A key component in understanding the ecological role of marine mammal predators is to identify how and where they capture prey in time and space. Satellite and archival tags on pinnipeds generally only provide diving and position information, and foraging is often inferred to take place in particular shaped dives or when the animal remains in an area for an extended interval. However, fast movements of the head and jaws may provide reliable feeding cues that can be detected by small low-power accelerometers mounted on the head. To test this notion, a harbour seal (Phoca vitulina) was trained to wear an OpenTag (sampling at 200 or 333 Hz with ±2 or ±16 g clipping) on its head while catching fish prey in front of four underwater digital high-speed video cameras. We show that both raptorial and suction feeding generate jerk (i.e. differential of acceleration) signatures with maximum peak values exceeding 1000 m s−3. We conclude that reliable prey capture cues can be derived from fast-sampling, head-mounted accelerometer tags, thus holding a promising potential for long-term studies of foraging ecology and field energetics of aquatic predators in their natural environments.

Pinnipeds are versatile top predators in marine food webs, and fine-scale information on their foraging behaviour is therefore critical for understanding top-down-mediated energy cascades. However, it has proven challenging to detect feeding events in free-swimming aquatic animals and, as a result, relatively little is known about the fine-scale feeding behaviour of many pinnipeds (Kuhn et al., 2009). With satellite and archival tags, foraging is typically inferred from movement patterns (e.g. area-restricted search) or from distinctive dive shapes (Kooyman, 2004), but without more detailed information, the accuracy of these methods may be difficult to assess. Moreover, such proxies provide little information about the quantity of prey taken. To directly observe foraging, cameras have been deployed on diving pinnipeds (Davis et al., 1999; Davis et al., 2001; Bowen et al., 2002; Hooker et al., 2002; Sato et al., 2002), but these are limited by battery power, and the need for a light source in deep dives may affect the behaviour of predator and prey. Actual prey ingestions have been measured with stomach temperature transmitters (Kuhn and Costa, 2006), but these sensors do not appear to be reliable for long intervals either because of changing conditions in the gut or because of passage of the sensor (Ropert-Coudert et al., 2000; Takahashi et al., 2004). Jaw opening and closing can be recorded by a mandibular sensor (Ropert-Coudert et al., 2004), but the logger may be unreliable over long recording periods where cabling to the jaw is likely to fail or affect the tagged animal.

Recent studies have shown promising use of head- and jaw-mounted accelerometers sampling at 32 Hz to measure head surge in foraging attempts of both pinnipeds (Skinner et al., 2009; Suzuki et al., 2009; Naito et al., 2010; Iwata et al., 2012; Naito et al., 2013) and penguins (Kokubun et al., 2011; Watanabe and Takahashi, 2013). Prey capture and engulfment involves rapid jaw movements in raptorial feeding and the retraction and lowering of the gular apparatus during suction feeding (Werth, 2000; Marshall et al., 2008). These movements are unique to feeding and should generate high-frequency acceleration signatures that are distinctive and so readily detected against other head movements. Here, we used fast super-cranial accelerometry on a trained male harbour seal, Phoca vitulina Linnaeus 1758, catching prey to show that the differential of the three acceleration axes, jerk (m s−3) (Simon et al., 2012), provides a reliable, easily computed and orientation-independent measure of both raptorial and suction feeding that can be recorded or relayed over long time periods from wild animals at sea.

Two experiments were conducted using different data collection parameters. In the first, an animal-attached triaxial accelerometer was set to sample at 200 Hz with a clipping level of ±2 g. A total of 124 trials were conducted over 27 days. After excluding prey captures in which engulfment was not visible on any of the video cameras, a set of 14 captures of dead fish, 10 of large live trout and 13 of small live trout was available for analysis. Because of the relatively low clipping threshold and the rapid head and jaw movements during capture (see supplementary material Movie 1), most of the captures had brief intervals in which the measured acceleration in one or more axes was clipped. Only 11 captures of dead fish and one with a small live trout were unaffected by this limitation. In the second experiment, the tag was therefore configured for a sampling rate of 333 Hz and a clipping level of ±16 g. A total of 20 trials were conducted with these settings, of which nine captures of large 18–23 cm live trout happened in front of the cameras, permitting analysis.

