Underwater Ambient Noise in a Baleen Whale Migratory Habitat Off the Azores

Assessment of underwater noise is of particular interest given the increase in noise-generating human activities and the potential negative effects on marine mammals which depend on sound for many vital processes. The Azores archipelago is an important migratory and feeding habitat for blue (Balaenoptera musculus), fin (Balaenoptera physalus) and sei whales (Balaenoptera borealis) en route to summering grounds in northern Atlantic waters. High levels of low frequency noise in this area could displace whales or interfere with foraging behavior, impacting energy intake during a critical stage of their annual cycle. In this study, bottom-mounted Ecological Acoustic Recorders were deployed at three Azorean seamounts (Condor, Acores and Gigante) to measure temporal variations in background noise levels and ship noise in the 18-1,000 Hz frequency band, used by baleen whales to emit and receive sounds. Monthly average noise levels ranged from 90.3 dB re 1 μPa (Acores seamount) to 103.1 dB re 1 μPa (Condor seamount) and local ship noise was present up to 13% of the recording time in Condor. At this location, average contribution of local boat noise to background noise levels is almost 10 dB higher than wind contribution, which might temporally affect detection ranges for baleen whale calls and difficult communication at long ranges. Given the low time percentatge with noise levels above 120 dB re 1μPa found here (3.3 % at Condor), we woud expect limited behavioural responses to ships from baleen whales. Sound pressure levels measured in the Azores are lower than those reported for the Mediterranean basin and the Strait of Gibraltar. However, the currently unknown effects of baleen whale vocalization masking and the increasing presence of boats at the monitored sites underline the need for continuous monitoring to understand any long-term impacts on whales.


INTRODUCTION
Marine acoustic pollution has become an issue of special concern over recent decades. Measurements from the North Atlantic show that average noise at 50 Hz has increased about 5.5 dB per decade from 1950 to 1970 (Ross, 2005) and about 2.8 dB from 1966 to 2013 (Širović et al., 2016). A similar trend has been found in the North Pacific with noise increasing at an average rate of 2.5-3 dB per decade at 30-50 Hz since the 1960s (Andrew et al., 2002;McDonald et al., 2006;Chapman and Price, 2011). This rise has been mainly due to shipping and together with seismic surveys has become one the principal sources of ambient noise below ∼1 kHz. (Wenz, 1962;Andrew et al., 2002;McDonald et al., 2006;Hildebrand, 2009;Klinck et al., 2012;Nieukirk et al., 2012). Shipping noise contribution can be at very low frequencies below 200 Hz (Ross, 1976), when is given by the summation of many distant large ships scattered throughout an ocean basin. When a ship passes nearby, however, it increases temporarily and substantially noise levels at that location at much greater frequencies since propagation removes the high frequency portion of the spectrum (Wenz, 1962;Hildebrand, 2009).
Baleen whales emit sounds with fundamental frequencies below 1 kHz (Richardson et al., 1995) which overlap with peak power in ship noise (Wenz, 1962;Hildebrand, 2009). The production and reception of baleen whale vocalizations have been associated to vital biological processes such as feeding, mating, group cohesion and social interaction (e.g., Payne and Webb, 1971;Dudzinski et al., 2002) which make these animals especially vulnerable to this source. Noise in the environment can limit the range for successful detection of signals through masking, thus significantly affecting the acoustic communication in large whales (Samaran et al., 2010;Ponce et al., 2012;Hatch et al., 2012;Erbe et al., 2015). Blue whales (Balaenoptera musculus) have shown increased source levels of their D calls (<100 Hz) as well as increased multiple callers when ships are nearby (McKenna, 2011;Melcón et al., 2012) and North Atlantic right whales (Eubalaena glacialis) call louder with increasing background noise levels (Parks et al., 2010). Following the mounting evidence of noise impact on marine mammals, the U.S. National Research Council (NRC) established the 120 dB re 1 µPa as the noise level above which marine mammals might be adversely affected by sound (NRC, 2005). Vessel avoidance behavior has been documented for some species of baleen whales at received sound pressure levels (SPLs) of 92.8-148.6 dB re 1 µPa, but especially above 120 dB re 1 µPa (Richardson et al., 1995;Richardson and Würsig, 1997;Southall et al., 2007). In addition, in the presence of shipping noise, North Atlantic right whales have been shown to exhibit increased stress levels (Rolland et al., 2012) and humpback whales (Megaptera novaengliae) changed their foraging activity (Blair et al., 2016).
