Effects of anthropogenic noise on fishes at the SGaan Kinghlas-Bowie Seamount Marine Protected Area

Abstract

Underwater noise from anthropogenic sources has been increasing dramatically for the past few decades and little is known about its effects on fishes. The objective of this study is to describe the occurrence and characteristics of fish sounds in the SGaan Kinghlas-Bowie Seamount Marine Protected Area (SK-B MPA, British Columbia, Canada) and to correlate them with the corresponding anthropogenic soundscape. Here we present preliminary results of the detection of fish sounds at SK-B MPA between July 2011 and July 2013. An automatic detector was used on nearly 40,000 acoustic samples (4,754.5 hours in total) to search for fish sounds. About 1.2% of the data were highlighted as containing fish-like signals. Manual verification of these detections revealed that 95.5% were false positives and the remaining sounds were of unknown origin. Eighty detections were highly stereotyped and are suspected to be produced by fish, but no identification has been confirmed yet. Systematic manual inspection of sub-sampled acoustic data is yet to be performed to determine if the detector missed any fish sounds. Future deployments should select areas based on the presence of known fish habitat occurrence, and install autonomous recorders optimized to reduce equipment self-noise and flow noise biases.

Full text

Published by the Acoustical Society of America
Volume 27 http://acousticalsociety.org/
Fourth International Conference on
the Effects of Noise on Aquatic Life
Dublin, Ireland
10-16 July 2016
Effects of anthropogenic noise on fishes at the
SGaan Kinghlas-Bowie Seamount Marine
Protected Area
Amalis Riera
Biology Department, University of Victoria, Victoria, BC, Canada; ariera@uvic.ca
Rodney A. Rountree
Independent Scholar, Waquoit, MA, USA; rrountree@fishecology.org
Xavier Mouy
JASCO Applied Sciences, Victoria, BC, Canada; Xavier.Mouy@jasco.com
John K. Ford
Fisheries and Oceans Canada, Pacific Biological Station, Nanaimo, BC, Canada; john.ford@ubc.ca
Francis Juanes
Biology Department, University of Victoria, Victoria, BC, Canada; juanes@uvic.ca
Underwater noise from anthropogenic sources has been increasing dramatically for the past few decades
and little is known about its effects on fishes. The objective of this study is to describe the occurrence and
characteristics of fish sounds in the SGaan Kinghlas-Bowie Seamount Marine Protected Area (SK-B
MPA, British Columbia, Canada) and to correlate them with the corresponding anthropogenic
soundscape. Here we present preliminary results of the detection of fish sounds at SK-B MPA between
July 2011 and July 2013. An automatic detector was used on nearly 40,000 acoustic samples (4,754.5
hours in total) to search for fish sounds. About 1.2% of the data were highlighted as containing fish-like
signals. Manual verification of these detections revealed that 95.5% were false positives and the
remaining sounds were of unknown origin. Eighty detections were highly stereotyped and are suspected
to be produced by fish, but no identification has been confirmed yet. Systematic manual inspection of
sub-sampled acoustic data is yet to be performed to determine if the detector missed any fish sounds.
Future deployments should select areas based on the presence of known fish habitat occurrence, and
install autonomous recorders optimized to reduce equipment self-noise and flow noise biases.
© 2016 Acoustical Society of America [DOI: 10.1121/2.0000245]
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 1
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

