Introduction

Surprisingly large shifts towards a simplification of species composition and ecosystem function, often known as regime shifts, occur within human-dominated landscapes of the terrestrial (Walker 1993), freshwater (Carpenter et al. 1999) and marine realms (Bellwood et al. 2004). The consequences of simplification of species diversity are often well known, generally resulting in less productivity than comparable assemblages with more species (Cardinale et al. 2006; Stachowicz et al. 2007). Yet, there remains less appreciation of change among other ecosystem properties that may yield and maintain such large shifts. Indeed, the question of how regime shifts affect other key ecosystem functions remains relatively understudied. One such function is the biological sounds that ecosystems produce (soundscapes, sensu Krause 1987) as potential orientational cues for dispersing taxa that replenish populations (Simpson et al. 2005; Vermeij et al. 2010; Lillis et al. 2015; Stanley et al. 2015) and as biological cues to evaluate habitat quantity and quality (Piercy et al. 2014).

In human-dominated landscapes, the positive and negative feedbacks that hasten or maintain an ecosystem in a particular state are partly a function of the ability for recovery of inhabitants that comprised the previous state (Scheffer and Carpenter 2003). We propose that soundscapes could act as positive or negative feedback by mediating the arrival of species associated with alternate ecosystem states. A soundscape is a mix of biological, geophysical and anthropogenic sounds that reflect various landscape features and ecosystem processes (Pijanowski et al. 2011). Soundscape ecology remains a relatively young field of research undergoing rapid progress (Servick 2014), with particular value for applied ecology, including the study of regime shifts. The quality of individual signals produced by soniferous animals is affected by stress levels (Buchanan et al. 2003), habitat condition (Laiolo et al. 2008) and climate (Botero et al. 2009; Narins and Meenderink 2014). Hence, ecosystems produce distinct acoustic signatures associated with their biodiversity (Sueur et al. 2008) and their ecological condition such as coral cover (Piercy et al. 2014; Tucker et al. 2014).

Sound represents a multidimensional space that carries decision-making information needed for locating resources (Farina et al. 2011). This information informs both resident and dispersing species about the composition of the local species community, the amount of available resources (such as, food, territory and partners), and the presence of potential predators (Slabbekoorn and Bouton 2008; Farina et al. 2011; Simpson et al. 2011; Parmentier et al. 2015). Sound is one of the most reliable directional cues in the ocean because it propagates to long distances with little attenuation and in a predictable fashion. In the marine realm, a variety of marine larval fish species respond to specific soundscapes in a species-specific manner to mediate two possible settlement strategies. Some species orient towards a habitat by being selectively attracted to its soundscape while others select habitats by avoidance of non-preferred soundscapes (Simpson et al. 2011; Parmentier et al. 2015). From the perspective of a dispersing propagule, the spatial heterogeneity of soundscape can therefore be used as a qualitative and directional guide towards the most appropriate adulthood habitat.

We hypothesized that ecosystems, which shift in state, regardless of the identity and scale of driver, produce distinct soundscapes that reflect their particular state. If regime-shifted soundscapes exist, such a finding would support the proposition that sound may well represent a key function that changes with regime shifts. In this study we focus on regime shifts in kelp forests and seagrass beds as model systems for assessing the consequences of locally and globally driven change. Locally driven shifts were assessed in systems in which nutrient pollution drives shifts from a highly diverse and productive state to a less diverse and less productive state (Neverauskas 1987; Connell et al. 2008; Gorman et al. 2009). Globally driven shifts, such as resulting from ocean acidification, were assessed in systems in which natural CO2 vents mimic future change due to increasing human greenhouse gas emissions (Nagelkerken et al. 2016).

Materials and methods

Study sites

Eutrophication is the main driver of ecosystem regime shifts in marine systems on the metropolitan coast of Adelaide (South Australia) (Connell et al. 2008; Gorman et al. 2009). Rocky reefs in this region are typically dominated by canopy forming kelp forests consisting of Eklonia radiata (Connell and Irving 2008). However, excessive nutrient input into this naturally oligotrophic system has driven a shift from kelp to algal turf domination over the last 40 years (Connell et al. 2008). Kelp forests are highly productive carbon storage systems (Copertino et al. 2005) and create a complex three dimensional habitat with their blades and convoluted holdfasts. Snapping shrimps (family Alpheidae), which are the most ubiquitous and noisy soniferous animal in coastal systems (Johnson et al. 1947a; Knowlton and Moulton 1963), commonly inhabit the holdfasts of these canopy-forming algae (Thiel and Vásquez 2000). A previous study that investigated the crypto fauna assemblages living in E. radiata holdfasts in southern Australia found snapping shrimps present in 26 % of the holdfasts collected (62/240) (Connell unpublished data, see Fowler-Walker et al. 2005 for experimental design).