In the long term, behavioral disturbance and physiological stress caused by noise could lead to population-level effects. Changes in vocal behavior in response to noise during feeding, socializing (Di Ioro and Clark, 2010) and breeding (Miller et al., 2000) may have energetic costs, and potential avoidance of noisy foraging/breeding/resting areas  could reduce energy intake and disrupt behavior at key life stages. These effects could have a negative impact at a population level by affecting growth, survival and reproductive success of individual animals. However, determining a causal link between noise exposure through effects on individual vital rates to population consequences is extremely difficult and further studies are needed and models developed to answer these questions.
Although research on noise levels and the impacts on marine life have been increasing over recent years (Williams et al., 2015), most studies have focused on whales' feeding grounds and coastal continental areas (Parks et al., 2010;Dunlop, 2016) with fewer studies on open ocean waters (Dziak et al., 2015;Bittencourt et al., 2016). In the central Atlantic area, only one measurement has been made north of the Azores archipelago  and only one study has been published documenting airgun seismic noise in mid-Atlantic waters .
The region around the Azores is a migratory habitat for several species of baleen whales. Blue and fin (B. physalus) whales interrupt their journeys to northern latitudes to feed in the archipelago every spring and early summer (Silva et al., 2013. Sei whales (B. borealis) travel through the archipelago in spring on their way up to the Labrador Sea but they do not seem to forage routinely in the area . Moreover, preliminary acoustic data suggest the presence of fin whale (Silva et al., 2011) and blue whale (unpublished data) calls also during the winter. This finding is in accordance with a study documenting winter calling by fin and blue whales around the mid-Atlantic ridge, south of the Azores . Therefore, the region around the Azores may be an important habitat for these species in the central North Atlantic and noise pollution should be carefully monitored to inform effective management of human activities in these waters.
This work investigates low-frequency underwater noise levels at an important baleen whale habitat in the North Atlantic, the Azores archipelago by: (a) investigating the spatial and temporal variability within the 18-1,000 Hz frequency band (calling range of most baleen whales), (b) determining the contribution of local ship and wind driven noise (c) describing noise levels above 120 dB re 1 µPa, reported to cause behavioral responses to baleen whales (NRC, 2005) and (d) discuss potential effects of these results on baleen whales in the Azores. In addition, we investigated variability of noise levels in one-third octave bands centered at 63 and 125 Hz, which have been specifically proposed by EU Marine Strategy Framework Directive (MSFD) as a measure of noise from distant shipping (2008/56/EC, European Commission 2008).

Deployment Locations
The three deployment locations are seamounts and were chosen for their distinct importance for baleen whales and their differences in anthropogenic usage. Condor seamount, located 17 km southwest of Faial Island, became a scientific observatory in 2008 when local authorities, researchers, fisherman and other stakeholders agreed on designating it as a protected area for scientific research. Since 2010, demersal fisheries are banned, tuna and big game fishing are permitted upon prior authorisation and scientific and recreational activities (such as shark diving) are allowed Ressurreição and Giacomello, 2013). Açores seamount, located 40 km southwest off Faial Island, is frequented by small commercial fishing and recreational activities, although to a much lesser extent than Condor (Figure 1). Gigante seamount, located 100 km west-northwest of Faial Island along the Mid-Atlantic Ridge, is used by commercial fisheries and lies close to major marine traffic lanes.
The areas around Condor and Açores seamounts are frequently used by blue and fin whales for foraging (Silva et al., 2013 and by sei whales for migrating . Gigante seamount is close to a transit area for the three species, and occasional feeding may also occur there. Other species of baleen whales may also occasionally occur in these areas .