1.INTRODUCTION
Underwater noise from anthropogenic sources has been increasing dramatically for the past
few decades (McDonald et al., 2006). Its impacts on marine mammals have been widely studied,
but little is known of the effects of noise on fishes and invertebrates (Popper and Fay, 2011;
Popper and Hastings, 2009).
Marine anthropogenic noise at low frequencies overlaps with fishes’ hearing range and peak
sound production which can reduce their communication space and result in habitat loss
(Hawkins et al., 2015). In addition to increasing stress levels and impairing the ability of fish to
detect prey and predators, intense noise can cause temporary hearing loss and reduced survival
(Simpson et al., 2016).
The main objective of this project is to study the effects of anthropogenic noise on fishes at
the SGaan Kinghlas-Bowie Seamount Marine Protected Area (SK-B MPA). This is an offshore
volcanic formation off the west coast of Haida Gwaii in the Pacific Ocean (Fig. 1) that rises from
a depth of 3000 m to within 25 m of the surface (Canessa et al., 2003). The MPA encompasses
three seamounts: Bowie, Hodgkins, and Davidson, and it is divided into three zones for
management purposes (DFO, 2011); from the peak of the Bowie seamount to the 457 m
bathymetric contour line (Zone 1), the remainder of the Bowie seamount (Zone 2), and the
Hodgkins and Davidson seamounts (Zone 3). The value of the MPA for protection includes a
rich, biodiverse, and productive ecosystem thanks to the surrounding oceanographic interactions
such as upwelling and turbulent mixing of surface waters. It also serves the possible functions of
biological oasis for unique plant and animal communities and of staging post for migrating
marine mammals and seabirds (Canessa et al., 2003). The fish community is dominated by
rockfish (25 species of Sebastes sp.), sablefish (Anoplopoma fimbria), and prowfish (Zaprora
silenus). Some of these species are known to produce sound (Wall et al., 2014). Prowfish is a
rare species that usually occurs in somewhat deeper water but is found in large numbers at
shallower depth at SK-B MPA (Canessa et al., 2003).
To investigate the effects of anthropogenic noise on fishes, we first need to determine the
presence of fish sounds and their temporal distribution at this location. Patterns in fish sound
production will then be compared to patterns in ocean vessel traffic and associated noise
exposure. This will be done in collaboration with the ongoing MEOPAR-funded (Marine
Environmental Observation Prediction and Response) NEMES (Noise Exposure to the Marine
Environment from Ships) project that is mapping ocean vessel traffic and modeling associated
noise exposure to predict gradients/spatial distribution of noise along the BC coast, including the
SK-B MPA. Here we present preliminary results of the detection of fish sounds at SK-B MPA
between July 2011 and July 2013.
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 2
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

2.METHODS
A.BOWIE SEAMOUNT ACOUSTIC RECORDINGS
Acoustic data were collected at SK-B MPA (Fig. 1) by means of three consecutive AURAL
M2 acoustic recorder (Multi-Electronique Inc., Rimouski, QC, Canada) deployments between
July 2011 and July 2013. Details for the three acoustic deployments are provided below in Table
1.
Figure 1. Location of Bowie Seamount in relation to other seamounts in the Northeast Pacific Ocean (left,
adapted from Canessa et al., 2003). Top and bottom right show the location of the three deployments (red
dots) in relation to the seamount bathymetry, with increased resolution (bottom right).The purple lines on
the top right figure delineate the MPA’s three zones: Zone 1 where the acoustic recorders were moored,
Zone 2 for the rest of Bowie Seamount and Zone 3 showing part of the Hodgkins Seamount.
Table 1. SK-B MPA acoustic data.
Deployment Period Coordinates Depth (m) Duty cycle
(min on / off) Sampling
rate (KHz)
D1 Jul 2011- Jan 2012 53.305 135.623 235 9 / 6 16
D2 Jan - Apr 2012 53.307 135.627 233 9 / 6 16
D3 Jul 2012- Jul 2013 53.308 135.627 232 4.5 / 10.5 16
B.AUTOMATIC DETECTIONS OF FISH SOUNDS
An automated detector developed by JASCO Applied Sciences was used to detect potential
fish sounds from these data. The algorithm employed is similar to the one described in Moloney
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 3
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

et al. (2014) and Dewey et al. (2015). The various processing steps of the detector are indicated
in Figs. 2 and 3.
Figure 2. Automatic detection of fish sounds: diagram of the automated processing.
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 4
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

Figure 3. Automatic detection of fish sounds. Top: example of spectrogram with fish sounds, center:
detection of acoustic events in the spectrogram, bottom: classification of fish sounds using a random forest
classifier.
The algorithm first calculated the spectrogram and normalized it for each frequency band.
Next the spectrogram was segmented to detect acoustic events between 10 Hz and 8 kHz. For
each event, a set of 40 features representing salient characteristics of the spectrogram was
extracted, several of which were calculated following Fristrup and Watkins (1993) and Mellinger
and Bradbury (2007), and were based on the spectrogram, frequency envelope, and amplitude
envelope of the signal. Extracted features were presented to a classifier to determine the class of
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 5
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

