Acoustic impact of a wave energy converter in Mediterranean shallow waters

In this study, underwater noise from a full-scale wave energy converter system (ISWEC), installed on the coast of Pantelleria Island (central Mediterranean Sea), was characterized. The noise was measured using an autonomous acoustic recorder anchored to the sea bottom 40 m from the ISWEC hull. Acoustic monitoring continued for 15 months, starting 7 months before (PRE), 2 months during (INST) and 6 months after the ISWEC installation (POST). The levels of noise, assessed with power spectrum density and octave and third-octave band sound pressure levels (BSPLs), were higher during the POST period than during the PRE period at lower frequencies up to 4 kHz and increased with wave height. During the ISWEC activation for energy production (POST_ON) in the wave height range 1–2.9 m, the BSPLs increased much more at lower frequencies up to 4 kHz (the median BSPLs at 63 Hz for the PRE, POST, and POST_ON conditions were 73, 106, and 126 dB re 1μPa, respectively). Considering the biophonies that make up the soundscape of the area, we examined the possible masking of fish choruses due to ISWEC noise and highlighted that at a distance of 1000 m, the 800 Hz peak frequency was 10 dB above the ISWEC signal. Within this distance from ISWEC, a possible masking effect is supposed.


Results
Noise levels before and after the IsWeC installation. Table 1 Table 1). At higher frequencies, the correlation between wave height and BSPL became weakly negative (except for the INST condition) with generally low values ( Fig. 2 and Table 1). Table 2 shows the median (25 th -75 th percentiles) BSPLs for different condition considering three wave height ranges (0-0.9 m, 1-2.9 m, and 3-6 m) and results of Kruskal-Wallis multiple comparison test to assess significant differences in BPLs between condition (PRE, INST, POST, POST_ON). The differences are all significant for the www.nature.com/scientificreports www.nature.com/scientificreports/ lower frequencies up to 1000. In Fig. 3, the median (25 th -75 th percentiles) BSPLs are plotted for all conditions (PRE, INST, POST, POST_ON) and for the three wave height ranges. The differences in the BSPLs among the conditions decrease at higher frequencies (Fig. 3).  (Table 2). Above 4 kHz, the differences in the BSPLs for different conditions decrease, and all the median values are between 100 and 110 dB (Fig. 3). Figure 4 shows the power spectral density (dB re 1 μPa 2 /Hz, 40 m distance) of PRE and POST condition (Fig. 4, Top) and the ISWEC during energy conversion at different flywheel speeds (Fig. 4, Below). The peak frequency and its amplitude change with the speed of the flywheel. The peak frequencies are below 50 Hz and reach 140 dB for a flywheel speed of 334 rpm. The sound pressure levels at different flywheel speeds of 334, 236 and 146 rpm are 143, 137, and 128 dB (SPL, dB rms re 1 μPa, 40 m distance, a wave height of 2.3 m, and a frequency band of 20-96000 Hz). The source levels (SL, dB rms re 1 μPa, 1 m distance, a wave height of 2.3 m, and a frequency band of 20-96000 Hz) corresponding to these flywheel speeds were assessed using the transmission loss model and are 173, 167, and 158 dB, respectively. Scatter plots for the band sound pressure level (BSPL, dB rms re 1 μPa) versus wave height at different frequencies in octave bands from 1/3 63 Hz up to 500 Hz. Each point represents the mean BSPL value calculated for each file 2 minutes long. The blue and red points indicate the BSPL values before (PRE, n = 1708) and after (POST, n = 1120) the ISWEC installation, respectively. The black points indicate recordings during the energy conversion (POST_ON, n = 24). The lines represent the linear regression equations.
www.nature.com/scientificreports www.nature.com/scientificreports/ ISWEC masking of fish choruses. Figure 5A shows the PSD calculated for different conditions (ISWEC_ Post_ON speed of 295 rpm, fish choruses, background noise and an example of a boat passage). The energy of the fish choruses is associated with frequencies between 600 and 1100 Hz. During energy conversion, at the peak frequency of the chorus (800 Hz), the ISWEC PSD is higher than the PSD of the fish at approximately 4 dB (see the zoom in Fig. 5B). However, the noise produced by a boat passage is higher than that of both the ISWEC and fish choruses from 100 Hz up to 10 kHz (Fig. 5A).
