice, 2010). The majority of fish sounds are low frequency (<3 kHz, mostly <1 kHz) and are made up of repetitive elements such as sound pulses (Amorim, 2006). Sounds made by fish species differ in fundamental frequency, dominant frequency, number of pulses and frequently in pulse period (Amorim et al., 2008; Colleye et al., 2011; Ladich, 2013). Fish sounds also differ from those made by other marine organisms such as cetaceans. Fish sounds frequently have a short duration (<5 s), are made of broad-band pulses, and often present multiple frequency harmonics (Fish & Mowbray, 1970). Sounds produced by cetaceans range from low frequencies (<1 kHz) up to 200 kHz, frequently have a long duration (>10 s), and often exhibit strong frequency modulation (Richardson et al., 1995). For example, Odontocetes produce a wide variety of whistles and broadband sounds (clicks) with main energies ranging from a few kHz (thus well detectable by humans) to ultrasonic frequencies of >100 kHz for echolocating prey and screening the environment. Baleen whales produce high intensity, lower frequency sounds (mostly <1 kHz, but can reach 24 kHz) lasting less than 1 s to over 30 s, spanning from simple growls to loud complex modulated songs (Mellinger et al., 2007; Au & Hastings, 2008). Fish sounds also differ from invertebrate sounds; the latter are typically broadband pulses with frequencies between 2 and 12 kHz (Au, 1998; Radford et al., 2008; Bittencourt et al., 2016), although some invertebrates may also produce much lower frequency sounds (e.g.,  Staaterman et al., 2011; Di Iorio et al., 2012). These differences allow discrimination of fish sounds in marine soundscape studies and offer a non-invasive way (through acoustic monitoring) to assess biodiversity of acoustic communities (Farina & James, 2016). Here, we consider the definition of acoustic community proposed by Farina & James (2016), i.e., a temporary aggregation of species that interact acoustically either in aquatic or terrestrial environments, producing sounds with internal or extra-body tools.

Owing in part to cost reduction and to technological improvements, Passive Acoustic Monitoring (PAM) of fish sounds has been increasingly used (e.g.,  Wall, Lembke & Mann, 2012; Tricas & Boyle, 2014; Rice et al., 2016). Besides having a negligible impact on biota, PAM supports long-term field studies of seasonal activities, and has been shown to be a useful tool in the conservation and management of vocal species (Mann & Lobel, 1995; Luczkovich, Mann & Rountree, 2008; Parmentier et al., 2018). In addition, monitoring the diversity of sounds made by marine organisms, including fish, can help assess ecosystem health (Bertucci et al., 2016). But PAM also presents limitations in its effectiveness. PAM will only record animal sounds when their sound pressure levels (or particle motion levels) are higher than the ambient noise levels, which means that taxa that produce low amplitude vocalizations (e.g., gobies; Lugli & Fine, 2003) will not be detected. Most importantly, there is a paucity of data on the sounds made by fish in their natural habitat and numerous fish sounds have not yet been identified (Tricas & Boyle, 2014; Ruppé et al., 2015; Mouy et al., 2018).

Seamounts and nearshore bank areas of the Azores are important hotspots of marine biodiversity, harboring fish species of commercial and conservation interest (Pitcher et al., 2007), several of which are known to vocalize (e.g., dusky grouper Epinephelus marginatus, ocean sunfish Mola mola, tarpon Megalops atlanticus, (Fish & Mowbray, 1970; Bertucci et al., 2015). Despite its large fish biodiversity (about 460 marine fish species were recognized in the Azores, (Santos, Porteiro & Barreiros, 1997), only 20 species present in waters of this archipelago have been reported as soniferous (Table 1) suggesting that more remain to be identified. Because seamounts are subjected to increasing anthropogenic pressure, caused by overfishing and marine pollution (e.g., plastic and noise pollution), it is paramount to develop effective tools, such as PAM, to monitor ecosystems in support of conservation and management. With this objective in mind we (1) listed the known vocal fish species occurring in Azorean waters and compiled their sounds; (2) characterized new sound types and proposed their association to known fish sounds; and (3) built a fish sound database to be used as a steppingstone both for future research and for conservation and management.

Materials & Methods

Acoustic recordings and analysis

Ecological Acoustic Recorders (EARs, Lammers et al., 2008) were bottom-moored on the Condor seamount at an approximate depth of 190 m, 5–10 m from the seafloor, and at a depth of 36 m in Princesa Alice bank, on the seafloor. The EAR is an autonomous acoustic recorder provided with a Sensor Technology SQ26-01 hydrophone that has a flat frequency response (±1.5 dB) from 18 Hz to 28 kHz and a response sensitivity between −193 and −194 dB re 1 V/µPa. (varying between deployments) at Condor and −193.6 dB at Princesa Alice.

