Fisheries science JSFS , tập 77, số 2, 2011 3

Fish Sci (2011) 77:161–167
DOI 10.1007/s12562-011-0323-1

ORIGINAL ARTICLE

Fisheries

Swimming angle and target strength of larval Japanese anchovy
(Engraulis japonicus)
Yusuke Ito • Hiroki Yasuma • Reiji Masuda
Kenji Minami • Ryuichi Matsukura •
Saho Morioka • Kazushi Miyashita

•

Received: 30 June 2010 / Accepted: 16 December 2010 / Published online: 18 February 2011
Ó The Japanese Society of Fisheries Science 2011

Abstract The swimming angle of larval Japanese anchovy
(Engraulis japonicus) was measured in a tank, and target
strength (TS) was calculated using a theoretical scattering
model. The mean swimming angle was 12.8° (SD ±22.1).
Increased speeds of flow led to increased mean swimming
angles. The mean swimming angle at flow of 5 cm s-1 was
higher than at other speeds. TS values were estimated
using a distorted-wave Born approximation model for two
cases. Average values were 1–3 cm s-1 (11.5° ± 22.1) and
5 cm s-1 (16.6° ± 21.7) for cases 1 and 2, respectively.
For case 1, TS ranged from -92.0 to -74.7 dB with a
mean of -79.4 dB at 120 kHz. For case 2, TS ranged from

-92.2 to -75.2 dB with a mean of -79.9 dB. The mean
TS in case 2 was lower than that in case 1, with the
maximum difference being 1.0 dB at 120 kHz (standard
Y. Ito (&)
Laboratory of Marine Ecosystem Change Analysis, Graduate
School of Environmental Science, Hokkaido University,
3-1-1 Minato-cho, Hakodate 041-8611, Japan
e-mail:
H. Yasuma Á K. Miyashita
Laboratory of Marine Ecosystem Change Analysis, Field
Science Center for Northern Biosphere, Hokkaido University,
3-1-1 Minato-cho, Hakodate, Hokkaido 041-8611, Japan
R. Masuda Á K. Minami
Fisheries Research Station, Kyoto University, Nagahama,
Maizuru, Kyoto 625-0086, Japan
R. Matsukura
National Research Institute of Fisheries Engineering,
FRA, 7620-7 Hasak, Kamisu, Ibaraki 314-0408, Japan
S. Morioka
Fisheries Research Institute, Tokushima Agriculture,
Forestry and Fisheries Technology Support Center,
1-3 Hiwasa, Minami, Kaifu, Tokushima 779-2304, Japan

length 22.0 mm). However, there were no significant differences between the regression lines of cases 1 and 2.
Thus, changes in flow speed altered the swimming angle of
larval Japanese anchovy, but had little influence on TS.
Keywords Larval Japanese anchovy Á Swimming angle Á
Target strength

Introduction

The Japanese anchovy (Engraulis japonicus) is one of the
most important coastal fisheries species in Japan. The larvae are an especially important resource in near-shore
fisheries. According to annual fishery statistics, commercial
tow-net harvests were valued at 30 billion yen in 2007 [1].
Therefore, data on the distribution and abundance of larvae
are important for near-shore fisheries. Accurate quantitative data on larvae are particularly important, not only
for appropriately managing larvae fisheries but also for
predicting the recruitment and management of adults.
Various studies have used commercial catch data to
estimate the abundance and distribution of larval anchovies, often linked to environmental parameters such as
current, salinity, and precipitation [2–4]. However, such
analyses must use data collected over a long period
(months or years), and it is thus difficult to obtain quantitative estimates within a single fishery season. Conversely,
acoustic observations using quantitative echo sounders can
provide quantitative data in the short term and have been
used for stock assessments of many species [5, 6].
In acoustic surveys, a quantitative echo sounder provides reflections from a fish school at various echo intensities. This acoustic reflection is converted to quantitative
data (e.g., number of individuals, biomass) using the target

123


162

strength (TS). Recently, the TSs of various species have
been reported for use in estimating abundance in the field
[7, 8]. A split-beam echo sounder can be used to measure
TS if a target is not too small in comparison with wave
length. In larval anchovy, however, field measurement of
TS is difficult. Acoustic reflections from larval anchovies

are weak because of the small size of the fish and they form
schools during the day influence the way the target detects
them. Therefore, the TS of larval Japanese anchovy should
be estimated using a theoretical sound-scattering model.
Many theoretical models have been developed for calculating the TS of fish [9]. Most of these models use an
approximate geometric configuration to represent the
swim-bladder (in swim-bladdered fish) or body (in swimbladderless fish) of the target species, which can be
obtained from the relationship between TS and swimming
angle (pitch or yaw). The Japanese anchovy is a physostomous fish. The swim-bladders of their larvae are filled
with gas during the night, which helps to reduce energy
consumption. However, during the day, they discharge gas
from their swim-bladders and form schools [10]. Miyashita
[11] reported that acoustic surveys of larval anchovy
should be conducted on high-density schools during the
day. Moreover, larval anchovy TSs of 50 and 200 kHz
were estimated using the DWBA-based deformed cylinder
model (DWBA model) [12, 13] that was developed for
swim-bladderless species. This study suggested that the
dorsal average TS is very sensitive to changes in tilt angle.
Therefore, the TS should be determined using the tilt angle
distribution after observation of swimming behavior.
In this study, we observed the swimming angles of
larval Japanese anchovy using a video camera and estimated the TS as a function of tilt angle (TSavg) for use in
acoustic surveys. We used the swimming angle observation
results to calculate the TSavg at 38 and 120 kHz (which are
the main frequencies used by coastal research vessels in
Japan) using the DWBA model. We also considered the
availability of TSavg as a scale factor in acoustic surveys,
using the relationship between TSavg and the body lengths
of larval anchovies.

Materials and methods
Observation of swimming behavior
In January and December 2008, live larval anchovies were
provided from fixed shore nets in Wakasa Bay. During both
experimental periods, the larvae were transferred as soon as
possible to a black fiberglass cylindrical tank (500 L,
116 cm diameter, 77 cm deep) at the Fisheries Research
Station of Kyoto University. One day later, we chose live
larval anchovies and transported them to an experimental

123

Fish Sci (2011) 77:161–167

116cm
13cm

Side
Side strut
strut

10cm

45cm
77cm
70cm
Video camera

97.5cm

Fig. 1 Schematic diagram of the experimental tank used for the
penned Japanese anchovy larvae

tank that was maintained at ambient temperature (10°C) in
an environment that blocked natural light. A 12-h photoperiod was maintained without a dawn or dusk transition in
light intensity. During the day, incident light at the surface
measured approximately 350 lx; no light was provided
during the night.
We conducted the swimming experiment during the day,
to simulate the environmental conditions in the bay, and to
adjust the flow in the experimental tank, seawater was added
as needed. The speed of flow was measured before and after
video recording using an Electromagnetic Flow Velocity
sensor (AEM1-D; ALEC Electronics, Tokyo) within the
camera field angle. Video recording began 10 min after the
speed of flow was stable and lasted for 1 h at each speed.
Video recordings of the swimming behavior of 40
individuals were collected using two underwater cameras
(T-WATER-2200c; HERO, Tokyo) (TL 39.0 ± 3.4 mm)
in January. In December, two digital HD video camera
recorders (HDR-SR12; SONY, Tokyo) were used to give
high quality image results for 12 individuals (total length
39.2 ± 1.8 mm). Recording systems were placed at a
depth of 45 cm (Fig. 1), and two additional underwater
cameras were used to check the relationships between
individuals and the camera lenses. In this environment, we
observed the swimming behavior of larval anchovies. The
obtained video recordings were converted to photographs
at 1-min intervals for each experimental period. Swimming
angles were calculated from these photographs using image

editing software (SCM Measure; Moritex, Tokyo) (Fig. 2).
There were two requirements for measurements: the centerline of the individual being measured could not be
bending, and the individual had to be perpendicular to the
camera. The second parameter was confirmed using
the dorsal camera. The swimming angle was computed


Fish Sci (2011) 77:161–167

163

rbs ¼ jfbs j2 , rpos is the position along the axis of the
deformed cylinder, and k is the acoustic wave number
given by k = 2p/k, where k is the acoustic wavelength.
The subscript sw refers to the surrounding seawater, the
subscript animal refers to larval anchovy, Jl is a Bessel
function of the first kind of order l, and ac is the crosssection radius of the cylinder and incident wave. We digitized and obtained 200 sets of rpos and ac from the dorsal
images of specimens. We tried to measure the density
contrast (g) and sound-speed contrast (h) of the larval
anchovy using the density-bottle [16] and time-of-flight
[17] methods. These values were applied to the DWBA
model as parameters (g = 1.068, h = 1.037) (Ito et al.,
unpublished). TS (dB) is defined as TS = 10 log10(rbs),
and the average TS was defined as the mean TS, ranging
from -90° to 90° at 1° steps (tilt angle: head-down, headup position).
The tilt-averaged TS (TSavg) was calculated using the
probability density function (PDF) of fish tilt angle f(bs) in
Eqs. 2 and 3 [15]:

Dorsal image

Target

Lateral image

0q
Non- target

+ǰ
-ǰ

Target

rAvg ¼

Zp=2

rðhÞf ðhÞdh

ð2Þ

Àp=2

Fig. 2 Typical example of a photograph obtained from the observation experiment. The upper panel shows a dorsal image and the lower
shows a lateral image; h indicates the swimming angle

according to the above parameters. The angle was that
between the centerline of the fish, an imaginary line running from the root of the tail to the tip, and the true horizontal. We defined positive angles as those for which the
fish was head-up and negative angles as those for which the
fish was head-down.
Theoretical model

A total of 200 larval anchovies ranging from 18.0 to
35.7 mm SL (SL = 1.22 9 TL - 2.41, R2 = 0.99) were
used for the TS calculation. The sound scattering from
specimens was estimated using the DWBA model. A
modification of the Matlab codes described in McGehee
et al. [14] ver. 6.1 (MathWorks, Natick, MA, USA) was
used to estimate the TS of the deformed cylinder, given as


Z
2
ksw
ac 1 þ h2
fbs ¼
À
2
4kanimal gh2
~
rpos

Âe2ikanimal Ár~pos


Jl ð2kanimal ac cosbtilt Þ
~pos 
dr
cosbtilt

ð1Þ

where f(bs) is the complex backscattering amplitude, the
relation to backscattering cross-section rbs is given by

TSAvg ¼ 10 logrAvg

ð3Þ

where h represents swimming angle; f(h) was assumed to
be a truncated normal distribution function. The truncations were made at "h À 3Sh and "h þ 3Sh , where "
h and Sh
denote the mean and standard deviation of the tilt angle,
respectively. Since the number of samples was different in
each photograph, strictly speaking, we should extract the
same number of samples from each photograph. However,
we were not able to do so because the experiment was
limited. Therefore, mean swimming angle and standard
deviation were calculated using the limited samples in this
study.