the sound detected. The classification was performed using a random forest classifier (Breiman,
2001), which was trained using several thousands of manually annotated vocalizations in
recordings collected at different locations in British Columbia (Mouy et al., 2015). The random
forest was defined with five classes: “killer whale”; “humpback whale”; “fish”; “Pacific white-
side dolphin”; or “other”. For this study, only fish classifications were analyzed. Fish sounds
used for the training of the classifier were all collected in the Strait of Georgia in 2014 by a
JASCO tetrahedral hydrophone array deployed on the Ocean Networks Canada VENUS
observatory (Moloney et al., 2014; Dewey et al., 2015). Fish sounds included both pulses and
grunts as illustrated in Fig. 4. Only about 100 fish sounds were used to train the fish class of the
classifier due to the limited availability of the manually annotated fish sounds. Consequently, the
fish classification is less mature than the other classes and needs to be improved. The present
collaboration with JASCO Applied Sciences will provide feedback for the fish automated
detector and any fish sounds that are positively indentified within the SK-B MPA acoustic data
will be used to adjust the classifier.
Figure 4. Examples of fish sounds collected in the Strait of Georgia that were used for training the random
forest classifier.
C.MANUAL VERIFICATION OF THE DETECTOR OUTPUT
Manual verification of the signals highlighted by the automatic detector was performed with
Raven Pro 1.5 (Cornell Lab of Ornithology, Ithaca, NY). Each sound was examined in 15-second
spectrograms to provide context. Spectrogram parameters were 1024 samples FFT, 85% overlap,
51% brightness, 81% contrast. Default initial view displayed the first 2.8 KHz, the top part of the
spectrogram (up to 8.2 KHz) being inspected only when needed. The first 200 Hz were filtered
out for all of Deployment 3 due to intense flow noise that made aural verification difficult. This
high-pass filter was applied only when needed throughout Deployments 1 and 2.
The automatic detector output was delivered in Selection Table format, which when opened
in Raven Pro 1.5 displayed a selection box around each sound it identified as fish. Raven Pro 1.5
allows toggling between selection boxes within a set of open WAV files. This made it possible to
move quickly between selections that were separated in time throughout the data. Each selection
was reviewed manually and annotated with a general “Class” (Unknown, Noise, Humpback
Whale, Sperm Whale, Fish) and “Sound Type” for a more specific and variable description of
the sound (e.g., “Bump”; “Current”; “Seismic”; “Biological?”; “Beginning of call”; “Mooring
noise”; “Vessel noise”; “Possible Fish”; “Echolocation Click”).
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 6
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

3.RESULTS
The detector found 6,387 possible fish sounds in 487 WAV files (Table 2) which were
manually inspected (only 1.2% of the total data). 95.5% of the automatic detections turned out to
be false positives (Fig. 5a and b), and the remaining sounds could not be positively identified to
any satisfactory category (Fig. 6). Twenty-eight percent of these unidentified sounds were
suspected to be of possible fish origin and will need to be further measured, classified and
characterized in order to confirm their source. Half of these possible fish sounds were
stereotyped and sounded like a deck of cards being shuffled (Fig. 7). While verifying these
detections, more card-shuffle sounds that had not been marked by the detector were observed on
the 15-second screen, indicating that they were missed by the detector and suggesting that the
detector needs to be tuned to these types of sounds. Details on number of detections and their
verified type per deployment are presented in Table 2.
Table 2. Automatic detector output and manual verification for acoustic data collected at SK-B MPA
between July 2011 and July 2013.
Deployment Total
WAV
files
WAV
files with
at least 1
detection
Total
detections
Detections
per WAV file
(that had
detections)
Detector
false
positives
Unknown
detections Possible
fish
D1 14,101 279 5127 4-110 4,960 167 69
D2 9,404 84 263 2-27 146 117 11
D3 16,383 124 997 4-14 996 1 0
Total dataset 39,888 487 6,387 6,102 285 80
Figure 5. False Positives. a. Total number of false positives (N = 6,102). b. Details on false positives that
were not attributed to current or seismic noise (N = 171).
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 7
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

Figure 6. Unidentified Sounds (N = 285).
Figure 7. Compilation of four card shuffle sounds. These were originally separated by more than half a
second. Sound between signals has been removed to show several examples together. Mean peak frequency
is 83.3 Hz, bandwidth 16-2704 Hz, mean duration 0.2 s. N=40.
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 8
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