Based on visual analysis of all the prey captures, a total of 16 were judged to be primarily raptorial feeding, while 15 were categorized as suction feeding. Raptorial feeding occurred mostly in captures of large prey, whereas smaller prey were caught by suction (Table 1). In both feeding mechanisms the absolute jerk in the z-axis

List of symbols and abbreviations

     
  • t0

    time of visible initial jaw opening

  •  
  • t1

    time of seal–prey contact

  •  
  • t2

    time of prey engulfment

  •  
  • x-jerk

    x-axis jerk

  •  
  • y-jerk

    y-axis jerk

  •  
  • z-jerk

    z-axis jerk

was highest, followed by the x-axis, then the y-axis. However, in suction feeding, the duration of the prey capture (t2t0, see Materials and methods) was shorter, and the amplitude of the jerk lower (Table 1). Fig. 1 shows an example of a raptorial prey capture of a large trout. Here, the jaw opening is followed by a sudden rise in jerk amplitude (Fig. 1A, image 1, and Fig. 1C). Subsequent jerk peaks are associated with capture and handling of the fish (Fig. 1A, images 2–8).

To test whether feeding jerks could be distinguished from the jerk recorded in intervals before and after feeding, we divided each capture session into three time windows of 250 ms each and computed the root mean square (RMS) of the norm jerk in each section: a pre-capture time window starting 1 s before t0 (jaw opening), a capture window starting at t0, and a post-capture window starting 1 s after t0. The RMS measure was chosen because it is relatively insensitive to brief intervals of clipping in the individual accelerometer signals (supplementary material Fig. S2). Results of a one-way ANOVA and multiple comparison test show that the RMS jerk during the feeding window differed significantly from the before and after windows for all fish types (Fig. 2, Table 1). Furthermore, engulfment of live fish generated significantly larger RMS jerk values compared with the RMS jerk during captures of dead fish (t-test, P<0.005). A similar analysis of raptorial and suction feeding did not, however, provide any significant difference. All data in the above analyses were found to be normally distributed by a Chi-square goodness-of-fit test.

The median sampling rate required to generate at least 90% of the observed peak broadband jerk was 73, 95 and 64 Hz, for prey captures of dead fish, live fish (non-clipped data) and clipped live fish, respectively (Table 1).

Foraging strikes in any predator targeting nekton inevitably involve sudden movements irrespective of the way in which prey are acquired. Here, we tested whether prey engulfment movements of the head and jaws of a pinniped produce fast, distinct changes in acceleration that can be measured by a small head-mounted tag sampling at high rates. We identified the same surge (i.e. x-axis) acceleration signature reported to serve as a good proxy for successful prey captures in other studies, but we show also that the RMS of the norm jerk over a short window (250 ms here) can provide a reliable and distinctive signal for detecting raptorial or suction-feeding events (Fig. 1, Table 1). Movements were more powerful in trials with live fish, which involved primarily raptorial feeding. Larger fish also required more handling as indicated by the comparably larger difference in time from fish–seal contact to engulfment (t1t2) found in these trials (Table 1, Materials and methods). Increased hunting and handling effort are also represented in the pre- and post-feeding RMS values in Fig. 2, opening the possibility that the magnitude and duration of the jerk signal may provide information about the type and size of prey, as well as the mode of capture, but utilization of this potential would require confirmation across a number of animals.

Triaxial on-animal accelerometer data provide dense information about the movements of animals and, as a result, can be complex to analyse. Existing methods for detecting foraging impulses require information about the orientation of the animal, the orientation of the tag on the animal and the time scales of events in order to choose filters and axes to process. In comparison, the norm of the jerk is a very simple processing method that does not require explicit time scale or tag orientation information. This makes the method both simple to implement for in situ processing and broadly applicable to other taxa.

Table 1.

Results for all fish

Results for all fish
Results for all fish
Fig. 1.

Example of prey capture of a large live trout. The jaw opening time (t0) corresponds to time 0 on the x-axis. (A) Still images of initial jaw opening (1), capture and handling (2–8). Measured triaxial acceleration (B) and jerk (C) are shown over the same time interval. The timing of the images is marked on the jerk (C).

Fig. 1.

Example of prey capture of a large live trout. The jaw opening time (t0) corresponds to time 0 on the x-axis. (A) Still images of initial jaw opening (1), capture and handling (2–8). Measured triaxial acceleration (B) and jerk (C) are shown over the same time interval. The timing of the images is marked on the jerk (C).