Implementing the transmission loss model for the ISWEC Post_ON condition, we found that at a distance of 1000 m from the ISWEC (Fig. 5B), the noise generated is strongly reduced (see blue dashed line). In detail, at that distance, the PSD of the noise produced by the ISWEC during energy conversion has a decrement of 13 dB at 800 Hz and is 10 dB below the peak frequency of the fish choruses (Fig. 5B).

Discussion
Wave energy devices and the associated cables could act as a potential source of cumulative stressors on aquatic ecosystems 23,35 . Anthropogenic activities can increase ambient noise 23,26 , and operating wave energy converters contribute to vibrations and low-frequency long-duration noise. Because ambient noise increases with increasing wind speed and wave height and wave energy converters act with moving water masses, we first evaluated the increase in noise in relation to wave height. We found that noise increases more with wave height during the    www.nature.com/scientificreports www.nature.com/scientificreports/ mooring installation (INST) and in the presence of the ISWEC infrastructure (POST) than without the ISWEC (PRE) (see Table 1 and Fig. 3). This is probably due to the anchoring system consisting of metallic chains that are more stimulated by higher waves (Figs 7, 10). In the presence of the ISWEC infrastructure (POST), the level of noise is higher than the background noise (PRE), for frequencies up to 4 kHz. Above 4 kHz, the differences between BPLS (PRE vs POST) decrease even if they are still significant (see Fig. 3 and Table 2).
The process of energy conversion (Post_ON) provided additional noise compared to the POST condition at lower frequencies up to 8 kHz, with a maximum difference of 20 dB for the 63 Hz BSPL (see Fig. 3 and Table 2). Therefore, the noise of the infrastructure is produced both by the vibrating bulkhead with its anchoring system and by the activation of the energy conversion system. During energy conversion (POST_ON), the noise changes with flywheel speed, with a maximum SPL corresponding to the maximum speed considered (334 rpm; SPL 143 dB rms re 40 m and a peak frequency of 140 dB at 50 Hz; see Fig. 4).
On Pantelleria Island (Fig. 8), the ISWEC was installed through a complex chain system to maintain a steady infrastructure. The noise generated by this system occupies frequency bands up to 8 kHz (see Figs 3,10). Instead, noise emitted during energy conversion, likely generated by the hydraulic pump and gyroscopic units, overcomes the ambient noise by at least 53 dB at the frequency of the 63 Hz BSPL (see in Fig. 3 and Table 2 the differences between the PRE and POST ON conditions, wave height range 1-2.9 m).
A number of recent works have shown that noise recorded from wind turbines and wave energy converters is audible by marine species such as fish 36,37 , crustaceans 38,39 , and pinnipeds 5 , but this noise is probably out of the audible zone of toothed whales 5,40 . The effects of the generated noise depend on the sensitivity of each species, their ability to habituate to the noise, and their behavioural state. In this work, the position of the ISWEC near the coastline and harbour is probably outside the migration routes of whales 41 , and the noise generated by the ISWEC overlaps with high-intensity fish choruses occurring in the summer season. Whalberg et al. 37 highlighted that the extent of the impact of sound pressure or the relative particle motion on fish communication, behaviour and fitness is still unknown. Considering the potential communicative function of these sounds 32 , the masking of fish signals might have consequences at the individual and population levels (i.e. the period during which choruses were recorded could correspond to courtship and spawning).