Results
Theoretical TS
Typical examples of the relationship between variations in
TS and body tilt angle at 38 and 120 kHz obtained by the
DWBA model are shown in Fig. 3. The variation in TS
versus tilt angle showed peaks at about 0° (dorsal) at both
frequencies, and the maximum TS at 120 kHz was higher
than those at 38 kHz in all specimens. Moreover, these
peaks were relatively sharp, especially for the higher frequency, suggesting that slight changes in fish tilt angles
have a major effect on TS. The maximum TSs at 120 kHz
were -85.6 and -68.1 dB in the smallest and the largest

123


164

Fish Sci (2011) 77:161–167

(a)

Target strength (dB)

-60

SL: 18.0mm

-80
-100
-120

㧦 38 kHz
㧦 120 kHz

-140
-90

-60

-30

0

30

60

90

Tilt angle (degree)

Target strength (dB)

(b)

-60

Fig. 4 Swimming angles of larval Japanese anchovies. Middle
squares indicate the mean; whiskers show the standard deviation.
Numbers in parentheses indicate the maximum and minimum values

SL: 35.7mm

-80

the swimming angles in cases 1 and 2. There were no
differences in either the ranges of values between cases 1
and 2 or their standard deviations (22.1 and 21.7, respectively). In contrast, the mean swimming angles differed by
more than 5° (case 1: 11.5°, case 2: 16.5°).

-100

㧦38 kHz
-120
-90

㧦120 kHz
-60

-30

0

30

60

90

Tilt-averaged TS

Tilt angle (degree)
Fig. 3 Typical variations in the target strength (TS) of minimum
(a) and maximum larval length (b) as a function of fish pitch angle,
estimated by the distorted-wave Born approximation (DWBA) model
at 38 (dotted lines) and 120 (bold lines) kHz. Positive angles are headup and negative angles are head-down

individuals, respectively. In individuals with larger standard length, the tilt angle had a considerable impact on TS.
Relationship between swimming angle and flow speed
The flow speeds in the experimental tank were 1, 2, 3, and
5 cm s-1, respectively. We extracted 2,637 angles (2,102

in January, 535 in December) from photographs. Figure 4
shows box plots of the swimming angles at each flow
speed. In this study, the mean and standard deviation of the
swimming angle of larval anchovies tended to rise as the
flow speed increased. However, the standard deviation was
not related to the flow speed. Larval anchovies were
affected by flow variations (P \ 0.001, ANOVA), and the
mean swimming angle at 5 cm s-1 was higher than those at
the other flow speeds (P \ 0.01, Tukey-Kramer test).
Therefore, we separated the calculations of TS into two
cases. One case combined the data from flow speeds of
1–3 cm s-1 (case 1), and the other used data from flow
speeds of 5 cm s-1 (case 2). Figure 5 shows histograms of

123

The values of TS were calculated with respect to fish tiltangle distribution (PDF), which was used for two values
(case 1 and case 2) at two frequencies (38 and 120 kHz).
TS values are plotted in Fig. 6 as functions of fish SL on a
logarithmic scale. Ranges of TSavg and the equations of the
regression lines (the TS-length equation) are shown in
Fig. 6 and in Table 1. In general, the results of TS can be
expressed in terms of the body length L using the following
equation:
TS ¼ m log10 L þ b

ð4Þ

where m and L are constants for a given species. This
equation has been a generally accepted description of the

way in which mean TS depends on fish length [18]. The
slope m and intercept b can be estimated by linearly
regressing the TS on log L. In this study, the regression line
was fitted to the estimated data using cases 1 and 2. The
results were TSavg ¼ 60:9 log10 L À 107:4 (R2 = 0.93) for
case 1, and TSavg ¼ 60:1 log10 L À 107:5 (R2 = 0.92) for
case 2 at 120 kHz. The mean TSavg was -79.4 for case 1
and -79.9 dB for case 2. The difference in each case was
0.98 dB (SL 22.0 mm) at the maximum. Analysis of
covariance (ANCOVA) was used to test for differences
in the regression lines. The slopes and intercepts of the
relationships did not differ among cases (p \ 0.05). TSavg
at 38 kHz was weak because L was too small for a
wavelength.


Fish Sci (2011) 77:161–167

165

Fig. 5 Frequency distributions
of the swimming angles of
larval Japanese anchovy under
flow rates of 1–3 cm s-1 (a) and
5 cm s-1 (b)

(a)

(b)

1-3cm/sec

30

30

n = 682
Frequency (%)

Frequency (%)

n = 1955
Mean: 11.5

20

S.D.: 22.1
10

0
-90

-60

-30

0

30

Tilt angle (degree)
-80

Target strength (dB)

: Case 1
: Case 2
-90

-100

38 kHz
-110
0.2
.

0.4

0.3

0.5

0.6

Log SL (cm)

Target strength (dB)

-70

: Case 1
: Case 2

-80

-90

120 kHz
-100
0.2
.

0.3

0.4

0.5

5cm/sec

0.6

Log SL (cm)
Fig. 6 Relationship between TS and log of standard length (in mm)
for each case. TS is 38 kHz for the upper panel and 120 kHz for the
lower panel. The circles and crosses indicate calculated TSs for cases
1 (flow speed 1–3 cm s-1) and 2 (flow speed 5 cm s-1), respectively.
Equations for the regression lines in each panel are given in Table 1

Discussion

In this study, we used individuals with total lengths of
28.3–45.4 mm for the observation experiment. The swimming angles of these individuals tended to be head-up for
each speed of flow. However, the relationship between

60

90

20

Mean: 16.5
S.D.: 22.0

10

0
-90

-60

-30

0

30

60

90

Tilt angle (degree)

swimming angle and body length was not clear. We conjecture that the swimming angle of the larvae differs with
body length. Hunter [19] showed that the swimming speed
of northern anchovy (Engraulis mordax) larvae obviously
changed with increases in body length. Additionally, Batty
[20] noted that the speed of larval herring increased with
length. Thus, the swimming abilities and swimming angles
of larvae would also change with increases in body length.
In the juvenile stage, the swimming angle was almost 0°,
and there was no influence of flow speed on the mean
swimming angle (1–5 cm s-1 in four steps, one-way
ANOVA, P [ 0.05) (Ito et al., unpublished). Thus, the
differences in swimming angle between larvae and juveniles are likely caused by body growth. Blaxter et al. [21],
for example, reported that metamorphosis entails the
development of fins, resulting in improved swimming
ability. Therefore, as the body grows, the swimming angle
will be close to 0°. However, the mean swimming angle of
individuals in this study would not be changed by body
growth because larval anchovies complete metamorphosis
at about 40 mm in length. Individuals measuring 20 mm or
less in length may swim at various swimming angles.
Therefore, an experiment examining the swimming angles
of smaller fish is necessary to improve the precision of the
TS. However, our results suggest that the influence of the
TS of small individuals is low because the TS of the largest
individual was significantly different compared with that of
the smallest individual due to changes in fish tilt angles.
We suggest that swimming angle gives larval anchovy
the ability to swim against the flow and is determined by

body length. In this study, body lengths ranged from 28.3
to 45.4 mm and from 36.4 to 41.3 mm in January and
December, respectively. The mean swimming angles were
13.6° and 9.6° for each period (t test P \ 0.05). Thus, the
swimming angle obviously changed with changes in the
body length. However, with the exception of the data
gathered at 5 cm s-1, these values were not significantly
different between periods. In individuals with small body
lengths, the swimming angle relates to the flow. In this
study, the flow at 5 cm s-1 affected the mean swimming

123


166

Fish Sci (2011) 77:161–167

Table 1 Equations for the
linear regression in Fig. 5 and
mean target strength at each
frequency

n

Case 1 (1–3 cm/s)
y represents TS (dB), and
x represents the log of standard
length in centimeters

Case 2 (5 cm/s)

200
200

angle, and the TSavg was calculated using the PDF for two
cases (1–3, 5 cm s-1). However, the regression lines of the
calculated TSavg for the two cases did not differ significantly (Table 1). The Japanese anchovy is widely distributed in the northwest Pacific under various environmental
conditions, and swimming angles under these conditions
have not yet been clarified. Therefore, in the future,
swimming behavior must be observed under varying
speeds of flow. Additionally, it is necessary to continuously
measure the flow speed during the experiment period.
In this study, the TS of larval anchovy clearly depended on
changes in the swimming angle, which was in turn determined by body length. In a previous study, the swimming
angle of pelagic fish was set at the TS of larvae because the
swimming angle of the larvae was not known. As a result, the
TSavg will become a source of error when abundance is
estimated. Therefore, reliable swimming data must be used
to calculate TSavg for use in acoustic surveys. Additionally,
this study shows that when the swimming angle has a large
standard deviation it has a little influence on TSavg, if a higher
standard deviation of swimming angle (Fig. 6).
There are also some additional parameters that affect
TS. Mikami et al. [22] measured the density (g) and soundspeed contrast (h) of E. pacifica, which has no swimbladder, and thus no air bubbles in the body, between
spring and autumn. Their results showed that the maximum
TS values for E. pacifica, calculated from a theoretical
scattering model, changed by about 5 dB from spring to
autumn. Furthermore, Matsukura et al. [23] suggested that
g and h are affected by seasonal changes in lipid profiles. In

this study, the larval anchovy g and h values were fixed at
1.068 and 1.037, respectively. However, these values must
be adjusted to suit different conditions (i.e., season, area),
and we must further examine such parameters in the future.
As mentioned above, these factors are important when
attempting to estimate the TS of larval Japanese anchovy.
In particular, the influence of the swimming angle was
high. This is the first paper to provide swimming angle
data. The TSavg values given in Fig. 6 are recommended
for use in acoustic surveys of fishing grounds. Additionally,
schools of the larvae of this species can be distinguished
from those of other species using the volume back-scattering strength difference method.

123

Frequency
(kHz)

Mean
TSavg (dB)

y=px?q
p

q

r2

SE

38

-94.2

67.3

-125.3

0.93

1.5

120

-79.4

60.1

-107.4

0.92

1.4

38

-94.5

66.2

-125.1

0.92

1.5

120

-79.9

60.1

-107.5

0.92

1.4

Acknowledgments We thank the captains and crew of Kanagasaki
Maru No. 8 for their cooperation in collecting specimens. We also
thank Yukio Ueta, Keisuke Mori Fisheries Research Institute, Tokushima Agriculture and the Forestry and Fisheries Technology Support
Center for their support in collecting specimens. This study was
supported in part by the Fisheries Agency of Japan under the project
‘‘Research and Development Projects for Application in Promoting
New Policy of Agriculture Forestry and Fisheries.’’ We thank this
institution for their support.