Acoustic Data
Bottom-mounted Ecological Acoustic Recorders (EARs; Lammers et al., 2008) were deployed at the three seamounts at an approximate depth of 190 m. The EAR consists of a sensor Technology SQ26-01 hydrophone with a response sensitivity of −193.14/−194.17 dB re 1 V/µPa (varying between deployments) for Condor and Açores and −193.64/−193.14 dB for Gigante and a flat frequency response (±1.5 dB) from 18 Hz to 28 kHz. A Burr-Brown ADS8344 A/D converter was used with a zero-to-peak voltage of 1.25. A total system gain of 47.5 dB re 1 µPa was used during all recordings resulting in a noise floor of 89 dB re 1 µPa (18-1,000 Hz), 65.5 dB re 1 µPa (63 Hz octave band) and 66.7 dB re 1 µPa (125 Hz octave band). Dynamic range of the instrument was of 57 dB re 1 µPa reaching saturation at 146 dB re 1 µPa.
EARs recorded from March 2008 to October 2012 at Condor, from November 2011 to October 2012 at Açores and from April 2008 to February 2011 at Gigante with several gaps due to equipment failure or maintenance duties. Sampling rates and duty cycles were constrained by battery life and disk space limitations, given programmed deployment durations (Table 1).

Noise Measurements
Recordings with sample rates of 50 kHz were re-sampled to 2 kHz using Adobe Audition 3.0 software (Adobe Systems Incorporated, CA, USA) to standardize all acoustic data from 18 to 1,000 Hz, which is the bandwidth dominated by anthropogenic noise (Wenz, 1962) and overlaps the vocalizing range of balaenopterids. Self-system tonal noise within the frequency band of interest was identified only in recordings with sampling rates of 2,000 Hz which correspond to deployments at Condor and Açores from 2011 and 2012. 1-Hz Spectrogram Power Density (SPD) plots were made for each month to precisely identify which frequency bins were affected so they could be removed before computing broadband SPLs. Given that all self-system noise identified was highly tonal, removing these few frequency bins is likely to have a negligible effect on averaged broadband SPLs and on the characterization of shipping noise, which spreads across a wide range of frequencies. Moreover, self-system noise removed was found in frequencies well above the one-third octave bands analyzed in this study (63 and 125 Hz). From the SPD plots we can say that data were not clipped since there is no flat line of data points at high noise levels clustered at the limit value of 146 dB where the system saturates (see Figure 4 from results section).
Each month of recordings was grouped and concatenated to form a single file to be analyzed with Matlab code written by Merchant et al. (2015). The time-series of every signal was divided into m 1-s segments of consecutive samples overlapping in time (50% overlap). Each segment was then multiplied by a Hann window and transformed to the frequency domain via the Discrete Fourier Transform (DFT). Spectra were then averaged to a 90-s resolution via the standard Welch method (Welch, 1976). The power spectrum (P) was then computed from the DFT, which for the m th segment, of signal X at frequency f and for N number of samples in each segment is given by: For each deployment, calibration data from the EAR, including the hydrophone sensitivity (M h ), system gain (G) and the zeroto-peak voltage of the analog-to-digital converter (V ADC ), were used to calculate a correction factor (S(f)) computed by: where N bit is the bit-depth of the digital signal (16 bits). S(f) was then used to obtain SPLs in the bandwidth from 18 to 1,000 Hz by: where p ref is the reference pressure of 1 µPa for underwater measurements, f low and f high are the lower and upper bounds of the frequency range under consideration and B is the noise power bandwidth of the window function, which corrects for the energy added through spectral leakage.

Noise Data Analysis
The effect of different duty cycles on the calculation of monthly average background noise levels was investigated by concatenating a full month of data (May 2008 from Condor), treating it as a continuous recording, and then subsampling it according to the different duty cycles used in this study (Table 1).