We identified an area between Horseshoe reef and Port Noarlunga (South Australia) where kelp forests and turf dominated areas are present in close proximity and with comparable abiotic factors (Connell et al. 2008) (depth between 1.5 and 9 m depending on site and tide, sedimentary rocks). In this area we selected three kelp forests (continuous canopy patch for a minimum diameter of ~20 m punctuated by small gaps of 1–2 m) and three algal turf dominated (uninterrupted turf forming algae for a minimum diameter of ~20 m punctuated by small kelp remnants) sampling stations where we made audio recordings (Table S1). Recordings at each control and degraded site were replicated three times during different months (Table S2).

Seagrass meadows (Posidonia spp. and Amphibolis spp.) extensively cover sandy substrate in the Gulf St. Vincent (South Australia). However, excessive eutrophication caused by sewage discharge resulted in large areas of seagrass meadows to be overgrown by epiphytes and ultimately die causing the system to shift to a bare sandy substratum (Neverauskas 1987; Walker and McComb 1992). Based on the most recent (year 2013) seagrass cover maps and aerial images (Hart 2013) we identified an area in front of Semaphore beach where extensive patches of seagrass have receded due to eutrophication (sandy patch minimum diameter: 500 m). Surrounding areas with the same water depth (12–14 m), distance to shore and sediment type (Jason Tanner personal communication) were covered by seagrass meadows that we used as controls. We selected three seagrass and three sand sampling stations for our audio recordings (Table S1). Recordings at each control and degraded site were replicated two times (Table S2).

Natural CO2 vents

Three natural CO2 vents in temperate waters–Vulcano Island and Ischia Island (Italy) and White Island (New Zealand) (Table S1)—were selected to assess the effect of ocean acidification on marine soundscapes, resulting from end-of-century levels of anthropogenic CO2 as forecasted under a business-as-usual emission scenario (Meinshausen et al. 2011) (Table S2). Natural CO2 vents, unaffected by elevated temperatures (Table S3), are present in the northern part of Levante Bay at Vulcano Island (Italy). A pH gradient was present with values that range from extreme (>2000 µatm pCO2) to roughly ambient levels along a 700-m stretch of rocky coast along the northern side of Levante Bay (Boatta et al. 2013; Nagelkerken et al. 2016; Rossi et al. 2016a). We selected an elevated CO2 and a control area (Table S3) along this gradient for our sound recordings, which consisted of three temporal replicates for each area (Table S2). Similarly, at the island of Ischia (Italy), shallow CO2 vents are present around Castello Aragonese where sound recordings were done at two well characterized sites based on previous studies (Hall-Spencer et al. 2008; Kerrison et al. 2011) (Table S1). White Island is a volcanic island located in the Bay of Plenty, North Island of New Zealand. CO2 vents are located along the north-east rocky coast of the island (Brinkman and Smith 2015; Nagelkerken et al. 2016; Rossi et al. 2016a). We selected an elevated CO2 and a control site for the sound recordings, which consisted of two temporal replicates for each site (Table S2). Details of the sampling design are shown in Table S2 and the characterization methods of the carbonate chemistry (as shown in Table S3) are described in the Supplementary methods.