References
1. Annual Statistics on Fishery and Aquaculture Production 2006
(2008) Statistics Department, Ministry of Agriculture, Forestry

and Fisheries, Tokyo
2. Saiura K, Takeda Y (2001) Spring fishing ground formation of
anchovy, Engraulis japonica, larvae in 1999 and 2000 in the Kii
Channel (in Japanese). Fish Biol Oceanogr Kuroshio 2:109–118
3. Morioka S (2006) Trial of spring shirasu (Engraulis japonicus
larvae) fishery forecast in Kii Channel by using multiple regression analysis (in Japanese). Fish Biol Oceanogr Kuroshio 7:21–27
4. Mitani I (1988) The biological studies on the larvae of Japanese
anchovy, Engraulis japonica Hottuyn, in Sagami Bay (in Japanese). PhD thesis, University of Hokkaido, Hokkaido
5. Honda S (2004) Abundance estimation of the young cohorts of
the Japanese Pacific population of walleye pollock (Theragra
chalcogramma) by acoustic surveys (In Japanese). PhD thesis,
University of Hokkaido, Hokkaido
6. Yasuma H (2004) Studies on the acoustical biomass estimation of
myctophid fishes (in Japanese). PhD thesis, University of Tokyo,
Tokyo
7. Zhao X, Wang Y, Dai F (2008) Depth-dependent target strength
of anchovy (Engraulis japonicus) measured in situ. ICES J Mar
Sci 65:882–888
8. Foote KG, Traynor JJ (1988) Comparison of walleye pollock
target-strength estimates determined from in situ measurements
and calculations based on swimbladder form. J Acoust Soc Am
83(1):9–17
9. Simmonds J, MacKennan D (2005) Fisheries acoustics, 2nd edn.
Blackwell, Oxford
10. Uotani I (1973) Diurnal changes of gas bladder and behavior of
postlarval anchovy and other related species (in Japanese with
English abstract). Bull Jpn Soc Sci Fish 39:867–876
11. Miyashita K (2003) Diurnal changes in the acoustic-frequency
characteristics of Japanese anchovy (Engraulis japonicus) postlarvae ‘‘shirasu’’ inferred from theoretical scattering models.
ICES J Mar Sci 60:532–537

12. Stanton TK, Chu D, Wiebe PH (1993) Sound scattering by several zooplankton groups. II. Scattering models. J Acoust Soc Am
103:236–253
13. Chu D, Foote KG, Stanton TK (1993) Further analysis of target
strength measurements of Antarctic krill at 38 and 120 kHz:


Fish Sci (2011) 77:161–167

14.

15.
16.
17.
18.

19.

comparison with deformed cylinder model and inference of orientation. J Acoust Soc Am 93:2985–2988
McGehee DE, O’Driscoll RL, Martin Traykovdky LV (1998)
Effects of orientation on acoustic scattering from Antarctic krill
at 120 kHz. Deep-Sea Res 45:1273–1394
Foote KG (1980) Averaging of fish target strength functions.
J Acoust Soc Am 67(2):504–515
Foote KG (1990) Speed of sound in Euphausia superba. J Acoust
Soc Am 87:1405–1408
Greenlaw CF (1977) Backscattering spectra of preserved zooplankton. J Acoust Soc Am 62:44–52
Foote KG (1979) On representing the length dependence of
acoustic target strengths of fish. J Fish Res Board Can 36(12):
1490–1496
Hunter JR (1972) Swimming and feeding behavior of larval

anchovy, Engrauis mordax. Fish Bull 70:821–838

167
20. Batty RS (1984) Development of swimming movements and
musculature of larval herring (Clupea harengus). J Exp Biol
110:217–229
21. Blaxter JHS, Staines ME (1971) Food searching potential in
marine fish larvae. Fourth European marine biological symposium. Cambridge University Press, Cambridge
22. Mikami H, Mukai T, Iida K (2000) Measurements of density and
sound speed contrasts for estimating krill target strength using
theoretical scattering models (in Japanese with English abstract).
Nippon Suisan Gakkaishi 66(4):682–689
23. Matsukura R, Yasuma H, Murase H, Yonezak S, Funamoto T,
Honda S, Miyashita K (2009) Measurement of density contrast
and sound-speed contrast for target strength estimation of Neocalanus copepods (Neocalanus cristatus and Neocalanus plumchrus) in the North Pacific Ocean. Fish Sci 75:1377–1387

123


Fish Sci (2011) 77:169–181
DOI 10.1007/s12562-011-0331-1

ORIGINAL ARTICLE

Fisheries

Codend selectivity for jack mackerel and whitefin jack
and unequal split parameter estimates observed in trouser
trawl experiments
Mohamed Salah Mahjoub • Seiichi Takeda

Toshifumi Hayashi • Daisuke Shiode •
Takafumi Arimoto • Tadashi Tokai

•

Received: 15 September 2010 / Accepted: 27 January 2011 / Published online: 22 February 2011
Ó The Japanese Society of Fisheries Science 2011

Abstract Codend selectivity for the jack mackerel Trachurus japonicus and the whitefin jack Kaiwarinus equula
were evaluated based on data from trouser trawl experiments carried out in the East China Sea, using a test codend
of 60 mm diamond mesh and a control codend made of
minnow net with a square mesh of 9 mm bar length.
Between-haul variations in parameters and the mean
selection curves were tested with the catch data in the
SELECT approach, and then the model of between-haul
variation in the split parameter with the mean selection
curve was chosen as the best fit using Akaike’s information
criterion model selection. The 50% retention lengths and
the selection ranges were 11.4 and 3.36 cm for jack
mackerel and 8.83 and 0.93 cm for whitefin jack, respectively. The selection curve for whitefin jack was sharp,
whereas that for jack mackerel was relatively wide. As the
estimated split parameters indicated, about 80% of the
whitefin jack entered the control codend, but 85 and 90%
of the jack mackerel entered the control codend in the
second and third hauls, respectively. The inequality in
the split parameter is discussed from the viewpoint of the
animal’s swimming behavior and water movement based
on underwater video observations.
Keywords AIC model selection Á Codend selectivity Á
Fish girth Á Jack mackerel Á SELECT Á Swimming ability Á

Trouser trawl Á Underwater video observation Á
Whitefin jack

M. S. Mahjoub Á S. Takeda Á T. Hayashi Á D. Shiode Á
T. Arimoto Á T. Tokai (&)
Tokyo University of Marine Science and Technology,
Minato, Tokyo 108-8477, Japan
e-mail:

Introduction
Fishing gear selectivity plays an important role in the
exploitation of fish stocks and fisheries management. A
trouser trawl that has two codends (one test codend and
another small-mesh control codend) attached to the aft end
has often been utilized to estimate codend selectivity [1–3].
The trouser trawl method has some advantages in experimental fishing operations; for example, compared with the
covered codend method and the twin trawl method, trawl
handling more closely resembles standard commercial
fishing [3]. It is also designed to ensure that similar numbers and size ranges of all species pass down into both
codends, in contrast with the other paired gear tests (the
alternate haul method and the parallel haul method) [3].
However, it was also pointed out that larger fish are caught
more in the test codend than in the control one of a trouser
trawl [1, 3]. Millar and Walsh [4] demonstrated that significantly more fish entered the test codend with a larger
mesh than the control codend in a trouser trawl fishing
experiment, and estimated the split parameter with codend
selectivity parameters by the SELECT method. However,
only a few studies on the cause of this have been
performed.
The jack mackerel Trachurus japonicus is a commercially valuable species in Japan [5]. For example, in 2007,

the annual landing and production of jack mackerel in
Japan were 170,389 tons and 36,721 million yen, respectively [6]. Jack mackerel is caught by trawl as well as purse
seine in Japan, and it is one of the important target species
of the large trawl in the East China Sea. The whitefin jack
Kaiwarinus equula is also caught by the large trawl there,
and both species belong to the Carangidae family. However, the jack mackerel has a spindle-shaped body that is
not as compressed as that of the whitefin jack. According to

123


170

the definition of fish swimming specialization from Videler
and He [7], jack mackerel are more hydrodynamic and can
be considered faster swimmers than whitefin jack. On the
other hand, fish body shape, in particular fish girth, is one
of the factors that affect codend selectivity [8–10]. This
study carried out a trouser trawl with two codends: a test
codend with a nominal mesh size of 60 mm and a control
codend made of fine minnow netting. Based on the data
obtained, it was possible to estimate not only logistic curve
parameters for expressing the mesh selectivity of the codend with the 60 mm mesh, but also the split parameter
indicating the proportions of each species of fish that enter
the test codend. The cause of an unequal split between the
test and control codends in the trouser trawl is also discussed in terms of the animal’s swimming behavior, based
on underwater video observations made in front of the
mouths of the two codends.

Materials and methods

Fishing experiments
A series of trouser trawl fishing experiments were conducted
onboard the research and training vessel Umitaka Maru
(1886 gross tonnage), Tokyo University of Marine Science
and Technology, in the East China Sea in October 2009. A
pair of codends consisting of a test codend of nominal 60 mm
mesh size and a control codend containing a minnow-net bag
were used (Fig. 1). The test codend of 60 mm nominal mesh
size with a diamond mesh had a mesh opening of 54 mm,
which is the current legal minimum mesh size for the large
trawl in the East China Sea, according to an Ordinance of the
Ministry of Agriculture, Forestry and Fisheries of Japan. The
control codend was made of minnow net with a square mesh
of 9 mm bar length to avoid clogging, and was covered by a
strengthening bag with a diamond mesh net with a 28 mm
mesh opening in order to stop the minnow net from bursting.
The trawl net used had a total net length of 39 m and a net
mouth with a 28 m headrope and a 37.2 m groundrope. Two
underwater video system units (DCR-SR100 video camera,
Sony Corp.; blimp-shaped underwater housing, Goto
Aquatics Corp. [11]) were installed facing each other on the
upper panel ahead of the codends to observe the behavior of
animals in front of the codend mouths (Fig. 1). The video
camera shot downward at about 20°. Two flashlights
(Excursion LS24, Tektite Industries Inc.) were attached on
both sides of video system #1. The auto-iris video lens of
system #2 was illuminated with excessively bright light from
the two flashlights, so the video feed from system #2 was out
of focus and too bright to be able to observe the animal’s
behavior. Accordingly, video from system #1 was utilized in

this study.

123

Fish Sci (2011) 77:169–181

Three hauls were performed at fishing grounds at a
depth of about 120 m on 11 October 2009. The water
temperatures at the grounds were 17.4°C near the bottom
and 26.4°C at 5 m depth according to a conductivity–
temperature–depth (CTD) device. The warp length was
400 m. After the warp was let out, the net was towed for
30 min at a vessel speed of approximately 3.5 knots,
before it was hauled at a winch speed of 0.44 m/s at a
vessel speed of \2 knots.
Catch numbers (the number of individuals caught) in the
test and control codends were recorded for each species,
apart from the three-spot swimming crab Ovalipes punctatus, which is usually dominant in the catch in the fishing
grounds at around 120 m depth. The whole catch of the
swimming crab was measured in trays when subsampling,
and one tray was randomly chosen for carapace width
measurements. Lengths of all fish in the codends were
measured to the nearest millimeter. Fork lengths were
measured for jack mackerel, and total lengths for whitefin
jack and other fishes. Fish length data were grouped into
0.5 cm intervals for further selectivity analysis. Fish girth
was also measured to the nearest millimeter to examine the
relationship between the mesh size and fish girth at a certain retention probability.
Data analysis
Differences in catch number between the test and control

codends were statistically tested for significance by the chisquared test for several of the main species in each haul.
Length distributions of jack mackerel and whitefin jack
were analyzed with the SELECT method [4, 12, 13] in
order to estimate the selectivity of the test codend, because
there were sufficient data to allow this estimate. For the
dominant species, except for jack mackerel and whitefin
jack, there were insufficient data to permit selectivity
estimation, so differences in length distributions for the test
and control codends were statistically tested with the
Kolmogorov–Smirnov test.
The SELECT approach [4, 12, 13] was applied to
estimate mesh selectivity parameters from the data
obtained using the paired gear test, including the trouser
trawl test. For a fish length of lj in the ith haul, the proportion /ij of the catch in the 60 mm mesh test codend
NLij compared to the total catch (the catch in the test
codend NLij and the catch in the control codend NSij) is
defined as follows:
/ij ¼

NLij
:
NLij þ NSij

ð1Þ

The probability of a fish of length l being retained in the
test codend is usually expressed with the logistic function
r(l), which has the parameters a and b [3]:

Fish Sci (2011) 77:169–181

171

Fig. 1 Schematic diagram of
the trouser codends of the trawl
used in the fishing experiments
and the locations of the two
underwater video systems

Test codend:
54 mm mesh opening diamond mesh

5m
29

31

59

62

Video
camera #2

Video
camera #1

200

700

35

87

Strengthening bag:
28 mm mesh opening diamond mesh

The probability that a fish enters the test codend rather
than the control codend is denoted by the split parameter p,
and the probability that it enters the control codend is
1 - p. The proportion of the catch /(l) is expressed as
/ðlÞ ¼

p expða þ blÞ
:
ð1 À pÞ þ expða þ blÞ

ð3Þ

Using the logistic parameter estimates of a and b, the
length at which 50% of the fish are retained l50 and the
selection range S.R. can be calculated with the following
two equations:
l50 ¼ Àa=b

ð4Þ

S:R: ¼ l75 À l25 ¼ 2 ln 3=b:

ð5Þ

It is widely accepted that there are between-haul
variations in selectivity parameters [14, 15] and in split
parameters [16]. In this study, we assumed three models for

s

5m

200
bars

9 mm bar length square mesh

ð2Þ

ba r

0
10 ars
b

Control codend of minnow net:

expða þ blÞ
:
1 þ expða þ blÞ

s

200
bars

5m

rðlÞ ¼

bar

70

74

Eq. 1 in order to examine variations in the logistic and split
parameters between hauls, as follows.
Model with between-haul variations in parameters
(BHV model)
The logistic parameters for codend selectivity and the split
parameters were estimated for each haul. Thus, the proportion of the catch /i(l) is defined as
/i ðlÞ ¼

pi expðai þ bi lÞ
;
ð1 À pi Þ þ expðai þ bi lÞ

ð6Þ

and the parameters were determined by maximizing the

following natural log-likelihood function:
X
Li ðai ; bi ; pi Þ
i

¼

XXÂ
À
ÁÃ
NLij ln /i ðlj Þ þ NSij ln 1 À /i ðlj Þ :
i

ð7Þ

j

When every term of Li(ai, bi, pi) is maximized, the value
of the total maximum log-likelihood is also maximized.

123


172

Fish Sci (2011) 77:169–181

The number of parameters to be estimated was three times
the number of hauls in the experiment.
Model with between-haul variations in the split parameters

with mean selection curve (BHV&MSC model)
In this model, it is assumed that the logistic parameters for
codend selectivity were common for the three hauls, but
the split parameters pi varied among the three hauls. The
proportion of Eq. 1 was expressed as follows:
/i ðlÞ ¼

pi expða þ blÞ
:
ð1 À pi Þ þ expða þ blÞ

ð8Þ

The following log-likelihood was then maximized:
XXÂ
À
ÁÃ
Lða; b; pi Þ ¼
NLij ln /i ðlj Þ þ NSij 1 À /i ðlj Þ :
i

Results

j

ð9Þ
The parameters to be determined in this model were the
two logistic parameters a and b for the mean selection
curve of codend selectivity, and the split parameters pi of
each haul.

Model with no between-haul variations in parameters
(NBHV model)
This model assumed that both of the logistic parameters for
codend selectivity and the split parameters were common
for the three hauls. Only three parameters were then estimated, a, b, and p:
/ðlÞ ¼

p expða þ blÞ
:
ð1 À pÞ þ expða þ blÞ

ð10Þ

The following log-likelihood function was maximized:
XXÂ
À
ÁÃ
Lða;b; pÞ ¼
NLij ln /ðlj Þ þ NSij 1 À /ðlj Þ :
ð11Þ
i

Solver in Excel (Microsoft, Redmond, WA, USA) was
implemented to maximize the functions of Eqs. 7, 9, and
11 [17, 18]. Akaike’s information criterion (AIC) was
employed for model selection [19]. The full model is
usually defined as the model that completely coincides with
the data, and in this study it was the proportions /ij
calculated for each length at each haul in Eq. 1. When the
full model has a maximum likelihood value, the number of

estimated parameters /ij is also maximized. The AIC of
each model was compared with that of the full model to
test the goodness of fit of the model to the data, instead of
using a likelihood ratio test.

Catch numbers and length distributions
of dominant species
In the three hauls, eight species were dominant in terms of
catch number: jack mackerel, whitefin jack, John dory Zeus
faber, spiny red gurnard Chelidonichthys spinosus, lesserspotted leatherjacket Thamnaconus hypargyreus, swordtip
squid Loligo edulis, three-spot swimming crab, and
pandalid shrimp Plesionika grandis (Table 1). Statistically
significant differences in catch number between the test
and control codends (v2 test, P \ 0.05) were found for the
14 pairs out of the 24 pairs in total (eight species and three
hauls). Among the other ten pairs, it was unclear whether
differences were significant for jack mackerel and swordtip
squid in the first haul, while the other eight species had
catch numbers of \25. For whitefin jack and three-spot
swimming crab, the catch number of the test codend was
statistically significantly larger than that of the control
codend in each of the three hauls. Similarly, for John dory,
spiny red gurnard, and lesser-spotted leatherjacket, there

j

Table 1 Catch numbers for each of the main species caught in each haul
Common name

Species name

Haul number
#1
Test

#2
Control

Test

#3
Control

Test

Control

John dory

Zeus faber

27**

7

7

5

8

4

Spiny red gurnard

Chelidonichthys spinosus

4

2

12

6

21*

8

Jack mackerel

Trachurus japonicus

21

27

102

814**

24

367**

Whitefin jack

Kaiwarinus equula

94**

18

147**

42

215**

108

Lesser-spotted leatherjacket

Thamnaconus hypargyreus

52**

22

24*

11

6

8

Swordtip squid

Loligo edulis

19

29

9

13

20*

7

Three-spot swimming crab

Ovalipes punctatus

1177**

792

1140**

661

949**

571

Pandalid shrimp

Plesionika grandis

23

314**

0

2

0

0

2

v test; * P \ 0.05; ** P \ 0.001

123

Fish Sci (2011) 77:169–181

173

was more catch in the test codend in almost all hauls. In
contrast, significantly larger catch numbers were found
in the control codend for jack mackerel in the second
and third hauls and for the pandalid shrimp in the first
haul.
In the subsample for the three-spot swimming crab,
crabs with carapace widths of less than 7 cm were more
abundant in the test codend sample than in the control
codend sample at length classes of \7 cm, although the
carapace width ranges of the crab were similar to each
other: from 4.0 to 10.5 cm in the test codend and from 4.0
to 9.5 cm for the control (Fig. 2). No statistical difference
was observed in the length distributions of the crab
between the test and control codends (Kolmogorov–
Smirnov test, P [ 0.10). There was also no statistical
difference in length distributions of John dory, spiny red
gurnard, and lesser-spotted leatherjacket between the test
and control codends (Kolmogorov–Smirnov test,
P [ 0.10). For instance, the total length of lesser-spotted
leatherjacket varied from 9 to 14 cm in the test codend
and from 9 to 13.5 cm in the control. The bodies of
pandalid shrimps with carapace lengths of around 2 cm
were small compared with the 54 mm mesh opening of
the test codend, suggesting that almost all of the shrimps

entering the test codend escaped through the mesh, while
shrimps of this size were retained in the control codend
(9 mm bar length). For swordtip squid of around 5 cm in

Fig. 2 Length distributions of the six
dominant species other than the jack
mackerel Trachurus japonicus and the
whitefin jack Kaiwarinus equula
caught in the test and control codends.
N total number of individuals caught in
the test and control codends

100

Ovalipes punctatus

80

Haul # 1
N =1969

60

mantle length in the first haul, a lower catch number was
observed in the test codend than in the control one, and
this also implied that squid of this size escape through the
mesh of the test codend; in other words there is mesh
selection for the test codend with the larger mesh. However, the catch number of the squid was not enough to
estimate the selectivity curve in these three hauls. The
species for which the three hauls provided enough data to

estimate selectivity curves were jack mackerel and
whitefin jack in this fishing experiment.
Model selection in SELECT for jack mackerel
and whitefin jack
The catch number for jack mackerel in the first haul was
less than in the two others (Table 1). The fishes caught in
the control codend ranged between 6.0 and 22.5 cm in fork
length, whereas those in the test codend ranged from 10.5
to 22.5 cm (Fig. 3). For jack mackerel with fork lengths
that were larger than 11 cm, the length distributions of both
codends were similar to each other in the three hauls,
although the catch number in the control codend was much
larger than that in the 60 mm test codend. In the test
codend, there was no catch of jack mackerel with a fork
length of below 11 cm, and hence the plot of the proportion
/l of the catch number in the 60 mm-mesh test codend
against fork length was almost zero (Fig. 4). As fork length

Zeus faber

20

Haul # 1
N =34

10

0

5 10 15 20 25 30 35 40

Total length (cm)

20

Catch number

Test

0

40

0
100

Thamnaconus
hypargyreus
Haul # 1
N =74

10
0

20

Haul # 2
N =35

10

Loligo edulis

20

Haul # 2
N =1801

80

Haul # 1
N =48

10

60

0

40

0

5

0
100
80

0

0

5

10 15 20 25

Total length (cm)

10 15 20 25

Mantle length (cm)

20

240

Plesionika grandis

200

Haul # 3
N =1520

60
40

160

20

Chelidonichthys
spinosus

10

Haul # 3
N =29

20
0

Control

20

0

5

10

15

Carapace width (cm)

0

0

5 10 15 20 25 30 35

Total length (cm)

Haul # 1
N =337

120
80
40
0

0

5

10

Carapace length (cm)

123


174

Fish Sci (2011) 77:169–181

Fig. 3 Length distributions of
the jack mackerel Trachurus
japonicus and the whitefin jack

Kaiwarinus equula caught in the
test and control codends. N total
number of individuals caught in
the test and control codends

Trachurus japonicus

Haul # 1
N = 48

5

Kaiwarinus equula

20

10

Haul # 1
N = 112

10

0

160

0

Haul # 2

N = 916

140
120

40

Haul # 2
N = 189

Catch number

100
80
60

20

40
20
0

0
80

Haul # 3
N = 381

60

Haul # 3
N = 323

60

Control
Test

Control
Test

40
40

20

20

0

0

0

5

10

15

20

0

25

5

10

15

20

Total length (cm)

Fork length (cm)

Deviance residual

Proportion of catch in
the test codend

1
0.8
0.6
Haul # 1
N = 48

0.4

0.4

0.2
0

5

10

15

20

25

BHV
model
D = 13.70
BHV&MSC
model
D = 13.88
NBHV
model
D = 44.60
0

4.75

9.75

14.75

19.75

24.75

0.2

0

0
0

2
1
0
-1
-2
2
1
0
-1
-2
2
1
0
-1
-2

Haul # 3
N = 381

0.4

0.2

0
2
1
0
-1
-2
2
1
0
-1
-2
2
1
0
-1
-2

0.6

Haul # 2
N = 916

5

10

15

20

2
1
0
-1
-2
2
1
0
-1
-2
2
1
0
-1
-2

BHV
model
D = 12.92
BHV&MSC
model
D = 13.91
NBHV

model
D = 16.00
0

4.75

0

25

9.75 14.75 19.75
Fork length (cm)