To test for statistical differences in SPLs between different duty cycles, a first order autoregressive model was fitted to the SPL time series of mean SPL per sample for each duty cycle. Then, based on the estimated parameters and corresponding standard errors, 95% confidence intervals for each duty cycle mean SPLs were derived, assuming a Gaussian distribution for the parameter estimates. Number of samples (N) was the number of files resulting from the different duty cycles applied. To investigate differences in noise levels between locations, only Condor's data from 2011 and 2012 were analyzed to compare with time series of similar length from Açores and Gigante. For every location, the arithmetic mean (AM) over the  period considered was calculated. For N samples p 2 rms , AM is given by: rms,i is the ith value of the mean squared pressure given by: To test for statistical differences in SPLs between locations, a first order autoregressive model was fitted to the SPL time series of each location (containing daily averaged SPLs). Then, based on the estimated parameters and corresponding standard errors, confidence intervals (95%CI) for each location mean SPLs were derived, assuming a Gaussian distribution for the parameter estimates. Number of samples (N) was the number of days. Variability in noise levels for every location was analyzed using the coefficient of variation (CV), which allows comparison between datasets with different means.
Within each location, temporal variability of noise levels was explored by calculating hourly and monthly averaged, median and 5th, 75th, and 95th percentiles SPLs for the frequency band of 18-1,000 Hz. Also, hourly and monthly averaged one-third octave bands centered in 63 and 125 Hz were calculated to specifically measure the contribution of distant ship noise to ambient noise as suggested by the MSFD (2008/56/EC, European Commission 2008).
Seasons were defined according to the location (North-East Atlantic) as follows: Spring: March-May, Summer: June-August, Autumn: September-November and Winter: December-February.
To allow for comparisons, average SPLs were calculated for the three noisiest months in Condor (July-September, 2010), Açores (May-July, 2011) and Gigante (May-September, 2008) in the frequency band of 10-585 Hz to be compared to SPLs found in the Mediterranean by Castellote et al. (2012). Also, median levels in the frequency band of 10-25,000 Hz were measured in Condor (July-September, 2010) to compare it with levels found in another oceanic archipelago by Bittencourt et al. (2016).

Ship Noise Analysis
In the absence of an operative antenna in the area for receiving information from Automatic Information System (AIS) installed in ships during the recording period, a methodology was used to study the contribution of local ship noise to general background noise levels. Using the broadband (18-1,000 Hz) noise background levels for every recording, an Adaptive Threshold Level (ATL; Merchant et al., 2012b) was obtained to identify local intermittent ship noise. The ATL was calculated by computing the minimum SPL in a certain period of time (W) and summing a tolerance above this minimum, a threshold ceiling (C) in dB re 1 µPa: Due to differences in background noise levels and duty cycles in this study compared with Merchant et al. (2012b), two different periods of time (W) (1 and 7 hours of recordings) and 4 different threshold ceilings (C) (from 4, 6, 8, and to 12 dB) were tested. Firstly, an appropriate time period, W, was selected by visually inspecting plots of SPL values and thresholds and selecting the one that best discriminated wind-wave driven noise from intermittent noise. Once W had been specified, results from the ATL applying different values of C were compared to visually confirmed boats in the spectrogram for one chosen month per location (July 2012 for Condor, May 2012 for Açores and July 2010 for Gigante). Those parameters that resulted in the best compromise between visually confirmed boats detected by the ATL (true positives) and detections by the ATL not corresponding to boats (false positives) were selected.
Once W and C were set, ATL was calculated for every month and location. Time with levels above the threshold was summed and divided by the total recording time to obtain the Percentage of Time with noise levels Above the Threshold Level (PT-ATL). The PT-ATL was then used to investigate spatial variations in boat presence using a Kruskal-Wallis ANOVA and a post-hoc Dunn test for multiple comparisons.