From the available deployments, 10 at Condor and 4 at Princesa Alice, we selected recordings from three in Condor (deployments 2, 7 and 10) and from one in Princesa Alice (deployment 3), based on duration and recording quality. The EARs were programmed to record on duty cycles: Condor deployment 2 and 7, 90 s of sound recorded every 900 s at a sampling rate of 50 kHz; deployment 10, 3,600 s every 12,600 s, at 2 kHz; Princesa Alice deployment 3, 90 s every 900 s, at 50 kHz.

Recordings from the following months were analysed. Condor: April, June, August and November 2010; June and August 2012. Princesa Alice: June 2010. From these recordings the Acoustic Complexity Index (ACI) was calculated using the plug-in SounsdscapeMeter in the WaveSurfer software (Pieretti, Farina & Morri, 2011). ACI presents higher values in sounds with inner variability over time (as sounds of biological origin) and low values for more constant sounds (e.g., passing vessels) (Pieretti et al., 2017). Within each month, a subsample of 5 days (24 h periods) was selected from those presenting higher ACI values, expected to be associated with a higher biophony (Pieretti, Farina & Morri, 2011; Bolgan et al., 2018).

Sound recordings were downsampled to 8 kHz and analyzed with Adobe Audition 3.0, both aurally and visually with spectrograms (FFT 2048 points, Hamming window, frequency range up to 4 kHz, fixed display settings). To inspect low amplitude sounds, a 25 dB amplification was used. Fish sounds were discriminated based on their similarity to reported fish calls, in frequency, relative energy level, duration, and timing (e.g., Parsons et al., 2016a; Parsons et al., 2016b). Because most sounds occurred in sequences and since in many sounds we could not analyse its fine structure due to the likely distance of the sound emitter to the hydrophone, we focussed on identifying sound sequence types rather than sound types. The following parameters were measured for seven types of sound sequences that had at least 14 occurrences, and were identified as fish calls: sequence duration (time elapsed from the start of the first sound in a sequence to the end of the last sound, s); number of sounds; sound duration (the mean duration of a sound in a sequence, s); sound period (mean time elapsed between the peak amplitude of two consecutive sounds within a group, s); peak frequency (frequency at which the sound presents its highest energy in the power spectrum, Hz); minimum and maximum frequency (the lower and the higher frequency of each sound in the spectrogram, Hz) and signal to noise ratio—SNR (ratio between the sound Root Mean Square (RMS) amplitude and the background noise RMS amplitude). Each sound sequence (Fig. 2) was defined based on frequency, duration and temporal patterns. Raven Pro Sound Analysis Software 1.5 (Cornell Lab of Ornithology, USA) was used to measure the temporal acoustic parameters. WaveSurfer was used to draw spectrograms and oscillograms. Fourteen to 20 sound sequence per sound sequence type with SNR > 1.03 were selected and characterized. In addition, 1 sound per soniferous species present in Azores was also characterised and used for comparison with sound sequences. Of the 20 known soniferous species (Table 1) only 9 species were included in this analysis (see Figs. 3 and 4) due to the lack of available sounds. Pomatoschistus pictus was not considered since it inhabits shallow waters and its sounds can only be recorded when a fish is very close (a few centimetres) to the hydrophone, thus being unlikely detected by the EARs. Note that caution should be taken when doing this comparison as we only considered one sound per identified vocal species and the analysis is thus not considering intraspecific variability.

We used the software PRIMER 6.0 to explore multivariate similarity profiles among the seven sound sequences and sounds from identified species using a Bray–Curtis Similarity matrix, followed by a Cluster plot and a non-metric multidimensional scaling approach (nMDS) with a 2D Stress of 0.13. The nMDS and the cluster plot were derived from the similarity matrix to evaluate similarities among sound sequences and sounds from known species and to investigate the acoustic parameters that contribute to those similarities (Clarke & Gorley, 2006; Sun et al., 2013). We used a similarity profile test (SIMPROF; Clarke & Warwick, 2001; Clarke, Somerfield & Gorley, 2008) to determine significant differences between the clusters.

The variables used in these multivariate analyses were peak frequency, maximum frequency, minimum frequency, sequence duration, sound duration and number of sounds. Variables were standardized prior to analyses. For this effect the mean was subtracted from each data point and divided by the standard deviation. Two units were further added to eliminate negative values that were not accepted by the analysis.