The differentiation used in computing the jerk emphasizes fast movements such as those produced by smaller muscles within the head during prey capture. Slower movements such as manoeuvres and stroking tend to produce smaller jerk signals even though the amplitude of the movements and the muscle mass involved may be much greater. The norm of the jerk is also completely independent of the orientation of the tag and so is unaffected by the direction of approach of the predator towards the prey or of the way the tag is attached to the head, provided that the attachment is sufficiently rigid. As a result, the jerk signal associated with raptorial and suction feeding may provide a more easily detected and less ambiguous measure for prey captures than does head surge.

Compared with other methods for detecting foraging activity, triaxial accelerometers offer a number of important advantages. Many tags now include these miniature low-power devices and, as we have demonstrated, foraging accelerations can be detected by a tag attached to the rear of the head, obviating the need for jaw sensors and cables. A supra-cranial placement of a small tag is also ideal for other sensors such as GPS and for radio telemetry of data. Accelerometers are straightforward to use, but require the selection of two parameters: the sampling rate and the full-scale sensitivity (or clipping level). Key to reliable detection of rapid foraging movements is a wide sensing bandwidth necessitating a high sampling rate. Previous studies of accelerometry on pinnipeds have used a sampling rate of 32 Hz for which the bandwidth is <16 Hz. Here, we used a sampling rate of 200 and 333 Hz, which enabled the detection of muscle movements with time constants of tens of milliseconds. Through decimation we can show that a sampling rate of more than 70 Hz is required on average, no matter the engulfment method, to capture 90% of the jerk (Fig. 1C). Although the higher sampling rate means that more data are collected by the tag per unit of time, the benefit of more readily detected foraging signals may mean that data compression methods such as event counting are more effective, increasing the quality of the data that are ultimately stored or telemetered.

Fig. 2.

Prey engulfment jerk. Boxplot of pre-engulfment (A), during engulfment (B) and post-engulfment (C) jerks for all prey engulfments. Groups consist of dead, small and large fish, sampled at 200 Hz and large fish sampled at 333 Hz with a clipping level of 2 and 16 g, respectively. The number of prey captures is indicated for each group. All groups during feeding that are significantly different from before and after feeding (one-way ANOVA) are marked by an asterisk.

Fig. 2.

Prey engulfment jerk. Boxplot of pre-engulfment (A), during engulfment (B) and post-engulfment (C) jerks for all prey engulfments. Groups consist of dead, small and large fish, sampled at 200 Hz and large fish sampled at 333 Hz with a clipping level of 2 and 16 g, respectively. The number of prey captures is indicated for each group. All groups during feeding that are significantly different from before and after feeding (one-way ANOVA) are marked by an asterisk.

The clipping level of an accelerometer determines both the maximum absolute acceleration that can be measured and, because the resolution of the sensor is fixed, the smallest change in orientation that can be detected. Accelerometers with clipping levels of 2 g are often used in tags as these provide detailed records of orientation. However, our results suggest that these devices will often clip during foraging strikes when head mounted. Although higher clipping level accelerometers are available, the RMS jerk processing method we propose appears to be robust to modest levels of clipping (see supplementary material Fig. S2).

We conclude that the RMS jerk calculated as the norm of the differential of the triaxial acceleration provides a reliable and widely applicable measure of both raptorial and suction feeding. Moreover, the duration and temporal sequence of jerks may offer the potential for separating prey sizes and feeding mechanisms, and provide quantitative measures of prey capture success. Given the low power consumption of accelerometers, this processing method enables the timing and method of prey ingestion to be sampled over periods of months and relayed from the wild via low bandwidth telemetry. Such long records of foraging behaviour will help us to understand how free-ranging aquatic predators search for and acquire energy from their dynamic environment in time and space.

