A large-aperture low-cost hydrophone array for tracking whales from small boats

B. Miller, and S. Dawson

Citation: The Journal of the Acoustical Society of America 126, 2248 (2009); doi: 10.1121/1.3238258
View online: https://doi.org/10.1121/1.3238258
View Table of Contents: https://asa.scitation.org/toc/jas/126/5
Published by the Acoustical Society of America

ARTICLES YOU MAY BE INTERESTED IN

Construction, calibration, and field test of a home-made, low-cost hydrophone system for cetacean acoustic
research
Proceedings of Meetings on Acoustics 11, 010001 (2010); https://doi.org/10.1121/1.3573502

Estimating source position accuracy of a large-aperture hydrophone array for bioacoustics

The Journal of the Acoustical Society of America 109, 397 (2001); https://doi.org/10.1121/1.1329619

Passive acoustic methods for fine-scale tracking of harbour porpoises in tidal rapids
The Journal of the Acoustical Society of America 141, 1120 (2017); https://doi.org/10.1121/1.4976077

Handbook of hydrophone element design technology

The Journal of the Acoustical Society of America 64, S167 (1978); https://doi.org/10.1121/1.2003987

Design and field test of a low-cost-portable linear array for marine mammal localization
Proceedings of Meetings on Acoustics 19, 010002 (2013); https://doi.org/10.1121/1.4800535

Male sperm whale acoustic behavior observed from multipaths at a single hydrophone
The Journal of the Acoustical Society of America 118, 2677 (2005); https://doi.org/10.1121/1.2033567
A large-aperture low-cost hydrophone array for tracking whales
from small boats
B. Miller and S. Dawson
Department of Marine Science, University of Otago, P.O. Box 56, Dunedin 9054, New Zealand
共Received 25 February 2009; revised 13 August 2009; accepted 1 September 2009兲
A passive sonar array designed for tracking diving sperm whales in three dimensions from a single
small vessel is presented, and the advantages and limitations of operating this array from a 6 m boat
are described. The system consists of four free floating buoys, each with a hydrophone, built-in
recorder, and global positioning system receiver 共GPS兲, and one vertical stereo hydrophone array
deployed from the boat. Array recordings are post-processed onshore to obtain diving profiles of
vocalizing sperm whales. Recordings are synchronized using a GPS timing pulse recorded onto each
track. Sensitivity analysis based on hyperbolic localization methods is used to obtain probability
distributions for the whale’s three-dimensional location for vocalizations received by at least four
hydrophones. These localizations are compared to those obtained via isodiachronic sequential bound
estimation. Results from deployment of the system around a sperm whale in the Kaikoura Canyon
in New Zealand are shown. © 2009 Acoustical Society of America. 关DOI: 10.1121/1.3238258兴
PACS number共s兲: 43.30.Wi, 43.30.Sf 关WWA兴 Pages: 2248–2256

I. INTRODUCTION Many of the more advanced localization algorithms in-

volve acoustic ray-tracing and can make use of acoustic mul-
Passive acoustic localization, via arrays of hydrophones, tipath detected in the recordings.4,5,22 Ray-tracing and acous-
has been used to study marine mammals for over 40 years.1–3 tic multipath localization algorithms require detailed
Instrumentation used in previous passive acoustic studies of knowledge of the sound velocity profile. Ray-tracing tech-
sperm whales has ranged from a single hydrophone,4–7 stereo niques can be computationally intensive especially when the
hydrophone arrays,8–11 multihydrophone towed and vertical hydrophone positions are not fixed with respect to each
arrays,12,13 and hydrophone arrays deployed simultaneously other. Additionally, multipath localization algorithms require
from multiple boats.14 Recent efforts to lower the entry bar- detailed knowledge of local bathymetry.5,23 If computation
riers for scientists interested in passive localization include time is limited. or if detailed bathymetry and sound velocity
the system presented by Hayes et al.15 in 2000. This array of profiles are not available, then alternative algorithms must be
relatively inexpensive passive sonar buoys made mostly employed instead.
from commercially available off-the-shelf components was In an effort to improve on hyperbolic localization tech-
successfully used to track blue whales over several kilome- niques, Spiesberger24 introduced a geometric surface called
ters. In 2001 Møhl et al.14 presented a similar unlinked sonar an isodiachron. Isodiachronic localization is a more general
array, but instead of using buoys with single hydrophones, form of hyperbolic localization that allows for the effective
hydrophone arrays were deployed from multiple boats to sound speed between the source and each receiver to differ.
track diving sperm whales.1 By deploying a relatively deep By using isodiachrons with sequential Monte Carlo methods,
vertical array, they were able to reconstruct three- one can estimate not just the location of the sound source,
dimensional 共3D兲 sperm whale tracks. but also the positions of the receivers and effective sound
Study design and hardware choice limit which software speeds. Spiesberger25 dubbed this technique sequential
techniques are appropriate for analysis of the raw data. bound estimation. Like many Monte Carlo methods, this
Beamforming techniques are most appropriately applied for technique is computationally intensive, but yields greater
short aperture towed or vertical arrays and usually yield precision and accuracy than hyperbolic least squares error
animal locations in one or two dimensions 共bearing and estimation under some circumstances.26 Furthermore, se-
range兲. To obtain 3D animal locations, hyperbolic multilat- quential bound estimation does not require detailed knowl-
eration is usually applied to a multi-receiver large aperture edge of the speed of sound, or the bathymetry, though it can
array.2,14,16–19 These hyperbolic localization techniques were make use of this information if it is available.25 In the ab-
originally developed for the LORAN system20 which used sence of measured sound speed profile 共SSP兲, it allows one
radio beacons to localize ships and aircraft, and a key as- to estimate a range of SSPs. Conservative estimates should
sumption was that the propagation speed of the transmission be accurate 共i.e., contain the true value兲, but may not be as
remained constant between the source and all receivers. The precise as using a measured SSP.
assumption of a constant speed of sound propagation is not In Sec. II, we present a passive sonar system that can
always valid for underwater applications, especially in a best be described as a hybrid between the approach taken by
place like Kaikoura, New Zealand, where several water Møhl et al.14 and Hayes et al.15 This system was designed for
masses converge and complex physical oceanographic pro- use from a single small 共6 m兲 boat with the purpose of lo-
cesses are at work.21 calizing diving sperm whales which at Kaikoura produce