Since the relation between fish sound abundance and diversity has not been validated for the present study sites, we compared mean ACI values (for the frequencies 15–2,000 Hz) between two months with contrasting fish sound abundance and diversity (August 2009 and June 2010 for Condor).

Results

Discussion

Many fish species produce species-specific sounds (Amorim, 2006; Fine & Parmentier, 2015) that can be detected (Vieira et al., 2015) and discriminated (Lillis, Eggleston & Bohnenstiehl, 2014) from other sound sources present in aquatic soundscapes. In this study several fish species present in the Azorean archipelago were identified as sound producers (20 species from 14 families) or potential sound producers (79 species from 24 families) based on the literature. Consistently, we found a considerable diversity of putative fish sounds during the analysis of acoustic recordings from Princesa Alice and Condor, contributing for the marine soundscape of the surveyed Azorean ecosystems. We described 20 sound sequences. From these, seven were sufficiently abundant to be characterized for several acoustic parameters.

Exploratory multivariate analyses based on the Bray-Curtis Similarity index agree with an ad-hoc classification made through comparison of the sound spectrograms and by listening to the sounds, suggesting this method produced a valid classification. The only exception being sound #7 that presented similarities with sounds #1 and #10 regarding their spectrograms (Figs. 3 and 4). This suggests that these three sound sequences might belong to closely related species, the same species or even be variants of the same sound type. The remaining sounds are more likely to be produced by different species. In particular sound #12, with energy at higher frequencies, appears to be quite different from the other groups. We cannot ensure, however, that all the identified sound sequences were made by fish, although they exhibited general characteristics common in fish sounds, i.e., they were low frequency short duration sounds, with no frequency modulation, and temporal patterns within the range of other reported sounds (e.g., Fish & Mowbray, 1970; Amorim, 2006; Parsons et al., 2016a; Parsons et al., 2016b).

Previous studies have shown that the Acoustic Complexity Index (ACI) is a useful tool to track variations in the soundscapes (e.g., Pieretti et al., 2017). Our results showed that higher values of the ACI might not indicate a higher abundance and diversity of fish sounds since we did not observe a relation between the abundance and diversity of fish sounds and the ACI values (Fig. 5). In fact, the ACI apparently increased in the presence of cetacean sounds or boat noise. Although this index is known to respond well to biological sounds in recordings with low background noise, it may fail in the presence of anthropogenic or environmental noise (Lin, Fang & Tsao, 2017), under continuous noise such as that created by fish choruses (Bolgan et al., 2018), or in the case reported here where the fish sounds are sparse and with a low signal to noise ratio. In fact, better results can be obtained when using ACI in combination with other methods (Phillips, Towsey & Roe, 2018). This paper supports the approach that multiple acoustic indices are required to understand a soundscape.

All 20 sound-producing fish species present in the Azores partially or fully overlap their depth distribution range with the study sites. Of the 79 species occurring in the Azores that potentially produce sounds only six are unlikely to be found in either study sites due to their depth distribution range, all others thus being potential source candidates for the recorded fish sounds. Known acoustic detection distances are usually short for fish, so detected sounds can be assumed to have been produced relatively close to the recorders. Indeed, fish sound propagation distances can vary from a few centimeters (e.g., gobies; Lugli & Fine, 2003) to a few meters (e.g., toadfish and damselfish; Fine & Lenhardt, 1983; Myrberg Jr, Mohler & Catala, 1986; Alves, Amorim & Fonseca, 2016; sweeper; Radford et al., 2015). Exceptions are sciaenids which have been estimated to be detectable from tens to few hundred meters (Locascio & Mann, 2011; Parsons et al., 2012). Most of the above acoustic detection distances, however, have been reported for fish calling in shallow water, thus facing strong propagation constraints due to the frequency cutoff phenomenon (Rogers & Cox, 1988). In deeper waters such as Condor, fish sounds will likely propagate to longer distances than in shallow water (Mann, 2006) but proximity to the recorder will still be a constraint for monitoring most soniferous species.