In our recordings, the fish chorus peak frequency (800 Hz) was overcome by more than 3 dB by the ISWEC noise (Fig. 5). Then, by using the transmission loss model, we highlighted that at a distance of 1000 m, the ISWEC noise was reduced 10 dB below the fish chorus peak at 800 Hz (Fig. 10). However, the PSD measured for a vessel passage was much more intense at all frequencies (except frequencies lower than 100 Hz), overcoming a fish chorus of approximately 20 dB. Even though in this study we could not evaluate whether the ISWEC noise affects the perception of conspecific sounds in fish, our results suggest that the noise of the ISWEC during its activity (POST_ON condition) can have an effect within a 1000 m radius. This is a first step study based on a simple power spectrum model, and further information will be necessary to measure the masking effect regarding fish calls 26,42 . Indeed, in order to determine masking effects of anthropogenic noise, more baseline studies should focus on hearing abilities (i.e. audiograms, SNR required for detection, discrimination, recognition of conspecific calls) and the communication space (effective vocalization radius) of fish species 26 .
In conclusion, underwater noise radiated from a full-scale wave energy converter system (ISWEC) was assessed in the shallow waters near the coast of Pantelleria Island in the central Mediterranean Sea. The noise of the ISWEC is higher at lower frequencies up to 4 kHz, especially when the ISWEC is active for energy production. The noise arises from both the anchoring system and the hull during energy conversion. The noise of the ISWEC infrastructure increases with wave height. ISWEC noise power spectrum amplitude exceeds that of the fish chorus power spectrum, therefore it is potentially masking fish choruses and, according to a transmission loss model, at a distance of 1000 m the fish peak is 10 dB higher than the ISWEC noise. Improvements in the materials in the anchoring system and in the bearings of moving parts could reduce the ISWEC noise and the noise of wave energy converter systems in general. Moreover, in steady operation for energy production, scheduled interruptions could be planned to avoid masking of fish choruses during dusk in the summer season, which could corresponds to courtship and spawning time.    www.nature.com/scientificreports www.nature.com/scientificreports/

Materials and Methods
IsWeC description. The ISWEC (inertial sea wave energy converter) is a floating all-enclosed converter retained by a slack mooring ( Fig. 6 and Table 3). The electro-mechanical system is completely sealed into the central body, granting intrinsic reliability and reduced maintenance. Wave power extraction is obtained by inertial torques provided by a gyroscope on an internal precession axis. An electric PTO (power take off) is connected to the precession motion, converting mechanical energy into electrical energy.
The ISWEC, deployed at Pantelleria in August 2015, is a device with a rated power of 100 kW and consists of a steel hull carrying two independent gyroscopic units (Fig. 6). Each unit consists of a flywheel inside a vacuum chamber to reduce the spinning losses. The vacuum chamber is designed to sustain the gyro and transfer its actions to the PTO. The gyro is supported by two radial roller bearings for the radial load and spherical roller thrust bearing for the axial load. Due to the relevant loading conditions, the bearings are provided with an oil cooling and lubrication system. www.nature.com/scientificreports www.nature.com/scientificreports/ The flywheel rotation requires energy, and its speed is controlled by software to increase the ratio between energy production (depending on the wave conditions) and energy absorption. However, considering the aims to evaluate the isolated contribution of different flywheel speeds to the generated noise, we used different flywheel speeds with the same wave height.
IsWeC mooring system. The ISWEC is equipped with three mooring lines (Fig. 7): a main mooring line on the fore for the wave actions and two mooring lines on the sides to keep the device aligned with the main wave direction and restrict the yaw to avoid the electrical cable connection. In the installation considered in this work, the only mooring line present was the main mooring line, since during the first deployment, there was not an electrical connection to the grid. It is plausible expects an increase of noise coming from additional metallic chains if all three mooring lines are present.
The main mooring line provides the necessary compliance for normal operation and simultaneously restrains the device position in the area. The mooring line is constituted by four anchors (hall-type) positioned on the seabed in a circle of diameter of 100 m. A chain departs from each anchor, and the free ends of the four chains are joined together and pulled upwards by a jumper, which pre-stresses the system, reducing the dragging and abrasion of the seabed. From the jumper, a chain reaches a clump weight, and then the last chain line, ending in a bifurcation, links the weight to the ISWEC.
