SPATIAL DYNAMICS OF PELAGIC FISH AGGREGATION IN THE JAVA SEA BASED ON HYDRO-ACOUSTICS

This study was based on a concept called acoustic population estimation. This is defined as a population of echoes which is based on the hyphothesis that a synthesis of biological and behavioural characteristics of the species (within a community) characterizes the community (Gerlotto & Marchal, 1987). The acoustic data representing the acoustic population contains a set of informaiion that can be used to represent the natural population. This study describes the community and behaviour of pelagic populations in the Java Sea. The acoustic data were collected during two large scale cruises in October 1993 and February 1994. The surveys represented the northwest and southeast monsoon periods. Acoustic shoal feature were derived from the acoustic data by means of MOVIES-B software. Graphical presentation and mapping were performed to represent the interrelation of shape and position dimension variables of the Jhoals. Totals of 350 and 143 shoals were obtained from October 1993 and February 1994 cruises respectively. Based on these variables analyses can be performed. Global observations on the shoal types reveal a possible zonation being generated by similarity of certain characters. Zones of different types of aggregation can be recognized in term of their spatial distribution, with each zone representing specific agregations reflecting fish of particular communities. The internal migration of fish within the Java Sea may endanger stratiflcation of the pelagic community into three parts during the two seasons, i.e. western, central and eastern zones. KEYWoRDS: spatial distribution, acoustic, pelagic, shoal, stratification, Java sea


INTRODUCTION
In marine ecosystems, most of fish species are concentrated in response to particular features of environment (such as current, shelf edge, transparency etc.) by forming structures with specification mainly depending on species and age, or a combination of these.Spatial dependence or continuity distribution such as a tendency for organisms to form patchiness, clumped in some areas, and random or sparse in other areas, is a common phenomenon in marine ecology (Maravelias ef a/., 1996).However, in marine ecosystems the dynamics of patchiness have proven less amenable to direct observation being limited by ability to observe directly the processes underlying patch dynamic in the sea (Greene et at.,19g4).
In this case, natural movement of pelagic fish in three-dimensional space and hydrological environment of the sea are parts of the problems (this is not in case in terrestrial systems).
The particular form of this structure is called as shoal or school, (Swartzman,1gg7) understanding this factual phenomenon in relation to environmentaifactors being triggered by seasonalvariability would provide auxiliary information for the study of spatial distribution, The possibility of an impact of shoal behaviour on acoustic biomass estimates is most likely (Fr6on ef a/., 1 993).From another point of view, constraints in breaking down into species of estimated biomass have encouraged scientists to attempt to classiff fish shoals according to species.This could be a crucialproblem as the pelagic fish communities are composed of several species (e.9. the catch composition from purse seines in the Java Sea indicate that at least 9 species may be present in samples).Severalacoustic methods have been developed for species identification purposes of the sparse and aggregated shoals.Simmonds et al. (1992) give a comprenhensive review on the classification of these methods.
Most of the identification methods used in these studies are based on the spatial structure, but have limitations, Le. the fish may not stay in a space in a random manner (Fr6on, ef a/., 1990).ln applying this to tropicalwaters, the fishing-based species identifi- cations are not clearly represented by acoustic characteristics.In order to overcome this problem, an in- direct method which relates to the generalcharacteristics of the pelagic oommunity can be used.lf the global variation becomes the main consideration, a concept of the "acoustic population" (Gerlotto & Marchal, 1987) can be used to describe the commu-1 Rescarcher at the Rcaearch Institute for Mailne Fiaheries, Muara BEru nity and behaviour of the pelagic populations in the Java Sea.Acoustic population is defined as a population of echoes.lt is based on the hyphothesis that a synthesis of the biological and behavioural characterictics of the species (within the community) characterizes the community; therefore community features may be described by acoustic data representing the acoustic population (Gerlotto, 1993).The basic principle is that the acoustic population con- tains a set of information that can be used to described the natural population.
However, most of the natural characters do not have a direct causal relation to one or more of acoustic characters, withg perhaps only the average size of the fish in a natural population specifically reflected as target strength (one of the acoustic population characters).The main dependence factors determining the acoustic population features are the anatomy, physiology, fish orientation and position as well as their distribution.Detailed explanations of their rela- tions to acoustic echoes have been discussed ex- haustively elsewhere (Foote, 1979; Foote & Ona,   1985a: 1985b; Foote 1980a;1980b; Fr6on efal.,1996;1993).
In this study, the main objective was not to iden- tify the species or group of species, but to map the community features while emphasizing spatial stratification based on the average value of the acoustic characters.ldentification is therefore based on the strata.This is analogous to an approach commonly used in the classic ecological appraisal based on community structure, e.g.certain levelof disturbance of the whole population may be reflected by changes in the community structure (Sainsbury, 1982).In this case, the general concept of diversity may be used to identify the community structure (Pielou, 1977).