To test the efficiency of the methodology at detecting the presence of vessels, a comparison was made between monthly PT-ATLs and the number of days per month with boat presence in Condor. Data on boat presence were obtained from logbooks that contained information on the number of boats and type of activity conducted at the Condor seamount area per day from 2008 to 2012. The type of boats' activities recorded were: recreational activities, such as big-game fishing and shark diving, with data logged by the operators themselves; scientific research, based on information provided by scientists conducting research at Condor; and tuna fishing, based on data recorded by onboard observers under the Azorean Fisheries Observer Programme (POPA).

Contribution of Wind-Wave and Vessel-Driven Noise
An analysis of wind-wave driven noise and intermittent ship noise was implemented to compare the relative contribution of natural and anthropogenic sources to background noise levels in this region. Windiest months were selected for the three locations and daily averaged SPLs and wind speeds calculated. Days with maximum and minimum SPLs coincided with maximum and minimum wind speeds. For every month, averaged SPLs were calculated from 10 min sound files free of ship noise (visually inspected spectrograms) selected from 2 days, one with maximum and one with minimum wind speed. Similarly, for all months and for all locations, average SPLs were calculated for periods of time above the threshold and compared to those with minimum wind conditions. Differences between quietest average and noisiest average were then calculated for the wind and for the ship contribution. Daily averaged wind speeds (km/s) were obtained from Weather Underground historical data (www.wunderground.com) for each location.
Noise Levels above 120 dB re 1 µPa Since baleen whales have been shown to avoid vessels at noise levels above 120 dB re 1 µPa (Richardson et al., 1995;Richardson and Würsig, 1997;Southall et al., 2007), percentage of time with SPL above this level was also calculated for every month of study. To do so, broadband average SPL for every month and location were used to calculate the amount of time with noise levels above 120 dB re 1 µPa and divide that by total recording time.

Ship Noise Analysis
The only anthropogenic noise source found in the recordings was ship noise, which had the maximum energy above 100 Hz for boats with higher noise levels, or in the bandwidth of 10-100 Hz for boats with lower noise levels.
The ATL only detected local boat noise that increased noise levels significantly and intermittently (Figure 2). The most adequate time period (W) to calculate minimum SPLs was 1 h of recordings for both duty cycles as this discriminated well between wind-driven and intermittent noise. The best compromise considering a minimum of 90% of visually confirmed boats in the spectrogram detected by the ATL, and a maximum of 5% of false positives was obtained using a threshold ceiling C = 4 dB for duty cycle of 3,600 s every 12,600 s, and C = 8 dB for the 90 and 30 s duty cycle ( Table 2). False positives were mainly caused by loud biological sounds consisting of low frequency clicks produced by delphinids and sperm whales.
Logbook data from Condor was compared to the acoustic recordings resulting in 19 months of simultaneous data. There was a weak correlation (R 2 = 0.354, p < 0.05, n = 19) between boat presence from logbooks and PT-ATL from 2008 to 2012, mainly because the high peak in PT-ATL in June was not matched by a higher presence of boats (Figure 3). Removing June 2012 from the analysis resulted in a stronger correlation (R 2 = 0.582, p < 0.001, n = 18). Recordings from this month were visually inspected and boat noise detected by the ATL was confirmed to be mainly present during daylight hours.

Spatial Variability in Ambient Noise Levels and Peak-Generating Vessels
The different duty cycles used in this study did not significantly affect the average monthly SPLs. Differences between the assumed "continuous" recording ( The arithmetic mean of SPLs was calculated for the 18-1,000 Hz band for Condor, Açores and Gigante over 9, 11, and 14 months, respectively. Açores had the lowest value (92.9 dB re 1 µPa), followed by Gigante (95.9 dB re 1 µPa), with higher mean SPL in Condor (97.6 dB re 1 µPa). Higher variability was found  Results are for July 2012 in Condor, May 2012 in Açores, and July 2010 in Gigante.