The main mooring line is a non-linear restraint system that adapts its restraint force to the ISWEC functionality. The mooring exerts small force for moderate to medium wave conditions when the ISWEC energy production is active, and the pitching motion must not be reduced by the mooring forces. On the other hand, the mooring increases its restraint force in extreme weather conditions, where the ISWEC is shut down to prevent damage.   (Fig. 8). In this area, the sea bottom is not uniform and consists of a mix of Mediterranean seagrass (Posidonia oceanica), sand and rocks (Fig. 9).
The Island of Pantelleria is the tip of the deep volcanic Pantelleria Graben 43 located in the central western side of the Strait of Sicily, a high biodiversity hot spot in Mediterranean Sea 44 , The island is strongly exposed to the prevailing north winds 45 . The deep North-Western Coast satisfies the requirements for a good installation site of a wave energy converter since, in the area, the power of the waves was estimated to be 62.5 MWh/m/y 46 . The island is totally dependent on energy generated through diesel, and the energy costs for the users are double those for the continental areas.
Acoustic data acquisition. An underwater acoustic recorder was placed at a 25 m depth, with a hydrophone height of 5.0 m from the sea bottom and 40 m from the ISWEC (Fig. 9).
The acoustic data were collected from 24 November 2014 to 25 January 2016. Since the infrastructure was deployed in August 2015, we obtained a set of data before the installation (from November 2014 to June 2015), during the installation of the mooring system and ISWEC (June-August 2015), and with the ISWEC present (from August 2015 to January 2016). We used an autonomous recorder (SM2, Wildlife Acoustics, US) with a ultrasonic hydrophone with a recording bandwidth of 8 to 150000 Hz and a declared sensitivities of -170 ± 5 dB re 1 V/μPa in the band of 25-200 Hz, -166 ± 1 dB re 1 V/μPa in the band of 100 Hz-15000 Hz, and -170 ± 5 dB re 1 V/μPa in the band 15-100 kHz. A 35 kg weight and a small sub-surface buoy were used to maintain the vertical arrangement in the case of strong currents or bad weather (a photo with all components of the mooring is shown in Fig. 9). The buoy was connected to the upper part of the recorder with a thin rope (the distance between the buoy and hydrophone was 2 m). All the components were connected with non-metallic ropes to avoid noise due to moving parts.
We used two sampling strategy. In the fist, used for PSD analysis, we set the sampling frequency to 192000 Hz, we sampled two day over three. In the days of recording we sampled using a duty cycle of 2 minutes for recording (.wav files) and 58 minutes of no recording. For the second sampling strategy, used for the preliminary soundscape analysis, we set the sampling frequency at 48000 Hz and recorded in a continuous way. We sampled one day over three. For both sampling strategies we used a resolution of 16 bits and no pre-amplification or filtering was applied during the recordings (except for the automatic anti-aliasing filter of the recorder).  www.nature.com/scientificreports www.nature.com/scientificreports/ Using an acoustic releaser, the recorder was recovered for maintenance every 3 months to change the batteries and storage memory. For each maintenance event, recorder was non-operative for two/three days (except for the months of April and October 2015 when the recorder did not work).

Acoustic analysis. Band Sound Pressure Level (BSPL).