General Theory
This study deals only with the utilisation of a certain part of the acoustic data collected during acoustic cruises of RA/ Bawal Putih.Two cruises were chosen from the 16 cruises made during the Pelfish Project, namely cruises number 34 and 41, which were carried out on October 1993 and February 1994, covering most of central and eastern parts of the Java Sea.These cruises represent the conditions during the southeast and northwest monsoon (Figure 1).Integration data were recorded during the cruises, and saved in the format of INES-MOVIES.The informa- tion content of the data are the date, time, position and digitized signal received by echo-sounder.
The broad spatialarea coverage and the periods of the two cruises, allow study of the behaviour of shoals relative to the environment as influenced by monsoonalclimate, based on the acoustics data collection.The acoustic technique is valuable as a tool for describing ecological phenomena as well as for direct estimation of fish stock assessment.In this study, it is performed in two fashions.Firstly, for describing the relationship between partition of shoals, their parameters (shape, density) and environment condition; and secondly to evaluate their relations to size partition in the Java Sea.Evaluation was focused on the variability of spatial distribution of the shoal structure and comparison of the characters from dif- ferent subareas and seasons.The use ofthe acoustic method was ntostly based on the treatment of data produced by an echo-integration technique applied during the cruises.
In our case, we assumed that the acoustic characters were random variables generated from natural populations, regardless of any ftrnction or process from "natural population space" to "acousflc population space".Aggregation or shoaling can be regarded as common phenomenon for most of the pelagic spe- cies, but how far the characteristics of "acoustic shoaf' represent those of natural shoals is unknown.In the context of the acoustic population concept, the pat- tern of acoustic shoal dimensions can be assumed to represent the general characteristics of the natural shoals.The MOVIES-B method may be used to identify characteristics of shoals based on severaldescriptors without having to verify shoalcomposition using catch data.The important step in applying this con- cept is to define appropriate variables that can be used to classify the shoaldata concerning the spatialstructure of the acoustic shoal.In this study, we do not perform species identification, due to the non- availaibility of fishing composition data corresponding to the identified shoals.This study emphasizes the following points: (a) a general description of the spatial distribution, (b) the structure of shoal dimension, (c) a generalclassification and zonation based on the similarity of the dimension variables and (d)  the relationships between classes of descriptors.
The definition emphasizes the spatial characteristics and focuses on the orientation and the ntovement of pelagic fish in distribution and space in relation to their life history in the Java Sea.Behaviour of aggregation is assumed to be related to their response to environment conditions.In term of acoustic detection' a shoal is defined as a set of samples which forms an echogram feature (Weill ef a/., 1993).The behaviour of the shoal in the acoustic detection will be a characteristic feature recognized by the acoustics sys- tem used.Shoal by shoal integration was also per- A Biosonic dual beam vertical echo_sounder oper_ ating at l20kHzfrequency was utilized.Some infor_ mation on operationalissues (r.e.speed of the ship, |g3lilg_qno her position) were integrated through INES-MOVIES and Biosonic f interfades.The tNES_ !!!!!!_S_Oata processing system was designed by IFREMER to perform echo^integration of verticatecho sounding (Diner, 1995).The data acquisition of this system can be described as Figure 2.However, the use of this technique is limited by the fact that the verticalecho-sounder measures only the height and the length cf a shoal passing under the vessel.This provides good vertical resolution but horizontally gives a less accurate measure.For this reason, the circuar shape of the shoal is assumed (Scalabrin & Mass6, 1 993) Weill ef a/ (1993) described the atgorithm of this software for defining a shoalfrom echogram features based on "contiguity of sample criteria;.ln a vertical sense, it is contiguity along the same,,ping,'and hori- zontally, it is from one ping back to the previous one (Figures 3 and 4).Verticalsegments are defined for each ping, and are composed by the sample having an amplitude value above the echo-integration threshold defined manually by the user.The criteria of vertical contiguity is that each segment has a vertical contiguous sample of less than half the pulse duration.lt means that a sample of more than 0.5 t.c' h will not be accepted as a shoal (where [ = pulse duration (s), c = speed of sound in water (m/s) and h = the vertical resolution representing the number of vertical samples per meter).Horizontal contiguity will be verified by a search being made to find at least one sample having the same depth as one segment sample values from the previous ping.Contiguous numbers of segments are considered as the same shoaldetection, and the detection of a shoal will terminate if the number of samples is not increased after a new ping.These criteria may result in a shoal (an acoustic deteetion shoa/) that is not really a natural one.lt may be part of natural big shoal detected as cluster of shoals.A definition of a natural shoal of Radakov (1973) is that a shoal is an active aggregation which may frequently change its shape as a response to external stimuli.A disoontinuity may exist within a natural shoal and it may be detected as separated