FIGURE 3 | Number of days with boats registered in logbooks (bars)
and PT-ATL from the acoustic data (red line).
in Condor (CV = 0.098) followed by Gigante (CV = 0.085) and Açores (CV = 0.071). Median values of noise levels were higher for Condor and Gigante (93.1dB re 1 µPa and 91.6 dB re 1 µPa, respectively) and lower for Açores (90.1 dB re 1 µPa). Averaged noise levels for the 63 and 125 Hz one-third octave bands were also lower in Açores (70.2 dB re 1 µPa and 74.6 dB re 1 µPa, respectively), while the highest levels at the 63 Hz band were found in Gigante (73.6 dB re 1 µPa) and for the 125 Hz band in Condor (79.5 dB re 1 µPa; Table 3). There was no overlap in the 95%CI of SPL within the 18-1,000 Hz, the 63 Hz and the 125 Hz one-third octave bands for Condor, Açores and Gigante, suggesting differences in average ambient noise levels between the three locations were highly significant (see Supplementary  Table 1 for details on 95%CI values). Differences in average noise levels for the 63 and 125 Hz onethird octave bands are supported by the spectral characteristics of sound for every location. Looking at the noisiest months, we can see that Gigante showed higher levels of noise from ships <100 Hz (Figure 4C) while Condor and Açores had higher ship noise levels >100 Hz (Figures 4A,B).

Temporal Variability in Ambient Noise Levels and Peak-Generating Vessels
In Condor seamount, percentage of time with boats peaked during the summer months (June-August) extending to September in some years ( Figure 5D). This is well illustrated by increased broadband and 125 Hz octave band noise levels at these periods (Figures 5A,C). Açores seamount also showed higher PTL-ATL during summer months, especially in June, with another peak seen in November ( Figure 6D). These peaks are well reflected in the higher average broadband and 63 and 125 Hz octave band noise levels (Figures 6A,C). In Gigante, values of PTL-ATL tended to be greater in summer months or in September, although differences to other seasons were not as obvious as in Condor and Açores ( Figure 6D). In this case, temporal patterns in broadband and one-third octave bands SPLs did not match those of boat time (Figures 6A,C,D). Note that fluctuations in the 75th percentile noise levels are still highly affected by wind-driven noise in all locations (Figures 5B, 6B) and only the 95th percentile is affected by the presence of boats in accordance to the low PT-ATL found in all areas (Figures 5D, 6D). Average levels (AM) are more affected by brief and high amplitude events such as ships and might not represent the real average of noise data which has a highly skewed distribution. However, average levels are a robust metric that can be used to assess shipping noise if presented in combination with other metrics that identify loud events, such as the PT-ATL used here (Merchant et al., 2012a).
Noise Levels above 120 dB re 1 µPa Percentage of time with noise levels >120 dB re 1 µPa was higher in Condor with a maximum of 3.3% in July of 2010. In 13 out of 32 months sampled, noise levels were always <120 dB and for the remaining months, time with boats varied from 0.007 to 0.4%, with greater percentages in summer and autumn. In Açores, noise levels >120 dB were recorded only in the noisiest months, May

DISCUSSION
This work provides the first long-term characterization of lowfrequency underwater noise levels at an important baleen whale mid-ocean habitat and discusses potential adverse effects on this cetacean group.
Noise levels at Condor seamount were higher than at Gigante and Açores seamounts and rises in monthly average broadband noise levels were mainly due to the presence of intermittent loud events such as boats. Median noise levels were more affected by wind-driven noise. In the absence of boats or with few boats, median and average levels were similar and wind became a major contributor to background noise.
The ATL methodology developed by Merchant et al. (2012b); to detect intermittent loud noise events attributed to boat presence has been successfully tested and applied in this study. Time period (W) over which minimum SPLs are calculated and threshold ceiling (C) are parameters that need readjustment depending on environmental acoustic characteristics and system duty cycle. We found that increasing C caused a decrease on the percentage of true and false positives but the extent of this variation differed between locations, depending on the acoustic characteristics of the environment. In our case, the same W worked well for all duty cycles but the adequate threshold ceiling was lower for higher duty cycles than for lower duty cycles ( Table 2). In general, environments with a high presence of loud intermittent events and systems with lower duty cycles should require smaller W and higher C than quieter places and higher duty cycles in order to detect these events above minimum SPLs. The high correlation between boat presence at Condor seamount from logbook data and PT-ATL values from 2009 to 2011 indicates that this methodology can be used to describe boat presence in the study area. The lack of correlation in 2012 is likely explained by the limitations in boat detection distance using these methods and the fact that not all boat activity was registered in logbooks. We suspect that the high PT-ATL values found in June 2012 could be due to an increase in recreational activities.