24.75

5

10

15

9.75

14.75

20

25

19.75

24.75

BHV
model
D = 8.74
BHV&MSC
model
D = 12.15
NBHV
model
D = 18.61
0

4.75

Fig. 4 Proportion /ij of jack mackerel Trachurus japonicus caught in
the test codend compared to the total catch, along with the estimated
curve /(l) and the deviance residuals for each model, versus fish

length l. Dash-dot lines BHV model, solid lines BHV&MSC model,
dotted lines NBHV model, N number of fish caught, and D model
deviance

increased above 11 cm, the proportion /l rose and seemed
to reach a constant value (Fig. 4). No lack of fitness was
found, because all three models had smaller AIC values

than the full model that completely coincides with the data
(Table 2). However, the curves of the NBHV model

showed poor fits to the plots and large biases in the

123


Fish Sci (2011) 77:169–181

175

deviance residuals, especially for the first and third hauls
(Fig. 4). The estimated /(l) curves for the BHV model and
the BHV&MSC model showed good fits to the plots, and
their deviance residuals did not show any specific bias
(Fig. 4). The BHV model seemed to have smaller sum of
deviance residuals (Fig. 4), but it also had a larger number
of parameters than the BHV&MSC model (Table 2). The
AIC value of the BHV&MSC model was the smallest
(140.6) for jack mackerel (Table 2). The BHV&MSC
model was therefore the best. This means that there were
still variations in the split parameter among the three hauls,
while there were no differences in codend selectivity
among the three hauls (Table 2).
The catch number of whitefin jack caught in the 60 mmmesh test codend was larger than that in the control one,
whereas the control codend had a larger catch of jack
mackerel than the test one, as stated above (Table 1). The
total length ranged between 3.5 and 21 cm in the control
codend and between 8 and 22.5 cm in the test codend
(Fig. 3). Fish with total lengths that were smaller than 8 cm
were only caught in the control codend, so the plot of the
proportion of the catch number in the 60 mm-mesh test

codend was zero for total lengths below 8 cm (Fig. 5). In the
second and third hauls, the proportion /l increased for total
lengths of 8–10 cm, and was constant at lengths [10 cm.
Aside from the BHV model curve in the first haul in the
three models (Fig. 5), the estimated curves /(l) showed
good fits to the plots because their AIC values were smaller
than that of the full model, implying no lack of fitness
(Table 2), and no specific bias was found in the plot of
deviance residuals versus total length (Fig. 5). However,
in the first haul, the BHV model produced a curve that started
to increase from a total length of 2 cm, but the curve seemed
to be inappropriate because of a lack of data in the initial
part of the length range. When comparing the AIC values

among the models, the BHV&MSC model gave better
estimates, with an AIC value of 187.8 (smaller than the
other models) for whitefin jack and for jack mackerel
(Table 2).
Selection curves for jack mackerel and whitefin jack
The selection curves of the 60 mm mesh test codend for
jack mackerel and whitefin jack based on the parameters
estimated in the BHV&MSC model are shown in Fig. 6.
For jack mackerel, the 50% retention length l50 and the
selection range S.R. of the 60 mm-mesh test codend were
11.4 and 3.36 cm, respectively. The retention probability
increased from almost 0 to close to 1 within the fork
length range from 5 to 18 cm (Fig. 3). The values of the
split parameter were 0.53, 0.15 and 0.09 for the three
hauls (Table 3). This implies that most jack mackerel
entered the control codend in the second and third hauls,

during which large numbers of jack mackerel were caught
in the control codend.
The selection parameters for whitefin jack were a length
corresponding to 50% retention of 8.83 cm and selection
range of 0.93 for the 60 mm mesh test codend. The
selection curve indicated a steep increase in retention
probability from 7 to 11 cm in total length (Fig. 5). The
values of the split parameter for whitefin jack were 0.84,
0.81, and 0.76 for the three hauls (Table 3), which means
that whitefin jack entered the test codend more, in contrast
to jack mackerel.
Relationships between fish length and girth
There were linear relationships between fish length and
girth for both species (Fig. 7), and the regression lines were
estimated as follows:

Table 2 AIC-based model selection for jack mackerel and whitefin jack
Species

Model

Maximum
log-likelihood

Total number
of parameters

Total AIC

Jack mackerel Trachurus japonicus

Full model

-45.31

61

212.6

Whitefin jack Kaiwarinus equula

BHV model

-62.99

9

144.0

BHV&MSC model

-65.29

5

140.6a

NBHV model

-84.92

3

175.8

Full model

-39.45

62

202.9

BHV model

-87.11

9

192.2

BHV&MSC model

-88.92

5

187.8a

NBHV model

-91.09

3

188.2

Full model: model where the proportion values for all length classes are estimated
BHV model: model with between-haul variations in parameters
BHV&MSC model: model with between-haul variations in split parameters with the mean selection curve
NBHV model: model with no between-haul variations in parameters
a

The smallest AIC

123


176

Fish Sci (2011) 77:169–181

Deviance residual

Proportion of catch in
the test codend

1

1

Haul # 1
N = 112

0.8
0.6

1

Haul # 2
0.8 N = 189

Haul # 3
0.8 N = 323

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0

0

0

5

10

20

15

25

0

5

10

15

20

25

0

2
1 BHV
model
0
-1
-2 D = 20.10

2
1 BHV
model
0
-1
-2 D = 33.28

0
5
2
1 BHV
model
0
-1
-2 D = 41.91

2
1 BHV&MSC
model
0
-1
D = 20.78

-2

2
BHV&MSC
1
model
0
-1
D = 35.42
-2

2
BHV&MSC
1
model
0
-1
D = 42.74
-2

2
NBHV
1
model
0
-1
-2 D = 22.71
4.75
0

2
NBHV
1
model
0
-1
-2 D = 35.13
4.75
0
9.75 14.75
Total length (cm)

9.75

14.75

19.75

24.75

Fig. 5 Proportion /ij of whitefin jack Kaiwarinus equula caught in
the test codend compared to the total catch, along with the estimated
curve /(l) and the deviance residuals for each model, versus fish

1
0.8

r(l) =

0

0

15

Fork length (cm)

for jack mackerel

ð12Þ

g ¼ 0:814lT þ 0:455

for whitefin jack:

ð13Þ

Here, g denotes fish girth, lF fork length and lT total
length in cm.
In the present study, the fork length was considered for
jack mackerel, but the total length is also often used. The
regression line of the total length lT to fork length lF was
hence estimated to be
À
Á
lT ¼ 1:0624lF þ 0:833 R2 ¼ 0:99; N ¼ 61
ð14Þ
Behavior of animals in front of the codends
Underwater video footage taken in front of the codend
mouths was successfully recorded by video camera #1 in

123

S.R. = 0.93

r(l) =

0.2

S.R. = 3.36

g ¼ 0:579lF À 0:640

24.75

0.4

l50 = 11.39

10

19.75

l50 = 8.83

0.8
0.6

5

14.75

25

1

0.6

0.2

9.75

20

Kaiwarinus equula

exp(−7.449 + 0.653l)
1+ exp(– 7.449 + 0.653l )

0.4

15

length l. Dash-dot lines BHV model, solid lines BHV&MSC model,
dotted lines NBHV model, N number of fish caught, and D model
deviance

Trachurus japonicus

Retention probability

Fig. 6 Codend selection curves
of the 60 mm mesh test codend
for the jack mackerel Trachurus
japonicus and the whitefin jack
Kaiwarinus equula

19.75

2
NBHV
1
model
0
-1
-2 D = 45.43
24.75
0
4.75

10

20

0

0

5

10

exp(– 20.754 + 2.350l )
1 + exp(–20.350 + 2.350l)

15

20

Total length (cm)

the first and third hauls, and the durations of underwater
video recording from shooting the net to hauling up onto
the surface of the sea were 77 and 71 min in the first haul
and third hauls, respectively (Fig. 8). The first sight of a
sand cloud or caught animals in the video footage occurred
about 20 min after shooting the net—almost at the same
time as the warp out was finished. The area that the two
flashlights illuminated in the net was too narrow to allow
the identification to species of all of the animals passing the
video camera when towing at a depth of 120 m. However,
some animals were identified under the illumination as
John dory, jack mackerel, whitefin jack, lesser-spotted
leatherjacket, blunthead puffer Sphoeroides pachygaster,
three-spot swimming crab, and so on. Apart from jack
mackerel, almost all animals that came into the space ahead


Fish Sci (2011) 77:169–181

177

Table 3 Estimates of the mean selection curve parameters and the
split parameters of each haul as well as their standard errors in the
BHV&MSC model for jack mackerel and whitefin jack
Parameters

Jack mackerel

Whitefin jack

-7.45 (4.89)

-20.8 (5.17)

Logistic parameters
a
b
Split parameters

0.654 (0.469)

2.35 (0.628)

p1

0.534 (0.0902)

0.848 (0.0338)

p2

0.147 (0.0331)

0.813 (0.0291)

p3

0.093 (0.0305)

0.762 (0.0296)

Selection parameters
l50
S.R.

11.39 (1.05)

8.83 (0.236)

3.36 (2.410)

0.93 (0.250)

Values in parentheses are standard errors

of the codends passed into either codend without any
struggle to swim, while a few animals attempted to swim in
a random direction and then fell into either codend. The
fish that entered the codend were spun in the water flow.
This indicates that turbulence and eddies occurred ahead of

the accumulated catch in the codends.
It was also observed that the bottom net of the test codend showed flapping motions more frequently than the
inner bag net of the control codend while towing, which
implies that more water entered the test codend with the
larger mesh than the control codend with the 9 mm bar
length square mesh covered by a strengthening bag with a
28 mm diamond mesh opening. According to the video
observations, more animals appeared to pass into the test
codend. The only species among the eight dominant species that was countable on the video image was the threespot swimming crab, because of its large and characteristic
body shape. Even during hauling, three-spot swimming
crabs that were caught in the belly and the wing of the net
still dropped into the codends. When the crabs were
countable in the first and third hauls, the numbers of threespot swimming crabs passing into the test codend and the

Discussion
In most of the hauls, aside from the pandalid shrimp and
the jack mackerel, the other dominant species were caught
more in the test codend than in the control codend

Trachurus japonicus
10

Kaiwarinus equula

g = 0.5797 lF - 0.6404
R2 = 0.8547
Girt h (cm)

Girth (cm)

Fig. 7 Linear regression lines
between fish length (total length
lT and fork length lF) and girth
for the jack mackerel Trachurus
japonicus and the whitefin jack
Kaiwarinus equula. N number
of individuals measured

control one in the video image were counted as 374 and
174, respectively. Although the number of crabs in the
video image was at most one-sixth of the total number
caught, the ratio of the number of crabs entering the test
codend to that of the control was [2, which was similar to
the ratio of the crab catch number in the test codend to that
in the control codend (Table 1).
Extreme bias was observed in the split parameter for
jack mackerel (Table 3), which suggested that most jack
mackerel entered the control codend. During towing at the
bottom, a fish school was observed to keep swimming in
front of the mouths of the codends; these fish were identified as jack mackerel under the brighter illumination
when hauling (Fig. 8). In the net, the jack mackerel always
swam in the towing direction with rhythmic tail beats to
maintain their position. Some of the jack mackerel in the
school became caught up with the codends during towing,
but they still swam in the towing direction. When the
towing speed dropped while hauling, the school of jack
mackerel swam forward through the codend mouth, and
individual jack mackerels were dispersed uniformly in
front of the codends in the third haul, during which more
jack mackerels were caught. Small jack mackerels gradually accumulated in the control codend, as they were likely

to swim in the relatively calm water in the control codend.
Large jack mackerels kept swimming in front of the codend
mouths until the net was hauled up to the sea surface
(Fig. 8). In the third haul, the net twisted just before the
codends were hauled over the sea surface, which caused all
of the jack mackerel swimming in front of the codends to
enter the control codend while the test codend mouth
looked completely closed.