Adjustment of threshold
Experimentaltrials had been performed using several settings for eliminating the bottom and surface noises.The best setting then was decided by considering a logical interpretation on the echogram feature.Some of these were used as input parameters for playing back the data by using the MOVIES-B software.We concluded that for the data and system of apparatus used, the most reasonable thresholds appropriate for a shoalwould be: -Color palette > 3% -Minimum back-scattered energy > 100 mv2 -Minimum number of pings > 3 -Minimum number of samples > 30 RESULTSAND DISCUSSION

Distribution of shoals
It must be kept in mind that there are different criteria for shoals detected by the instrument and naturalones, thus possibly generating substantial bias in further analyses, In the system used in this study the coverage is defined by the acoustic beam and the Where k=sounder constant speed of the vessel.lf a shoal is largerthan the coverage area, then only part of the shoal will be detected by the acoustic system.In the case of shoal detection by MOVIES-B, a discontinuity that defines the border of shoalmay be false, and two close shoals identified by this system may really be a single shoal.
Factors generating these biases (Fr6on et at.,   1993;1996) include deformation of the shoal due to avoidance to vessel passage in the near surface layer (Fr6on & Gerlotto, 1988) and diel variabilig.
Evaluation will be focused on the possibility of the spatial gradients atong tongitudinat and latitudinal directions.An attempt at deeper analysis is also carried out emphasizing distribution along the .straighttransect" from Karimunjawa to Matasiri Bank (legs KB, BM and q) in comparison with south-north legsln the western, central and eastern parts of the area surveyed.Overlays of information on shoal distribution by depth and classified size with bottom con- tou rs are performed in two-dimensional presentations of vertical coverage, i.e. by depth and longitude.ln this part we also re-analyse shoal data using the outputs of the MOVIES system (measured in deviation, Qd with unit in mV,)which were played back by the acoustic team of the Pelfish Project in Jakarta.Some of these data had been analysed and presented elsewhere (Petit ef a/., 1995).Totats of 929 and 912 shoals in Cruises 41 and 34 (rather than the 350 and 143 shoals analysed here) were identified and ex- tracted by performing the MOVIES system with set- ting thresholds defined by the team.However, these differences can be explained as follows.The MOVIES software does not automatieally recognize the shoal (as does MOVIES-B), therefore, the shoals have to be determined by matching the result of visualobservations from the echogram feature (at given thresholds) and the values of deviation (Qd) obtained by playing back the raw data using this software.lf an aggre-Figure 3. Morphological and spatial parameters computed by Movies-B (Diner, 1995) P|of gation as marked in the echogram results in "sharp" deviation, it will be interpeted as a shoal.This manual observation resulted in shoals being identified at thrice the rate of those identifed by MOVIES-B where at- most all of these shoals have Qd factor values more than 0.01 m\f .ln order to clarity the definition of mea- surement unit of Qd, it should be born in mind that the value of Qd is relative, of which the multiplierfactor of 1000 was defined when playing back the data in order to ease scale reading of the echogram.
The numberof shoals derived by this protocol de- creases as the minimum value of ed rises.This simu- lation indicates that the minimum ed value used by this protocol is too low, and the value of 0.05-0.06mV2 gives results similar to those of MOVIES-B.At the Qd minimum criterion, the two softwares do not give the same number of shoals identified.Using MOVIES-B, we introduce a magnitude of 100 m\tr (equivalent to 0.035 mV, in ed) as the minimum en- ergy for a shoal.At minimum ed of 0.04 mW for MOVIES the number of shoals identified is stillgreater than those for MOVIES-B.As these softwares tend to give different results we could not compare the influence of other thresholds on the shoal identifica- tion.A comparison of the measurement values of the two softwares is beyond of the scope of this study.Despite this, the acoustic data can still be used to describe the gerreral distribution of schooling pat- terns.

