IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 28, NO. 1, JANUARY 2003 3
A Digital Acoustic Recording Tag for Measuring the
Response of Wild Marine Mammals to Sound
Mark P. Johnson and Peter L. Tyack
Abstract—Definitive studies on the response of marine mam-
mals to anthropogenic sound are hampered by the short surface
time and deep-diving lifestyle of many species. A novel archival
tag, called the DTAG, has been developed to monitor the behavior
of marine mammals, and their response to sound, continuously
throughout the dive cycle. The tag contains a large array of
solid-state memory and records continuously from a built-in
hydrophone and suite of sensors. The sensors sample the orienta-
tion of the animal in three dimensions with sufficient speed and
resolution to capture individual fluke strokes. Audio and sensor
recording is synchronous so the relative timing of sounds and
motion can be determined precisely. The DTAG has been attached
to more than 30 northern right whales (Eubalaena glacialis)
and 20 sperm whales (Physeter macrocephalus) with recording
duration of up to 12 h per deployment. Several deployments have
included sound playbacks to the tagged whale and a transient
response to at least one playback is evident in the tag data.
Index Terms—Effects of noise, marine animals, tags,underwater
acoustic measurements.
I. INTRODUCTION
VER THE PAST century, economic and technological de-
velopments have increased the human contribution to ambient
noise in the ocean. Although shipping is the overwhelmingly
dominant source of manmade noise in the ocean [1], a wide va-
riety of artificial sound sources also contribute to the ambient
sound field, examples being air guns, used in seismic explo-
ration, sonar, and acoustic navigation, and telemetry sources.
There is growing evidence that man-made sounds can disturb
marine mammals, and this issue has received increasing atten-
tion [2], [3].Observedresponses include silencing,disruption of
activity, lengthening of song, movement away from the source,
and perhaps even stranding [3, ch. 9]–[5]. The zone of influ-
ence of a sound source depends upon its level, its frequency
spectrum, its significance to the animal, and upon the condi-
tions for sound propagation near the source [3, ch. 10]. Sound
carries so well underwater that animals may be affected many
tens of kilometers away from a loud source [6], [7], and there
is no a priori reason to rule out effects at even greater ranges.
Marine mammals rely on sound for communication, orientation,
and detection of predators and prey; disruption of any of these
Manuscript received May 2001; revised August 2002. This work was sup-
ported in part by the Office of Naval Research under Grants N00014-99-10831
and N00014-99-10819, in part by the National Marine Fisheries Service
(NMFS) under Grant NA87RJ0445, in part by the Strategic Environment
Research Development Program (SERDP) under Grant CS-1188, in part by the
International Fund for Animal Welfare, and in part by the Ida and Cecil Green
Technology Development Award.
The authorsare withthe Woods HoleOceanographic Institution,Woods Hole,
MA 02543 USA.
Digital Object Identifier 10.1109/JOE.2002.808212
functions would interfere with normal activities and behavior.
This raises the concern that, along with short-term impacts of
single sources, increasing noise may have long-term impact as
a form of habitat degradation.
Research on the effects of noise on large whales has suffered
from a lack of methods to observe behavior in sufficient de-
tail. Many deep diving species are visible only 5% of the time,
when they are breathing at the surface, so visual observations
are seldom adequate. Passive acoustic monitoring of whales is
often hindered by a lack of knowledge of the species repertoire
and whether a change in vocal output can be expected in re-
sponse to noise. Moreover, in animals that are thought to silence
in response to noise (e.g., sperm whales [8]), passive acoustic
tracking may be impossible following an exposure, making it
difficult to assess the magnitude of response.
Acoustic recording tags represent a new technological solu-
tion for monitoring disturbance reactions of marine mammals.
The concept here is to measure the sound environment of the
animal in tandem with physiological or behavioral information.
By comparing the timing of any change in behavior to the sound
as heard bythe animal, causality can be established in controlled
experimentalexposuresof sound. The extent of the response can
then be gauged against received sound level, the first step to-
ward determining suitable exposure limits for a given sound.
Early examples of acoustic recording tags are those of Burgess
et al. [9], developed for deep-diving elephant seals, and Fletcher
et al. [10], who used such a tag to study the effects of an ATOC
transmitter on the dive patterns of seals. Behavioral and phys-
iological sensing on both of these tags was limited to depth
of dive although Burgess found a heartbeat sound in the audio
record from elephant seals [9]. In an effort to extend behavioral
sensing, Tyack, Johnson, and Nowacek combined a miniature
digital audio tape (DAT) recorder with an orientation sensor in
1998 [11]. Various versions of this tag were deployed on wild
bottlenose dolphin and northern right whales, proving the feasi-
bility of on-animal sound and orientation measurement.
An effort funded by the Office of Naval Research (ONR)
to reduce the size and increase the capabilities of acoustic
recording tags resulted in the development of the DTAG in
1999. This tag uses FLASH memory in place of moving mag-
netic tape or disks to record data and so can be encapsulated in
plastic. A low-power digital signal processor combines audio,
acquired from a hydrophone, with sensor measurements, and
streams the data to the nonvolatile memory array. The sensor
suite comprises acceleration, magnetic field, and pressure
sensors and is tailored to measuring orientation at sampling
rates of up to 50 Hz, much higher than traditional time–depth
recorders.
0364-9059/03$17.00 © 2003 IEEE
4 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 28, NO. 1, JANUARY 2003
In this paper, the design of the new tag is presented together
with results demonstrating its potential for studying the re-
sponse of marine mammals to acoustic stimuli. The following
section describes the tag and the means by which it is attached
to wild marine mammals. Section III deals with estimation
of orientation from the DTAG sensors. Although orientation
measurement is a well-established technique, being the main-
stay of navigation systems in underwater vehicles and aircraft,
its application to marine mammals is new. Section III focuses
on the issues that arise in estimating the orientation of marine
mammals with sub-fluke-stroke time resolution and the sup-
porting visual observations needed to determine movement in a
global frame. Some quality metrics implicit in the measurement
are also highlighted. The remainder of the paper is devoted to
a series of examples taken from tag deployments on northern
right whales and sperm whales. Two possible responses to
controlled sound playbacks are examined to demonstrate the
power of the new tag in parameterizing responses, and the
confounds implicit in such fine-scale analyses.
II. DTAG D
ESIGN
The governing constraints on the DTAG design are that it
be small, lightweight, pressure tolerant and have a substantial
recording time. Although the large whales of primary interest
here can carry a big tag, the problems involved in delivering
and attaching the tag to the animal, scale strongly with tag size
and weight. Options for tag delivery include a long pole, gun,
or crossbow, all of which require a small, light payload. The tag
must also be tolerant to the pressure experienced during deep
dives. Evidence exists for sperm whale divesinexcessof2000m
[12] corresponding to a hydrostatic pressure of over 20 MPa.
The recording time of the tag is determined by its memory
capacity and audio sampling rate. For the typical controlled
sound experiment, a recording time of at least 4 h is needed.
This allows at least two dives after tag delivery to establish
a behavioral baseline for the animal, even for animals with
hour-long dives, followed by a sound playback of 1 h and
a 1-h post-exposure period. Based on the frequency range
of their vocalizations, suitable sampling rates for baleen
and sperm whales are in the 10–50-kHz range. A min-
imum memory capacity of 200–1000 MB is thus required
(e.g., 12 b per sample
16 kHz 4h 350 MB). More
memory is desirable as then a longer interval can be left
between tagging and sound playback, increasing the chance
of observing a natural response. A long recording time after
an exposure is also desirable to better observe the return to
baseline behavior.
A tag design meeting the above constraints is illustrated in the
block diagram of Fig. 1 along with a photograph of the com-
plete electronics package. The design centers on a low-power
programmable digital signal processor (DSP) which combines
data from the audio and sensor circuits and stores the result to
a memory array. Although the sampling-rates of the audio and
sensor signals necessarily differ, synchrony is maintained by ac-
quiring a precise number of audio samples between each sensor
sample. Use of a DSP also enables real-time filtering and com-
pression of the signal streams when required. When the tag is
Fig. 1. Simplified block diagram of the DTAG (above) and photograph of the
encapsulated electronics package (below). The leads on the left of the tag are
for battery and hydrophone.
not recording, the DSP performs the user interface and data-of-
floading functions. An infrared interface with a data rate of
0.5 MB/s is used to program the tag and to recover data. As
initially designed, the DTAG had 400 MB of FLASH memory
but increases in component density now allow up to 3 GB in the
same size tag.
The audio subsystem in the tag consists of a piezo-ceramic
hydrophone, a preamplifier, anti-alias filter and analog-to-dig-
ital converter (ADC). A 12-b ADC is used with sampling-rate
programmable between 2–200 kHz. The anti-alias filter cut-off
frequency and preamplifier gain are also programmable. With
gain set to 12 dB, the dynamic range of thetag audio recordingis
from 80 dB (noise floor) to 152 dB (onset of clipping) re 1
Pa.
With this gain, good fidelity recording of distant animals and
vessels is obtained at the expense of clipping during loud vocal-
izations from the tagged whale.
The tag has sensors for depth, temperature, and orientation.
Depth is determined from a pressure sensor with a resolution
of 0.5 m
H O over a range of 0–2000 m. Orientation, param-
eterized by the Euler angles: pitch, roll, and heading, requires
two sets of sensors. Pitch and roll are measured by capacitive
accelerometers (Analog Devices ADXL202). These sense both
the dynamic acceleration of the tagged whale and its orientation
with respect to the gravity vector. The spatial freedom enjoyed
by a submerged animal necessitate a 360
measurement range
for pitch and roll requiring, in turn, a three-axis accelerometer.
The heading sensor uses a three-axis magnetometer to measure
the direction of the earth’s magnetic field relative to the tag.
Each magnetometer axis comprises a low-power magnetoresis-
tive bridgesensor (Honeywell HMC1021). To estimate heading,
JOHNSON AND TYACK: DIGITAL ACOUSTIC RECORDING TAG FOR MEASURING THE RESPONSE OF MARINE MAMMALS TO SOUND 5
Fig. 2. Complete tag including plastic fairing, floatation, and two suction cups
(front of tag is to the left).
the three magnetometer signals are corrected for pitch and roll.
This process, called gimballing in analogy to the floating nee-
dles used in traditional compasses, effectively transforms the
magnetic-field measurement to that which would be made on
a horizontal surface with the same heading. The method of ori-
entation estimation is the subject of Section III.
To avoid a heavy pressure housing, the DTAG circuit boards
are encapsulated in epoxy resin. This provides electrical isola-
tion but does not fully insulate the circuitry from hydrostatic
pressure. It is thus crucial to select components that are robust
to high pressure and to eliminate air bubbles during potting. A
low-viscosity resin was selected for this reason. The complete
encapsulated DTAG has been tested to a water depth of 2000 m
and functioned satisfactorily. Although the components in the
tag are pressure tolerant, their performance may vary with pres-
sure. For example, the accelerometers have an output offset of
about 0.002 m
s per meter of water depth. The power con-
sumption of the DTAG is about 150 mW while recording and
this can be metwith a single 3-Wh lithiumpolymer rechargeable
cell. Polymer cells have a solid electrolyte and so are inherently
pressure tolerant. To verify this, we have discharged polymer
cells under pressure (2000 m H
O) and have found no deviation
from the discharge characteristic at atmospheric pressure.
Due to the short recording time of the DTAG, being designed
for daily playback experiments, and the desire to minimally dis-
turb the tagged animal, a noninvasive suction-cup attachment
has been developed. Suction cups have been used widely with
marine mammals [13] and attachment durations of tens of hours
have been reported. For the DTAG to track orientation accu-
rately, a rigid connection is required between the whale and
the tag using at least two suction cups as shown in Fig. 2. A
near-vertical force on the tag is needed to attach it to a whale
and a system devised by Moore [14] for ultrasound inspection
of right whale back fat has been adapted to deliver the DTAG.
The system consists of a 12-m carbon-fiber pole, cantilevered
in a bow-mounted oarlock, as shown in Fig. 3. The mounting
provides four degrees of freedom: the pole may be rotated and
Fig. 3. Delivery of the tag to a northern right whale using a 12-m cantilevered
pole developed by Moore [14].
can slide through the oarlock, and the oarlock itself swivels and
tilts. The key advantage of the long pole is that it is possible to
deliver the tag without encroaching over the flukes of the an-
imal. As suction cups often release from an animal in a matter
of hours due to leakage, our attachment includes a pump pow-
ered by the pressure changes during the whales’ dive cycle to
maintain vacuum in the cups. An active release is also included
consisting of a nickel–chromium wire which seals a valve in the
air line to each suction cup. The wire corrodes rapidly in sea-
water when made anodic and is controlled by a clock circuit in
the DTAG.
The tag electronics, battery, very high frequency (VHF) radio
beacon, and suction cups are housed in a thermoformed polyeth-
ylene hull to minimize drag. The hull is flooded to allow water
to reach the pressure and temperature sensors. The rear section
of the hull is filled with syntactic foam sufficient to float the tag
tail-up when not attached, keeping the VHF antenna above the
sea surface. The volume of the tag is about 1 liter and the dry
weight is 500 g, making it acceptable for pole delivery.
Two versions of the DTAG, the two-cup design shown in
Fig. 2 and a three-cup version, have been attached to over 30
northern right whales during three years of field work in the
Bay of Fundy, Canada. The longest attachment time was 21 h
and 33% of attachments lasted sufficiently long for a playback
experiment (4 h). The DTAG has also been attached to about
6 IEEE JOURNAL OF OCEANIC ENGINEERING, VOL. 28, NO. 1, JANUARY 2003
Fig. 4. Three frames of reference involved in determining whale orientation
showing the rotational transformations between frames. In the earth frame,
x
is
northward,
y
is westward, and
z
is upward. In the whale frame,
x
is rostrally
directed and is the long axis of the animal. In the tag frame,
x
is the long axis
of the tag pointing noseward.
20 sperm whales over three years in the Mediterranean Sea and
the Gulf of Mexico. The longest attachment on sperm whales
was 10 h, and 41% of attachments lasted longer than 4 h. A
majority of early releases on right whales and some releases on
sperm whales resulted from social interactions between animals
in which the tag was rubbed off. Breaching of the tagged whale
was another less frequent cause of early release. Rubbing and
breaching are behaviors which few nonimplanted tags could be
expected to survive and, in fact, may also lead to failure of im-
planted tags. We view movement or release of the suction cup
tag, when stressed, not as a problem but a feature which reduces
the chance of discomfort or minor injury to the tagged whale.
III. O
RIENTATION ESTIMATION
A key innovationof the DTAGis its ability to measure the ori-
entation of the tagged animal as a function of time. Orientation
is deduced from the three-axis accelerometer and magnetometer
signals and is expressed in terms of the Euler angles, pitch, roll,
and heading, with reference to the fixed (earth) frame [15]. As
the tag may be placed anywhere on the back of a whale, the
tag axes do not generally coincide with the whale axes. There
are thus three frames involved: the earth frame, the tag frame,
and the whale frame, and these are related as shown in Fig. 4.
Note that the definitions of heading and pitch in Fig. 4 differ
from their standard Euler definitions: we have chosen heading
to follow the compass convention and pitch is positive for a
nose-upward tilt. Also, heading refers to magnetic heading and
requires compensation for declination angle to obtain the true
heading. The goal is to determine the orientation of the whale,
i.e., the angles of rotation,
, , and , relating the whale frame
to the earth frame. This is achieved in two steps. First the tag
measurements are corrected for the orientation of the tag on the
whale; then the whale frame angles,
, , and , are deduced
from the corrected tag data.
If the whale is moving at a constant velocity, and measure-
ment noise and sensor miscalibration are ignored, the three-axis
acceleration and magnetic-field measurements made by the tag
can be expressed as
(1)
where
, and , ,
, are the sensor outputs and the subscript indicates
the tag frame.
is the gravity vector in
the earth frame and
is the acceleration due to gravity.
is the magnetic-field vector in the
earth frame, where
and are the magnetic-field intensity and
inclination angle, respectively. The rotation matrices in (1) are
defined as
and
These matrices are not symmetric so the multiplication order in
(1) is important.
The tag orientation on the whale is parameterized in (1)by the
pitch, roll, and heading,
, and , of the tag with respect to
the whale. Values for
, , (which may vary with time if the
tag slides on the whale), can be measured from photographs of
the tag but can also be deduced readily from visual and tag data.
The location of the nares in most whale species necessitate a
pitch and roll close to 0 when surfacing for breath. If the heading
of the whale during surfacing is recorded by visual observers,
a sequence of known whale orientations result which can be
compared to the tag data todetermine
, and .Oncethetag
orientation has been established,
and can be converted
to their whale-frame equivalents, that is, the sensor signals that
would be measured if the tag was perfectly aligned on the whale
(2)
where
for a rotation matrix. Combining (1) and (2)
(3)
The whale pitch and roll can be estimated from the whale frame
signals by
and (4)
Note that a four-quadrant arctangent is required in (4) to esti-
mate roll over the (
180 –180 ) range. For reasons to be dis-
cussed, pitch is constrained to (
90 ,90 ). Having calculated
and , the heading can be determined by premultiplying the
second line of (2) by
to get
(5)
where
is the magnetometer measurement that would be
made on a horizontal (i.e., gimballed) surface with the same
JOHNSON AND TYACK: DIGITAL ACOUSTIC RECORDING TAG FOR MEASURING THE RESPONSE OF MARINE MAMMALS TO SOUND 7
heading as the whale. Heading can then be estimated by
, again using a four-quadrant arctangent to
realize a (
180 –180 ) range.
The pitch, roll, heading parameterization of orientation is not
unique: any orientation can be described by two sets of
, , and
, say and , both of which produce the
same
and . These are related as
, and . The ambiguity is a physical one.
There are always two combinations of pitch, roll, and heading
which result in the same orientation. For example, if the an-
imal is upside-down, it may have attained this attitude by rolling
180
, or by changing heading by 180 and pitching 180 . The
ambiguity can be resolved by constraining
to ( 90 ,90 ) and
accepting jumps in roll and heading if the animal does pitch be-
yond 90
. An additional ambiguity arises when the animal is at
90 pitch: in both cases, and become 0 and the roll
is indeterminate.
The above results are for the unrealistic case of perfect,
noise-free sensors. In practice, a number of factors degrade
the orientation estimate. These include variations in sensitivity
and zero offset between sensor axes, sensitivity changes due
to changing pressure or temperature, measurement noise,
and acceleration of the animal. The most significant of these
is acceleration, which may result from swimming, rapid
maneuvering, or even from waves slapping the tag when the
animal surfaces. For the large whales of primary interest here,
acceleration of the tag due to body undulations during swim-
ming is likely to be small, with peak magnitude of less than
0.1 g for a 15-m animal with a 6-s fluke rate [16]. However,
startle responses or fast feeding movements may produce large
transient accelerations. Although use of a low-pass filter on
the acceleration signals will reduce such transients, substantial
orientation errors are still possible. Fortunately, there are three
implicit quality metrics that facilitate detection of distorted
measurements and which can provide a sample-by-sample
estimate of orientation accuracy. First, the two-norm of the
accelerometer vector,
, should, in the absence
of acceleration, miscalibration, and measurement noise, equal
. Likewise, the two-norm of : , should
equal
, the magnetic-field intensity. Finally, referring to (1)
and recalling that rotation matrices are unitary, the dot-product
of
and should equal , i.e., the product
of
, , and the sine of the magnetic-field inclination angle.
Although
and vary geographically, they do so gradually and
will be essentially constant over a deployment area. Therefore,
any deviation of
, ,or , from their nominal values
sheds doubt on the accuracy of the orientation estimate. Epochs
of high acceleration and so high orientation error can be
identified readily by examining the time series of
, , and
.
Although deviations in
and can be used to locate in-
accurate orientations, the precise level of error is generally dif-
ficult to estimate. However, for small errors, e.g., due to sensor
noise, the orientation error can be predicted directly from
and . Assume that the sensor noise is independent but identi-
cally Gaussian distributed in each sensor axis with variance
and in the accelerometer and magnetometer signals, respec-
tively. With low noise levels, i.e.,
g and b,
Fig. 5. Sperm whale ascent example showing (a) depth; (b) whale frame pitch;
and (c) pitch error estimate.
the standard deviations of , , , , and can be estimated
using a Monte Carlo method as
std
std std
std std
(6)
Thus, the pitch accuracy is independent of orientation, while
the roll error is increased at high pitch angles due to the
scaling factor in (3). Due to the gimballing operation, heading
accuracy is dependent on the pitch and roll accuracy as well as
the magnetic-field inclination angle. When the inclination angle
is large, as it is in high latitudes, the horizontal components of
the magnetic field become small, reducing the accuracy of
.
The key result in (6) is that std
and std , which can be
measured directly from the tag data, can be used to estimate the
sensor noises,
and , and therefore, the accuracy of the
orientation estimate.
In addition totheir diagnostic function,
and canbe used
to identify and reducecertain errorsin the sensorsaftera deploy-
ment. Miscalibration of the sensors, for example, due to drift or
uncompensated pressure and temperature effects, will lead to
consistent errors in the orientation estimate correlated with ori-
entation. Provided that the orientation varies widely and often
throughout the time series (i.e., is persistently exciting [17]), a
locally-linearized least squaresmethod can be used to determine
the amount of offset, pressure signal or temperaturesignal to add
to each sensor output so as to minimize the variance of
and
. For the tag recordings described in the following section,
the standard deviations of
and were improved from about
0.1 to 0.02 g and from 0.5 to 0.1
T, respectively, using least
squares fitting.
IV. E
XAMPLE RESULTS
The DTAG has been deployed in more than ten field ex-
periments, four involving the northern right whale, Eubalaena
glacialis, and five focusing on sperm whales, Physeter macro-
cephalus. The right whale deployments took place in the Bay of