2248 J. Acoust. Soc. Am. 126 共5兲, November 2009 0001-4966/2009/126共5兲/2248/9/$25.00 © 2009 Acoustical Society of America
loud broadband clicks at a mean click rate of 1.3 clicks/ s cable. The cable chosen was TMB Proplex CAT5e which has
about 60% of the time 共including their time at the surface兲.27 a light Kevlar strength member allowing a maximum work-
Our system consists of four free-floating buoys and one long ing pull of 140 N. Previous experience revealed that record-
vertical cabled array, all deployed from the same boat. Our ings made with hydrophones shallower than 20 m, resulted
system is based on commercially available off-the-shelf in increased surface noise, as well as distortion from surface
hardware, offers ease of deployment and recovery of buoys echo multipaths. A 2 kg lead weight was attached to the end
from a single platform, yet it allows measurements of three of the hydrophones to speed deployment, maintain hydro-
dimensional movements of nearby vocalizing whales. The phone depth, and reduce hydrophone drift with respect to the
system also allows measurement of bearings to whales that buoy. Each hydrophone was connected to one channel of the
are many kilometers away. We describe the approach we recording device via a waterproof connector embedded in the
have taken in hardware and software, and discuss the advan- buoy lid, while modulated GPS data were recorded on the
tages and limitations of operating this array from a 6 m boat. other channel.
Additionally, we show results from deployment of the sys- The GPS used on each buoy was the Garmin GPS-
tem around a sperm whale in the Kaikoura Canyon in New 17HVS. This low cost OEM GPS was chosen because of its
Zealand. waterproof housing and its ability to output an accurate tim-
ing signal in addition to the raw GPS carrier phase. Via post-
II. METHODS processing 共with data from a suitable base-station兲, the raw
A. Array design carrier phase can be used to obtain highly accurate 共subme-
ter兲 position information. The GPS position output was con-
The design was heavily influenced by the limited deck nected to an FSK modulator while the GPS timing signal
space on a small vessel. The components of the array had to was connected to an amplitude modulator. Detailed descrip-
be compact, robust, and quickly deployed. We opted for a tion of the FSK modulation is beyond the scope of this paper,
modular approach in order to reduce maintenance time 共de- for an overview of FSK modulation with respect to this ap-
fective components can be quickly replaced兲 and to ensure plication consult Møhl et al.14
that overall success in tracking did not depend too heavily on An Oceanic Veo 250 personal scuba diving computer
any one component. The modular approach also leaves open was used to record the depth of each hydrophone throughout
the possibility of adding additional instrumentation as future the duration of the deployment. The depth resolution of the
needs dictate. dive computers was 0.3 m. Depth sensors in conjunction
Design specifications required sufficient battery and with post-processed GPS positions allowed for more accu-
storage capacity to make relatively wideband recordings for rate estimation of the hydrophone position. This was neces-
several hours at a time. A further requirement common to sary when ocean currents and/or wind caused the hydro-
non-linked arrays is that recordings from different platforms phone to drift so that it was not directly beneath the GPS
had to be synchronized precisely. receiver. Testing at the field site revealed this to be necessary
for only the deeper boat-based array.
1. Buoys Each buoy had a Sirtrack VHF radio transmitter beacon
Similar to the instrument packages of Hayes et al.15 each mounted on a 1 m tall mast. The beacon was used to assist
of our buoys includes a hydrophone, a recording device, a with relocation and recovery at the end of a recording. Bea-
GPS receiver, a time synchronization device, and a battery cons were tracked with a Yagi aerial connected to an Icom
pack. Additional instrumentation included an optional fourth IC-R10 wideband scanner. Including a radio beacon on each
order bandpass filter 共passband 1–40 kHz兲, a depth logger buoy also helped with keeping track of drifting buoys during
attached to each hydrophone, and a VHF locator beacon at- a recording. This was especially useful over large deploy-
tached to a small mast on each of the buoys. ment areas and when sea state and weather conditions made
The recording device used in each buoy was the spotting buoys difficult.
M-Audio Microtrack 24/ 96 with a 16 Gbyte compact flash Within each buoy, the recorder, FSK circuit, bandpass
card as the recording medium. Each Microtrack had stereo filter, and two 12 V gel cell batteries were housed on a cus-
recording capabilities. We used a sample rate of 96 kHz 共16 tom frame which was made from 80 mm diameter PVC
bit兲 which gave a maximum record time on 16 Gbyte media drainage pipe. Slots were cut from the drainage pipe and
of over 11 h. While the Microtrack can record 24 data bits components secured to the frame via cable ties. The frame
per sample, ambient ocean noise and electrical noise within was placed within a watertight 100 mm diameter housing
the device itself effectively rendered this setting superfluous. also made from PVC drainage pipe. The gel cells were
Recording quality could be lowered to 44.1 kHz to enable up placed at the bottom of the frame with 1 kg of lead ballast to
to 24 h of recording, though this was not attempted during help the buoy maintain a vertical attitude in the water. Closed
this study. As compact flash cards increase in size and de- cell foam was glued around the top of the housing to provide
crease in price, recording duration can be increased. The Mi- additional floatation 关Fig. 1共B兲兴. Buoy dimensions were 1 m
crotrack recorders were powered using their internal battery in height and 100 mm in diameter. Deck space on the re-
which gave up to 4 h of operating time. search vessel was limited so the four buoys were stowed
All hydrophones were built in-house as described by upright on deck in a purpose-built wooden rack 关Fig. 1共A兲兴.
Barlow et al.28 A single hydrophone was connected to each While we chose to package the instruments in a 1 m tall tube,
buoy with 20–30 m of shielded, harsh-environment ethernet the instruments could have fitted into an enclosure as small

J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat 2249
 depth resolution of our system was provided primarily from
the boat-based array, it was especially important that the
boat-based array remain close enough to detect the animal
continuously throughout the recording. In addition to con-
stant feedback, the range and bearing estimate also provided
an independent check on the validity of localizations ob-
tained from the non-linked array.