Large scale distribution
A general comparison of the average shoal size of the two cruises can be expected to explain seasonal distribution of the pelagic shoals.We can see that during the southeast monsoon (October), the number of shoals and theiraverage sizes are biggerthan those during the northwest monsoon (February).As shown by the summary of shoal data and the distribution along the transects, tlre shoals also tend to be closer to each other in the eastern part of the area (Figure 5).Gradients of shoal distribution in the west-east direction are clearly observed in the October cruise, when highly clumped shoal aggregations were frequently found in the easten part.There are three sub_ areas where the shoal clustering formations occur close together with denser biomass; sub-area of Masalembu-Matasiri bank, sub-area extending from Kangean lsland to the near continental edge of eastern part of Matasiri Bank and the coastal area in the North coast of Central Java.
Diurnal migratory behaviour occurred in most of the area covered during the two seasons.tn October, a diel pattern is more apparent than in February.During the dry season (October), the fishes tend to form IFR Joumal Vol. 7 No.l.2A01 smaller shoals and maintain a closer distance during night-time, as indicated by the lower average value of Qd and smaller distances between the shoals.Dur- ing the day{ime, however, bigger shoals were appar- ently found in most of the area, but were less numer- ous than the smaller ones.In February no specific pattern was observed, except for some extreme values of Qd in the eastern part of the survey area (legs n and p) and near coastal area in the north coast of CentralJava during the February cruise.

Distribution in longitude
Forthese analyses the same east-west transects as used in the previous part were chosen (l.e. from   Karimunjawa to Matasiri Bank: legs KB, BM and q with steaming distance of 440 and 331 nautical mile for cruises 41 and 34 respectively) and night and day discriminations were performed.Unfortunately the positions of the night and daytime tracks were not exactly the same and no particular track was covered repeatedly during day and night to know diurnal movements precisely.Generalobservations can be focused on the nearest position by ignoring this difference and combining all times of observations.As shown in Figure 5, different patterns of distribution are exhibited during day and night-time, and diurnalpatterns seems to be apparent, particularly in the Masalembu-Matasiri Bank.During northwest monsoon (in February), a high aggregation of shoals was observed during day and night in the sub-area between Masalembu and Matasiri Bank and the shoals tended to occupy the mid-layer during both the day and night.In the western part, the occurrence of shoals was rare and their sizes were smaller.ln the other season (October), higher densities and numerous shoals were found along the transect with similar patterns of concentration (Le. a tendency of fish to form shoals in the same longitude during the two season).During the day, denser shoals tended to occupy the near bottom layer, while in the night shoals of smaller densities were observed during the October cruise in the eastern sull-area.In Fehruary this variability was not clear, with high density shoals existing in the sub- area of Masalembu-Matasiri Bank allthe time at the same layer.In general, based on the evenness of shoal occurrence and the gradient of aggregation pattern in west-east direction for the two periods of the survey, we can infer that the longitude 1 14' E, or the western slope of Masalembu Bank, is a particular border for a different tyoe of shoal clustering (see also Figure 6).Petit ef a/.(1995;1997) characterized the high density shoals as "oceanic community" and the smaller ones, with a relative low density and an homogeneous geographical distribution, as "pelagic type" shoals.The following analyses are based on measure- ments or classes of descriptors produced by MOV- IES-8, the descriptors being extracted from acoustic data for each shoal identified by the software.This step is aimed to evaluate the structure of the shoals in the context of spatial distribution and aggregation.
We still use the same legs as previously applied for shoal data.In this study, not all of the variables pro- duced by MOVIES-B are used, because of specific correlations between cbrtain variables.For example, the relationship between Qd and Energy are com- pletely linear.We can state that backscattering vol- ume, Rv (in dB) is proportionalto the internaldensity of the shoal, and deviation Qd (in mV'z) is proportional to the "biomass" of the shoal.In this case Rv and Qd/ Area will exhibit a logarithmic relationship.The size of the shoal is expressed as an energetic variable (r.e.biomass measured in deviation, Qd), as well as in geometric variables (r.e.length, height, perimeter and cross sectional area).The position of the shoal is defined as a vertical postion at given latitude or longitude including minimum depth (shoaldepth), al- titude from sea bed (altitude) and relative altitude (percentage of shoal depth to bottom depth).
In these cases, the shoal data derived from the MOVIES-B would be more explainable as these data contain more variables than those from the older version.The internal structure of a shoal can be characterized by the variability of descriptors variables (Scalabrin & Mass6, 1993).
A common ordination technique is to enable the multi-dimensional data structure to be projected in descriptor variables (R space) and in individual shoal space (R space).To do this, we emphasize the vari- ability of the value ratherthan the profile structure or shape of the data.For this reason, principal component analysis (PCA)would be more appropriate than correspondence analysis to summarize the multi-dimensional data and to provide a geometric of low dimensional representation.
Figure 7 shows a two-dimensional plot of the first two principal factors.The variance explained by these components for the cruises 34 and 41 data reveals that the geometric descriptor variables are well represented, l.e. more than 50% of the total variance.High variance of some individual shoals is attributed to high observed value of certain descriptors (particularly the perimeter and length) as indicated in the Rn-space figures.ln term of "distance" between points repre- senting the individual shoals, a grouping "cloud" of points or shoals was evidently observed during the February cruise.ln the same presentation, the data of cruise 34 seem to be more homogenous and the difference of the mean of descriptors between legs of this cruise also indicates a lower fluctuation compared with those of cruise 41 .lnspection of the positions of individual shoals belonging to these groups (during cruise 41) reveals a possible preference of the fish on specific areas (Figure 5).Unfortunately, there is no additional information available to explain this pattern.
However, it is possible to speculate on two factors playing an important role in governing the shoalstructure.The first factor is salinity.Higher salinity water exists in the deep layers in the western part of the area and probably in the eastern area near the continental edge as it is influenced by the waters coming from the Macassar Strait.
The group A and C (as coded as "9" shown in Fig- ure 8)would be composed by the oceanic species.In this, attention can be focused on places where higher concentrations of shoals are observed over relatively short distances (as shown in energetic dimensions (Qd and Rv) generated from the two softwares).The high variation of the values of descriptors between 113' and 114", and 116' and116'30' E in the longitudinal direction transect, and in the south of west leg (l.e. in the coastal subarea of north of Central Java) and in the northeast leg (in the near continental shelf) may not be readily ascertainable.However, many factors may determine this variation, but fishes' responses to sea-water conditions at a small-scale may be exhibited by various types of shoals with particularcharacteristics.An attempt to clarify the distribution can be based on the typology of the shoals.

Shoal typology and stratification
This typology will allow the classification of shoals into strata with homogenous characteristics.Each stratum should contain individualswhich are relatively homogenous with respect to shoal size or morpho' logic dimension variables (deviation, back scatter- ing volume, length, height, perimeter, area of shoals) and posifion dimension variables (altitude, altitude relative to bottom depth, and depth).
Based on these measurements, stratification can be performed by using a statistic (r.e.similarity distance) for defining homogeneity criteria.Descriptive statistical techniques may be performed entphasizing classification, and aimed to show a general clustering of the acoustic shoal to describe the spatial distribution correspond ing to the co-ordinates.
For this purpose, analysis can be performed in two steps: (a) to classify each shoal into strata, and (b)to map a new variable representing these dimen- sions on their correponding positions.(2)R"space Some shoals having these attributes are consid_ ered as out-liers and are excluded from the data in_ put.The algorithm is to examine iteratively all shoal partitions and to select one which is the closest to the centroids being calculated as a moving average of the distance.As shown in Table 1, the size dimen_ sion appears to be the dominant variable in determining this classification.The pre-defined strata tend to correspond to their biomass (ed) and length dimen_ sion, so that the shoals may be categorized as big, medium and sma//.ln this paragraph we use the terminofogy "Upe 1" and "type 2,, to discriminate the different strata derived from cluster analysis without specifying their characteristics. .t rl ,$"r' "l l89q e9 .

Classification
This step is not specific, but means that the clus_ tering will result in a general division of shoals into different strata with respect to the above measure_ ments.In order to clariff further interpretation, we define only three strata of shoals, as more than this number of strata artificially produces very few individuals per group.ln general, computation using the K_means method will produce clusters of greatest possible dis_ tinction based on the Euclidian distance attributed to the variables in Table 1.
However, re-filtering the shoaldata is stillneeded to relegate the extreme and unreasonable values of Qd and the length of shoal due to the inability of the In attempting to observe the pattern.ofshoaltype distribution we focus on the big shoalcategory.Ob- servations on the shoal typology reveal possible zonation generated by similarity of certain shoal characters, even though this evaluation is limited by the lack of the same track observed during both day and night.
In February: there are three zones, one is in the eastern part between 1 15'30'-1 17' E, the second one is in the northern part, North of Masalembu Bank, and the third is in the middle area between 109'-a1 13' Et (Bawean lsland).
In October: the first zone is in the eastern part near the continental slope, the second one is Masalembu Bank and the last one is the area extending from Matasirito Kangean lsland.Karimunjawa Bank and in the middle of the Java Sea (see also Figure 6).The big night-shoals occupied Masalembu and Matasiri Bank, while the medium category shoals seemed to dominate the sub-areas in the North of Kangean lslands.In February, the small types sparsely occupied almost all parts of the area, with only a few big shoals found in Matasiri Bank and in the near slope to the north of the Kangean lslands.

Spatial stratification
In this part, the term of zonation and stratification should not be confused with the patchiness of shoal distribution pattern.These terms are aimed at explaining the different patterns of shoal structure at different locations in the Java Sea.In the previous analysis, description of the change of pattern by sub-areas is based on the real position of the acoustic shoals.In this part, the pattern is described by interpolation of Table 1.Average value of shoal dimension by groups (remarks: the groups defined by K-means method) To'o.t strata snoats Qd RV Atttude Unoal AIIIUOe Lengm Helgnl trlong /{ea rerrrlrer,E (m\F) (dB) (m) depth (m) relatir,e (%) (m) (m) ation (m') r (m)   132 1 0.15 -46.6 11 .7 29.8 17 2 0.29 -51.1 6.3 45.2 6 3 0.66 -52.1 13.8 46 In term of patchiness in the occurrence of big shoals, a tendency towards aggregation is evidently observed-shoals coded as 9 in lhe cloud ofpolnfsA, (Figure  8).In October, big shoals found in the near slopes during the day time, in the North of Madura, in generalized variables by performing ordinary krigging to create graphical presentation.
Since the natural shoal characteristics can be assumed to be represented by those acoustic shoals, mapping the acoustic shoal variables on the actual chart may spatially describe their stratification.lt will, however, face the problem of defining the appropriate variables in conjunction with their position and coordinates.For this reason a canonical correlation analysis will be peformed, which gives a new variable (canonicalscores) representing both sets of variables with a mathematical solution of maximizing the cor- relation between the first variable set (size dimension variables) and the second variable set (position and co-ordinate variables).
In this part we perform two steps of analysis.The first is to determine the important variable that characterize the acoustic shoal and the second to deter-

Size and position variables relationship
Two sets of variables are used in the analysis, namefy a position dimension set composed of longi- tude, latitude, depth, altitude, relative altitude of shoals and bottom depth; and a size dimension set consisting of Qd, Rv, height, length, perimeter and area of shoals, One would expect that the canonical variates would represent combinations of variables.
High canonical correlations between the position dimension and the size dimension set of variables appear to be significant for further interpretation, as ^,r:fl::: these correlations account for a large proportion of the variability in the data (r's are greater than 0.85 for all shoaltypes) (Table 2).However, in this analysis, we use the first two roots for describing the variability.The height of the shoal contributes to the first canonical variate (also to the second one for some shoal types), while the second variates tend to correspond to Qd and Rv.On the other hand, the bottom depth and shoalaltitude significantly explain the variability of the position set of variables.
These variables dorninate the contribution to the total explained variance for the size dimension set of variables.Height of shoal and bottom depth can be called influence variables in determining the factor structure of shoal data, but, it is not necessary to conclude also that these measurements are the most important in describing the shoalstructure.

