Vous êtes sur la page 1sur 9

Behavioral Ecology

doi:10.1093/beheco/arr038
Advance Access publication 10 June 2011
Original Article
Optimal foraging theory predicts diving and
feeding strategies of the largest marine predator
Thomas Doniol-Valcroze,
a
Veronique Lesage,
a
Janie Giard,
b
and Robert Michaud
b
a
Department of Fisheries and Oceans, Maurice-Lamontagne Institute, 850 Route de la Mer, Mont-Joli
(Qc) G5H 3Z4, Canada and
b
Group for Research and Education on Marine Mammals, 108 de la Cale
Se`che, Tadoussac (Qc) G0T 2A0, Canada
Accurate predictions of predator behavior remain elusive in natural settings. Optimal foraging theory predicts that breath-hold
divers should adjust time allocation within their dives to the distance separating prey from the surface. Quantitative tests of these
models have been hampered by the difculty of documenting underwater feeding behavior and the lack of systems, experimental
or natural, in which prey depth varies over a large range. We tested these predictions on blue whales (Balaenoptera musculus),
which track the diel vertical migration of their prey. A model using simple allometric arguments successfully predicted diving
behavior measured with data loggers. Foraging times within each dive increased to compensate longer transit times and optimize
resource acquisition. Shallow dives were short and yielded the highest feeding rates, explaining why feeding activity was more
intense at night. An optimal framework thus provides powerful tools to predict the behavior of free-ranging marine predators
and inform conservation studies. Key words: aerobic dive limit, blue whale, central-place forager, dive-time budget, feeding
behavior, optimal foraging. [Behav Ecol 22:880888 (2011)]
INTRODUCTION
A
ir-breathing predators in aquatic environments face the
challenge of feeding underwater while managing oxygen
stores. Diving performance is linked to body mass in endo-
therms via simple scaling relationships (Schreer and Kovacs
1997; Halsey, Butler, and Blackburn 2006; Brischoux et al.
2008). Dive-time budgets (e.g., surface-to-dive-duration ratio),
however, are not easily explained by mass and metabolic rate
alone (Halsey, Butler, and Blackburn 2006), suggesting that
ecological aspects such as prey depth need to be taken
into consideration. Models based on optimal foraging theory
(Stephens and Krebs 1986; Perry and Pianka 1997) have been
proposed to explain observed patterns of dive-time budgets
across taxa (Mori 1998; Stephens et al. 2008). Although de-
veloped for constant-rate foragers, these models can be adap-
ted to predict time allocation in the dive cycle of other diving
predators like single- and multiple-prey loaders.
Because diving predators return repeatedly to the surface to
breathe, they can be studied under the framework of central-
place foraging (Orians et al. 1979) with the surface acting as
the central place (Houston and McNamara 1985). As distance
to food increases, central-place foragers should compensate
travel costs by increasing their energy gain, which divers
achieve by increasing time spent foraging at depth (Mori
1998), thereby increasing prey encounters and captures
(Thompson and Fedak 2001). Because oxygen is acquired at
the surface with diminishing returns, recovery times increase
rapidly with the lengthening of the preceding dive (Kooyman
and Ponganis 1998). Thus, shallow dives yield a higher net
rate of oxygen acquisition and deeper dives result in a higher
proportion of time spent recuperating at the surface (Kramer
1988). To maximize the proportion of time spent in the food
patch, both surface time and bottom time should increase
with target depth (Houston and Carbone 1992).
Predictions of how divers respond to prey depth and forag-
ing costs have been tested in birds qualitatively (Carbone and
Houston 1994; Carbone and Houston 1996; Walton et al.
1998) and quantitatively (Mori 2002; Halsey et al. 2003; Cook
et al. 2008). Tests on large divers, however, have been ham-
pered by the difculty of documenting underwater feeding
behavior and the lack of systems, experimental or natural,
in which variations in prey depth are large enough to examine
how predators adjust their allocation of time within dives.
Blue whales (Balaenoptera musculus) represent an ideal model
species for testing predictions of optimal models in a natural
setting. They track the diel vertical migrations of euphausiids
between the surface and deeper water layers (Fiedler et al.
1998; Calambokidis et al. 2007; Oleson et al. 2007), making
it possible to study the time allocated to foraging over a large
range of target depths. Moreover, their narrow trophic niche
ensures that they feed on the same prey regardless of the time
of day, allowing meaningful comparisons of feeding behavior
across the diurnal cycle.
For any thorough study of foraging strategies, it is crucial to
detect feeding events equally well regardless of depth or time
of day. Previous studies of rorquals and other pelagic feeders
have relied mostly on visual examination of time-depth proles
for the detection of feeding events, usually identied by the
occurrence of vertical excursions greater than an arbitrary
threshold, called wiggles (Croll et al. 2001; Schreer et al.
2001). However, not all feeding events are characterized by
a wiggle. Blue whales perform discrete feeding events called
lunges, which are characterized by an acceleration when ap-
proaching prey and a sudden deceleration when opening
their mouth to engulf quantities of water and food equivalent
to over 100% of their body volume (Goldbogen et al. 2010).
Lunges close to the surface and vertical lunges during descent
Address correspondence to T. Doniol-Valcroze. E-mail: thomas
.doniol-valcroze@dfo-mpo.gc.ca.
Received 29 June 2010; revised 14 December 2010; accepted 9
February 2011.
Crown copyright 2011. Published by Oxford University Press on behalf of
the International Society for Behavioral Ecology. All rights reserved.
For permissions, please e-mail: journals.permissions@oup.com

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

or ascent phases have been documented previously in rorqual
whales and are not detectable on time-depth proles (Wood-
ward 2006; Calambokidis et al. 2007; Goldbogen et al. 2008).
Fortunately, the kinematics of feeding lunges can be used to
pinpoint the moment and location of each feeding attempt
(Goldbogen et al. 2008).
Accurate predictions of predator behavior remain elusive in
natural settings, particularly in the marine environment. Our
objective was to test whether a simple optimal model could pre-
dict the diving and feeding behavior of free-ranging blue
whales. Specically, we predicted that surface and foraging
times increase with target depth to compensate longer transit
times and that the number of discrete feeding events, assessed
using a novel automated method, increases accordingly.
MATERIALS AND METHODS
Data collection
We deployed velocity-time-depth recorders (VTDR; Wildlife
Computers, Redmond, WA) and radio transmitters on blue
whales in the St Lawrence River estuary (Quebec, Canada)
during the time of greatest abundance and presumed feeding
activity in the region (AugustSeptember). Tags were de-
ployed from a 5 m rigid-hulled inatable boat using a pole
or crossbow and were attached to whales with suction cups.
VTDRs recorded time, depth, and instantaneous swim speed
with a pressure transducer resolution of 0.25 m for the rst
15 m. We also deployed one digital acoustic recording tag (D-
tag, Johnson and Tyack 2003), which lacked a velocity meter
but recorded the animals pitch, roll, and heading as well as
ambient noise (including ow noise) every second. Whales
were radio-tracked from a distance of 5001000 m to minimize
disturbance. We recorded surface activity and tracked whales
until tags were released due to corrosion of a magnesium cap
and subsequent entry of air or water under the suction cup or
until nightfall. Following their release, tags were located using
the radio-transmitters signal and retrieved by boat.
Dive characteristics and feeding events
Data were corrected for electronic drift using the software In-
strument Helper (Wildlife Computers). We validated zero-
offset corrections by comparing depth and respiration patterns
of a large number of dive sequences with observations of
breathing events recorded from the research vessel. This ne
degree of correction allowed us to investigate near-surface div-
ing and feeding activity, a section of whale habitat often dis-
carded in other studies due to limitations in tag resolution
and availability of data for cross-validation. The rst few dives
following tag deployment were scrutinized for any atypical be-
havior indicative of a reaction to tagging.
Estimates of swim speeds for the D-tag were obtained follow-
ing Goldbogen et al. (2008). We used the PRAAT software
(Boersma and Weenink 2009) to extract ow noise at 150
Hz (at which contrasts due to speed changes were strongest).
We established the relationship between ow noise (con-
verted to dB re. 1 lPa
2
Hz
21
) and swim speed calculated for
segments steeper than 45 by dividing vertical velocity by the
sine of pitch angle. We used the resulting calibration curve to
calculate swim speed every second.
Because visual techniques are ill-suited for the large data sets
needed for conclusive testing of biological hypotheses (Lesage
et al. 1999), we have used a novel method to automatically
detect feeding lunges of blue whales. Depth and swim speed
data were analyzed using a custom-made program in Visual
Basic (available from the authors) to identify dive phases and
calculate the number of wiggles and feeding attempts per dive.
Wiggles were dened as vertical excursions of more than 8 m
(following Croll et al. 2001) made of an ascent followed by
a descent without a return to the surface (0 m). Lunges were
identied independently of depth values using velocity reading
measured every second (Figure 1). Unlike Goldbogen et al.
(2006), we did not detect lunges using an absolute speed
threshold because velocity readings recorded by VTDRs can
underestimate true speeds and depend on their position on
the body (Baird 1998). Instead, we rst agged swim speeds
95th percentile of all velocity values recorded for the individ-
ual under scrutiny as indicators of potential accelerations typ-
ical of lunge feeding. Swim speeds following such an extreme
velocity were searched until 4 consecutive speeds , 95th per-
centile were detected. The abruptness of deceleration was de-
termined by comparing mean speed over the 10 s following the
last extreme swim speed to the mean speed during the accel-
eration period. A ratio 0.5 between mean acceleration and
deceleration was considered indicative of a feeding event.
The start of each deceleration determined the exact moment
of the opening of the mouth (Goldbogen et al. 2006) and the
depth of each lunge. The accuracy of our automated lunge de-
tection method was validated in 2 ways. First, we compared de-
tection of surface lunge-feeding activity with visual observations
from the research vessel. Second, using the D-tag record, we
compared lunges detected using ow noise with those identied
by rolling angles over a 45 threshold (Woodward 2006).
Target depth was dened as the mean depth of feeding ac-
tivity for each dive (Figure 1). Although bottom phase usually
corresponds to the time spent at a xed percentage of max-
imum depth (Lesage et al. 1999), we calculated it as the time
between rst reaching target depth and last passing that
depth during the ascent phase. Shallow dives constituting
the post-dive surface interval, and which contributed to oxy-
gen replenishment and dive recovery, were separated from
deeper dives using a K-means cluster analysis (Hartigan and
Wong 1979) based on depth and duration to separate dive
types into 2 categories. Total surface time was obtained by
summing durations of all short dives and surface intervals
following longer deeper dives.
Statistics were performed in the R programming language
(R Development Core Team 2008). Some scripts were modi-
ed from the diveMove package (Luque 2007).
Optimal model
For a detailed description of the model, see Houston and
Carbone (1992). Dives are divided into 3 components: s is
the round-trip travel time between the surface and the forag-
ing stratum, t is the time spent in the foraging stratum (bot-
tom time), and s is the time spent at the surface. The oxygen
used during these 3 periods depends on the different rates of
oxygen consumption, m
1
, m
2
, and m
3
, respectively. The oxygen
accumulated by a diver spending time s at the surface is as-
sumed to balance oxygen used during the dive:
K1 2e
2as
m
1
s 1m
2
t 1
where K is the total oxygen storage capacity and a is the initial
rate of oxygen replenishment. The optimal surface duration
s* is obtained by solving:
K1 2e
2as
2m
1
s 2aKe
2as
s

1s 0 2
Given s* and m
2
, Equation 1 can be used to yield estimates
of t*, the optimal foraging time.
Parameters
Several parameters are needed to generate predictions of optimal
time allocation and feeding strategies (Table 1). We obtained m
1
,
Doniol-Valcroze et al. Optimal foraging in a diving predator 881

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

a, and K from the allometric relationships described in Kleiber
(1975) and Stephens et al. (2008), assigning blue whales a mass
of 92 671 kg (Croll et al. 2001). Stephens et al. (2008) found
that a 5% increment to the allometric coefcient for O
2
stores
yielded values closer to the empirical observations of dive-pause
ratios in Halsey, Butler, and Blackburn (2006), likely because
some species have higher mass-specic O
2
storage capacities.
We noted that using this increment yielded similar O
2
stores
(4925 L) to those obtained by Croll et al. (2001) in their detailed
calculation of blue whale O
2
stores (5015 L).
Vertical travel speed is needed to express t as a function of
target depth. Although the cost of swimming is expected to
decrease in large animals, allowing them to swim faster (Eckert
et al. 1999), the speed of divers in optimal models is often
assumed independent of body size (Mori 2002). Stephens
et al. (2008) used a vertical speed of 1 m s
21
for animals of
all possible masses. Exploration of our data set showed that
blue whale vertical speeds during feeding dives (dives with 1
or more lunges) had a mean of 0.97 m s
21
(2575% quan-
tiles: 0.721.21). To improve comparability, we used 1 m s
21
.
The metabolic rate during foraging m
2
has not been mea-
sured in blue whales and needs to be estimated. Houston and
Carbone (1992) warned against estimating m
2
from surface
behavior alone, precisely because under optimal strategies,
the animal will choose the behavior that avoids costly surface
times. Acevedo-Gutierrez et al. (2002), who were the rst to
use optimal models to better understand the feeding behavior
of rorquals, calculated m
2
from the observed allocation of
time in their data. However, to estimate m
2
using optimal
models, one has to trust that whales employ optimal strategies
in the rst place. To avoid circular reasoning, we chose in-
stead to estimate m
2
from energetic considerations alone. In
a detailed simulation of the engulfment process of n whales,
blue whales closest relative, Potvin et al. (2009) estimated that
the energy spent by a 20 m, 50 tons n whale during the pre-
engulfment and engulfment phases of a lunge was 298 080 J
over a 13-s period. Because this energy was spent accelerating
during the pre-engulfment phase, and ghting drag and
pushing the engulfed was mass during the engulfment phase,
we assumed that it was spent on top of the normal energetic
budget of a swimming whale (i.e., on top of the 3.75 3 basal
metabolic rate). Potvin et al. (2009) assumed similar costs
during the purging phase, though it is unclear if these costs
applied to the entire between-lunges interval. We assumed
this was the case and that an extra cost of 22 929 J s
21
applied
to the entire foraging time. Goldbogen et al. (2010) found
that lunging cost was likely to scale allometrically with body
length because mass-specic engulfment volume (which gen-
erates most of the drag-related costs) increases with body
length. Given the similar body shapes of n and blue whales,
Figure 1
Examples of daytime (a) and
nighttime (b) foraging dives
by a blue whale in the St Law-
rence River estuary. Top panel
shows depth recorded every
second (bold line). Circles:
feeding attempts; dashed line:
surface; dotted line: target
depth. Bottom panel: speed re-
corded every second (thin
line). Note the occurrence of
a feeding lunge during the as-
cent phase of the daytime dive
(with no corresponding wig-
gle). In this example, the indi-
vidual performed 6 lunges
during the daytime dive versus
9 lunges during the same pe-
riod of time (13 min) at night-
time.
Table 1
Value and source of parameters used in equations of the optimal model
Parameter Description (units) Formula Value References
M Body mass (kg) 92 671 Croll et al. (2001)
K Total oxygen storage capacity (L) 0.03 M
1.05
4925 Stephens et al. (2008)
a Initial rate of oxygen replenishment (s
21
) 0.90 M
20.33
0.002 Stephens et al. (2008)
m
1
Metabolic rate while traveling (L O
2
s
21
) 0.00065 M
0.75
3.45 Kleiber (1975) and Stephens et al. (2008)
m
2
Metabolic rate while foraging (L O
2
s
21
) 1.78 m
1
6.13 Potvin et al. (2009) and Goldbogen et al. (2010)
882 Behavioral Ecology

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

we assumed that the scaling relationship for n whales of dif-
ferent lengths also applied to blue whales. Because mass-specic
engulfed volume scales with L
0.94
and the mass M scales with
L
2.6
(Goldbogen et al. 2010), lunging costs must scale with
M
1.36
.
For a 92 tons whale, this cost is 22 929 3 (M
blue
/M
n
)
1.36

53 068 J s
21
. And the metabolic rate during foraging is
m
2
3:75 33:39M
0:75
blue
153 068
120 589 J s
21
6:13L O
2
s
21
:
To predict the number of feeding events, we examined the
data to see if the rate at which feeding lunges were performed
changed over the course of a dive. We did not nd any indica-
tion to that effect and therefore assumed a constant rate. The
model predicts a bottom time of 0 at target depths of 0 m be-
cause theoretically, an animal does not need to dive when feed-
ing at the surface. Lunge feeders, however, need to take 1
lunge to have any feeding success. Thus, we set the minimum
number of lunges in the model to 1.
Testing model predictions
We compared predictions of the model over a range of target
depths with actual diving data for bottom time t, dive duration
(s 1 t), the number of feeding lunges, n and the feeding rate
n/(s 1 t 1 s), as well as surface time and dive:pause ratio. To
conform to the models assumption, we only retained dives of
approximately square shape, dened by a ratio between bot-
tom time and dive duration greater than 0.2 (Lesage et al.
1999). Because there are no absolute criteria to assess a -
models validity (Mayer and Butler 1993), we tested the t
of the model in several ways. First, we calculated the Pearsons
productmoment correlation (Guisan and Zimmermann
2000) and the modeling efciency (Mayer and Butler 1993)
as overall measures of agreement between observed and pre-
dicted values. We calculated the slope and intercept of the
linear regression of observed versus predicted values and
tested whether they deviated signicantly from 1 and 0, re-
spectively (Smith and Rose 1995). However, when performed
with large sample sizes, these tests may detect statistically sig-
nicant deviations that are not necessarily indicative of a poor
t (Smith and Rose 1995). Therefore, to better understand
potential biases in prediction accuracy, we computed the mean
squared deviation (MSD), which calculates the deviation of
each predicted value from the 1:1 line (rather than from the
regression line as in the case of the mean squared error, MSE).
MSD is more conservative than MSE and can be decomposed
into 3 additive components with distinct and clear interpreta-
tions (Gauch et al. 2003): squared bias (SB) resulting from
unequal means, non-unity slope (NU) which arises when the
slope of the least-square regression differs from 1, and lack of
correlation (LC) which indicates how scatter around the re-
gression line decreases the correlation coefcient.
Whales feed in bouts within which dives are likely to have
similar characteristics. Lack of independence between dives
of the same individual may result in spurious conclusions
due to autocorrelation. To avoid this, we drew 10 000 random
samples containing 10% of the dives and counted the draws
for which the slope and intercept of the regression between
observed and predicted values deviated signicantly from 1
and 0, respectively. Individual variability was assessed by exam-
ining dive variables for each tagged whale separately.
RESULTS
We deployed instruments on 10 blue whales from2002 to 2009.
Tags remained on whales for 225 h, and individuals were
tracked from the surface for 211 h, yielding 6501 dives over
139 h of data, of which 66 h included surface observations.
Maximum depth of nonfeeding dives was 154 m and maxi-
mum duration was 23 min. Seven individuals performed at
least one feeding lunge on the rst dive following tagging
and 2 others did so during the second dive, suggesting min-
imal impact of tagging on feeding behavior.
Feeding behavior
We identied 1703 dives with at least one feeding attempt and
a total of 2689 lunges (Table 2). All 6 deployments that re-
mained attached into the night showed feeding near the sur-
face (,20 m). Daytime feeding behavior was essentially
bimodal, taking place either in relatively shallow waters
(,40 m) or at depths of about 70100 m, with some lunges
occurring at intermediate depths during twilight. Foraging
blue whales in the St Lawrence Estuary lunged on average
1.58 (61.26 SD) times per dive, with a maximum of 15 lunges
in one single dive. Feeding occurred throughout the diurnal
cycle but lunges were twice as frequent at night than during
daytime (average of 1 lunge every 2.1 min at night vs. every
4.2 min during daytime). The number of lunges per dive was
lower and less variable at night (1.25 6 0.53 SD) than during
daytime (2.18 6 1.69 SD).
Forty-four surface lunges were observed from the research
vessel, all of which were detected by the automated lunge-
feeding detection algorithm. Similarly, the program accurately
detected all 35 lunges from the D-tag that were characterized
by a roll greater than 45. The algorithm proved robust at dis-
criminating against other types of speed changes, for example,
drops in speed when surfacing for air or high-speed pursuits
between individuals during social interactions. A feeding
lunge during the ascent to the surface occurred in 89.5% of
daytime dives (e.g., Figure 1).
Although there was a strong correlation between the num-
ber of wiggles and lunges per dive (P , 0.001, r
2
0.59),
wiggle counts underestimated the number of feeding at-
tempts performed by each individual by 2585%. The propor-
tion of missed feeding attempts per dive averaged 48% during
the day but increased to 98% during twilight and 100% at
night. This high proportion of missed events was largely due
to the consistent failure of wiggle counts to detect lunges
during the nal ascent to the surface and feeding activity at
shallow depths.
Performance of the optimal model
Model predictions of dive characteristics were generated over
the range of feeding depths documented for blue whales (e.g.,
Goldbogen et al. 2011). Characteristics of square feeding di-
ves spanning a range of target depths of 0134 m were com-
pared with model predictions. In agreement with theory,
bottom times and dive durations increased quickly at rst
then at a decreasing rate with increasing depth (Figure
2a,c). Pearson correlation coefcients of 0.74 and 0.87 (P ,
0.001) and model efciencies of 0.42 and 0.70 indicated
a good match with predicted values for these 2 variables.
The slopes of the linear regression between observed and
predicted values did not differ signicantly from 1 despite
the large sample size, showing that prediction accuracy did
not vary with target depth (Figure 2b,d). The intercepts both
differed signicantly from 0 with values of 240 s and 239 s,
respectively (P , 0.001). The similar values of the intercepts
show that dive durations were shorter than predicted solely
because their bottom time component was shorter. Decompo-
sition of the MSD indicated that scatter of observed values was
the largest contributor to the observed deviation from the 1:1
Doniol-Valcroze et al. Optimal foraging in a diving predator 883

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

line (7883%). The contribution of bias was ;20% and that
of slope rotation was negligible (,1%). Random resampling
conrmed these results with 96% of runs yielding slope coef-
cients not signicantly different from 1.
The number of lunges per dive also increased with foraging
depth (Figure 3a). Using the mean interval between lunges
(88 s) to predict the number of lunges from modeled values
of bottom times yielded a strong correlation with model pre-
dictions (q 0.73, P , 0.001) and good overall t (model
efciency 0.49). Neither the slope nor the intercept of the
regression line between observed and predicted values dif-
fered signicantly from 1 and 0, respectively (Figure 3b). Re-
sults were the same in 99% of random resampling runs for the
intercept and 90% for the slope. Decomposition of the MSD
showed that scatter of points was responsible for 90% of the
observed deviation (rather than bias or rotation).
Optimal surface times predicted by the model were signi-
cantly correlated with observed values (q 0.62, P , 0.001).
Model t, however, was poor (model efciency 215), espe-
cially for dives longer than 200 s, which were grossly overesti-
mated (Figure 4a). The dive:pause ratio predicted by the
model was almost invariant with target depth (Figure 4b).
Blue whales feeding at depths over 50 m conformed to this
prediction, but at a ratio 3 times higher than expected. Pre-
dicted dive efciency (bottom time divided by total dive du-
ration plus surface time, Figure 4c) was highest at the surface
Table 2
Characteristics of feeding dives (sample size, depth, dive duration, and number of lunges performed per dive, 6SD) of 10 St Lawrence blue
whales equipped with data recorders
ID
Day (.50 m) Day (50 m) Night
n Depth (m) Duration (s) Lunges/dive n Depth (m) Duration (s) Lunges/dive n Depth (m) Duration (s) Lunges/dive
Tag 0201 7 100 6 23 750 6 100 6.7 6 2 2 30 6 2.3 500 6 330 3 6 2.8 0 n/a n/a n/a
Tag 0301 0 n/a n/a n/a 49 30 6 8.6 390 6 150 3.2 6 1.2 26 3.4 6 1.3 90 6 16 1.6 6 0.5
Tag 0401 0 n/a n/a n/a 170 4.4 6 2.5 110 6 75 1.4 6 0.58 145 1.8 6 0.95 91 6 31 1.3 6 0.46
Tag 0402 11 86 6 11 860 6 300 7.7 6 3.7 50 12 6 6.1 210 6 140 2 6 1 105 0.98 6 2 81 6 50 1.1 6 0.27
Tag 0403 0 n/a n/a n/a 28 11 6 10 190 6 130 2.2 6 1.5 0 n/a n/a n/a
Tag 0404 8 91 6 24 690 6 260 4.8 6 3.5 5 27 6 19 540 6 42 2.6 6 2.5 12 1.1 6 0.72 79 6 20 1 6 0
Tag 0501 2 78 6 30 500 6 180 1.5 6 0.71 4 29 6 3.5 390 6 200 1.8 6 0.5 0 n/a n/a n/a
Tag 0502 22 76 6 17 520 6 130 4.9 6 1.6 145 15 6 7.7 110 6 62 1.3 6 0.63 83 13 6 4.9 140 6 89 1.5 6 0.93
Tag 0602 25 69 6 16 530 6 120 4 6 1.2 43 23 6 13 300 6 150 2.5 6 1.4 132 3.6 6 1.6 85 6 22 1.2 6 0.47
Tag 0901 15 75 6 9 460 6 81 3.8 6 0.94 154 8.4 6 5.1 110 6 68 1.3 6 0.58 0 n/a n/a n/a
Daytime feeding dives followed a bimodal distribution and were consequently separated into shallow (50 m) and deep (.50 m) depth
categories. Feeding dives occurring during twilight were performed across a wide range of depths (080 m) and were omitted for clarity. n/a, not
applicable.
Figure 2
(a) Observed and predicted
values of bottom time as a func-
tion of target depth, over the
range of feeding depths docu-
mented for blue whales. Each
circle represents one dive.
Solid line: model predictions.
(b) Linear regression of pre-
dicted and observed values
(solid line) and 1:1 line
(dashed line). (c) Observed
and predicted values of dive
duration as a function of target
depth. Each circle represents
one dive. Solid line: model
predictions; dotted line:
TADL. (d) Linear regression
of predicted and observed val-
ues (solid line) and 1:1 line
(dashed line).
884 Behavioral Ecology

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

and declined steadily with increasing depth. Feeding rates
decreased slowly with increasing depth and matched the over-
all trend of model predictions though at slightly higher val-
ues. Mean feeding rate at the surface was 0.72 lunges min
21
with maxima in excess of 1.5 lunges min
21
. Feeding deeper
than 40 m yielded an average rate of about 0.5 lunges min
21
.
There was variability in terms of where and when individuals
fed (Figure 5, Table 2). For instance, some individuals were
observed feeding only in shallow waters during the day (e.g.,
Figure 5b), whereas others fed deeper as well (e.g., Figure
5f,g,h). Despite this variability, all individuals conformed to
predictions of dive times and lunge-feeding effort.
DISCUSSION
For pelagic predators, prey behavior determines feeding
depth. Using VTDRs and a novel method to automatically
detect feeding attempts, we have found strong quantitative
agreement between the observed diving behavior of a free-
ranging predator and predictions from a central-place forag-
ing model. With increasing target depth, blue whales in-
creased their time in the food patch, which, combined with
longer transit times, resulted in longer dive durations. Longer
bottom times corresponded to an increase in the number of
feeding events, which, in turn, helped maintain a high feed-
ing rate. Observed data tted model predictions both for each
individual separately and when pooling individuals together,
even though these individuals were tagged at different times,
in different years, and fed at different sites. To our knowledge,
this is the rst time that a theoretical model based on values
derived entirely from the literature and simple allometric re-
lationships can yield quantitatively accurate predictions of
time allocation and feeding behavior of a large marine pred-
ator in a natural system.
The model correctly predicted lunge numbers over the
range of target depths observed in the data set. Beyond
150 m, the model predicted a monotonic response that we
were unable to test because St Lawrence blue whales did not
Figure 3
(a) Observed and predicted
values of the number of lunges
per dive as a function of target
depth, over the range of feed-
ing depths documented for
blue whales. The size of circles
indicates the number of data
points for each depth range
of 10 m. Solid line: model pre-
dictions. (b) Linear regression
of predicted and observed val-
ues (solid line) and 1:1 line
(dashed line).
Figure 4
Observed and predicted values
of (a) surface time, (b) surface
time/dive duration ratio, (c)
dive efciency measured as
bottom time/(dive duration
1 surface recovery time), and
(d) feeding rate over the dive
cycle as a function of target
depth. Each circle represents
one dive. Solid lines: model
predictions.
Doniol-Valcroze et al. Optimal foraging in a diving predator 885

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

feed deepenough. However, blue whales taggedona California
feeding ground performed 35 lunges on average over the
150300 m depth range (Goldbogen et al. 2011), which cor-
responds to our predictions and suggests validity of the model
across populations and larger depth ranges.
Body size is the main determinant of a divers theoretical aer-
obic dive limit (TADL). Yet blue and n whales, the Earths
largest predators, have relatively short dive times (Croll
et al. 2001). The high cost of lunging behavior has been pro-
posed as an explanation for this paradox (Acevedo-Gutierrez
et al. 2002). Our results show that this interpretation is in-
complete because it does not take into account the ecological
context (prey depth and optimization of energy intake). First,
in the optimal foraging framework, the TADL is not a unique
value but rather depends on the relative time spent foraging
versus traveling (Houston and Carbone 1992). When foraging
is more costly than traveling, the TADL actually decreases with
decreasing target depth (Figure 2c) because predators should
spend an important proportion of their time engaged in
costly foraging. The difference between observed behavior
and the TADL at any particular depth is therefore not as large
as previously suggested. However, optimal dive time and
TADL converge with increasing depth, suggesting the TADL
becomes more constraining at larger depths.
Second, it is in fact optimal to perform dives shorter than
the TADL, especially at shallow depths. Because breath-hold
capacity increases with body mass, large animals that dive
shallowly for ecological reasons could make use of the physi-
ological advantage that their size confers by performing long
dives even at shallow depths (Halsey, Blackburn, and Butler
et al. 2006). We have shown that this was not the case for blue
whales foraging in the St Lawrence Estuary, which performed
shorter dives with fewer lunges at shallow depths (Figures 2
and 3). Similarly, Goldbogen et al. (2011) found no support
for the hypothesis that the number of lunges performed by
Pacic blue whales should increase with decreasing dive
depth. These observations suggest that, in agreement with
optimal theory, blue whales perform shorter dives at shallow
depths because the additional recovery time needed at the
surface if they dived for their entire TADL would decrease
their overall feeding rate.
As predicted, feeding rates were consistently higher at shal-
low depths (Figure 4d), conrming that diving predators
should forage close to the surface when possible (Kramer
1988; Carbone and Houston 1996). Accordingly, St Lawrence
blue whales concentrated the majority of their feeding activity
at night, when krill was near the surface. Some individuals in
our data set also foraged at shallow depths during the day,
likely taking advantage of particular habitat conditions (e.g.,
currents forcing krill upward or over shallow banks). These
results suggest that, to a certain extent, diving predators may
judge habitat quality in terms of prey accessibility at shallow
depths rather than selecting solely based on prey density or
abundance. Acevedo-Gutierrez et al. (2002) used optimal
models to study the effect of high feeding costs on diving
behavior of rorqual whales, but small sample sizes prevented
them from fully considering the effect of prey depth. We con-
clude that while the cost of lunges undoubtedly reduces the
duration of feeding dives (via m
2
), the main reason for blue
whales to perform foraging dives shorter than their TADL is
that it is optimal to do so.
Decomposition of the model error (MSD) showed that most
lack of t was due to scatter. This could reect individual pref-
erences and variations in body mass among tagged whales, but
separate examination of individuals showed that all individuals
conformed to model predictions to some degree and that most
of the scatter came from intraindividual variation. Dive time at
a given depth may depend on the quality of food patches
(Mori 1998). Behavioral exibility should allow individuals
to take advantage of a high-quality patch by feeding more
than predicted and paying the oxygen debt later (e.g., the
individual that took 15 mouthfuls in one dive). Thompson
Figure 5
Observed and predicted values
of dive duration as a function
of target depth for 8 individual
blue whales. Each circle repre-
sents one dive. Solid line: model
predictions. (a) tag 0301; (b)
tag 0401; (c) tag 0402; (d) tag
0403; (e) tag 0404; (f) tag 0502;
(g) tag 0602; and (h) tag 0901.
Note: tags 0201 and 0501 con-
tributed only 9 and 6 feeding
dives and thus were not repre-
sented here.
886 Behavioral Ecology