For each 2-minute file, the power spectral density (PSD) (dB re 1 µPa 2 /Hz), with Welch's overlapped segment averaging estimator method 47 , was calculated. The calculation was performed by dividing the acoustic data into Hann windows of 192000 samples (∆t = 1 s), with a signal superposition of 50%. From these PSDs, the sound pressure level (SPL, dB re 1 μPa) and the octave band sound pressure level (BSPL, dB re 1 μPa) were calculated integrating over the specific band frequency. In total, 11 octave frequency (Hz) bands were considered: 63 (44- Soundscape Components of the study area. A preliminary visual qualitative analysis of the acoustic recordings ( Fig. 10) was conducted on a subsample of 3 days per month (randomly selected) to confirm the presence of the main components of the soundscape already described by Buscaino et al. 28 and Ceraulo et al. 48  Flywheel mass 10 ton Table 3. ISWEC main parameters. www.nature.com/scientificreports www.nature.com/scientificreports/ Island (approximately 140 km from Pantelleria island) and for the Sicilian Southwestern Coast, Cape Granitola (approximately 95 km from Pantelleria island), respectively. All these sites in the Sicilian Channel (Fig. 8) present some common characteristics: they are in very shallow waters, near the coast, and have a sea bottom consisting of a mix of Mediterranean seagrass, sand, and rock.
We found that the measurements at frequencies below 1000 Hz are dominated by noise generated by waves and are louder during the winter; conversely, snapping shrimps dominate the measurements at higher frequencies from 4000 up to 96000 Hz and increase their acoustic activity during the summer. During summer, fish choruses dominate the measurements in the frequency range up to 2000 Hz, especially during dusk and night hours until dawn (Fig. 10). These intense choruses are created by fish signals that mainly occupy a frequency range between 600 and 1200 Hz with a peak frequency of 800 Hz (Figs 5, 10, 11). Their acoustic features and temporal patterns are comparable to the fish signals already described in a Mediterranean Posidonia meadow 32,48 . These signals are also acoustically comparable to fish sounds recorded in shallow water of southwester Arabian Sea 34 and in the Great Barrier Reef (Australia) 33 . Relying on these precedent studies, fish sounds could be assigned to Terapon theraps (for which our study area should be a suitable habitat; see fishbase suitable habitat map) or to a species belonging to Scorpaenidae family. A description of the main sound characteristics for our study site is given in Table 4 and in Fig. 11. Anthropogenic noise is represented by vessel passages (Fig. 10).

ISWEC masking of fish choruses.
After identifying the soundscape components, we analysed the potential masking of fish choruses by ISWEC noise. We focused our analysis on fish choruses because the peak energy of the ISWEC noise overlaps in frequency with the fish choruses.
Here, we could not evaluate whether the ISWEC noise affects the perception of conspecific sounds by fish (no data are available on the hearing capability of these fish). Therefore, we considered masking as noise overlap when its energy is in the frequency region of the fish signal 26 .
For each file, the power spectral density (PSD)(dB re 1 µPa 2 /Hz) was calculated following Welch's overlapped segment averaging estimator method 47 . The calculation was performed by dividing the acoustic data into Hamming windows of 192000 samples (∆t = 1 s), with a signal superposition of 50%.
We calculated the PSD on selected files containing (a) the highest intensity of fish chorus signals assumed to be the closest to the hydrophone; (b) ISWEC noise during activation for energy production (ISWEC_Post_ON) with a flywheel speed of 295 rpm (the most suitable speed for the median wave height wave in Pantelleria); (c) sea background noise; and (d) an example of a boat passage.
We hypothesized as acceptable the reduction of the ISWEC noise when the peak frequency of the fish chorus is above the ISWEC noise by at least 10 dB. Based on this result, we applied a transmission loss (TL) model to the total PSD levels recorded for the Post_ON condition to calculate the distance at which the ISWEC noise has a negligible masking effect on the fish signals. www.nature.com/scientificreports www.nature.com/scientificreports/ The transmission loss (TL) model was applied to the total PSD levels recorded for the Post_ON condition. We explored the variation in the PSD during noise propagation through the seawater at 40 m (the distance between the ISWEC and the recorder) and at the distance to obtain a negligible masking effect on the fish choruses at 800 Hz. The model considered the attenuation due to geometric spreading and absorption processes related to the dissolved salts 49,50 . The geometric spreading model was used to take into account spherical spreading until the maximum water depth (D = 25 m) and cylindrical spreading for the remaining distance. For ranges R > D, the transmission loss TL becomes: where R is the distance from the source, and α is the absorption coefficient calculated as: