ICES Journal of Marine Science, 53: 147–153. 1996
Spatial organization of pelagic fish: echogram
structure, spatio-temporal condition, and biomass in
Senegalese waters
Pierre Petitgas and Jean Jacques Levenez
Petitgas, P. and Levenez, J. J. 1996. Spatial organization of pelagic fish: echogram
structure, spatio-temporal condition, and biomass in Senegalese waters. – ICES
Journal of Marine Science, 53: 147–153.
Understanding the spatial distribution of fish schools is expected to enable future
improvement in the reliability of acoustic abundance estimates as well as catch data.
The analysis of echograms will provide detailed morphological descriptions of fish
aggregations together with characteristics of their habitat. The data set studied here is
a series of comparable echograms taken during monitoring surveys and showing
important fluctuations in biomass. Echoes are coded into different consistent morphological types. First, we study how the occurrence of these types varies with diel cycle,
sea-bottom depth, and fish density by computing a chi-square table. Second, characteristics of inter-school spatial structure are analysed by computing indicator variograms. Third, location of high-density values within the areas where schools are
present is studied by computing indicator cross-variograms. A descriptive model of
habitat occupation is proposed where some parameters are related to biomass level
and others are not.
? 1996 International Council for the Exploration of the Sea
Key words: geostatistics, schools, spatial structure, survey design.
Address correspondence to: P. Petitgas, ORSTOM, HEA, 911 avenue Agropolis,
BP 5045, 34032 Montpellier, Cedex 1, France. [Fax: (33) 67419430; email:
[email protected]].
Introduction
In echo-integration surveys of pelagic fish resources, the
measured backscattered acoustic energy is averaged over
all individual samples made through the water column
and along unit distances of the ship’s course. Thus, the
structural information present on the echogram is not
used when performing biomass estimation. Since it is
common that dense schools will constitute a large percentage of the biomass, survey reliability largely depends
on encountering a sufficient number of these schools.
MacLennan and MacKenzie (1988) and Marchal and
Petitgas (1993) have envisaged biomass estimation by
separating school counts from school biomass. This
approach showed that a major factor in acoustic surveys
is the imprecision in the right tail of the school biomass
histogram. Additional knowledge of the dynamics of
schools that may generate such high values should
improve the reliability of acoustic estimates. Here, we
investigate for a tropical data set the relations between
the spatial distribution of echo traces and four parameters: sea-bottom depth, diel cycle, local density, and
total biomass.
1054–3139/96/020147+07 $18.00/0
Materials
Acoustic surveys and echo integration
Pelagic biomass monitoring surveys have been undertaken in Senegal since 1983 by CRODT (Oceanographic
Research Centre of Dakar-Thiaroye) and ORSTOM
(French Institute of Scientific Research for Development). We consider here the surveys carried out from
Dakar to Roxo Cape (14)45*N to 12)20*N) during the
cold season. These surveys were made from the same
research vessel, the same acoustic equipment, the same
echo integration procedures, and the same survey
design.
The acoustic equipment was a Biosonics echo-sounder
working at 120 kHz. Echo integration was performed by
layers using the same adjustment in all surveys (Levenez
et al., 1985). The ESDU (elementary sampling distance
unit – the distance over which the echo integral is
accumulated to give one sample) was 1 nautical mile
(nmi). The survey design was made of parallel regularly
spaced transects oriented east–west across the continental shelf. The inter-transect distance was 5 nmi. Backscattered acoustic energy was converted to fish density
? 1996 International Council for the Exploration of the Sea
148
P. Petitgas and J. J. Levenez
(t per square nautical mile of sea surface) using the
average target strength given by Levenez (1990).
Echograms
The echograph used for all the surveys was a Ross Fine
Line 250C working with dry paper. In all surveys,
adjustments such as paper speed, echo-sounder ping
repetition rate, and vessel speed were the same. Thus, all
echo traces are comparable from a morphological point
of view.
Methods
Echogram coding
The morphology of echo traces was consistent from
survey to survey. Nine types of echo trace were defined
and named by analysing various transects in all surveys
(Fig. 1). The nine types and the null type enable a
complete description of the echo traces present in each
ESDU. The different echo types are as follows:
-
-
-
Null: the ESDU is empty of echo traces.
Fish scattered echoes (FS): the echo traces do not
form any aggregated structure.
Schools: these will be denoted by three letters, the
first being S for schools. The distinction between
small schools and schools was necessary because
small schools may be in great numbers in an ESDU,
giving an aspect of aggregated but also scattered
biomass. Small schools (SSM) were defined as being
less than 5 m high and having a compact internal
appearance. Pole-shaped schools (SPO) are fully in
contact with the bottom, they are compact inside and
look like poles. Tower-shaped schools (STW) are
fully in contact with the bottom, they have a fluffy
internal appearance and have a large section that
makes them look like towers. Fluffy schools (SFL)
are pelagic, oval-shaped and with a variable internal
density giving a fleecy or fluffy appearance. Compact
schools (SCO) are pelagic, of varying shape, but
showing clear borders and a compact appearance
inside. Zig-zag schools (SZZ) are pelagic, compact
inside, irregularly shaped, and can be present as
isolated schools or form layers.
Macro structures: these are very large aggregates
showing dimensions of 1 to 2 nmi. This echo type is
not present for all surveys and is rare. ESDUs
containing this type were not used in the present
analysis.
Layers: these are thin layers of scattered but dense
echoes. They show important depth variations on a
very small scale. Vertical temperature measurements
showed that this layer structure is associated with a
thermocline. It was thus named Layer associated with
a Thermocline (LT).
The echoes present in each ESDU were coded in
presence/absence for the following vector of nine echo
types: (NULL, FS, SSM, SPO, STW, SFL, SCO, SZZ,
LT). This coding was performed for all ESDUs of four
surveys, two showing high biomass (1985 and 1993),
one showing low biomass (1989), and one showing
intermediate biomass (1992).
Density estimation of an echo type and a school
intersect
Average density in each echo type for each survey was
estimated by considering the ESDUs containing just this
echo type.
Height and diameter of schools were measured on the
echogram and were corrected for beam width and pure
length distortion as proposed by Johannesson and Losse
(1977). School intersect area was estimated by multiplying height by length. In the ESDUs containing just one
school, horizontal surface density in the school intersect
expressed in kilograms per square metre of sea surface
was estimated by rescaling the ESDU biomass to the
school area. When more than one school was present,
average density per school was estimated. Density (kg
m "2) in the intersects of compact schools (SCO) was
estimated. This echo type was the school type contributing the most to the biomass and was frequently found
alone in the ESDUs.
Contingency and chi-square tables
Eight spatio-temporal situations were defined by combining the following classes: coastal and offshore waters,
day and night, rich and poor ESDU density. Coastal
waters were defined by the depth limit of 25 m used by
Fréon (1991) for the area. The time limits between night
and day were defined at 0700 h and 1900 h, from field
experience. Rich ESDUs were defined as having a density higher than 100 t per nautical mile square. This
value corresponded to the beginning of the long tail of
the combined density histogram for all the data of the
four selected surveys.
The distribution of the occurrence of the nine echo
types in the eight spatio-temporal situations was analysed by computing a contingency table and deriving the
corresponding chi-square table. Each line refers to a
spatio-temporal category and each column to an echo
type. The table produced is the signed contribution of
each cell to the total chi-square statistic. This allows us
to see where the most important features in the data lie
by simple comparison between cell values (Saporta,
1990). Interactions between echo types were not
considered.
Variograms and cross-variograms of indicators
Experimental variograms may characterize spatial structure of natural phenomena (Matheron, 1971; Journel
Spatial organization in pelagic fish
Figure 1. Echo types defined for characterizing each ESDU of the echograms.
149
150
P. Petitgas and J. J. Levenez
Table 1. Chi-square table when crossing echo types (columns) with spatio-temporal situations (rows) for the four selected
surveys. Each cell number is the cell’s contribution to the total chi-square value per thousand (‰). Last column shows each
row contribution and last row shows each column contribution, in percent. The total chi-square values is 2631. Symbols
for spatio-temporal situations are: N=night, D=day, C=coastal waters, O=offshore waters, L=low ESDU density, H=high
ESDU density (i.e. greater than 100 t per nautical mile square). Symbols for echo types are: FS=scattered fish, SSM=small
schools, SPO=pole schools, STW=tower schools, SFL=fluffy schools, SCO=compact schools, SZZ=zig-zag schools,
LT=thermocline-associated layer.
NCL
NCH
NOL
NOH
DCL
DCH
DOL
DOH
NULL
FS
SSM
SPO
STW
SFL
SCO
SZZ
LT
"24.2
"16.0
"4.8
"24.1
78.6
"11.0
75.6
"15.4
25.0
79.3
13.3
3.8
0.1
"4.6
"7.0
"44.8
"21.3
17.4
"20.1
"9.5
0.3
"1.5
1.4
6.1
9.5
0.7
4.9
"1.9
0.3
"5.0
0
0
2.1
0
32.6
4.2
14.5
98.3
"12.2
"1.5
"0.7
4.3
"8.8
"2.5
14.3
"0.1
1.1
"1.4
14.9
"7.8
"0.5
0.7
0.1
2.7
"15.6
"1.5
"5.3
2.4
"7.3
15.2
1.7
81.7
13.1
"9.1
"3.1
"6.4
5.7
"0.6
24.6
"0.4
58.7
10.9
"3.7
"5.8
40.8
4.7
"11.3
"4.5
"2.7
"3.1
7.7
and Huijbregts, 1978). In general, the variogram curve
will increase, then reach a sill (an asymptotic plateau in
the variance). This highlights spatial correlation between
pairs of samples up to the distance where these vanish,
which is the range. A flat variogram indicates no spatial
structuring.
Variograms and cross variograms were computed for
indicator variables. The indicator, Ia(x), of some characteristic a takes at point x the value 1 if a is true,
otherwise it takes the value 0. In space, the points where
a=true define the geometrical sets A. The variogram of
Ia(x), ãa(h), measures the probability of a segment of
length h having one extremity inside a set A and the
other outside. The range is related to the average
diameter of the sets A. The sill is related to the number
of sets A on the area studied. It is equal to p(1"p),
where p denotes the average of Ia(x). Let Ia denote the
indicator of the presence of schools in the ESDUs. Its
variogram will indicate spatial structure of the areas
where schools are present.
Now consider two characteristics a and b. The points
where these are true form geometrical sets Aa and Ab
respectively. The cross-variogram, ãab(h), between indicators Ia and Ib measures the probability of a segment of
length h having one extremity outside one of the sets and
the other extremity inside the other set. In the case where
sets Ab are included in sets Aa, and assuming symmetry
in h, the spatial setting of sets Ab within sets Aa can be
characterized by a conditional probability which has the
following form (Rivoirard, 1993):
Prob(Ib(x) =1PIa(x) =1, Ia(x+h) =0)=ãab(h)ãa(h) "1
An increase of this ratio with increasing distance highlights border effects, as sets Ab will be positioned on
average in the middle of sets Aa. When the ratio is flat,
sets Ab are situated as much on the sides as in the middle
16.9
14.9
8.0
5.5
11.2
7.5
14.4
21.6
of sets Aa. Let Ib* denote the indicator of high density
ESDUs. The product Ib =Ib**Ia is the indicator of the
rich ESDUs that contain schools. Sets Ab will be
included in sets Aa. The previous ratio will indicate how
high density ESDUs that contain schools are positioned
in the areas where schools are present.
Results
Relationship between echo type occurrence and
spatio-temporal condition
Table 1 shows the chi-square table characterizing the
distribution of the nine echo types between various
temporal, spatial, and depth-related situations. The
value of the chi-square statistic is 2631, which results in
a highly significant test for 56 degrees of freedom.
The spatio-temporal situations contributing most to
the chi-square statistic are coastal waters at night (lines 1
and 2) and offshore waters during day (lines 7 and 8).
The echo types contributing the most to the chi-square
are null, scattered fish (FS), tower schools (STW),
compact schools (SCO), and zig-zag schools (SZZ).
Echo-type occurrence varies with the coast-to-offshore
gradient, the day–night cycle, and the biomass level in
the ESDU. At night, the coastal waters are characterized
by the absence of null ESDUs and the presence of
scattered fish and tower-shaped schools. The night offshore waters are characterized by the absence of null
ESDUs and presence of SFL in rich areas, and in poor
areas by the absence of SFL and the presence of LT
echoes. During the day, the coastal waters are characterized by the presence of null ESDUs in poor zones,
and in rich zones by the presence of SCO and SZZ. The
day offshore waters are characterized by the absence of
FS and the presence of null ESDUs in poor areas, and in
rich areas by the presence of schools of SPO, SCO, and
Spatial organization in pelagic fish
Table 2. Statistics of fish density per ESDU (t per nautical
square mile). n=number of ESDUs, mean=average,
s.d.=standard deviation, f0 =percentage of empty ESDUs,
max=maximum ESDU density value.
n
Mean&s.d.
f0
Max
1985
1989
1992
1993
1110
123&375
12.5
7240
916
17&59
36.5
1155
964
75&190
18.4
3062
804
142&454
19.1
7027
Table 3. Statistics of echo types and compact schools (SCO)
computed on non-zero ESDUs. Echo diversity=average
number of different echo types per ESDU. SCO area (%)=
average number of ESDUs containing compact schools. SCO
nb=average number of compact schools per ESDU. SCO
(b>5)=average number of ESDUs containing compact schools
with surface density higher than 5 kg m "2.
Echo diversity
SCO area (%)
SCO nb
SCO (b>5)
1985
1989
1992
1993
1.60
13.5
1.75
0.52
1.61
8.8
1.28
0.19
1.54
8.2
1.45
0.25
1.63
10.8
1.56
0.64
SZZ. Small schools are not characteristic of any situation; their presence is slightly increased offshore during
the day in the poor zones.
Chi-square tables computed for each survey showed
the same pattern in the distribution of echo types.
Thus, echo-type occurrence was not related to biomass
level.
Relationship between echo-type density and
biomass
Tables 2 and 3 show respectively statistics of within
ESDU density and SCO for the four surveys. Table 2
shows that extreme ESDU density values (zero and
maximum) are related to biomass level. Table 3 shows
that echo-type diversity per ESDU, the area occupied
by SCO schools in positive zones, and the average
number of SCO schools per ESDU are not related to
biomass level. But the occurrence of dense schools
(SCO) follows the biomass level. The combined histogram for all values of surface density in schools (SCO)
of the four surveys was computed. It showed a long
tail starting near the value of 5 kg m "2, which we
considered as defining the threshold of the dense
school class.
Table 4 shows the average density in each echo type
for each survey. Echo-type density varies with biomass
level. In the low biomass situation of 1989, density in all
echo types was low. In the intermediate biomass level
151
Table 4. Average density in the ESDUs containing just a single
echo type (in t per nautical square mile). Values are shown
when more than five ESDUs could be used to compute the
average. Symbols for echo types are: FS=scattered fish,
SSM=small schools, SPO=pole schools, STW=tower schools,
SFL=fluffy schools, SCO=compact schools, SZZ=zig-zag
schools, LT=thermocline-associated layer.
FS
SSM
SPO
STW
SFL
SCO
SSZ
LT
1985
1989
1992
1993
58.0
36.8
—
—
131.8
210.5
—
—
13.2
7.5
—
—
12.3
34.6
83.4
4.3
59.4
37.2
—
—
84.7
126.6
162.7
34.9
49.7
44.4
—
403.4
52.0
181.1
236.6
116.1
situation of 1992, density is lower only in the dense echo
types SCO, SZZ, and LT. Thus, echo types are related to
one another
Spatial structure of school presence and its
relation to rich ESDUs
For each survey we computed the variogram in 2D on
the echo integration samples for the indicator, Ia, of the
presence of SSM and/or SCO, because these echo types
contributed the most to the biomass. The variograms are
given in Figure 2. A consistent structure is observed
with a range between 3 and 7 nmi and a sill close to
0.25. School presence is spatially organized in clusters.
Whatever the biomass level of the survey, clusters of
school presence had constant dimensions (variogram
range) and the number of clusters per unit area stayed
constant (variogram sill).
Then we considered the indicator, Ib, of rich ESDUs
(density greater than 100 t per square nautical mile) that
also contained small schools (SSM) and/or compact
schools (SCO). The cross variogram of indicators Ib and
Ia was computed in 2D for each survey, then divided by
the variogram of indicator Ia for each distance. The
ratios stayed flat, unstructured, and erratic (not shown).
High densities associated with school presence are not
situated on average in the middle of the areas of school
presence. Precise prediction of high-density values from
echogram structure is not to be expected.
Discussion
Echo-trace morphologies (echo types) were related to
sea-bottom depth and diel cycle, but did not depend on
biomass level. Where fish were present, the average
number of schools per sea-surface unit stayed constant
across the years, as did the diversity of echo types. These
152
P. Petitgas and J. J. Levenez
Variogram
0.3
0.2
0.1
0
5
10
15
20
Distance (nmi)
25
30
Figure 2. Variogram of the indicator of small schools (SSM) and/or compact schools (SCO), computed on non-zero ESDUs for
the four surveys.
parameters were not related to biomass level. Spatial
structure of school presence in areas of non-empty
ESDUs was consistent across the years and showed no
relationship with biomass level. The presence of schools
was clustered. Average cluster dimensions stayed constant, as did the number of clusters relative to the area of
presence. Biomass in each echo type followed the general
level of population biomass, as did the area of fish
presence and the occurrence of dense schools. Rich
ESDUs containing schools were placed at random inside
the areas of school presence.