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

and Fedak (2001) proposed that simple giving-up rules can
help divers assess patch quality and recognize a poor patch
early in the dive. In this case, it can be benecial to give up
before reaching the optimal foraging time and start again in
an area of higher prey concentration. The occasional occur-
rence of wiggles not accompanied by feeding events suggests
that vertical excursions can be exploratory movements or
lunges that were aborted because of poor patch quality. Con-
current measures of krill vertical distribution and density (ob-
tained from in situ measurements or model simulations)
would allow the estimation of feeding efciency in terms of
net energy gain. Adding this parameter in the optimal model
would presumably explain some of the observed deviance and
improve its explanatory power.
Recovery surface times following deep dives were shorter
than predicted by the model (Figure 4a). Stephens et al.
(2008) recognized the sensitivity of model predictions to a,
the initial proportional rate of oxygen replenishment, and
used an ad hoc coefcient to obtain realistic dive times for
divers of intermediate mass (150 kg). It is possible that this
value of a does not provide a realistic measure of the O
2
exchange performance by large whales. Our denition of sur-
face time could also underestimate time allocated to O
2
re-
covery, for instance, if gas exchange continues during the rst
few seconds of a dive (before lung collapse) or if an O
2
debt is
accumulated and not repaid until the end of a dive bout, as
in Steller sea lions, Eumetopias jubatus (Fahlman et al. 2008).
Conversely, some short dives were followed by longer than
expected post-dive periods, suggesting that other factors, for
example, CO
2
build-up due to unequal rates of exchange
between O
2
and CO
2
, can constrain surface times (Boutilier
et al. 2001). Additional efforts are clearly needed to model
respiratory gas exchanges in cetaceans.
Blue whales obtained very advantageous dive/surface ratios
during numerous shallow dives by limiting their recovery time
to a single breath at the surface (i.e., near-instantaneous surface
times of 13 seconds). At larger depths, their diving strategy
allowed them to maintain a stable ratio (Figure 4b). This ratio,
however, was 3 times higher than expected, presumably because
surface times were overestimated by the model. Dive efciency,
measured as bottom time divided by total dive duration plus
surface time, was predicted to peak at the surface and to de-
crease with increasing depth (Figure 4c). At depths over 20 m,
blue whales followed this pattern at slightly higher values than
predicted but not at shallowdepths for whichdive efciency was
lower than expected because of shorter bottom times.
Blue whales performed the predicted number of lunges de-
spite bottomtimes being slightly shorter than model predictions.
The average difference between predicted and observed bottom
times (;40 s) was roughly half of the mean interval between
lunges. Thus, whales save time by performing the last lunge
during the ascent to the surface, essentially transferring the time
necessary to process water and food into the incompressible
transit time. This is also true for shallow dives with only one
lunge, in which case blue whales were likely coupling respiration
with the purging phase of the lunge, as suggested by Goldbogen
et al. (2011). Combining foraging, ascent and even respiration
thus represents additional strategies to maintain a high lunge
rate and maximize energetic efciency.
Many marine predators have to optimize a short seasonal
window of feeding opportunity. Recent results from bio-
logging studies in natural systems have provided an increasing
body of evidence that divers employ strategies based on opti-
mal decisions to maximize foraging efciency. Dive time allo-
cation in diving birds has been shown to vary with distance
(Heath et al. 2007) and prey density (Mori et al. 2002; Cook
et al. 2008) in accordance with theoretical predictions. Diving
seabirds also select optimal stroke frequency patterns during
vertical movements (Mori et al. 2010). Differences in the op-
timal diving depths of penguins can help explain coexistence
of sympatric species (Mori and Boyd 2004b). Seals also seem
to conform to optimal theory when allocating time within
dives (Boyd et al. 1995) and choosing time spent in food
patches of different quality (Mori and Boyd 2004a). Hump-
back whales have been observed foraging shallower than the
depth of maximal prey density (Goldbogen et al. 2008), tting
predictions of optimal models (Mori 1998).
Direct measures of how optimal strategies increase feeding
success, however, are rare and altogether missing in large diving
predators like cetaceans. We have shown for the rst time that
a simple model of optimal time allocation using allometric argu-
ments can explain not only the dive time budgets of a large ma-
rine predator but also its feeding strategies as a function of the
distance betweensurface and prey. Allometric relationships have
considerable potential for explaining patterns across taxa but
often fail to address specic ecological situations. As suggested
by Stephens et al. (2008), combining optimality and allometry
can better explain the actual foraging choices made by air-
breathing divers. With the increasing availability of data loggers
placed on free-ranging animals, this framework opens new ave-
nues of study to better understand the behavior of marine
predators. Such models could help predict responses of pred-
ators to environmental changes and anthropogenic pressures,
placing them directly at the interface between ecology and
conservation.
We thank Paul Couture for Visual Basic programming, Robin Baird,
and Michel Moisan for tag development, Yves Morin, Daniel Lefebvre,
Renaud Pintiaud, Michel Moisan, Caroline Guimont, Sean Thompson,
and Jeremy Winn for help with eldwork, and Becky Woodward for
providing access to the D-tag data. We also thank Sebastien Lemieux
Lefebvre, Arnaud Mosnier, and Frederic Bailleul for advice on this
project. Finally, we thank Dr Sue Healy, Dr Jeremy Goldbogen, and
one anonymous referee for their constructive comments on the man-
uscript. This work was supported by the Species at Risk and Oceans
Management programs of Fisheries and Oceans Canada, and by the
SaguenaySt Lawrence Marine Park.
REFERENCES
Acevedo-Gutierrez A, Croll DA, Tershy BR. 2002. High feeding costs
limit dive time in the largest whales. J Exp Biol. 205:1747.
Baird RW. 1998. Preliminary calibration of velocity meters on a captive
killer whale. Newport (OR): Free Willy Keiko Foundation.
Boersma P, Weenink D. 2009. Praat: doing phonetics by computer.
v.5.1.02. . Available from: http://www.praat.org.
Boutilier RG, Reed JZ, Fedak MA. 2001. Unsteady-state gas
exchange and storage in diving marine mammals: the harbor
porpoise and gray seal. Am J Physiol Regul Integr Comp Physiol.
281:R490R494.
Boyd IL, Reid K, Bevan RM. 1995. Swimming speed and allocation of time
during the dive cycle in Antarctic fur seals. Anim Behav. 50:
769784.
Brischoux F, Bonnet X, Cook TR, Shine R. 2008. Allometry of diving
capacities: ectothermy vs. endothermy. J Evol Biol. 21:324329.
Calambokidis J, Schorr GS, Steiger GH, Francis J, Bakhtiari M, Marshal
G, Oleson EM, Gendron D, Robertson K. 2007. Insights into the
underwater diving, feeding, and calling behavior of blue whales
from a suction-cup-attached video-imaging tag (CRITTERCAM).
Mar Technol Soc J. 41:19.
Carbone C, Houston AI. 1994. Patterns in the diving behavior of the
Pochard, Aythya-Ferinaa test of an optimality model. Anim Behav.
48:457.
Carbone C, Houston AI. 1996. The optimal allocation of time over the
dive cycle: an approach based on aerobic and anaerobic respiration.
Anim Behav. 51:1247.
Cook TR, Lescroel A, Tremblay Y, Bost CA. 2008. To breathe or not to
breathe? Optimal breathing, aerobic dive limit and oxygen stores in
deep-diving blue-eyed shags. Anim Behav. 76:565576.
Doniol-Valcroze et al. Optimal foraging in a diving predator 887