4.CONCLUSION AND NEXT STEPS
There are more than 32,000 known species of fishes (www.fishbase.org) but not all of them
have the ability to produce sound. The mechanisms for fish sound production include sonic
muscles attached to the swim bladder and stridulation mechanisms consisting in rubbing skeletal
elements such as teeth, fin rays, and vertebrae which lead to a great variation of sounds (Ladich,
2015). This variation combined with the very limited knowledge about which species of fishes
are soniferous and what their sounds are like (Hawkins et al., 2015), particularly in waters of
British Columbia (Wall et al., 2014), makes it difficult to identify fish sounds captured by
passive acoustic recorders (Rountree et al., 2006; Rountree, 2008).
The complete lack of confirmed fish sounds found within the SK-B MPA acoustic recordings
does not necessarily indicate absence of sound-producing fish, although it highlights the
challenges of using passive acoustic methods to detect fish sounds and it potentially affects the
next stage of the study which is to compare patterns in fish sound production to the
anthropogenic soundscape. However, the analysis has not been finalized yet. The unknown
sounds found by the detector that are possible fish need to be further measured, classified, and
investigated in order to confirm their source, especially the potential thump or knock sounds
referred to in Fig. 6 and the card shuffle-like sounds (Fig. 7).
In addition, systematic manual inspection of acoustic data and comparison with automated
detections is yet to be performed in order to also quantify false negatives (fish-like sounds that
the detector may have missed). For example, several additional cards-shuffle-like sounds that
had not been marked by the detector were observed, and there might be more. The detector might
not have highlighted all card-shuffle sounds present in the dataset because it was tuned primarily
to knock and groan types of sound. Our preliminary results indicate that these types of sounds
may be uncommon in the study site, while other types of sounds like the card-shuffle sound were
more common. Retuning the detector to optimize for these types of sounds may improve our
ability to detect suspected deep-sea fish sounds. It is also our intention to look for similar type of
sounds in other datasets to assess their occurrence, and use them to better tune the detector for
deep-sea fishes. The more examples we can get of fish sounds produced in the deep sea, the
better we can train the detector.
Since the peak energy of typical fish sounds is most often encountered in the 20-200 Hz
frequency range, these sounds may have been masked by flow noise that was present throughout
a major part of the recordings. Future deployments should select areas based on the presence of
known fish habitat occurrence, rather than solely on geology, and install autonomous recorders
optimized to reduce equipment self-noise and flow noise biases.
Unfortunately, attempts to evaluate the impact of noise on fishes and invertebrates is
compounded by our limited data on sound production in marine and aquatic ecosystems
(Rountree et al., 2002; Rountree et al., 2006; Luczkovich et al., 2008). Sound reference
collections are particularly limited in the eastern Pacific (Wall et al., 2014) and more generally in
deep-sea habitats (Rountree et al., 2012). Therefore, we need more information on sound-
producing fish species, particularly in the northeastern Pacific area, and we need catalogues of
vocalizations produced by known fish species in order to identify sounds captured by passive
acoustic methods. Addressing knowledge gaps on sound type, conditions, and seasonality of fish
sound production will be highly valuable for future studies.
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 9
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