Experiments were carried out at the Fjord and Belt in Kerteminde, Denmark, with a trained adult male harbour seal (P. vitulina; 13 years old, 80 kg) housed in a net pen. Head accelerations during prey captures were measured using a triaxial accelerometer (‘OpenTag’, Loggerhead Instruments, Sarasota, FL, USA), sampling at 200 or 333 Hz (16 bits). The tag was calibrated for sensitivity and frequency response using a Brüel and Kjær Vibration Exciter Type 4809 and a pre-calibrated Brüel and Kjær Accelerometer Type 4381. The seal was trained to wear the datalogger (dimensions 7.5×3.5×2.2 cm, 55 g in air, 3 g in water) on top of its head attached by means of a small, custom-made elastic hood (supplementary material Fig. S1). The hood fitted snugly around the head and neck, holding the tag firmly against the dorsal surface of the skull. In each trial, the seal swam towards and acquired individual prey items released from a custom-made fish dispenser, and then returned to station. Small (12–13 cm) and large (15–25 cm) live trout, Oncorhynchus mykiss (Walbaum 1792), 12–13 cm dead sprat, Sprattus sprattus (Linnaeus 1758), and 15–16 cm capelin, Mallotus villosus (Müller 1776) were used as prey in the experiments. The prey captures were filmed using four GoPro HD Hero2 cameras (120 frames s−1) in underwater housings (Eye of Mine Action Cameras, Carson, CA, USA) arranged so as to image captures from different angles to ensure that the timing of mouth opening and prey contact could be established. All recorders were synchronized before and after a session, and the data were subsequently analysed in Matlab 7.5 (MathWorks, Natick, MA, USA) with custom-written scripts. Three events were identified in the videos from each prey capture: the time of the first sign of jaw opening (t0), the time of first fish–seal contact (t1) and the time of complete engulfment (but not necessarily deglutition) of the fish (t2). Each prey capture was classified to be either primarily suction or raptorial feeding by five observers tasked with judging whether the fish appearing in the videos were actively drawn into the mouth or not. Prey capture events were grouped according to fish type and feeding mechanism (suction or raptorial). The jerk was computed as the differential of the acceleration for each axis and the total jerk was taken as the norm of the triaxial jerk (i.e. the square-root of the sum of the squared value in each axis) at each time instant. In Matlab, this is achieved with the following instruction:
(1)
where A is a three-column matrix containing the measured triaxial acceleration time series and fs is the sampling rate in Hz. The RMS jerk was calculated as the square-root of the sum of the squared jerk over an averaging window of 250 ms. Sampling rates required for generating 50% and 90% of the maximum jerk peaks were also calculated for each capture by decimating the sampled acceleration prior to jerk computation using a 12-length symmetric FIR filter (Orfanidis, 2010) with a cut-off frequency of 0.4 of the new sampling rate.

Dr D. Mann kindly shared prototypes of the OpenTag. We thank the Fjord and Belt staff for their dedicated help and support, and the staff at the workshop of the Department of Bioscience, Aarhus University, for assisting with the construction of the recording setup. The authors acknowledge helpful discussions on processing methods with A. Kato, T. Costa and Y. Ropert-Coudert and thank Alex Werth and an anonymous reviewer for helpful critique.

Funding

This work was funded by the Carlsberg Foundation through a grant to P.T.M. M.J. was funded by the Office of Naval Research and the Marine Alliance for Science and Technology Scotland. D.M.W. was funded by the Oticon Foundation, Denmark.