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Croll DA, Acevedo-Gutierrez A, Tershy BR, Urban-Ramirez J. 2001.
The diving behavior of blue and n whales: is dive duration shorter
than expected based on oxygen stores? Comp Biochem Physiol A
Mol Integr Physiol. 129:797.
Eckert R, Randall D, Burggren W, French K. 1999. Animal physiology:
mechanisms and adaptations. Paris: De Boeck.
Fahlman A, Svard C, Rosen DAS, Jones DR, Trites AW. 2008. Metabolic
costs of foraging and the management of O-2 and CO2 stores in
Steller sea lions. J Exp Biol. 211:35733580.
Fiedler PC, Reilly SB, Hewitt RP, Demer D, Philbrick VA, Smith S,
Armstrong W, Croll DA, Tershy BR, Mate BR. 1998. Blue whale
habitat and prey in the California Channel Islands. Deep Sea Res
Part II Top Stud Oceanogr. 45:17811801.
Gauch HG, Hwang JTG, Fick GW. 2003. Model evaluation by compar-
ison of model-based predictions and measured values. Agron J. 95:
14421446.
Goldbogen JA, Calambokidis J, Croll DA, Harvey JT, Newton KM,
Oleson EM, Schorr G, Shadwick RE. 2008. Foraging behavior of
humpback whales: kinematic and respiratory patterns suggest a high
cost for a lunge. J Exp Biol. 211:3712.
Goldbogen JA, Calambokidis J, Oleson EM, Potvin J, Pyenson ND,
Schorr G, Shadwick RE. 2011. Mechanics, hydrodynamics and en-
ergetics of blue whale lunge feeding: efciency dependence on krill
density. J Exp Biol. 214:131146.
Goldbogen JA, Calambokidis J, Shadwick RE, Oleson EM, McDonald
MA, Hildebrand JA. 2006. Kinematics of foraging dives and lunge-
feeding in n whales. J Exp Biol. 209:1231.
Goldbogen JA, Potvin J, Shadwick RE. 2010. Skull and buccal cavity
allometry increase mass-specic engulfment capacity in n whales.
Proc R Soc B Biol Sci. 277:861868.
Guisan A, Zimmermann NE. 2000. Predictive habitat distribution
models in ecology. Ecol Modell. 135:147186.
Halsey L, Woakes A, Butler P. 2003. Testing optimal foraging models
for air-breathing divers. Anim Behav. 65:641.
Halsey LG, Blackburn TM, Butler PJ. 2006. A comparative analysis
of the diving behaviour of birds and mammals. Funct Ecol.
20:889.
Halsey LG, Butler PJ, Blackburn TM. 2006. A phylogenetic analysis of
the allometry of diving. Am Nat. 167:276.
Hartigan JA, Wong MA. 1979. A K-means clustering algorithm. Appl
Stat. 28:100.
Heath JP, Gilchrist HG, Ydenberg RC. 2007. Can dive cycle models
predict patterns of foraging behaviour? Diving by common eiders in
an Arctic polynya. Anim Behav. 73:877884.
Houston AI, Carbone C. 1992. The optimal allocation of time during
the diving cycle. Behav Ecol. 3:255.
Houston AI, McNamara JM. 1985. A general theory of central place
foraging for single-prey loaders. Theor Popul Biol. 28:233.
Johnson MP, Tyack PL. 2003. A digital acoustic recording tag for
measuring the response of wild marine mammals to sound. IEEE
J Oceanic Engineering. 28:3.
Kleiber M. 1975. The re of life: an introduction to animal energetics.
Huntington (NY): Krieger.
Kooyman GL, Ponganis PJ. 1998. The physiological basis of diving to
depth: birds and mammals. Ann Rev Physiol. 60:19.
Kramer DL. 1988. The behavioral ecology of air breathing by aquatic
animals. Can J Zool. 66:89.
Lesage V, Hammill MO, Kovacs KM. 1999. Functional classication of
harbor seal (Phoca vitulina) dives using depth proles, swimming
velocity, and an index of foraging success. Can J Zool. 77:74.
Luque SP. 2007. Diving behaviour analysis in R. R News. 7:814.
Mayer DG, Butler DG. 1993. Statistical validation. Ecol Modell. 68:2132.
Mori Y. 1998. Optimal choice of foraging depth in divers. J Zool.
245:279.
Mori Y. 2002. Optimal diving behaviour for foraging in relation to
body size. J Evol Biol. 15:269276.
Mori Y, Boyd IL. 2004a. The behavioral basis for non-linear functional
responses and optimum foraging in Antarctic fur seals. Ecology. 85:
398410.
Mori Y, Boyd IL. 2004b. Segregation of foraging between two sympat-
ric penguin species: does rate maximization make the difference?
Mar Ecol Prog Ser. 275:241249.
Mori Y, Takahashi A, Mehlum F, Watanuki Y. 2002. An application of
optimal diving models to diving behaviour of Brunnichs guille-
mots. Anim Behav. 64:739745.
Mori Y, Takahashi A, Trathan PN, Watanuki Y. 2010. Optimal stroke
frequency during diving activity in seabirds. Aquat Biol. 8:247257.
Oleson EM, Wiggins SM, Hildebrand JA. 2007. Temporal separation
of blue whale call types on a southern California feeding ground.
Anim Behav. 74:881894.
Orians GH, Pearson NE, Horn DJ, Mitchell DR, Stairs GR. 1979. On
the theory of central place foraging. Analysis of ecological systems.
Columbus (OH): Ohio State University Press. p. 155.
Perry G, Pianka ER. 1997. Animal foraging: past, present and future.
Trends Ecol Evol. 12:360364.
Potvin J, Goldbogen JA, Shadwick RE. 2009. Passive versus active engulf-
ment: verdict from trajectory simulations of lunge-feeding n whales
Balaenoptera physalus. J R Soc Interface. 6:10051025.
R Development Core Team. 2008. R: a language and environment for
statistical computing. Vienna (Austria): R Foundation for Statistical
Computing. ISBN 3-900051-07-0. [cited 2010 December 22]. Available
from: http://www.r-project.org.
Schreer JF, Kovacs KM. 1997. Allometry of diving capacity in air-
breathing vertebrates. Can J Zool. 75:339358.
Schreer JF, Kovacs KM, Hines RJO. 2001. Comparative diving patterns
of pinnipeds and seabirds. Ecol Monogr. 71:137.
Smith EP, Rose KA. 1995. Model goodness-of-t analysis using regres-
sion and related techniques. Ecol Modell. 77:4964.
Stephens DW, Krebs JR. 1986. Foraging theory. Princeton (NJ):
Princeton University Press.
Stephens PA, Carbone C, Boyd IL, McNamara JM, Harding KC, Houston
AI. 2008. The scaling of diving time budgets: insights from an opti-
mality approach. Am Nat. 171:305.
Thompson D, Fedak MA. 2001. How long should a dive last? A simple
model of foraging decisions by breath-hold divers in a patchy envi-
ronment. Anim Behav. 61:287.
Walton P, Ruxton GD, Monaghan P. 1998. Avian diving, respiratory
physiology and the marginal value theorem. Anim Behav. 56:
165174.
Woodward R. 2006. Locomotory strategies, dive dynamics, and func-
tional morphology of the Mysticetes: using morphometrics, osteol-
ogy, and Dtag data to compare swim performance in four species of
baleen whales [dissertation]. Orono (ME): University of Maine.
180 p.
888 Behavioral Ecology

b
y

g
u
e
s
t

o
n

J
u
l
y

4
,

2
0
1
4
h
t
t
p
:
/
/
b
e
h
e
c
o
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Vous aimerez peut-être aussi