B. Deployment and recovery

FIG. 1. 共A兲 Photograph of 4 buoys and boat-based array ready for deploy- Before deploying the buoys and boat-based array, sperm
ment from the research vessel. 共B兲 Buoy schematic showing arrangement of whales were tracked using a custom-built directional hydro-
the Microtrack recorder 共REC兲, GPS receiver, FSK modulation electronics,
and power supply 共BAT兲. phone. Whales were tracked until they surfaced in order to
obtain an identification photograph29,30 before deploying the
array. The photographed whale 共target whale兲 was typically
as 0.30 m long, making these buoys especially suitable for
tracked via directional hydrophone for 20–25 min to ascer-
operation from small vessels. The extra space inside each
tain its speed and heading 共if any兲 before deploying the first
buoy can be used for additional instrumentation or extra bat-
buoy. Because sperm whales in Kaikoura can travel several
tery packs.
kilometers from fluke up, to fluke up,31 it was important to
Hydrophones were secured to the rack next to each
have an estimate of the target whale’s position and general
buoy, and hydrophone cables were wound onto the rack in a
direction of movement before deploying the buoys.
figure eight fashion to prevent tangling. By keeping buoy
Buoys were typically deployed in either a triangle or
dimensions small and stowing them upright, deployment re-
square configuration surrounding the animal共s兲 of interest,
mained manageable even with limited deck space. A single
with the boat and stereo hydrophone array 共and with a bit of
manual switch within each buoy activated power to all elec-
luck the target whale兲 at the center of the polygon. Typical
tronics further facilitating speedy deployment.
deployment distances between adjacent buoys were 1.5–2
km. To deploy four buoys typically took 20 min; however,
2. Boat-based cabled array poor weather, other vessels, and navigational hazards such as
The boat-based stereo array consisted of two custom fishing gear increased deployment time.
built hydrophones spaced 5 m apart on 105 m of cable. The To deploy buoys, the boat engine was stopped, the buoy
recording device used on the boat was an Edirol R-4, four electronics turned on, and the Microtrack set to record. The
channel digital recorder operating at 96 kHz sample rate with hydrophone cable was deployed before placing the buoy
a sample resolution of 16 bits. Hydrophone and GPS data overboard, and the level meter on the Microtrack was
were recorded onto the first three channels, while the fourth checked to make sure that both the hydrophone and GPS
channel was used to record dictated commentary about the recording chains were functioning and that the recorder gains
situation. Commentary included descriptions of animal be- were set appropriately The waterproof lid was replaced, and
haviors, vocalizations, weather information, sea state, esti- the VHF locator beacon was attached before placing the
mated whale position, and movements of any other vessels in buoy over the side of the boat. To facilitate recovery, each
the area. buoy’s deployment location was marked on the vessel’s navi-
Using a deeper array proved crucial for obtaining accu- gation GPS.
rate 3D localizations of the target animal. The short distance Some care was required when setting the recording gain
between hydrophones of the boat-based array also facilitated for each buoy because this could not be changed after de-
tracking individual animals when several were vocalizing at ployment. Over the course of the recording, whale共s兲 can
the same time. The boat-based array functioned as a short- swim toward some buoys and away from others, causing a
aperture vertical array which was used to measure the verti- large change in received level. The dynamic range of the
cal bearing to vocalizing animals using a custom MATLAB Microtrack recording unit was not wide enough to accommo-
script. For short time scales 共tens of seconds兲, vocalizations date this change in received level without gain adjustment.
with widely different bearings are likely to come from dif- Had the 24-bit recordings on the Microtrack not been domi-
ferent individuals. Similarly, bearings coming from the same nated by electrical noise, then changing the bit resolution
individual would be expected to change gradually over short from 16 to 24 bits would have greatly helped handle this
time scales. These assumptions were used by the classifica- limitation. While hardware-based automatic gain control
tion algorithm to reduce ambiguities during analysis that oc- might have been a possible solution, we did not pursue this.
curred due to multiple vocalizing animals. Because the goal of this study was localization of whales, we
In addition to the recordings made using the stereo hy- opted for a higher gain setting in order to detect whales fur-
drophone array, a handheld directional hydrophone and com- ther away and maximize buoy separation. This choice came
pass were used to estimate the range and bearing to the target at the expense of clipping some loud whale vocalizations.
animal共s兲 throughout the deployment. Tracking the whale When the target animal was unlikely to be audible on a
this way provided feedback necessary to reposition hydro- minimum of four hydrophones, buoys could be repositioned;
phones that drifted too far from the target whale. Because the however, this required an interruption in monitoring with the