N = 445

8

6

14

g = 0.8138 lT + 0.4547
R2 = 0.9422

11

N = 466

8

Test

Test

Control

4

9

11

13

15

Fork length (cm)

17

Control

5

7

9

11

13

15

17

Total length (cm)

123


178

Fig. 8 Photos taken from video camera #1 of animal behavior in
front of the two codends, the test codend (left, starboard side) and the
control codend (right, portside). Top: a three-spot swimming crab is
flushed into the test codend during hauling in the first haul; middle: a
school of jack mackerel swimming in front of the two codend mouths
during hauling; bottom: the school of jack mackerel swimming there
just before the net was hauled up in the third haul

123

Fish Sci (2011) 77:169–181

(Table 1). From the underwater video observations, more
animals passed into the test codend due to the slightly
increased water flow into the test codend; Pope et al. [1]
pointed out that this is one of the causes of the inequality in
catch number between the test and control codends in
trouser trawls. The control codend used in this study was
made of 9 mm bar length square mesh net covered by a
strengthening bag with a 28 mm diamond mesh opening.
Such double bag nets with a smaller mesh reduce the rate

that the inner water can flow out through the mesh in the
control codend compared with a test codend with a larger
mesh. Thus, more water enters the test codend, bringing
more fish with low swimming abilities into the test codend.
Since the pandalid shrimps also could not swim away from
the oncoming codends, more shrimps pass into the test
codend. However, most of the pandalid shrimps were
caught in the control codend, with only a few being caught
in the test codend (Table 3). Even if the same number of
pandalid shrimps entered the two codends, almost all of the
shrimps escaped from the test codend through the larger
mesh, in contrast to the shrimps in the control, as the
shrimp bodies were small compared to the mesh opening of
the test codend.
The estimated value of the split parameter for whitefin
jack was about 0.8 (Table 3), which means that about 80%
of the fish entered the test codend. Jack mackerel and
whitefin jack belong to the Carangidae family, but
according to the definition of fish swimming specialization
from Videler and He [6], the jack mackerel is a typical
high-speed cruising fish, whereas whitefin jack is a swimming generalist with a compressed body shape. For jack
mackerel, especially in the second and third hauls, the
catch numbers in the control codend were dominant
(Table 1), and the split parameter values indicated that the
majority of the jack mackerel passed into the control
codend. As stated above, jack mackerel was the only
species that kept swimming inside the net in the same
direction as the towing direction, and the jack mackerel
that swam in front of the codend mouths entered the
opened mouth of the control codend when the codend was

hauled up to the sea surface. Several studies on the
swimming speed of the jack mackerel Trachurus japonicus
have been carried out [20–23]. It was reported that the
maximum swimming speed of the jack mackerel is 10
times the fish length per second [20], and the sustained
swimming speed obtained with the red muscle changed to a
burst speed obtained with the white muscle at a swimming
speed of 5.0–7.9 fork lengths per second [21, 22]. Nofrizal
et al. [23] demonstrated that the maximum sustained
swimming speed and the burst speed at water temperatures
of 10, 20 and 22°C were 2.4–3.4 fork lengths per second
and 8–10.3 fork lengths per second, respectively. The water
temperatures in °C were 17.4 at the bottom and 26.4 at the


Fish Sci (2011) 77:169–181

surface in the fishing grounds, and there were two modes in
the fork length distributions of jack mackerel caught in this
study (12.5 and 15 cm; see Fig. 3). The burst speeds of
jack mackerel with fork lengths of 12.5 and 15.5 cm were
estimated to be 1.25 and 1.55 m/s, respectively, which
were slower than the towing speed of 3.5 knots (1.8 m/s).
This explains why the jack mackerel kept swimming once
they entered either codend during towing. However, from
the underwater video observations, the jack mackerel still
swam in either codend during towing. The fish could swim
forward out of the codend while the net was being hauled at
the slower towing speed of \2 knots (1.0 m/s), and they
kept swimming in front of the codend for at least 25 min

until the net was hauled up to the sea surface. The towing
speed of 2 knots during hauling was assumed to be within
the speed range for jack mackerel with fork lengths of 12.5
and 15.5 cm that allows them to keep swimming for prolonged periods; the previous study indicated that the jack
mackerel can keep swimming for around 15–20 min at
such speeds [23]. This implies a potential energy-saving
strategy, where the towing speed can be set in anticipation
of the fish gradually falling back into the codend due to
fatigue after swimming for a prolonged period at this speed
[24]. According to the water flow speed around a codend
model, as measured in a circulator channel [25], the flow
speed inside the codend is slightly slower than the main
flow outside the codend. In addition, Winger et al. [26]
pointed out that there would be turbulence and eddies close
to the netting and ahead of the accumulated catch clogging
the mesh, which allows the fish to keep swimming. These
studies also support the idea that the jack mackerel can
keep swimming for longer than 30 min in a net moving at a
speed of 3.5 knots. The two flashlights attached to video
camera system #1 may have affected fish behavior, especially that of jack mackerel. As indicated by Walsh and
Hickey [27], in the presence of strong light, the fish behave
in a different way from their usual ordered pattern of
behavior. Inoue et al. [28] also confirmed that fish change
their swimming patterns due to unusual lighting conditions.
Arimoto et al. [29] also reported that the vision of jack
mackerel is influenced by ambient light as well as artificial
light. In the present study, for animals other than jack
mackerel, the effect of the light intensity on the swimming
behavior was limited because they entered either codend
without exhibiting any struggle to swim. In contrast, jack

mackerel swimming in the towing direction when the net
was being towed close to the bottom can be attributed to
the effect of the flashlights on the fish [29].
The mesh opening of the 60 mm mesh test codend was
54 mm, so the mesh perimeter was approximately 10.8 cm,
twice that of the mesh opening. The fish length corresponding to a girth of 10.8 cm was about 19.8 cm in fork
length for jack mackerel and around 12.7 cm in total length

179

for whitefin jack. The fish lengths corresponding to 95%
retention were estimated to be 15.9 cm in fork length for
jack mackerel and 10.1 cm in total length for whitefin jack
from the selection curve equations. These results indicate
that fish with girths larger than the mesh perimeter would
not be able to pass through the mesh. According to previous studies on the retention probabilities of trawl codends
in terms of the value of girth divided by the mesh perimeter, g/p, the retention probability is generally observed to
increase from 0 and to reach close to 1 for g/p values of
0.5–1.0 [30–32]. For whitefin jack, the fish length corresponding to 5% retention was 7.58 cm in total length,
which corresponded to a g/p value of 0.61 (i.e., larger than
0.5). This means that the selection curve for whitefin jack
was relatively sharp. In contrast, jack mackerel, which had
a girth that was half of the mesh perimeter (i.e., g/p = 0.5),
was 10.4 cm in fork length. Even though it was small
enough to escape through the mesh, the retention probability for jack mackerel that was 10.4 cm in fork length
was still 0.34, indicating a wide selection range of the
selection curve for jack mackerel. This suggests that there
was an insufficient sieving effect of the mesh on jack
mackerel with girths that were small enough to pass
through the mesh in this study. Previous studies [33–39]

have reported that the selection factor of diamond mesh
codend for Trachurus fish varies from 2.0 to 3.43, and that
of square mesh codend varies from 3.4 to 4.3, based on the
measurement of the total length of the fish (Table 4). The
l50 in fork length for the jack mackerel Trachurus japonicus estimated in this study corresponded to the l50 in total
length of 12.95 cm with the regression equation in Eq. 14;
the selection factor, defined as the fish length corresponding to 50% retention (mm) divided by the mesh opening
(mm), was then 2.40 (Table 4). The selection factor for
jack mackerel fell within the range from 2.0 to 3.43, but
was relatively small. According to the underwater video
observations made in this study, whitefin jack passed into
the codend without struggling to swim, and hence had more
time to encounter the mesh of the codend. In contrast, jack
mackerel kept swimming in the net until the net was hauled
up to the sea surface, which may explain why some jack
mackerel with small bodies were retained in the test codend
without passing through the mesh. The total duration of
55 min (30 min towing and 25 min hauling) for this study
was shorter than the usual duration of commercial fishing.
During the longer towing durations of commercial trawlers
at towing speeds of over 3.5 knots, jack mackerel cannot
keep swimming, so the fish would probably come into
contact with the mesh more times and hence show a
sharper selection curve.
In this study, the BHV&MSC model was found to be the
best one, and it provided reasonable selectivity parameter values. It suggested the possibility of between-haul

123

180

Fish Sci (2011) 77:169–181

Table 4 Comparison of the codend selectivity parameters for Trachurus fish
Species

Atlantic horse mackerel
Trachurus trachurus

Reference

Campos and Fonseca [33]

Campos et al. [34]

Mediterranean horse mackerel
Trachurus mediterraneus

Selection parameters in total length (cm)
l50inTL

SRinTL

SF

DM 64

14.4

3.3

2.3

DM 69

14.7

2.9

2.1

DM 79

16.0

3.7

2.0

SM 63

21.9

8.3

3.5

DM 55

18.0

3.8

3.3

DM 60

19.8

3.6

3.3

DM 71
SM 55

21.9
21.7

4.9
5.0

3.1
3.9

Campos et al. [35]
Tosunog˘lu et al. [36]

SM 63

27.3

3.4

4.3

DM 50

15.6

5.5

3.15

Aydın and Tosunog˘lu [37]

DM 44

14.7

4.6

3.29

SM 42

15.9

5.6

3.75

DM 40

13.70

2.10

3.43

SM 40

15.20

3.00

3.80

DM 39

9.71

2.75

2.51

SM 40

13.12

2.43

3.40

DM 54

12.95

3.57

2.40

Ordines et al. [38]
Sala et al. [39]

Jack mackerel Trachurus japonicus

Mesh shape and mesh
opening (mm)

Present study

DM diamond mesh, SM square mesh, l50inTL total length corresponding to 50% retention, SR selection range in total length, SF selection
factor = l50inTL/mesh opening

variations in split parameters. Usually, estimating the
parameters for fishing gear selectivity requires a large
sample size. When sample sizes obtained from a single
haul are inadequate, data pooling is often considered (e.g.,

in a covered codend experiment). However, in a trouser
trawl experiment, simple data pooling would result in
considerable bias in estimated selectivity parameters with
large variations in the values of split parameters between
hauls. Even in such a case, the BHV&MSC model proposed in this study is useful for estimating the parameters
of the mean selectivity curve as well as the split parameters
for each haul.
Acknowledgments The authors would like to thank the former
Captain Yuji Mine and the crew of RTV Umitaka-maru as well as the
students from the Tokyo University of Marine Science and Technology who participated in the trawl experiments for their invaluable
support. The authors are also grateful to two anonymous reviewers for
offering excellent suggestions.