The spatial structure of school presence was found to
agree with the cluster model proposed by Fiedler (1980).
Random placement of high density ESDUs in the areas
where schools are present is in accordance with the
disjunctive kriging results of Petitgas (1993). The relationship between biomass level and dense school occurrence and the stability in the number of school clusters
relative to the area of non-empty ESDUs agree with the
characteristics of the pelagic purse seine fishery for
Sardinella sp., observed by Fréon (1991) south of Dakar.
Inter-annual variations in the number of hauls per day
at sea were small, but the occurrence of large hauls (i.e.
rich schools) varied between years. Reduction in the area
of presence with biomass level does not entirely concur
with the observations of Ulltang (1980), because low
biomass was not confined to a specific and reduced area.
An increase in stock catchability with a decrease in
biomass level does not seem to be compatible with our
results.
The stability of the spatial structure of school presence could be used for designing surveys: the intertransect distance could be related to the variogram range
of the school indicator. The relationship observed
between density in the echo types and biomass level
indicates that echogram appearance could enable simple
classification of a survey into the rich or poor category,
and thus enable critical evaluation of the sampling
efficiency of rich schools.
Acknowledgements
The monitoring acoustic surveys were part of a project
between ORSTOM (France) and CRODT (Senegal). We
are grateful to B. Samb and his team (A. Sarre, I. Sow,
M. Sylla, and I. Sane), who collected, stored the data,
and discussed the survey results, and the crew of RV
‘‘Louis Sauger’’.
References
Fiedler, P. 1980. The precision of simulated transect surveys of
northern anchovy (Engraulis mordax) school groups. Fishery
Bulletin, 76: 679–685.
Fréon, P. 1991. Variations saisonnières et inter-annuelles de la
prise moyenne par calée dans la pêcherie sardinière dakaroise
et possibilité d’utilisation comme indice d’abondance. In
Pêcheries Ouest-Africaines, pp. 259–268. Ed. by P. Cury and
C. Roy. ORSTOM Editions, Paris. 525 pp.
Johannesson, K. and Losse, G. 1977. Methodology of acoustic
estimations of fish abundance in some UNDP/FAO resource
survey projects. Rapports et Procès-verbaux des Réunions du
Conseil International pour l’Exploration de la Mer, 170:
296–318.
Journel, A. and Huijbregts, Ch. 1978. Mining geostatistics.
Academic Press, London. 600 pp.
Levenez, J. 1990. Mesures de l’index de réflexion acoustique
(TS) de quelques poissons pélagiques tropicaux par la
méthode de la cage. Archive 117, Centre de Recherche
Océanographique de Dakar-Thiaroye, Dakar. 19 pp.
Levenez, J., Samb, B. and Camarena, T. 1985. Résultats de la
campagne Echosar VI. Archive 133, Centre de Recherche
Océanographique de Dakar-Thiaroye, Dakar. 39 pp.
MacLennan, D. and MacKenzie, I. 1988. Precision of acoustic
fish stock estimates. Canadian Journal of Fisheries and
Aquatic Sciences, 45: 605–616.
Marchal, E. and Petitgas, P. 1993. Precision of acoustic fish
abundance estimates: separating the number of schools from
the biomass in the schools. Aquatic Living Resources, 6:
211–219.
Matheron, G. 1971. The theory of regionalized variables and
their applications. Les Cahiers du Centre de Morphologie
Mathématique, Fascicule 5. Ecole Nationale Supérieure
des Mines de Paris, Centre de Géostatistique, Fontainebleau.
212 pp.
Spatial organization in pelagic fish
Petitgas, P. 1993. Use of disjunctive kriging to model areas of
high pelagic fish density in acoustic fisheries surveys. Aquatic
Living Resources, 6: 201–209.
Rivoirard, J. 1993. Relations between the indicators related
to a regionalized variable. In Geostatistics Troia 1992,
pp. 273–284. Ed. by A. Soares. Kluwer Academic Publishers,
Dordrecht. 1089 pp.
153
Saporta, G. 1990. Probabilités, analyse des données et
statistique. Editions Technip, Paris. 488 pp.
Ulltang, O. 1980. Factors affecting the reaction of pelagic fish
stocks to exploitation and requiring a new approach to
assessment and management. Rapports et Procès-Verbaux
des Réunions du Conseil International pour l’Exploration de
la Mer, 177: 489–504.