ACKNOWLEDGMENTS
This project was funded by Fisheries and Oceans Canada (DFO)’s Academic Research
Contribution Program (ARCP), by the Canadian Healthy Oceans Network II (CHONE) which is
a strategic partnership funded by the Natural Sciences and Engineering Research Council of
Canada (NSERC) with DFO as the major contributing partner, and by an NSERC Discovery
grant. We thank the DFO Species at Risk Program for funding the data collection. The Bowie
Seamount bathymetry maps were provided by Katie Gale. We gratefully acknowledge travel
support from the Effects of Noise on Aquatic Life Conference.
REFERENCES
Breiman, L. (2001). “Random Forests,” Machine Learning 45: 5-32.
Canessa, R., Conley, K., and Smiley, B. (2003). “Bowie Semount Pilot Marine Protected Area: an ecosystem
overview,” Can. Tech. Rep. Fish. Aquat. Sci. 2461, xi + 85 p
Dewey, R., Dakin, T., Mouy, X., and Urazghildiiev, I. (2015). “A regional hydrophone network: Monitor, detect and
track,” 2015 Underwater Acoustics Conference and Exhibition. Crete, Greece.
https://www.researchgate.net/publication/281283826_A_REGIONAL_HYDROPHONE_NETWORK_MO
NITOR_DETECT_AND_TRACK.
DFO (2011). “Sgaan Kinghlas Bowie Seamount Marine Protected Area monitoring indicators, protocols and
strategies,” DFO Can. Sci. Advis. Sec. Sci. Advis. Rep. 2010/036.
Fristrup, K. M. and Watkins, W. A. (1993). “Marine animal sound classification,” Woods Hole Oceanographic
Institute Technical Report WHOI-94-13. Woods Hole Oceanographic Institution. 32 pp.
http://darchive.mblwhoilibrary.org:8080/bitstream/handle/1912/546/WHOI-94-13.pdf?sequence=1.
Hawkins A. D., Pembroke, A. E., and Popper, A. N. (2015). “Information gaps in understanding the effects of noise
on fishes and invertebrates,” Rev. Fish Biol. Fish. 25, 39-64.
Ladich, F. (2015). “Sound Communication in Fishes,” [electronic resource] in Animal signals and communication
No. 4 (Springer, Vienna).
Luczkovich, J. J., Mann, D. A., and Rountree, R. A. (2008). “Passive acoustics as a tool in fisheries science,” Trans.
Am. Fish. Soc. 137, 533-541.
McDonald M. A., Hildebrand J. A., and Wiggins, S. M. (2006). “Increases in deep ocean ambient noise in the
Northeast Pacific west of San Nicolas Island, California,” J. Acoust. Soc. Am. 120, 711–718.
Mellinger, D.K. and Bradbury, J.W. (2007). “Acoustic measurement of marine mammal sounds in noisy
environments,” Proceedings of the Second International Conference on Underwater Acoustic
Measurements: Technologies and Results, Heraklion, Greece, pp. 273-280.
ftp://ftp.pmel.noaa.gov/newport/mellinger/papers/Mellinger+Bradbury07-
BioacousticMeasurementInNoise-UAM,Crete.pdf.
Moloney, J., Hillis, C., Mouy, X., Urazghildiiev, I., and Dakin, T. (2014). “Autonomous Multichannel Acoustic
Recorders on the VENUS Ocean Observatory,” Oceans - St. John's, 2014. Sep 14-19, 2014. IEEE. 1-6 pp.
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=7003201&url=http%3A%2F%2Fieeexplore.ieee.or
g%2Fxpls%2Fabs_all.jsp%3Farnumber%3D7003201.
Mouy, X., Ford, J., Pilkington, J., Kanes, K., Riera, A., Dakin, T., and Mouy, P. A. (2015). “Automatic Marine
Mammal Monitoring off British Columbia, Canada,” 7th International DCLDE Workshop. July 13-16,
2015, La Jolla, CA.
Popper, A. N., and Fay, R. R. (2011). “Rethinking sound detection by fishes,” Hear. Res. 273, 25-36.
Popper, A. N., and Hastings, M. C. (2009). “The effects of anthropogenic sources of sound on fishes,” J. Fish. Biol.
75 (3), 455-489.
Rountree, R. A., (2008). “Future technological development in passive acoustics underwater observation,” Marine
Technology Reporter, pp. 40-46. [electronic resource] in www.seadiscovery.com (November/December).
Rountree, R. A., Gilmore, R. G., Goudey, C. A., Hawkins, A. D., Luczkovich, J. J., and Mann, D. A. (2006).
“Listening to fish: applications of passive acoustics to fisheries science,” Fisheries 31 (9), 433-446.
Rountree, R. A., Juanes, F., Goudey, C. A., and Ekstrom, K. E. (2012). “Is biological sound production important in
the deep sea?,” in Effects of Noise on Aquatic Life, edited by A. N. Popper, and A. D. Hawkins (Springer-
Verlag, New York), pp. 181-183.
Rountree, R. A., Perkins, P. J., Kenney, R. D., and Hinga, K. R.. (2002). “Sounds of Western North Atlantic Fishes:
Data rescue,” Bioacoustics 12 (2-3), 242-244.
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 10
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17

Simpson, D. S., Radford, A. N., Nedelec, S. L., Ferrari, M. C. O., Chivers, D. P., et al. (2016). “Anthropogenic noise
increases fish mortality by predation,” Nature Comm. 7:10544, 1-7.
Wall, C. C., Rountree, R. A., Pomerleau, C., and Juanes, F. (2014). “An exploration for deep-sea fish sounds off
Vancouver Island from the NEPTUNE Canada ocean observing system,” Deep-Sea Res. I 83, 57-64.
A. Riera et al.
Noise effects on fishes at SK-B MPA
Proceedings of Meetings on Acoustics, Vol. 27, 010005 (2016)
Page 11
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 192.255.79.157 On: Sat, 17 Sep 2016 19:01:17