References
1. Pope JA, Margetts AR, Hamley JM, Akyuz EF (1975) Manual of
methods for fish stock assessment, part III. Selectivity of fishing
gear (FAO Fisheries Tech Paper 41). FAO, Rome
2. Broadhurst MK, Kennelly SJ (1995) A trouser-trawl experiment
to assess codends that exclude juvenile mulloway (Argyrosomus
hololepidotus) in the Hawkesbury River prawn-trawl fishery. Mar
Freshw Res 46:953–958

123

3. Wilemann DA, Ferro RST, Fonteyne R, Millar RB (1996)
Manual of methods of measuring the selectivity of towed fishing
gears. ICES Coop Res Rep 125:1–126
4. Millar RB, Walsh SJ (1992) Analysis of trawl selectivity studies
with an application to trouser trawls. Fish Res 13:205–220
5. Hara I, Tokai T (2004) Spawning ground and recruitment process

of jack mackerel—from the East China Sea to the coastal waters
of Japan. Kouseisya-Kouseikaku, Tokyo (in Japanese)
6. Statistics Department, Minister’s Secretariat of the Ministry of
Agriculture, Forestry and Fisheries (2010) Annual report of
fisheries and aquaculture statistics. Association of Agriculture &
Forestry Statistics, Tokyo (in Japanese)
7. Videler JJ, He P (2010) Swimming in marine fish. In: He P (ed)
Behavior of marine fishes—capture processes and conservation
challenges. Wiley-Blackwell, Ames, pp 5–23
8. Margretts AR (1954) The length–girth relationship in haddock
and whiting and their application to mesh selection. J Cons Int
Explor Mer 20:56–61
9. Margretts AR (1957) The length–girth relationships in whiting
and cod and their application to mesh. J Cons Int Explor Mer
23:64–71
¨ zbilgina H, Ferro RST, Robertson JHB, Holtrop G, Kynoch RJ
10. O
(2006) Seasonal variation in trawl codend selection of northern
North Sea haddock. ICES J Mar Sci 63:737–748
11. Inoue Y, Matsushita Y, Goto M (1998) Development of timer
controlled underwater video camera pack. Tech Rep Nat Res Inst
Fish Eng 20:23–31
12. Millar RB (1992) Estimating the size-selectivity of fishing gear
by conditioning on the total catch. J Am Stat Assoc 87:962–968
13. Millar RB, Fryer RJ (1999) Estimating the size selection curves
of towed gears, traps, nets, and hooks. Rev Fish Biol Fish
9:89–116
14. Fryer RJ (1991) A model of between-haul variability in selectivity. ICES J Mar Sci 48:281–290

Fish Sci (2011) 77:169–181
15. Hamabe Y, Harada S, Yamashita H, Tokai T (2010) Effect of
square mesh window on codend selectivity for blackthroat seaperch and white-spotted conger in offshore trawl net. Nippon
Suisan Gakkaishi 76:824–840 (in Japanese with English abstract)
16. Millar RB, Broadhurst MK, Macbeth WG (2004) Modeling
between-haul variability in the size selectivity of trawls. Fish Res
67:171–181
17. Tokai T (1997) Maximum likelihood parameter estimates of a
mesh selectivity logistic model through SOLVER on ME-Excel.
Bull Jpn Soc Fish Oceanogr 61:288–298 (in Japanese)
18. Tokai T, Mituhasi T (1998) SELECT model for estimating
selectivity curve from comparative fishing experiments. Bull Jpn
Soc Fish Oceanogr 62:235–247 (in Japanese with English
abstract)
19. Sakamoto Y (1991) Categorical data analysis by AIC (Mathematics and its Applications Series). KTK, Tokyo
20. Xu G, Arimoto T, Inoue M (1988) The measurements of swimming speed in jack mackerel Trachurus japonicus. Nippon Suisan
Gakkaishi 54:1493–1497 (in Japanese with English abstract)
21. Xu G, Arimoto T, Inoue M (1993) Red and white muscle of the
jack mackerel Trachurus japonicus during swimming. Nippon
Suisan Gakkaishi 59:745–751
22. Yanase K, Arimoto T (2007) Quantitative assessment of behavioural response to the netting panel in the jack mackerel Trachurus japonicus. J Fish Biol 71:265–271
23. Nofrizal H, Yanase K, Arimoto T (2009) Effect of temperature on
the swimming endurance and post-exercise recovery of jack
mackerel Trachurus japonicus as determined by ECG monitoring. Fish Sci 75:1369–1375
24. Wardle CS (1984) Fish behaviour, trawl efficiency and energy
saving strategies (FIIT working paper). FAO Fishing Technology
Service, Rome
25. Pichot G, Germain G, Priour D (2009) On the experimental study
of the flow around a fishing net. Eur J Mech 28:103–116
26. Winger PD, Eayrs S, Glass CW (2010) Fish behavior near bottom

trawls. In: He P (ed) Behavior of marine fishes—capture processes and conservation challenges. Wiley-Blackwell, Ames,
pp 67–102
27. Walsh SJ, Hickey WM (1993) Behavioural reactions of demersal
fish to bottom trawls at various light conditions. ICES J Mar Sci
Symp 196:68–76
28. Inoue Y, Matsushita Y, Arimoto T (1993) The reaction behaviour
of walleye pollock (Theragra chalcogramma) in a deep/lowtemperature trawl fishing ground. ICES J Mar Sci Symp
196:77–79

181
29. Arimoto T, Glass CW, Zhang X (2010) Fish vision and its role in
fish capture. In: He P (ed) Behavior of marine fishes—capture
processes and conservation challenges. Wiley-Blackwell, Ames,
pp 25–43
30. Tokai T, Omoto T, Matsuda K (1994) Mesh selectivity of
unmarketable trash fish by a small trawl fishery in the Seto Inland
Sea. Nippon Suisan Gakkaishi 60:347–352 (in Japanese with
English abstract)
31. Matsushita Y, Ali R (1997) Investigation of trawl landings for the
purpose of reducing the capture of non-target species and sizes of
fish. Fish Res 29:133–143
32. Liang Z, Horikawa H, Tokimura M, Tokai T (1999) Effect of
cross-sectional shape of fish body on mesh selectivity of trawl
codend. Nippon Suisan Gakkaishi 65:441–447 (in Japanese with
English abstract)
33. Campos A, Fonseca P (2003) Selectivity of diamond and square
mesh cod ends for horse mackerel (Trachurus trachurus), European hake (Merluccius merluccius) and axillary seabream
(Pagellus acarne) in the shallow groundfish assemblage off the
south-west coast of Portugal. Sci Mar 67:249–260
34. Campos A, Fonseca P, Erzini K (2003) Size selectivity of diamond

and square mesh cod ends for four by-catch species in the crustacean fishery off the Portuguese south coast. Fish Res 60:79–97
35. Campos A, Fonseca P, Henriques V (2003) Size selectivity for
four fish species of the deep groundfish assemblage off the Portuguese southwest coast: evidence of mesh size, mesh configuration and cod end catch effects. Fish Res 63:213–233
¨ zaydın O (2008) Selectivity of a 50-mm
36. Tosunog˘lu Z, Aydın C, O
diamond mesh knotless polyethylene codend for commercially
important fish species in the Aegean Sea. J Appl Ichthyol
24:311–315
37. Aydın C, Tosunog˘lu Z (2010) Selectivity of diamond, square and
hexagonal mesh codends for Atlantic horse mackerel Trachurus
trachurus, European hake Merluccius merluccius, and greater
forkbeard Phycis blennoides in the eastern Mediterranean. J Appl
Ichthyol 26:71–77
38. Ordines F, Massutı´ E, Guijarro B, Mas R (2006) Diamond vs.
square mesh codend in a multi-species trawl fishery of the western Mediterranean: effects on catch composition, yield, size
selectivity and discards. Aquat Living Resour 19:329–338
39. Sala A, Lucchetti A, Piccinetti C, Ferretti M (2008) Size selection
by diamond- and square-mesh codends in multi-species Mediterranean demersal trawl fisheries. Fish Res 93:8–21

123


Fish Sci (2011) 77:183–190
DOI 10.1007/s12562-010-0317-4

ORIGINAL ARTICLE

Biology

Embryonic development and morphology of eggs and newly

hatched larvae of Pacific herring Clupea pallasii
Tatsuya Kawakami • Hiroyuki Okouchi •
Masato Aritaki • Jun Aoyama • Katsumi Tsukamoto

Received: 22 June 2010 / Accepted: 28 November 2010 / Published online: 7 January 2011
Ó The Japanese Society of Fisheries Science 2010

Abstract The embryonic development and morphology
of eggs and newly hatched larvae of the Pacific herring
Clupea pallasii were described using laboratory-reared
specimens originating from the Miyako Bay stock. The
eggs were almost spherical in shape, 1.33–1.46 mm (mean:
1.38 mm) in diameter, and had a thick adherent chorion.
They had a segmented pale yellow yolk, no oil globule, and
a relatively wide perivitelline space. A subgerminal cavity
was observed during the gastrula period, whereas the
blastocoel did not appear. Mass hatching occurred by 271 h
45 min after fertilization, and the newly hatched larvae
were 7.1–7.7 mm (mean: 7.5 mm) in total length with
53–56 myomeres at 9.6°C. The embryonic development of
Pacific herring was substantially similar to that of zebrafish
Danio rerio, American shad Alosa sapidissima, as well as
Atlantic herring Clupea harengus, and generally followed
the basic developmental pattern of teleosts. However,
Pacific herring larvae hatched at a more developed stage
than some other clupeoids, such as Japanese sardine

T. Kawakami (&) Á J. Aoyama Á K. Tsukamoto
Atmosphere and Ocean Research Institute, University of Tokyo,
Kashiwa, Chiba 277-8564, Japan

e-mail:
H. Okouchi Á M. Aritaki
Miyako Station, National Center for Stock Enhancement,
Fisheries Research Agency, Miyako, Iwate 027-0097, Japan
Present Address:
H. Okouchi
Head Office of Fisheries Research Agency,
Yokohama, Kanagawa 220-6115, Japan
M. Aritaki
Seikai National Fisheries Research Institute,
Nagasaki, Nagasaki 851-2213, Japan

Sardinops melanostictus, and the progressed developmental stage at hatching could be interpreted as an advanced
adaptation.
Keywords Clupea pallasii Á Egg Á Embryonic
development Á Morphology Á Newly hatched larva Á
Pacific herring

Introduction
Pacific herring Clupea pallasii (Clupeidae, order Clupeiformes) inhabit shallow coastal waters and are widely
distributed in the temperate waters of the north Pacific from
the California coast to the Aleutian Islands and from the
Bering Sea across the Pacific Ocean to the Yellow Sea [1].
This species has been an important fishery resource, but the
annual catch has decreased drastically during the twentieth
century. For example, the annual catch of the Hokkaido–
Sakhalin population yielded 972,000 tons in 1897, but
declined steadily in the first half of the twentieth century,
and the catch virtually disappeared after the mid-1950s
[2, 3]. This tendency was observed in other Pacific herring

populations in the Prince William Sound in Alaska [4],
Cherry Point in Washington state [5] and California [6],
although it is unlikely that those declines were caused only
by overexploitation. Therefore, artificially reared Pacific
herring juveniles have been released into natural waters for
stock enhancement purposes in Japan since 1982 [2, 3], and
the number of hatchery-reared juveniles released reached
approximately 7,000,000 during 2008 [7].
Because of the importance of fisheries and the great
interest in enhancing fish stocks, many studies focusing on
early life history [8], early growth and survival [9], the
migration of released juveniles [10], and factors affecting