Zonation
Following Gittins (1986), plotting the canonical scores of the size dimension set of variables onto a chart with respect to their latitude and longitude may describe the nature of the spatial distribution.This procedurewas applied to differenttypes of shoals (Fig-  ure 9) and times of observations (Figure 10), in order to reduce the variation of the data.These plots intu- itively describe the spatial distribution of the shoals with respect to size dimension variables in an analogous way to the zonation of the comunity structure.
As shown in the figures, a possible zonation of shoaltypes generated by size descriptors seems to appearfor all seasons in the Java Sea.This zonation pattern may be sufficient evidence to classify this area according to morphological factors, as well as to the position setof variables.The high contribution of shoal height and back-scattering volume (Rv) to the canonical factor also demonstrate a similar pattern for the presentation.
Three zones are clearly identified, r.e. the western (north of Central Java), the middle (Karimunjawa- Bawean lsland) and the eastern parts (Masalembu, Matasiri, and Kangean lslands).Attention should be paid to the time of observation, as the size dimension variables may vary with time.For example, in October, most of medium shoals were observed during the night-time, while small shoals were equally common by day or night.ln general, this mapping would reflect a subdivi- sion of the Java Sea into three parts, with the middle area frequently represented by a consistent b/a nk area showing few shoals, especially for February.Different times of observation may also obscure the limit of zonation.The next step is to confirm the limit of zo-nation which can be performed by defining a border in between 113-114"E segment (in this case we use 113'30'E) based on the previous findings.Pufting dif- ferent labels for the shoals detected in the western and eastern part of this border, the superimposed plots of the first two components (individual or shoals space on size dimension variable space) reveal a clearer separation of another type of shoal (Figure 1 1).We can then contrast the shoaling patterns of the two sides sub-areas during October, where two groups of points tend to separate and correspond to the eastern part of the border (labeled as "1") and western area.In terms of Euclidian distance, we can infer that "eastern" shoals are characterized by their /engfh di- mension (perimeter, length, height, area), while the "western" shoals tend to be influenced by position dimension (altitude and relative altitude ).ln fact, the mean values of length dimension of the eastern shoals are greater than those of the western ones, as well as their variances (as indicated in the graph by longer vectors of length dimension variables).