Soundscape recordings and analysis

The most common source of marine biological sounds are from snapping shrimps snaps, sea urchins rasping and fish vocalizations all of which are known to peak at dusk during the so called “dusk chorus” and remain active until dawn (Staaterman et al. 2014). For this reason we kept the sampling time constant across all recordings (i.e. from 1 h before to 1 h after sunset). This sampling regime does not aim to comprehensively characterize the soundscapes of these habitats but rather to answer specific questions related to the effects of regime shifts. To enable comparison among sites with different water depths we positioned the hydrophone (held by hand from a craft) at one meter elevation above the substratum in all recordings with the exception of White Island where the same recording apparatus described above was enclosed in a waterproofed barrel fixed to an anchored buoy for overnight recordings (Rossi et al. 2016a). Here the hydrophone was deployed at 5 meters depth over a rocky seafloor whose sea surface ranged between 8 and 12 m depending on site and tide. The recordings were made with a Hydrophone (HiTech HTI-96-MIN with inbuilt preamplifier, manufacturer-calibrated sensitivity −164.1 dB re 1 V/μPa−1; frequency range 0.02–30 kHz; calibrated by manufacturers; High Tech Inc., Gulfport, USA) and a calibrated audio recorder (PCM-M10, 48 kHz sampling rate, Sony Inc., Tokyo). All recordings were high pass filtered at 300 Hz in order to remove the noise of breaking waves (present in White Island). Recordings of at least 7–8 min were obtained in the field in order to allow for unwanted sounds, such as passing boats, to be excluded from the analyses. For consistency, samples of 5-min duration were used in the analysis for all sites. Recordings were acquired at a sampling rate of 96 kHz and 32 bit. Power spectra and sound levels (RMS) were calculated with the package PAMguide (Merchant et al. 2015) in MATLAB® (Natick, MA, USA) and calibrated in SASLab Pro (Avisoft bioacoustics) using a reference signal at full (1 kHz) scale and the hydrophone sensitivity. Power spectra were calculated using the Hann window (length 1 s and 50 % overlap). The frequency of snapping shrimp snaps was measured on band-pass filtered recordings (2.5–30 kHz) using the pulse train analysis function in the software SAS lab pro (Avisoft bioacoustics), where an artificially generated full scale snap was inserted at the beginning of each recording in order to allow valid comparisons between recordings with different noise levels. Sound pressure level (SPL) and frequency of snaps were averaged across multiple days of sampling for each site. Additional details on the experimental design are shown in Table S2. The waveforms and spectrograms of the recordings were visually and aurally inspected in Audition CS6 (Adobe, San Jose, USA) for other sources of biological sound. Boat noise and artefacts were deleted if present.

Fish and invertebrate hearing range

Sound in water is characterized by two components: sound pressure and particle displacement. All fish are able to detect particle displacement with their otoliths and some are sensitive to both particle displacement and sound pressure due to a series of anatomical specializations such as gas chambers and connections between the swim bladder and the inner ear (Popper and Fay 2011). It is worth noting that some fish species can detect sound pressure even if not equipped with such anatomical specializations, suggesting that our morpho-functional understanding of fish hearing is still incomplete (Popper and Fay 2011). The same is true for invertebrates where often, despite showing behavioural and physiological evidence of hearing ability, we cannot associate this sensitivity to any obvious anatomical specialization (Montgomery et al. 2006). Studies suggest that the sound pressure sensitivity in fish ranges widely from a few Hertz to 1, 5 kHz or even higher in some species (Montgomery et al. 2006; Popper and Fay 2011). In order to capture this wide range of sensitivities we calculated the sound pressure levels in our recordings in three overlapping frequency bands (<1, <3,  <5 kHz).

Statistical analysis

The sound pressure level in the hearing range of fish (<1, <3 and <5 kHz), full bandwidth (<10 kHz) and the frequency of snapping shrimp snaps were averaged across days of sampling and then tested for the factor habitat condition with one-way ANOVAs using spatial replication (n = 3 sampling stations per treatment for each of the rocky reef and seagrass system) (Table S2). For ocean acidification, the three natural CO2 vents served as spatial replicates. In order to allow multiple comparisons (3) the Bonferroni correction was applied and α adjusted from 0.05 to 0.017. The interaction between the factor stressor (nutrient/CO2) and habitat condition (control/degraded) was tested with a two-way ANOVA on data standardized by habitat. Each habitat group (kelp, seagrass and mixed) was standardized to its mean before the computation. The link between full bandwidth sound pressure level and frequency of snapping shrimp snaps was analysed with a simple regression.

Results

Regime shifts triggered by nutrient pollution and ocean acidification correspond to reduced levels of sound pressure in kelp-dominated and mixed macroalgae-seagrass habitats (nutrient pollution on kelp forests, ANOVA, F(1,4) = 39.3, p = 0.003, ocean acidification on macroalgae-seagrass, ANOVA, F(1,4) = 24.7, p = 0.008) (Fig. 1a). Likewise, the rate of snapping shrimp snaps decreased following the same pattern suggesting that snapping shrimp sound is a key component of biological sound in both nutrient driven change to kelp forests (ANOVA, F(1,4) = 150.2, p < 0.001) and ocean acidification driven change in macroalgae-seagrass habitats (ANOVA, F(1,4) = 32, p = 0.005) (Fig. 1b). Despite the low levels of biogenic noise that characterized seagrass beds, nutrient pollution caused a reduction in the rate of snapping shrimp snaps and a marginal reduction in level of full bandwidth sound pressure (snaps, ANOVA, F(1,4) = 103.1, p < 0.001, full bandwidth sound pressure level, ANOVA, F(1,4) = 8.34, p = 0.045, α = 0.017 with Bonferroni correction). The reduction in number of snaps and full bandwidth sound pressure level was independent of the type of stressor when habitat specific differences were standardized (ANOVA, snap frequency: condition × stressor interaction, F(1,14) = 0.55, p = 0.46, sound pressure level: F(1,14) = 0.55, p = 0.46). Regression analysis incorporating all recordings across habitat types, seasons, and locations showed that the variation in rate of snapping shrimp snaps significantly explained 84 % of the variation (F(1,41) = 271.8, p < 0.001) in full bandwidth sound pressure level (Fig. S1).