123


184

embryonic development—such as intertidal exposure [11],
salinity [12], water temperature [13, 14], paternal effects
[14], suspended sedimentation and weathered crude oil
[15]—have been performed. The early development of
Pacific herring has also been reported [13, 16]. Uchida
et al. [16] described the morphology of eggs, larvae and
juveniles of Pacific herring based on wild specimens collected from the east coast of the Korean Peninsula, and
Kuwatani et al. [13] reported the embryonic development
and morphology of newly hatched larvae using laboratoryreared specimens that originated from adults caught in
northern Ishikari Bay. However, the morphological
description of Pacific herring eggs provided by Uchida
et al. [16] did not cover their whole embryonic period, and

Kuwatani et al. [13], who outlined the embryonic stages of
this fish, lacked precise information about egg and larva
size, a detailed morphology of newly hatched larvae, and
data on the time taken to reach the each developmental
stage. Because of these factors, it has been difficult to fully
evaluate the embryonic development of Pacific herring and
to make comparisons within or among species.
Knowledge about morphological embryonic development and comparative studies with their congeners must be
accumulated in order to determine the optimum incubation
conditions for Pacific herring eggs, which will to help
produce healthy artificial seedlings effectively. Moreover,
a detailed description of egg morphology can help to
identify unknown wild specimens, because the morphological similarity of eggs among different taxa is the main
obstacle to the identification of natural fish eggs. Comparisons of the developmental patterns among diverse taxa
would also contribute to phylogenetic and developmental
research aimed at understanding the similarities and differences among teleosts.
The objective of this study was to describe the detailed
egg morphology and the embryonic development of Pacific
herring using artificially inseminated eggs, and to improve
on the previous knowledge of the embryonic development
of this species, provided by Uchida et al. [16] and Kuwatani
et al. [13].

Materials and methods
Mature Pacific herring were caught in coastal set-nets near
the Tsugaru-ishi River in Miyako Bay, Iwate, Japan on 10
February 2003. The fish were immediately transferred to
the Miyako Station of the National Center for Stock
Enhancement, Fisheries Research Agency, Japan. In
Miyako Bay, artificially produced Pacific herring juveniles

that originated from the Mangoku-ura population, distributed along the Pacific coast of northern Japan, have been
released since 1984 [10]. Previous reports about the early

123

Fish Sci (2011) 77:183–190

development of this species used artificially fertilized eggs
from the Ishikari Bay population [13], and wild specimens
collected around the east coast of the Korean Peninsula
[16], which have different spawning grounds and are
considered to be populations that are genetically isolated
from the Mangoku-ura population [17].
Artificial insemination was performed at about 6 h after
the capture of adult fishes. Unfertilized eggs were stripped
from ovulating females and then placed into a plastic bowl
without water, and sperm suspension from 2 matured males
was immediately poured over them. Fertilized eggs were
adhered to hatching trays that consisted of a wooden frame
(300 9 450 mm) and a polyethylene net and incubated in a
10 m3 indoor tank with gentle aeration and flow-through
filtered seawater. The several hatching trays were transferred to a 0.5 m3 plastic tank with gentle aeration and
pouring filtered seawater at 10 days after fertilization. The
water flow was stopped about 220 h after fertilization. The
rearing water temperature was kept at 8.0°C on the first
day, 9.0°C on the second day and 10.0°C after the third day
after fertilization. Suitable rearing temperatures were estimated to be 3.5–10°C [13]. Although the eggs are tolerant
of higher water temperatures such as 15°C, and increasing
the water temperature shortens the duration of the embryonic stage [18], it costs to increase the water temperature.
Thus, the temperature regime in the present study was

thought to be a cost-effective method for mass seedling
production of Pacific herring. Water temperatures during
the rearing period varied between 8.0 and 10.8°C. Mean
water temperature was 9.6°C. The salinity of seawater was
approximated as 24.0–33.6 based on observation station
data for February in Miyako Bay during 2006–2008
(Miyako Station, National Center for Stock Enhancement,
unpublished data).
Thirty fertilized eggs were observed for about 3 h after
fertilization under a dissecting microscope (SMZ800,
Nikon, Tokyo, Japan), and their egg diameters were measured to the nearest 0.01 mm under a profile projector
(V12-B, Nikon, Tokyo, Japan). Several live eggs were
periodically observed and sketched every 2–14 h. In those
eggs, the chorion was then manually removed and the
detailed embryonic morphology was observed after being
fixed in a 5% buffered formalin solution. Twenty newly
hatched larvae were randomly sampled, immediately fixed
in 5% buffered formalin solution and observed under a
dissecting microscope, and the total length (TL), notochord
length (NL), preanal length (PAL) and long axis of the yolk
were measured to the nearest 0.1 mm. All drawings made
by camera lucida were of fixed specimens. Deformed larvae with notochordal bending, contortion of the body, and
reduction of the jaws and pectoral fins that were reported
from both laboratory-reared yolk-sac Pacific herring larvae
[18] and those collected from the wild [19] were counted to


Fish Sci (2011) 77:183–190

calculate the deformity rate at hatching, and were excluded

from the morphological observations.
To compare the developmental rate with that of previously reported studies, the degree-hours [20] from fertilization to each developmental event was calculated.
Because the water temperature was measured at several
hour intervals, the water temperature between observations
was complemented by measurements following each
observation.
The embryonic development of Pacific herring was
described based on terminology and staging that were
proposed for the common development of teleosts by
Kimmel et al. [21]. Furthermore, to clarify the similarities
and differences in embryonic development within and
among species, results were compared with previous
research into Pacific herring [13] as well as the embryonic
development of a congeneric species, Atlantic herring
Clupea harengus [22] and that of another clupeid, American
shad Alosa sapidissima [23].

Results
Fertilized eggs
The eggs adhered to the hatching tray by their thick pale
yellowish semitransparent adherent layer, and were almost
spherical in shape, with diameters ranging from 1.33 to
1.46 mm (mean ± SD: 1.38 ± 0.03 mm, n = 30) (Fig. 1a).
The yolk was segmented and the color was pale yellow
with no oil globule. The perivitelline space was relatively
wide and the yolk diameter ranged from 0.99 to 1.11 mm
(mean ± SD: 1.04 ± 0.03 mm, n = 30) and occupied
71.2–79.9% (mean ± SD: 75.3 ± 2.03%, n = 30) of the
egg diameter. The fertilization rate after about 21 h was
88.1% (385/437 eggs).

Embryonic development
The developmental sequence of Pacific herring corresponding well with that of Atlantic herring [22], American
shad [23] and zebrafish [21], and could be divided into five
periods (Table 1). However, some detailed developmental
events could not be compared and assigned to the stages
proposed in the previous studies because of difficulties in
correctly matching the results to the stages using percentage of eye pigmentation for the organogenetic period of
Atlantic herring [22], and problems identifying the primary
diagnostic features under a dissecting microscope for the
segmentation and pharyngula periods of zebrafish [21]. The
zygote, cleavage, blastula and gastrula periods could be
identified based on embryonic morphology, such as elevation of the blastodisc, cleavage of the blastodisc,

185

formation of blastoderm, and formation of the germ ring,
respectively (Table 1).
The blastodisc and perivitelline space were formed at 1
h 58 min after fertilization. Cleavage in Pacific herring was
meroblastic, and the first and second cleavages had
occurred by 3 h 38 min and 5 h 15 min, respectively
(Fig. 1a). At 17 h 53 min, a multilayered and highly
domed blastoderm was formed (Fig. 1b). Epiboly began by
30 h 13 min. At 42 h 30 min, the blastoderm continued to
spread by epiboly to cover 50% of the yolk, and the germ
ring and embryonic shield appeared (Fig. 1c). Although no
blastocoel was observed, a cavity developed between the
blastoderm and the yolk (subgerminal cavity). At 48 h
28 min and 54 h 20 min, the epiboly covered 70 and 90%
of the yolk, respectively (Fig. 1d, e). The subgerminal

cavity disappeared at 48 h 28 min.
At 66 h 20 min, ten myomeres appeared, and this was
regarded as the segmentation period (Fig. 1f). The number
of myomeres increased by about 5 every 6 h; the total
number reached 46–49 at 114 h 28 min. Appearance of
optic vesicles (Fig. 1f), Kupffer’s vesicle, lens and otic
vesicles (Fig. 1g), and the disappearance of Kupffer’s
vesicle were recognized at 66 h 20 min, 72 h 21 min, 90 h
23 min, and 96 h 5 min, respectively. The end of the tail
began to separate from the yolk sac, the finfold formed
around the tail at 90 h 23 min, and brain differentiation
was recognized (Fig. 1g). The embryo was subsequently
elongated almost around the yolk sac and the end of the tail
separated further from the yolk sac at 102 h 32 min
(Fig. 1h). At this time, the appearance of olfactory organs,
movements of the embryo, and the onset of the heartbeat
were also recognized. The gut was found to have formed at
114 h 28 min, and its end reached the ventral margin of the
finfold at 126 h 5 min. The number of preanal myomeres
(PAM) ranged from 44 to 46: almost the same as in the
adult fish.
During the pharyngula period, the total number of
myomeres reached the number seen in adult Pacific herring
(54–56) at 138 h 28 min (Fig. 1i). At this time, the embryo
was subsequently elongated beyond the eye, and part of the
head separated from the yolk sac such that it was level with
the anterior edge of the eye. Eye pigmentation was also
observed, and otoliths and pectoral fin buds appeared
below the sixth myomere. The guanophores on the eye and
the mesh-like pattern on the surface of the embryonic body

appeared by 186 h 15 min after fertilization (Fig. 1j). The
mouth opened and a row of melanophores appeared on the
anterior part of the ventral side of the body at 210 h 15 min
(Fig. 1k). At 216 h 29 min, the embryo was elongated
twice around the yolk sac. A row of melanophores and a
branched melanophore appeared on the ventral side of the
posterior part of the gut and the dorsal side of the posterior
end of the gut, respectively. The occurrence of many

123


186

Fish Sci (2011) 77:183–190

Fig. 1 Embryonic development of eggs and a newly hatched larva of
Pacific herring Clupea pallasii. a Four-cell stage, 5 h 15 min after
fertilization. b Blastula, 17 h 55 min. c 50% epiboly, 42 h 30 min.
d 70% epiboly, 48 h 28 min. e 90% epiboly, 54 h 20 min. f Appearance of optic vesicle, 10 myomeres present, 66 h 22 min. g Appearance of lens and otic vesicle, 30 myomeres present, 90 h 23 min.

h Beginning of body movements and heartbeat, 102 h 32 min.
i Appearance of melanophores on the eye and pectoral fin rudiments,
138 h 28 min. j Appearance of guanophores on the eye, 186 h
15 min. k Melanophores appeared on the body, 210 h 15 min.
l Appearance of melanophores on the caudal part, 234 h 10 min.
m Newly hatched larva, 7.6 mm in total length, 271 h 45 min

granular hatching glands was clearly observed on the
anterior part of the head region. Some branched melanophores appeared on the caudal part of the body at 234 h

10 min (Fig. 1l). The mouth gape expanded, and the jaws
and branchial arches were found to form by 264 h 40 min.

Hatching had started by 241 h 58 min and mass hatching
occurred around 271 h 45 min (the hatching period).
The beginning of the first division, appearance of an
optic vesicle, appearance of an otic vesicle, beginning of
body movements, start of heart functioning, origin of

123