2250 J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat
boat based array which effectively limited localization accu-
Time
racy to two dimensions 共x and y兲 during this time. While Modulated
GPS Data
PPS Edge
Detection Alignment
buoys could be repositioned by a dedicated support vessel, Spreadsheet
this would effectively double the operating costs of the array
and has not been attempted. More often instead of reposi-
PC Sound FSK PC Serial
tioning buoys the recording was terminated and the buoys Card(s) Demodulator Port(s)
recovered. After recovery, data were downloaded to a per-
sonal computer 共PC兲 for synchronization and analysis.
FIG. 2. Data flow for time alignment and demodulation of recorded GPS
data.
C. Synchronization
The standard procedure for localizing animals with an
instantaneous jitter was computed as the difference between
acoustic array involves computing time of arrival differences
nominal sample rate and instantaneous sample rate.
共TOADs兲 between each pair of hydrophones for each
vocalization.2,3,14,15,17,18,32–36 When using an unlinked array, Custom synchronization software written using Math-
all recordings made at each location must be synchronized works MATLAB was used to detect the sample number corre-
before TOADs can be computed accurately. Synchronization sponding to the leading edge of the amplitude-modulated
must address both jitter and clock drift. For the purposes of timing pulse. This edge detection software began by loading
this article, jitter can be thought of as very short-term 1 s of audio into memory. This audio was divided into ten
changes in the sample rate, while clock drift refers to long- consecutive sequences and the rms amplitude of each se-
term differences between the device’s clock and the GPS quence was computed. Due to the amplitude modulation, the
synchronization signal. Both of these errors arise from im- sequences containing the timing pulse had a different rms
perfections in the digital clock used for analog-to-digital amplitude than the rest of the signal. The earliest sequence
conversion in the recording unit. with a different rms amplitude contained the leading edge of
For our system, jitter, measured during synchronization, the pulse. This sequence was kept, while the others were
was typically on the order of 0.002% for all devices. Clock discarded. For each sample in the remaining sequence, the
drift was also measured and was typically between 0.5 and 2 rms value of the subsequent 20 samples was computed. The
ms/min for all devices. Measurement of the jitter and drift difference between sequential rms values was computed and
rates is not only necessary for accurate localization but also the sample with the largest change in rms amplitude corre-
provides a measure of the temporal fidelity of the audio de- sponded to the leading edge of the PPS. This process was
vice. Audio time alignment and jitter/drift correction was repeated for the duration of the recorded audio.
performed via a two-stage process. The first step involved Simultaneous to edge detection, the FSK-modulated au-
coarse alignment, which synchronized the start and end of dio track was played into a custom-made hardware demodu-
each recording to within 1 s and assumes constant drift and lator and the GPS data were recorded via a PC serial port.
no jitter. The second step 共fine scale alignment兲 provided Hardware FSK demodulation with concurrent software PPS
sample-accurate audio synchronization once every second detection allowed for synchronization of multiple buoys at
for the duration of the recorded audio. the same time. The audio sample number of the PPS edge,
The GPS position information stored in the FSK- latitude, longitude, UTC time, and raw carrier phase infor-
modulated audio signal included latitude, longitude, as well mation from each platform was written to a synchronization
as UTC date and time of the signal with time resolution of 1
data file for every second of audio processed 共Fig. 2兲. By
s. For coarse alignment, we extracted this information from
using a hardware-based demodulator circuit for each of the
the first and last seconds of the recording to compute the
five audio channels, 5 channel-hours of modulated GPS po-
GPS start and end seconds for the recording. Subtracting the
sition and timing information could be decoded in 1 h. This
ending GPS second from the starting second yields the GPS
proved to be significantly faster than our best attempts at
duration, tGPS. The average clock drift rate was computed as
共tGPS − trecording兲 / tGPS, where trecording was the total number of implementing software-based demodulation as described by
audio samples per channel divided by the nominal sample Møhl et al.14 and can work in real time provided that there
rate 共96 kHz兲. This coarse alignment does not account for are as many demodulators and serial ports as there are modu-
jitter or inaccuracies resulting from a non-constant clock lated GPS signals.
drift rate over the duration of the recording. To investigate Hydrophone depth sensors were activated via a water
these errors and account for them if they are present, we used contact switch. Depth data were synchronized with the GPS
the Garmin GPS 17 timing signal, which is a 1 Hz pulse data using either the audible tone made when the depth sen-
wave with a duty cycle of 0.1. The rising edge of this pulse sor was active for a predetermined amount of time or the
marked the start of the GPS second with a nominal accuracy sound made from the entry of the depth sensor into the water,
of ⫾1 µs 共Garmin GPS 16/ 17 Technical manual兲. When the both of which were audible in the recording for each hydro-
100 ms pulse was active, it reduced the signal amplitude of phone. At the end of preprocessing, the multichannel record-
FSK-modulated GPS data. The instantaneous sample rate of ings of the whales’ sounds, location data for each buoy and
the recording unit was computed by simply counting the the boat, and the depth data for each hydrophone were syn-
number of samples between successive pulse edges, and the chronized.

J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat 2251
D. Detection and localization 200 ms after the direct arrival was considered a surface echo
so long as the time lag of this peak did not correspond to a
Recently there have been many different techniques pro-
direct arrival from another bearing trace. All surface reflec-
posed for detection, classification, and localization of sperm
tions from a particular bearing trace were written to a sepa-
whale clicks. To obtain 3D whale positions, we implemented
rate log file. These surface reflections can be thought of as
a selection of detection, classification, and localization
arriving at a virtual hydrophone that mirrors the real hydro-
algorithms22,25,37,38 using Mathworks MATLAB 7.3, and adapt-
phone above the ocean surface.5,9,46 These virtual hydro-
ing the methods for use with our system as necessary.
phones were used as additional receivers and increase both
1. Detection the number of hydrophones in the array and the vertical hy-
drophone separation, thus increasing the localization perfor-
For detection of sperm whale vocalizations, audio re- mance of the array.16
cordings from each platform were bandpass filtered between
2 and 20 kHz, a band which contains most of the energy of
4. Classification „click association…
typical sperm whale vocalizations. Vocalizations were de-
tected from filtered recordings using Page’s test, which is an The inter-click interval from each of the bearing traces
energy detector.37 Specifically, we followed the algorithm was computed and used as input into a custom MATLAB pro-
outlined in Ref. 39, Part II, Sec. I. While there have recently gram that implemented the ⬙rhythm analysis⬙ algorithm de-
been numerous methods for detection and classification of scribed by Thode.22 This algorithm was necessary to associ-
sperm whale clicks,40–44 Page’s test was chosen because its ate vocalizations received at each buoy and virtual
implementation was intuitive, fast, and it has been used suc- hydrophone with vocalizations received from an individual
cessfully in previous studies involving localization of sperm whale at the stereo array. When a vocalization was matched
whales.39,45 Detection parameters that yielded good agree- at 4 or more real hydrophones, arrival time differences were
ment with visual inspection of the spectrogram for the first calculated between all hydrophone pairs by computing the
few minutes of audio were selected for use. The detection cross-correlation of the audio for each matching detection.
threshold, V1, was set to 16 共24 dB兲; the end of detection The time lags of the peak of the cross-correlation function
threshold, V0, was 1; and the exponential weighting on the were stored as TOADs and used in further localization analy-
noise, ␣, was set to 0.9 共notation follows Ref. 39兲. Automat- sis.
ing the detection process was necessary for analyzing the
large number of recorded sperm whale vocalizations. 5. Localization