Fig. 1
figure 1

Responses of soundscapes to regime shifts a full bandwidth root mean square sound pressure level (>300 Hz, <10 kHz) ± SE. b Mean rate of snapping shrimp snaps counted per minute ± SE. Error bars represent variability due to spatial replication (N = 3 per treatment). Stars indicate significant differences between bar under the brackets

Full size image

The sound pressure level was also analysed for three frequency bands relevant to fish hearing: 1, <3 and <5 kHz. Sound levels were reduced in the frequency range <5 kHz due to regime shifts triggered by nutrient pollution and ocean acidification on rocky and mixed rocky-seagrass bottoms (kelp forest, ANOVA, F(1,4) = 35, p < 0.004, ocean acidification, ANOVA, F(1,4) = 76.7, p < 0.001) but not on seagrass beds (Fig. 2). In the two lower frequency ranges, however, the sound levels did not decrease significantly due to habitat degradation. Power spectral density plots for nutrient enriched habitats are included in the supplementary materials (Fig. S2, while those for CO2 enrichment are published elsewhere (Rossi et al. 2016a).

Fig. 2
figure 2

Effect of regime shifts and ocean acidification on sound levels in three frequency bands relevant to fish hearing: a <1, b <3, and c <5 kHz. Error bars represent variability due to spatial replication (N = 3 per treatment). Asterisks indicate significant differences between bars under the brackets

Full size image

Discussion

We propose a new concept; that regime shifts, regardless of their driver, produce distinct soundscapes. We found that although each ecosystem was associated with distinct soundscapes, the shift from diverse and structurally complex macroalgae like kelp, to low lying turfs (Connell et al. 2008) or from seagrass to sand (Bryars et al. 2011), was associated with a reduction in biological sound. Importantly, although these ecosystem shifts are caused by resource enhancement, the driving mechanisms are different; i.e. local nitrogen enrichment (seagrass, Bryars et al. 2011; kelp, Gorman et al. 2009) versus CO2 enrichment at natural vents (kelp, Connell et al. 2013; kelp and seagrass, Nagelkerken et al. 2016). It was striking that regardless of the driver of regime shift (eutrophication or ocean acidification) the magnitude of changes in sound was remarkably similar. It is possible that despite the diversity in drivers of regime shifts, they not only result in similar ecological outcomes of reduced species diversity and productivity, but also reduced soundscapes and the orientational cues they offer. Common to regime shifts is a reduction in resources (Cardinale et al. 2006; Stachowicz et al. 2007), such as food and habitat to inhabiting species, which dispersing animals may be able to detect. As underwater coastal sounds in both temperate and tropical areas are largely biological (Johnson et al. 1947b), the degradation of the habitat is reflected as an impoverished soundscape. This also suggests that soundscapes are a promising, rapid and cost effective monitoring tool for ecosystem condition (Piercy et al. 2014; Tucker et al. 2014).

Soundscapes were dominated by snapping shrimp sound. Despite being more intense on rocky reefs than soft sediment, snapping shrimp crackle was not only ubiquitous, but it also dominated the soundscapes of all types of habitat and their shifts to lower diversity and complexity. This predominance is evidenced by the correlation (84 %) between the rate of snapping shrimp snaps and the total loudness of all seascape sounds. Kelp forests are not only more productive than turf dominated reefs (Copertino et al. 2005), but they also create habitat for cryptic animals like snapping shrimps with their convoluted holdfasts (Fowler-Walker et al. 2005). Similarly, soft sediment covered by seagrass has a much higher productivity compared to bare sand and creates habitat and shelter for a number of organisms (Hemminga and Duarte 2000). Although the quantification of the density of shrimps is difficult due to the highly cryptic nature of these animals, their snapping rate and intensity (i.e. within their main frequency band) are commonly used as an index of their abundance (Kennedy et al. 2010; Lillis et al. 2014; Piercy et al. 2014; Staaterman et al. 2014; Nedelec et al. 2015). In both kelp and seagrass ecosystems, a decrease in snapping reflected the regime-shifts that result in a simplification of habitat complexity and loss of habitat (i.e. kelp holdfasts).