Bowen
W.
,
Tully
D.
,
Boness
D.
,
Bulheier
B.
,
Marshall
G.
(
2002
).
Prey-dependent foraging tactics and prey profitability in a marine mammal
.
Mar. Ecol. Prog. Ser.
244
,
235
-
245
.
Davis
R. W.
,
Fuiman
L. A.
,
Williams
T. M.
,
Collier
S. O.
,
Hagey
W. P.
,
Kanatous
S. B.
,
Kohin
S.
,
Horning
M.
(
1999
).
Hunting behavior of a marine mammal beneath the Antarctic fast ice
.
Science
283
,
993
-
996
.
Davis
R. W.
,
Fuiman
L. A.
,
Williams
T. M.
,
Le Boeuf
B. J.
(
2001
).
Three-dimensional movements and swimming activity of a northern elephant seal
.
Comp. Biochem. Physiol.
129A
,
759
-
770
.
Hooker
S. K.
,
Boyd
I. L.
,
Jessopp
M.
,
Cox
O.
,
Blackwell
J.
,
Boveng
P. L.
,
Bengtson
J. L.
(
2002
).
Monitoring the prey-field of marine predators: combining digital imaging with datalogging tags
.
Mar. Mamm. Sci.
18
,
680
-
697
.
Iwata
T.
,
Sakamoto
K. Q.
,
Takahashi
A.
,
Edwards
E. W. J.
,
Staniland
I. J.
,
Trathan
P. N.
,
Naito
Y.
(
2012
).
Using a mandible accelerometer to study fine-scale foraging behavior
.
Mar. Mamm. Sci.
28
,
345
-
357
.
Kokubun
N.
,
Kim
J. H.
,
Shin
H. C.
,
Naito
Y.
,
Takahashi
A.
(
2011
).
Penguin head movement detected using small accelerometers: a proxy of prey encounter rate
.
J. Exp. Biol.
214
,
3760
-
3767
.
Kooyman
G.
(
2004
).
Genesis and evolution of bio-logging devices: 1963–2002
.
Mem. Natl Inst. Polar Res.
58
,
15
-
22
.
Kuhn
C. E.
,
Costa
D. P.
(
2006
).
Identifying and quantifying prey consumption using stomach temperature change in pinnipeds
.
J. Exp. Biol.
209
,
4524
-
4532
.
Kuhn
C. E.
,
Crocker
D. E.
,
Tremblay
Y.
,
Costa
D. P.
(
2009
).
Time to eat: measurements of feeding behaviour in a large marine predator, the northern elephant seal Mirounga angustirostris
.
J. Anim. Ecol.
78
,
513
-
523
.
Marshall
C. D.
,
Kovacs
K. M.
,
Lydersen
C.
(
2008
).
Feeding kinematics, suction and hydraulic jetting capabilities in bearded seals (Erignathus barbatus)
.
J. Exp. Biol.
211
,
699
-
708
.
Naito
Y.
,
Bornemann
H.
,
Takahashi
A.
,
McIntyre
T.
,
Plötz
J.
(
2010
).
Fine-scale feeding behavior of Weddell seals
.
Polar Sci.
4
,
309
-
316
.
Naito
Y.
,
Costa
D. P.
,
Adachi
T.
,
Robinson
P. W.
,
Fowler
M.
,
Takahashi
A.
(
2013
).
Unravelling the mysteries of a mesopelagic diet
.
Funct. Ecol.
27
,
710
-
717
.
Orfanidis
S.
(
2010
).
Introduction to Signal Processing
.
New Brunswick, NJ
:
Rutgers University
.
Ropert-Coudert
Y.
,
Baudat
J.
,
Kurita
M.
,
Bost
C.-A.
,
Kato
A.
,
Le Maho
Y.
,
Naito
Y.
(
2000
).
Validation of oesophagus temperature recording for detection of prey ingestion
.
Mar. Biol.
137
,
1105
-
1110
.
Ropert-Coudert
Y.
,
Kato
A.
,
Liebsch
N.
,
Wilson
R.
,
Muller
G.
,
Baubet
E.
(
2004
).
Monitoring jaw movements: a cue to feeding activity
.
Game and Wildlife Science
20
,
1
-
19
.
Sato
K.
,
Mitani
Y.
,
Cameron
M. F.
,
Siniff
D. B.
,
Watanabe
Y.
,
Naito
Y.
(
2002
).
Deep foraging dives in relation to the energy depletion of Weddell seal (Leptonychotes weddellii) mothers during lactation
.
Polar Biol.
25
,
696
-
702
.
Simon
M.
,
Johnson
M.
,
Madsen
P. T.
(
2012
).
Keeping momentum with a mouthful of water: behavior and kinematics of humpback whale lunge feeding
.
J. Exp. Biol.
215
,
3786
-
3798
.
Skinner
J. P.
,
Norberg
S. E.
,
Andrews
R. D.
(
2009
).
Head striking during fish capture attempts by Steller sea lions
.
Endanger. Species Res.
10
,
61
-
69
.
Suzuki
I.
,
Naito
Y.
,
Folkow
L. P.
,
Miyazaki
N.
,
Blix
A. S.
(
2009
).
Validation of a device for accurate timing of feeding events
.
Polar Biol.
32
,
667
-
671
.
Takahashi
A.
,
Dunn
M.
,
Trathan
P.
,
Croxall
J.
,
Wilson
R. P.
,
Sato
K.
,
Naito
Y.
(
2004
).
Krill-feeding behaviour in a chinstrap penguin compared to fish-eating in Magellanic penguins: a pilot study
.
Mar. Ornithol.
32
,
47
-
54
.
Watanabe
Y. Y.
,
Takahashi
A.
(
2013
).
Linking animal-borne video to accelerometers reveals prey capture variability
.
Proc. Natl. Acad. Sci. USA
110
,
2199
-
2204
.
Werth
A.
(
2000
).
Feeding in marine mammals
. In
Feeding: Form, Function and Evolution in Tetrapod Vertebrates
(ed.
Schwenk
K.
), pp.
475
-
514
.
San Diego, CA; London
:
Academic Press
.

Competing interests

The authors declare no competing financial interests.

Supplementary information