The monthly variability in PT-ATL values found in Condor reflects this area's main recreational uses during late spring and summer. Also, the annual decrease in PT-ATL values in winter can be explained by the designation of the temporary protected area for research in force since 2010. Since this was implemented, demersal fisheries, which operate year round, are prohibited, thus explaining the lower presence of boats during the winter. On the other hand, recreational activities are gaining importance, particularly shark diving, an activity which started experimentally in 2009 and that mostly operate in spring and summer (Ressurreição and Giacomello, 2013). The expeditions to dive with sharks at Condor were reported to double between 2011 and 2012 (Ressurreição and Giacomello, 2013), which might well explain why PT-ATL values increased in 2012. This activity also takes place in the Açores seamount, but to a lesser extent, which is reflected by a lower boat presence than in Condor. Gigante seamount shows a higher presence of boats throughout the year, as a result of the proximity of a marine traffic route used by commercial shipping and the presence of commercial fishing year-round in Gigante. Therefore, there is a great potential for using passive acoustic techniques to monitor boat activity in specific areas such as the ones in the study. This methodology, however, cannot be used to detect distant vessels and might not be adequate for areas with higher ship traffic where separation between continuous and intermittent events might not be possible.
For measuring the contribution of distant ship noise, the European MSFD /56/EC, European Commission 2008 suggests the use of one-third octave bands, centered at 63 Hz and 125 Hz, which are included as indicators to assess the Good Environmental Status (GES) of the marine environment. In this study, Gigante shows the highest noise levels in the 63 Hz onethird octave band which can be explained by the proximity of a shipping lane mentioned in the above paragraph. Noise levels measured in the 125 Hz octave band better reflect local boat presence at Condor and Açores while at Gigante the difference between the two octave bands (63 and 125 Hz) is not very clear. This is mainly due to the difference in the type of vessels and distance of those to the hydrophone at each location. Comparison of spectrum levels indicates that Gigante has higher noise levels below 100 Hz, which is typical of distant large vessels such as tankers, while Condor and Açores have higher levels above 100 Hz, which is characteristic of smaller boats. Performance of one-third octave bands with distant shipping could not be assessed in this study because AIS data were not available for this period.
The maximum percentage of time with presence of boats found in this study is relatively low (13%). However, our results show that contribution of local boat noise to background noise levels ranged from 16-19 dB, depending on the study area, and on average was nearly 10 dB higher than wind contribution. This value is similar to those described in the literature where it is documented that below 1 kHz ship traffic regularly increases noise levels by 25 dB above background levels (Bassett et al., 2012). These increases in ambient noise might be sufficient to mask baleen whale calls unless they are able to compensate vocally, which is known as the Lombard effect (Lombard, 1911). There is some evidence that several species of mysticetes can, but sometimes do not, modify their vocalization's characteristics in response to shipping noise. Blue whales have been found to change the interval, types and amplitudes of their calls (McKenna, 2011;Melcón et al., 2012) while male fin whales seem to change their song characteristics . Other baleen whale species such as gray whales (Eschrichtius robustus) also modify calling rates, received levels and percentage of calls (Dahlheim and Castellote, 2016), humpback whales sing shorter versions of their songs (Sousa-Lima et al., 2002) and North Atlantic right whales show short-and long-term changes in their calling behavior in response to increased low-frequency noise (Parks et al., 2007(Parks et al., , 2009(Parks et al., , 2010. However, other studies show that humpback whales respond to increases of noise levels produced by wind but do not compensate for higher levels of noise from vessels (Dunlop, 2016).