The ecological implications of impoverished biological soundscapes may depend on the extent to which a species relies on sound for navigation. Time of day may also play a role as the effects found in this study are likely to be less marked during daytime when snapping shrimps are less active. Many marine vertebrates (Simpson et al. 2005; Montgomery et al. 2006; Huijbers et al. 2012; Parmentier et al. 2015) and invertebrates (Vermeij et al. 2010; Lillis et al. 2015; Stanley et al. 2015) can use sound as directional cue towards settlement habitats or away from undesired habitats. Based on the reduction of soundscapes measured, we suggest that those organisms that rely more heavily on higher frequency sound (i.e. snapping shrimp sounds >500 Hz) are likely to be more penalized by shifted soundscapes. However, this effect remains hard to quantify because of our incomplete knowledge about sound pressure hearing sensitivities in fish and invertebrate larvae and the complexity of sound propagation in shallow waters.

Models suggest that ocean acidification levels expected by the end of the century will significantly decrease sound absorption in the ocean (Ilyina et al. 2010), potentially compensating for the loss in biogenic sound caused by ocean acidification. (Ilyina et al. 2010) predict that sound absorption will be attenuated at <10 kHz and will be most pronounced in the lower part of this range. However, by considering all important energy loss mechanisms the effect of pH on sound absorption is negligible in shallow waters and at the distances likely to be relevant for larval orientation (<10 km) because geometric spreading accounts for most of the sound loss from the source (Au and Hastings 2008; Reeder and Chiu 2010).

We propose that if regime shifts produce distinct soundscapes that reflect their particular state, sound may represent a functional change of regime shifts. Shifted soundscapes may act as one of the mechanisms that control resistance or resilience of ecosystems, notably where sound mediates the supply of keystone species. For example, shifted soundscapes could hamper the recovery of ecosystems by directly attracting a lower number of engineer species’ propagules such as coral and oyster larvae (Vermeij et al. 2010; Lillis et al. 2015). Analogously, chemical cues from coral or seaweeds reinforce the persistence of coral or seaweed dominated alternative states respectively by acting as attractor or repellent for dispersing coral and fish larvae (Dixson et al. 2014).

Another example is that of simplification of soundscapes could lead to reduced supply of herbivores that resist ecosystem shifts (e.g. Ghedini et al. 2015), and that would act as negative feedbacks to stabilise the shifted state characterised by low herbivory (i.e. resistance to positive change: Connell and Ghedini 2015). Shifted soundscapes could also reduce the settlement of keystone predators, such as lobsters (Ling et al. 2009; Stanley et al. 2015), which are crucial to limit catastrophic overgrazing by sea urchins and maintain the system in its kelp-dominated state. Either way, we propose that if shifted soundscapes exist as a functional ecosystem trait, then the possibility that they function as positive or negative feedbacks to dispersing taxa that govern regime shifts also exists.

Dispersing propagules may be able to compensate for lowered soundscape value by using alternative cues such as olfaction and vision (Dixson et al. 2010; Ferrari et al. 2012). However, these cues have a limited reach relative to sound which is one of the cues that propagate the furthest and most predictably with distance from the source (Montgomery et al. 2006; Simpson et al. 2013). Furthermore, the value of these cues is not immune to habitat degradation and is likely to similarly decrease (Dixson et al. 2014). This will be further exacerbated in the future, where regardless of the type of cue used, ocean acidification will disrupt propagule orientation by causing ineffective processing of all these sensory cues (Munday et al. 2009; Dixson et al. 2010; Ferrari et al. 2012; Rossi et al. 2015, 2016b).

In conclusion, the simplification of habitats within human-dominated landscapes has similarly profound effects on their soundscapes. The specific consequences remain elusive, but this concept is likely to attract considerable attention given its potential to mediate a diversity of key ecological processes and become a new cost effective monitoring tool. Importantly, our study reveals that regime shifts within human-dominated landscapes impoverish natural soundscapes, tending towards ‘sounds of silence’.