Can any of the described sound sequences be attributed to a known vocal fish species? Comparisons made between the reference fish sounds and the seven recorded sound sequences (Figs. 3 and 4) must be considered with care since only one sound from each species was available. We found some resemblance in the analysis between some sound sequences (e.g., #4 with Seriola dumerilli, sounds #1, #5 and #7 with Epinephelus marginatus, the sound #10 with Dactylopterus volitans) but a more careful inspection revealed that none of these species sounds matched the identified sound sequences. Sound sequences #1, #5, #7 and #10 displayed similarities with sounds produced by species of Batrachoididae, which generate tonal sounds with several harmonics and dominate soundscapes in different habitats across the world (Bass & McKibben, 2003; Maruska & Mensinger, 2009; Rice et al., 2016). Sound sequences #1, #5 are similar to grunt trains, #10 resembles boatwhistles from toadfishes and #7 resembles boatwhistles followed by grunts trains. For example, advertisement calls of Opsanus beta are composed by grunt and tonal elements (Thorson & Fine, 2002). No Batrachoididade species are known for the Azores, though Halobatrachus didactylus occurs in the eastern Atlantic, and both Opsanus tau and Opsanus phobetron in the western Atlantic (Amorim, 2006; Amorim, Simões & Fonseca, 2008; Fine, 1978). Sound sequence #1 sounds like an Ophidiiform (M Bolgan, pers. comm., 2019). Sound sequence #12 (duration: 0.56 s, peak frequency: 1,175 Hz; Table 4) exhibits similarities with the /kwa/ recorded in Posidonia oceanica meadows in the Mediterranean Sea, recently attributed to Scorpaena spp. (0.27s duration and 747 Hz peak frequency; Di Iorio et al., 2018; Bolgan et al., 2019). This sound presents several acoustic features which are typical of the “kwa” sound. In particular, peak frequency is always higher than 600 Hz, the pulse envelop presents a highly stereotyped amplitude modulation where cycle period corresponds to peak frequency; finally, the sound presents a typical pseudo-harmonic structure where the pseudo-harmonic interval corresponds to the inverse of the pulse period (Di Iorio et al., 2018; Bolgan et al., 2019). Interestingly sound sequence #12 also shows some resemblance in frequency and duration with the chatter sound made by cusk-eels Ophidion marginatum (Ophidiidae) (Sprague & Luczkovich, 2001; Mann & Jarvis, 2004), although they do not seem similar to the human ear. Both members of the Scorpaenidae and Ophidiidae can be found in Azores (Table 2).

Although we could not suggest a potential candidate for sound sequence #15, similar sounds were also found during boat-based recordings around Faial Island at depths between 2–10 m (R Carriço, pers. comm., 2017) indicating that its emitter may also inhabit shallow waters. Sound sequences #1, #2, #3, #4, #5, #7, #10 and #15 were recorded in both study locations, suggesting a wide depth distribution. In contrast, sounds #6 and #8 were recorded only in Condor (190 m) while sounds #12, #14, #17, #22, #28, #35, #38, #47 and #48 were recorded only in Princesa Alice (36 m) indicating that these soniferous fishes may be somewhat stenobathic.

Potential vocal fish species present in Condor at the studied depth (Table 2) produce sounds similar to the sound sequences that were recorded. For example, Phycis phycis, Phycis blennoides, Anthias anthias, and Molva macrophtalma are potential source species for the calls #6 and #8, while Pagellus bogaraveo and Helicolenus dactylopterus are potential source species for the sound sequences #1, #2, #3, #4, #5, #7, #10 or #15 (Menezes & Giacomello, 2013). On the other hand, Sphyraena viridensis and Seriola rivoliana have confirmed presence in Princesa Alice (Fontes & Afonso, 2017), being eventually potential source species for the sound sequences #12, #14, #17, #22, #28, #35, #38, #47 or #48.

Conclusions

The present study highlights the wealth of fish sounds awaiting to be described and challenges associated with PAM. For example, only some soniferous fish species produce loud conspicuous sounds easily detectable by PAM. Also most fish sounds are still unidentified making it difficult to identify the sources of most fish sounds detected with PAM. However, these monitoring studies can contribute to evaluate fish presence and abundance, to identify spawning seasons of species of commercial, conservation and scientific interest. It can also be used to investigate fish community structure (Harris, Shears & Radford, 2015; McWilliam & Hawkins, 2013). To increase the effectiveness of PAM, basic research is needed on sound source identity, behavioral context of sounds production, and spatial and temporal distribution of the sounds/species (Rountree et al., 2006; Sirovic et al., 2009; Wall et al., 2014). Two important approaches to overcome these main challenges are coupling PAM with in situ visual monitoring techniques (e.g., Mouy et al., 2018), and recording more species in laboratory conditions.

Our results provide baseline data on a collection of sounds, contributing to building a comprehensive open access library of both identified and unknown fish sounds that will boost the usefulness of PAM.

Supplemental Information