Spatial distribution
The spatial distribution evaluated in term of the number of shoals and their size (Qd) per unit distance or segment (for example in one degree) in the eastwest transect gives an indication of different aggregations of shoals in the eastern and western parts of the area.The straight line transect of Karimunjawa-Matasiri Bank is more representative of the direction of fish movement being followed by the fishermen, while the influence of morphological features maybe more important in the sub-area located near to Kalimantan and Java-Madura coast, and in the near continental slope.Detailed inspection on these points of the southnorth transects, reveals a possible high aggregation of schooling composed of coastal and migrant spe- cies in the western part (in the near coastline of Cen- tral Java) and near the continental break in the eastern part (leg p,).These are in accordance with the findings of Petit ef a/. (1995; 1997) who named these aggregations as coasfal and oceanic pelagic com' munities.However, these phenomena do not stand alone and should be seen in relation to biologicalfactors (species composition and size)and variabiity of the Java Sea as induced by seasonal climatic fac- tors.
In February the pattern is not as clear as of Octo_ ber, Following salinity gradients, the oceanic species tend to stay in the western part (Sadhotomo, rnpress), while the neritic species (perhaps the coastal ones too) prefer to occupy the eastern area from Masalembu-Matasiri Banks In this case, we can deduce that some oceanic species may stay in the western part where they have been considered by previous studies as another population entering the Java Sea from the Sunda Strait during the rainy season (Hardenberg,   1938;Widodo, 1988).Until now, there has been no confirmation of the hyphothesis on tlre origin of this stock.
Based on an assunrption that different patterns of behaviour relate to the different species forming the shoal, evaluations can be attempted for different patterns of aggregation in the sub-areas.Shoal clustering in denser concentrations is observed in the eastern part of the Java Sea (included the near continental slope and certain places beyond of the fishing zones) during the nor-thwest and southeast monsoons.
In certain places, such as the Masalembu-Matasiri   Bank, different patterns are clearly displayed during February and October.Also, as indicated by different pattern in diurnalvariability in some areas during the two seasons, we can suggest that different species make up the shoal.We can infer that a $pical oce- anic species would be more likely to undertake mi- gratron following a diurnal rhythmic in high light transmission water, as indicated by the shoal occupy the Masalembu-Matasiri Bank during October, while less structured vertical movements would be exhibited by the non-oceanic species during rainy season in the same place.ln February, this sub-area (the Masalembu- Matasiri Bank) will be favourable for neritic or coastal species and young fishes, as it is of considerably lower salinity in the eastern part of the area surveyed (Sadhotomo & Potier, 1995).Land mass impact on the eastern area as induced by heavy rainfall in Kalimantan during December -January would enrich these waters with nutrients from the coastal area.Also the current (flowing west to east) would not create a clear border of water mass gradient (contrary to those during the southeast monsoon in October).This latter condition causes discontuinities of water mass with some dispersed patchiness in the eastern area of the Java Sea (V$rtki, 1956).A more heterogeneous distribution of species and size naturally exists in this area (Sadhotomo & Potier, 1995), coincidentally, with more variation of migratory behaviour of shoals and target strength distribution are exhibited.Following this phenomenon, it can be suggested that the shoals during the rainy are less structured and connposed of nrore species than those during October.However, the above evaluation is in relation to a smali scale area, Le" the area covered by the cruise tracts.In the larger area, it is interesting to consider migration and its influence on fish distribution in rela- tion to the features of acoustic populations inside the Java Sea.
The migratory scheme relates to the concept of the Java Sea populations.Two types of migrations may be considered, r.e.seasonal and internal migrations.The seasonal migration the movement of a big mass of fish from other areas (the South China Sea, Makasar Strait, and Flores Sea).During the northwest monsoon, the neritic fishes (mainly D. russellii) enterthe Java Sea from the western and northern areas (the South China Sea) as planktonic and nectonic stages (Sadhotomo, 1998) During the southeast monsoon, the direction of migratory activity reverses (from east to west), and the migrants entering the Java Sea during this season are much more itnportant than those of the southeast monsoon.At the beginning of this season, the young fish that previously entered the Java Sea as planktonic stages probably mix with new recruits coming from the eastern archipelago and Makasar Strait.
From the end of the southeast monsoon (r.e.Oc- tober -Novernber)), a reciprocal movement to the east occurs, with the new immigrants coming from the eastern archipelago.This movement is marked by a shift of the aggregations along an gxig from Karimunjawa lsland to Matasiri lsland.These move- ment are defined as an ldenaLrnlllfallon.and it has been briefly discussed elsewhere (Sadhotomo'   lnpress).During the northwest monsoon, this migra- tion is continued, with the fishes moving in three di- rections: to the east (Flores Sea), northeast (southern part of Makasar Strait) and southeast direction.A small part of them migrates to the west as far as the north of Sunda Strait where the salinity of the area is still more than 31%o.Petit ef al. ( 1 997) indicated that the nocturnalbehaviour of the shoals along the north coastof West Java observed in April-May 1995 was different from those off the Central and East Java coast.
The diel nrovement pattern of these shoals was very similar to those of the shoals located in the sub-area of Matasiri Bank in October 1993.Genetic evaluation of D. russe//ii specimens collected in April 1995 (Perrin, 1998) also confirmed this possibility' In previ- ous studies (e.g.Hardenberg, 1938;Widodo, 1988)' this stock was usually regarded originating from the lndian Ocean population entering the Java Sea during northwest season.Consequently, the orientation of the pelagic communi$ distribution extends fiom the middle of the Java Sea to eastern areas.

Figure 1 .
Figure 1.Acoustic cruise tracks made during two seasons

Table 1 .
List of descriptors produced by Movies-B Figure 9. Contour of first canonicalcomponent of size descriptors set (1) and height of shoals (2) of cruise 34 as being superimposed on the chart Tatrle 2. Summary of canonical correlation analysis by types of shoal Biplot of the first two components derived from PCA (Remarks: 1 's and 0's denote shoals occu- pying East and West parts of 113"30' E border, respectively, dotted-circle lines are fitted by eye to indicate grouping of points)