Auditory masking reduces the effective communication space between sender and receiver . A model developed by Tennessen and Parks (2016) demonstrated that a right whale is not able to hear an upcall from another whale if a ship passes at less than 25 Km, unless the calling whale increases the amplitude of the calls by 20 dB. Despite differences in call source levels between right whales and blue, fin and sei whales, they share similarities in call frequency ranges (e.g., Parks and Tyack, 2005;Sirović et al., 2007;Romagosa et al., 2015). Detection ranges of calls for these three species might be affected by passing ships in similar ways, which is of concern given the dependence of Balaenoperids on long range communication (Payne and Webb, 1971). Although their calls have been mainly attributed to male reproductive displays, whales also produce sounds outside their breeding grounds and season (Clark et al., 2002;Oleson et al., 2007;Vu et al., 2012). Blue whales are known to produce D calls during foraging within groups (McDonald et al., 2001;Stafford et al., 2005;Calambokidis et al., 2008) and fin whales produce "20-Hz pulse" calls that are likely to have a social purpose or a contact maintaining function when produced irregularly or as call-counter calls (McDonald et al., 1995;Edds-Walton, 1997). Baleen whale long-range calls could also be used for orientation purposes, as suggested by Payne and Webb (1971). In the Azores, preliminary analysis of acoustic data shows that blue and fin whales produce these types of calls when they are seen in spring as well as reproductive songs during the winter (unpublished data). Sei whales also vocalize during their migratory journey (Olsen et al., 2009;Prieto et al., 2014) through the Azores producing a well-known downsweep call for this species (Romagosa et al., 2015). Several studies indicate that blue, fin and sei whales are present around the archipelago, including in the deployment areas, mostly from February to May Prieto et al., in press) which coincides with a lower presence of boats at Condor but with a higher presence of boats at Açores. As for the transiting area (Gigante), due to its more constant vessel traffic throughout the year, an overlap exists with the baleen whale northward migration in spring and summer, and possibly with the southward journey in late autumn.
The biological implications of masking for these three species, whether they compensate it by modifying vocal behavior or not, are still unclear. However, given the association of vocalizations to such vital processes (e.g., social, foraging, navigation, reproduction), masking could reduce the chance of finding partners to mate, the ability of finding food or even their navigational skills thus negatively affecting their reproductive success and ultimately its survival. Although some efforts have been made to develop masking models that can be incorporated into regulation strategies, more research is needed to better understand potential effects of this complex phenomena, hearing characteristics from different species and anti-masking strategies used by free-ranging animals (Erbe, 2002;Clark et al., 2009).
Ship noise can also cause behavioral responses to cetaceans and Southall et al. (2007) suggests using SPLs to assess it. This metric might not be the most appropriate way to look for consistent patterns of response but it is often measured or estimated because it is required by law in many European countries and the USA as part of their noise mitigation regulations. Also, many other variables such as location, nature and behavior of noise sources and characteristics and activity of the individual animal among others, can affect the nature and extent of responses (Ellison et al., 2012). Therefore, the percentage of time with SPL levels above 120 dB re 1 µPa was calculated based on the model from the NRC (2005) that established that marine mammals exposed to levels above this value might be affected by sound. The maximum monthly percentage time with levels above 120 dB re 1 µPa was 3.3% (at Condor seamount) which is very low considering that an animal is unlikely to remain in the same location for the entire month. However, deployment depth affect noise levels received by the hydrophone and percentages with levels above 120 dB re 1 µPa are certainly higher closer to the source which in this case is found at the surface. While this is a simplistic and limited approach, it can nevertheless give an initial sense of the time that noise levels in an area could induce behavioral responses on baleen whales.
Despite our findings suggesting the Azores is characterized by reduced underwater noise, we expect other areas in the archipelago closer to ferry routes, commercial shipping routes or routinely used by whale watching boats to be considerably noisier. Therefore, these measurements are representative only of these locations and further measurements and sound propagation modeling in other areas will be necessary to produce a detailed soundscape for the entire archipelago.