2. Bearing localization Once all TOADs have been computed, these data as well
as the hydrophone positions were used as input into a MAT-
TOADs were computed between both hydrophones from LAB program that implemented the hyperbolic localization
the boat-based stereo array using cross-correlation of the algorithm described by Spiesberger and Fristrup.38 To esti-
waveform of detected clicks. The time lag at the peak in the mate the localization precision, a separate sensitivity analysis
cross-correlation function was recorded as the TOAD of a was performed.
direct arrival. Because the distance between these hydro- For the sensitivity analysis, we assigned uncertainty to
phones was much smaller than the distance to the target each of the model inputs and created uniform probability
whale, these TOADs provide a measure of the angle of ar- distributions based on the measured data and estimated/
rival of the sound. These angles were plotted as a function of measured uncertainty for each of the model inputs. Variance
time to yield a bearing-time plot for the recording. Bearing for the horizontal hydrophone position was measured to be
tracks were traced by a human operator and traces were num- ⫾2 m based on a 48 h comparison of each GPS receiver to a
bered and assigned to an individual whale via a custom MAT- surveyed reference position. Variance for each hydrophone
LAB interface. Bearings were traced with the following crite- depth sensor was assumed to be ⫾0.3 m according to the
ria. Bearings that corresponded to an individual whale track manufacturer’s specifications. Effective sound speeds were
must change slowly and continuously over time. This con- allowed to vary across the range of sound speeds computed
straint eliminated noise sources from being selected as a with equations from Del Grosso47 using historical monthly
whale trace. Any ambiguities in a trace, such as the intersec- temperature, salinity, and depth data for the study area from
tion of multiple traces or gaps longer than 7 min, resulted in the World Ocean Atlas.48,49 TOAD variance was computed
the termination of a trace and the start of a new trace at a according to Spiesberger and Fristrup38 equation 41. We then
time after the ambiguities could be resolved. A recording drew 2000 samples from each of these random variables and
typically contained between 1 and 6 individual bearing traces used each set of samples as input to the localization algo-
at any one time. rithm to obtain a cloud of points that represents the whale’s
position.
3. Surface echo detection For the x, y, and z coordinates of the whales position,
Echo detection based on autocorrelation was performed probability distributions, Px, Py, and Pz, were estimated at
on vocalizations from each bearing trace from the vertical each time step from the output of the sensitivity analysis.
array. For each vocalization, the absolute value of the auto- Estimates of Px, Py, and Pz were calculated from the normal-
correlation of the waveform was computed. The largest peak ized histograms of each x, y, and z coordinate of the whale’s
in this autocorrelation function that occurred between 10 and position using bin widths of 1 m. The total volume for each

2252 J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat
 37.8

37.6
End

37.4

37.2

37
Northing (km)

36.8

End

36.6

36.4
End

36.2 End

36

FIG. 4. Joint X-Y whale position probability from isodiachronic sequential

35.8 Monte Carlo analysis. The whale circles as he dives initially and then heads
21.5 22 22.5 23 23.5 24
Easting (km) from northwest to southeast.

FIG. 3. October 30, 2007 array deployment geometry. Bold line shows the
E. Trial deployment
track of the boat, while normal lines show the track of the buoys. The thin
line with crosses is the whale track, which shown in detail in Fig. 4. On 30 October 2007, the array was deployed around a
single male sperm whale diving over the Kaikoura canyon
共Fig. 3兲. The array was deployed from the research vessel
localization cloud was computed as 共P̂x − P̌x兲共P̂y − P̌y兲共P̂z
Grampus, a 6 m aluminum boat, operating over the Kaikoura
− P̌z兲, where P̂ and P̌ denote the maximum and minimum canyon with a crew of two. Using the directional hydro-
values from the probability distributions. The total cloud vol- phone, three sperm whales were detected vocalizing; how-
ume is a measure of localization precision. A threshold vol- ever, only one sperm whale was estimated to be within the
ume of 1.77⫻ 106 m3, which is equal to the volume of a bounds of the array at the time of deployment. When pos-
sphere with a diameter of 150 m 共approximately 10 whale sible the range and bearing to the diving whale were mea-
lengths兲, was used to exclude localizations with low preci- sured at the surface using a hand bearing compass and laser
sion 共Fig. 6兲. range finder 共Bushnell Yardage Pro Compact 600兲.
Because hyperbolic localization can yield incorrect re-
sults in a stratified environment, isodiachronic sequential
III. RESULTS
bound estimation25 was used to spot-check the whale’s posi-
tion at 15 s intervals starting from the first vocalization. Directional hydrophone estimates of the whale’s posi-
While isodiachronic sequential bound estimation is more ac- tion indicate that the whale dived near the research vessel
curate than hyperbolic localization, our implementation of and moved toward the southeast. This is consistent with the
this method was computationally intensive and would have track generated by the isodiachronic sequential bound esti-
taken prohibitively long to analyze every vocalization this mate 共Fig. 4兲. However, estimates from the directional hy-
way. Performing the sequential bound localization every 15 s drophone were not precise enough to reveal that the whale
served as a quality control check on the hyperbolic localiza- descended along the path of a spiral, which can be seen in
tion results. the 3D tracks computed via sequential bound estimation
The same random variables created for the sensitivity 共Fig. 5兲. The precision of localization decreased as the whale
analysis were used as inputs into the isodiachronic sequential moved away from the center of the array 共Figs. 6 and 7兲.
bound localization algorithm to obtain a cloud of potential When localization algorithms could make use of surface re-
whale positions.25 The shape of this cloud reflects the opti- flected multipath the localization precision substantially in-
mal localization precision and accuracy of the system with- creased 共Fig. 6兲, which is consistent with results described by
out requiring the constraint of a homogeneous environment. Wahlberg et al.16
When localization clouds from the sequential bound estima- The median whale position from the marginal distribu-
tion are drastically different than those from the hyperbolic tions of the hyperbolic sensitivity analysis, Px, Py, and Pz,
localization algorithm, then the assumption of an isovelocity fell within the 95% confidence intervals from the isodiach-
sound speed is likely to be invalid.26 ronic sequential bound analysis 共Fig. 7兲. Maximum depth

J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat 2253
 8
−250 10
Hyperbolic Real Only
Hyperbolic Real + Virtual
Isodiachron Real Only
Start
Isodiachron Real + Virtual
−300
7 Sphere with diameter 15m
10
Sphere with diameter 45m
Z Position (m)

Sphere with diameter 150m

−350

6
10
−400

Volume (m3)
5
−450 10
0

20

4
40 10

60

80
3
80 100 10
X Position (m) 100 40 60
0 20
Y Position (m)
00:30 00:40 00:50
UTC Time (HH:MM)
FIG. 5. Whale trajectory in 3D from the beginning of the dive 共time
00:26:15–00:35:32兲. Whale spirals as he descends. Each cloud corresponds
FIG. 6. The total volume of each localization cloud. Open circles and filled
to one whale vocalization. For clarity successive clouds have different shad-
circles show volumes from hyperbolic sensitivity analysis using real hydro-
ings and the solid line connects the median point from each cloud.
phones and using virtual hydrophones from multipath surface reflections,
respectively. Squares and diamonds show isodiachron sequential bound
analysis again using real and virtual hydrophones. Solid line indicates the
was 599 m, while mean depth was 418 m which is compa- precision threshold. Dashed and dot-dashed lines show reference volumes
rable to sperm whale diving depths measured in other parts corresponding to spheres with diameters of 45 and 15 m 共3 and 1 whale-
of the world.34,50 length兲, respectively.

IV. DISCUSSION overall accuracy of the isodiachronic analysis. This trade-off

is only necessary when computing power or analysis time are
The array has been successfully used to localize diving limited.
sperm whales in the Kaikoura canyon. Results and error es- Our system does have a few additional limitations. A
timates obtained are consistent with those obtained from notable limitation is the limited dynamic range of the record-
other passive sonar systems used to track sperm whales in ing units, resulting from our choice of inexpensive off-the-
3D.17,19,34,45 Tracking whales in three dimensions has the po- shelf field recorders. Because the goal of our system was 3D
tential to yield information that may not be observable from localization rather than measurement of sonar emission pat-
tagging animals with a depth logger. The spiral at the begin- terns, we view this trade-off as acceptable since it allows
ning of the dive in Fig. 5 is a clear illustration of an advan- detection and localization over greater ranges. Automatic
tage of 3D tracking. As computer processors, analog-to- gain control or recording devices with wider dynamic range
digital converters, and digital storage become more powerful would address this issue. Another limitation of the system is
and affordable, so should the ability to create inexpensive the amount of time required to process the data from each
passive sonar systems with higher fidelity hardware and platform to obtain localization. Presently the largest portion
more sophisticated on-board software. of processing time is spent demodulating the GPS positions,
As computing power increases, it should become fea- which takes the same amount of time as the duration of the
sible to perform the more accurate isodiachronic analysis for recording. Lastly, a major limitation of the system is that
every vocalization instead of using it primarily as a quality high accuracy localizations only occur when the whale is
control check. By using the fast hyperbolic analysis method within the bounds of the array. For whales that can swim
to localize every click and validating the results with the several kilometers in a single dive, some luck is required to
slower isodiachronic analysis, we attempt to strike a compro- obtain high accuracy localizations over a full dive cycle.
mise between computation time and accuracy. By using con- However, even the subset of observations for which the
fidence intervals from the sequential bound analysis instead whale remains inside the bounds of the array has the poten-
of the hyperbolic analysis, we effectively trade the higher tial to yield important insights into the underwater behavior
temporal resolution of the hyperbolic analysis for higher of these whales.

2254 J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat
 23.4 field work. We thank Hamish Bowman for help with buoy
construction and design advice, Ross Vennell for assistance
Easting (km)
23.2
23
writing localization software, and Paul Denys for many con-
sultations about GPS postprocessing. Bertel Møhl and Niels
22.8
Kristiansen generously provided their FSK modulator and
22.6 demodulator designs, as did Aleks Zosuls for bandpass filter
00:25 00:30 00:35 00:40 00:45 00:50 designs. Funding for field work was provided by the New
Zealand Whale and Dolphin Trust. Funding for instrumenta-
tion was provided by Otago University.
36.8
Northing (km)

1
36.6 W. Cummings, B. Brahy, and W. Herrnkind, “The occurrence of underwa-
ter sounds of biological origin off the west coast of Bimini, Bahamas,” in
36.4 Marine Bio-Acoustics: Proceedings of a Symposium Held at the Lerner
Marine Laboratory, Bimini, Bahamas, edited by W. Tavolga, Pergamon,
36.2
New York 1964, pp. 27–43.
2
36 W. Watkins and W. Schevill, “Sound source location by arrival-times on a
00:25 00:30 00:35 00:40 00:45 00:50 non-rigid three-dimensional hydrophone array,” Deep-Sea Res. 19, 691–
706 共1972兲.
3
C. W. Clark, W. T. Ellison, and K. Beeman, “Acoustic tracking of migrat-
ing bowhead whales,” Proc. IEEE Oceans ’86 共IEEE, New York, 1986兲,
−300 pp. 341–346.
4
C. Laplanche, O. Adam, M. Lopatka, and J.-F. Motsch, “Male sperm
Depth (m)

−400 whale acoustic behavior observed from multipaths at a single hydro-

phone,” J. Acoust. Soc. Am. 118, 2677–2687 共2005兲.
−500 5
C. O. Tiemann, A. M. Thode, J. Straley, V. O’Connell, and K. Folkert,
−600 “Three-dimensional localization of sperm whales using a single hydro-
phone,” J. Acoust. Soc. Am. 120, 2355–2365 共2006兲.
00:25 00:30 00:35 00:40 00:45 00:50 6
M. Q. Rhinelander and S. M. Dawson, “Measuring sperm whales from
Time their clicks: Stability of interpulse intervals and validation that they indi-
cate whale length,” J. Acoust. Soc. Am. 115, 1826–1831 共2004兲.
FIG. 7. Whale location as a function of time. Dots show the median of the 7
N. Jaquet, S. Dawson, and L. Douglas, “Vocal behavior of male sperm
marginal probability distribution from the hyperbolic sensitivity analysis.
whales: Why do they click?,” J. Acoust. Soc. Am. 109, 2254–2259 共2001兲.
Squares show the 95% confidence limits from the isodiachronic sequential 8
J. Barlow and B. Taylor, “Estimates of sperm whale abundance in the
bound estimation.
Northeastern temperate Pacific from a combined acoustic and visual sur-
vey,” Marine Mammal Sci. 21, 429–445 共2005兲.
9
The principal advantages of our system are its low cost, E. K. Skarsoulis and M. A. Kalogerakis, “Ray-theoretic localization of an
impulsive source in a stratified ocean using two hydrophones,” J. Acoust.
portability, and ease of use from a small boat. The instrumen-
Soc. Am. 118, 2934–2943 共2005兲.
tation has no moving parts and can survive bumps and jostles 10
A. Thode, “Three-dimensional passive acoustic tracking of sperm whales
that occur at sea during difficult weather conditions. The 共Physeter macrocephalus兲 in ray-refracting environments,” J. Acoust. Soc.
system is portable and unlike a fixed hydrophone array, it can Am. 118, 3575–3584 共2005兲.
11
B. K. Nielsen and B. Møhl, “Hull-mounted hydrophones for passive
be deployed and repositioned around the target animals. Pro- acoustic detection and tracking of sperm whales 共Physeter macroceph-
cessing occurs on shore and requires a desktop PC with ad- alus兲,” Appl. Acoust. 67, 1175–1186 共2006兲.
equate storage space 共5–10 Gbytes/recording session兲. Each 12
V. Teloni, “Patterns of sound production in diving sperm whales in the
one of our buoys can be built using mostly off-the-shelf com- Northwestern Mediterranean,” Marine Mammal Sci. 21, 446–457 共2005兲.
13
A. Heerfordt, B. Møhl, and M. Wahlberg, “A wideband connection to
ponents with a total cost of the parts under US$1000. sperm whales: A fiber-optic, deep-sea hydrophone array,” Deep-Sea Res.,
While the main goal of our study is to detect and local- Part I 54, 428–436 共2007兲.
14
ize sperm whales in the Kaikoura canyon, the array could B. Møhl, M. Wahlberg, and A. Heerfordt, “A large-aperture array of non-
potentially be used to localize any loud sound sources in the linked receivers for acoustic positioning of biological sound sources,” J.
Acoust. Soc. Am. 109, 434–437 共2001兲.
area including baleen whale vocalizations, shipping traffic, 15
S. A. Hayes, D. K. Mellinger, D. A. Croll, D. P. Costa, and J. F. Borsani,
underwater explosions, or construction activity. On one oc- “An inexpensive passive acoustic system for recording and localizing wild
casion the system was used to localize concurrently not only animal sounds,” J. Acoust. Soc. Am. 107, 3552–3555 共2000兲.
16
a nearby diving sperm whale but also a singing humpback M. Wahlberg, B. Møhl, and P. T. Madsen, “Estimating source position
accuracy of a large-aperture hydrophone array for bioacoustics,” J. Acoust.
whale in Kaikoura at an approximate range of 8 km 共B. Soc. Am. 109, 397–406 共2001兲.
Miller, unpublished兲. 17
P. Giraudet and H. Glotin, “Real-time 3d tracking of whales by echo-
Long-term use of the system has the potential to provide robust precise TDOA estimates with a widely-spaced hydrophone array,”
Appl. Acoust. 67, 1106–1117 共2006兲.
insight into whether individual whales have different forag- 18
R. Morrissey, J. Ward, N. DiMarzio, S. Jarvis, and D. Moretti, “Passive
ing styles and how diving behavior changes with season. acoustic detection and localization of sperm whales 共Physeter macroceph-
Additionally the system may be used to investigate the ef- alus兲 in the tongue of the ocean,” Appl. Acoust. 67, 1091–1105 共2006兲.
19
fects of anthropogenic noise from sources such as whale L. F. Baumgartner M. F., J. Partan, K. Ball, and K. Prada, “Tracking large
marine predators in three dimensions: The real-time acoustic tracking sys-
watching platforms on sperm whale underwater behavior. tem,” IEEE Transl. J. Magn. Jpn. 33, 146–157 共2008兲.
20
J. P. Van Etten, “Loran c system and product development 共loran c prin-
ACKNOWLEDGMENTS ciples, ground station operation and navigation equipment兲,” Electrical
Communication 45, 100–115 共1970兲.
Thanks to Miranda van de Linde, Elanor Hutchison, Liz 21
F. Lariviere, “Sperm whale habitat off Kaikoura, New Zealand, a descrip-
Slooten, and all of our volunteers for providing help with tive study of physical and biological processes influencing sperm whale

J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat 2255
 distribution,” MS thesis, Otago University, Dunedin, New Zealand 共2001兲. 36
E.-M. Nosal and N. L. Frazer, “Track of a sperm whale from delays
22
A. Thode, “Tracking sperm whale 共Physeter macrocephalus兲 dive profiles between direct and surface-reflected clicks,” Appl. Acoust. 67, 1187–1201
using a towed passive acoustic array,” J. Acoust. Soc. Am. 116, 245–253 共2006兲.
共2004兲. 37
D. Abraham, “Passive acoustic detection of marine mammals using Page’s
23
C. Tiemann, A. Thode, J. Straley, K. Folkert, and V. O’Connell, “Model- test,” Applied Sequential Methodologies: Real-World Examples With Data
based passive acoustic tracking of sperm whale foraging behavior in the Analysis, edited by N. Mukhopadhyay, S. Datta, and S. Chattopadhyay
Gulf of Alaska,” J. Acoust. Soc. Am. 118, 1909–1909 共2005兲. 共Marcel Dekker, New York, 2004兲.
24
J. L. Spiesberger, “Geometry of locating sounds from differences in travel 38
J. Spiesberger and K. Fristrup, “Passive localization of calling animals and
time: Isodiachrons,” J. Acoust. Soc. Am. 116, 3168–3177 共2004兲. sensing of their acoustic environment using acoustic tomography,” Am.
25
J. L. Spiesberger, “Probability distributions for locations of calling ani- Nat. 135, 107—153 共1990兲.
mals, receivers, sound speeds, winds, and data from travel time differ- 39
W. M. X. Zimmer, P. L. Tyack, M. P. Johnson, and P. T. Madsen, “Three-
ences,” J. Acoust. Soc. Am. 118, 1790–1800 共2005兲. dimensional beam pattern of regular sperm whale clicks confirms bent-
26
J. Spiesberger and M. Wahlberg, “Probability density functions for hyper- horn hypothesis,” J. Acoust. Soc. Am. 117, 1473–1485 共2005兲.
bolic and isodiachronic locations,” J. Acoust. Soc. Am. 112, 3046–3052 40
O. Adam, “Advantages of the Hilbert Huang transform for marine mam-
共2002兲. mals signals analysis.” J. Acoust. Soc. Am. 120, 2965–2973 共2006兲.
27
L. A. Douglas, S. M. Dawson, and N. Jaquet, “Click rates and silences of 41
O. Adam, “The use of the Hilbert-Huang transform to analyze transient
sperm whales at Kaikoura, New Zealand,” J. Acoust. Soc. Am. 118, 523–
signals emitted by sperm whales,” Appl. Acoust. 67, 1134–1143 共2006兲.
529 共2005兲. 42
28 V. Kandia and Y. Stylianou, “Detection of sperm whale clicks based on the
J. Barlow, S. Rankin, and S. Dawson, “A guide to constructing hydro-
Teager-Kaiser energy operator,” Appl. Acoust. 67, 1144–1163 共2006兲.
phones and hydrophone arrays for monitoring marine mammal vocaliza- 43
J. P. Larue, G. E. Ioup, and J. W. Ioup, “Detecting sperm whale clicks in
tions,” NOAA Technical Memorandum NMFS, Technical Report No. 417,
the presence of ambient and shipping noise using higher order moments,”
National Oceanic and Atmospheric Administration, Southwest Fisheries
J. Acoust. Soc. Am. 115, 2487 共2004兲.
Science Center, La Jolla, CA, 2008. 44
29
T. Arnbom, “Individual identification of sperm whales,” Rep. Int. Whal. M. Lopatka, O. Adam, C. Laplanche, J.-F. Motsch, and J. Zarzycki,
Comm. 37, 201–204 共1987兲. “Sperm whale click analysis using a recursive time-variant lattice filter,”
30
S. Childerhouse, S. Dawson, and E. Slooten, “Abundance and seasonal Appl. Acoust. 67, 1118–1133 共2006兲.
45
residence of sperm whales at Kaikoura, New Zealand,” Can. J. Zool. 73, E. Nosal and L. Frazer, “Sperm whale three-dimensional track, swim ori-
723–731 共1995兲. entation, beam pattern, and click levels observed on bottom-mounted hy-
31
N. Jaquet, S. Dawson, and E. Slooten, “Seasonal distribution and diving drophones,” J. Acoust. Soc. Am. 122, 1969 共2007兲.
46
behaviour of male sperm whales off Kaikoura: Foraging implications,” A. M. Thode, “The use of acoustic multipath for localization of sperm
Can. J. Zool. 78, 407–419 共2000兲. whales,” J. Acoust. Soc. Am. 116, 2606 共2004兲.
47
32
L. E. Freitag and P. L. Tyack, “Passive acoustic localization of the Atlantic V. A. Del Grosso, “New equation for the speed of sound in natural waters
bottlenose dolphin using whistles and echolocation clicks.” J. Acoust. Soc. 共with comparisons to other equations兲,” J. Acoust. Soc. Am. 56, 1084–
Am. 93, 2197–2205 共1993兲. 1091 共1974兲.
33 48
J. L. Spiesberger, “Locating animals from their sounds and tomography of R. A. L. Antonov J I, T. P. Boyer, A. V. Mishonov, and H. E. Garcia,
the atmosphere: Experimental demonstration,” J. Acoust. Soc. Am. 106, World Ocean Atlas 2005, Volume 2: Salinity, 共U.S. Government Printing
837–846 共1999兲. Office, Washington, DC, 2006兲.
34 49
M. Wahlberg, “The acoustic behaviour of diving sperm whales observed A. V. M. Locarnini R A, J. I. Antonov, T. P. Boyer, and H. E. Garcia,
with a hydrophone array,” J. Exp. Mar. Biol. Ecol. 281, 53–62 共2002兲. World Ocean Atlas 2005, Volume 1: Temperature 共U.S. Government Print-
35
W. Zimmer, M. Johnson, A. D’Amico, and P. Tyack, “Combining data ing Office, Washington, DC, 2006兲.
50
from a multisensor tag and passive sonar to determine the diving behavior S. Watwood, P. Miller, M. Johnson, P. Madsen, and P. Tyack, “Deep-
of a sperm whale 共Physeter macrocephalus兲,” IEEE J. Ocean. Eng. 28, diving foraging behaviour of sperm whales 共Physeter macrocephalus兲,” J.
13–28 共2003兲. Anim. Ecol. 75, 814–825 共2006兲.

2256 J. Acoust. Soc. Am., Vol. 126, No. 5, November 2009 B. S. Miller and S. Dawson: Tracking whales from a small boat