224 
Fishery Bulletin 1 14(2) 
Table 1 
Number of tows per water mass 
for each of 5 
cruises 
(III— VII) of the Program ECOSAR (translates from Por- 
tuguese as 
‘Prospecting to investigate sardine biomass 
by acoustic methods”). The water 
masses 
included the 
South Atlantic Central Water (SACW), Coastal Water 
(CW), and Mixed Water (M). 
Cruise 
Water mass 
III IV 
V 
VI 
VII 
SACW 
5 9 
10 
1 
5 
CW 
6 3 
0 
5 
1 
M 
4 8 
2 
11 
8 
Total 
15 20 
12 
17 
14 
masses (Table 1) because of the different selections of 
samples, operational constraints, weather conditions, 
and other aims of the surveys. Nevertheless, fish sam- 
pling covered the entire length of the SBB in most of 
the cruises (Fig. 1). 
The water mass, either SACW or CW, in which each 
tow was carried out was identified on the basis of the 
mean temperature and salinity recorded at the mean 
depth of tow. Any record of salinity or temperature 
outside the usual bounds for these water masses may 
be indicative of mixing processes among SACW, CW, 
or Tropical Water; therefore, the water mass for such 
records was classified as M. 
Species composition and abundance The frequency of 
abundance (biomass) for each species for each cruise 
(% B), total frequency of abundance for all cruises 
(%B t ), and total frequency of occurrence (% FO T ; i.e., 
the number of occurrences of the species in relation to 
the total 78 tows) was calculated to infer the charac- 
teristics of the structure of the fish aggregations and 
assemblages. 
The effect of water masses on the fish assemblage structure 
After we assigned each of the tows to one of the water 
masses (CW, SACW, and M), the following procedures 
were performed to construct the response biomass ma- 
trix: 1) species occurring in <3% of tows were removed 
from the data set because they may have added noise 
rather than information to the statistical solutions 
(Legendre and Legendre, 1998) and 2) the abundance 
data were first transformed in order to scale reduc- 
tion due to high species-specific abundance variability 
among fish samples. As a result, the resulting matrix 
contained 78 tows and 36 species. 
A nonparametric, permutational multivariate analy- 
sis of variance (PERMANOVA; Anderson et al., 2008) 
was used to test for differences in fish assemblages 
among water masses (factor fixed, with 3 levels: CW, 
SACW, M) and among the cruises (factor random, with 
5 levels: ECOSAR III, IV, V, VI, VII cruises). In brief, 
PERMANOVA (analogous to multivariate analysis of 
variance [MANOVA]) is a routine that calculates the 
pseudo-F statistic for testing the response of variables 
(the response biomass similarity matrix) to factors in 
an analysis of variance experimental design on the ba- 
sis of any resemblance measure. Unlike tests where 
multinormality is assumed (e.g., MANOVA), PER- 
MANOVA obtains the significance of the test statistic 
by permutation. Therefore, this routine was quite suit- 
able for our data that were not normally distributed 
(Anderson et al., 2008). 
Because our design was unbalanced (Table 1), we 
used type-I sums of squares to conduct the partition- 
ing the variability of the total species data. A potential 
problem in an unbalanced design is that the order in 
which the terms of the main PERMANOVA model are 
inserted into the analysis may affect the result because 
the terms are not independent of one another. There- 
fore, we changed the order of main-effect terms and 
verified how this different change in order affected the 
results (Anderson et al., 2008). Terms of the main mod- 
el with negative components of variation were taken 
as probably having zero variance and, for that reason, 
were eliminated from the analyses and the data were 
re-analyzed (Fletcher and Underwood, 2002). A post- 
hoc permutational £-test from the PERMANOVA rou- 
tine was then applied to compare levels of the fixed 
factor (water mass) when it was significant. 
The method of PERMANOVA is sensitive to dif- 
ferences in within-group dispersions. A significant re- 
sult for a given factor for PERMANOVA could signify 
that the groups differ in their location in multivari- 
ate space, in their dispersion in it, or a combination of 
the 2 (Anderson et al., 2008). Therefore, we used the 
permutational analysis of multivariate homogeneity 
of dispersions (PERMDISP), which is a distance-based 
test of homogeneity of multivariate within-group dis- 
persions among groups of a single factor. A significant 
result of PERMANOVA and a nonsignificant effect of 
PERMDISP indicate that there is a genuine difference 
in location among the groups in the multivariate space 
(i.e., that there is a significant effect of factor). The 
PERMANOVA and PERMDISP routines were based on 
the Bray-Curtis similarity index and their significance 
on 9999 permutations. 
Similarity percentages analysis (SIMPER) was ap- 
plied with a cumulative contribution cutoff level of 90% 
to determine which species contributed to differences 
in fish assemblage structure among the water masses 
(Clarke and Warwick, 2001). 
These analyses were performed with the software 
PRIMER 6, vers. 6.1.11, with the add-on package PER- 
MANOVA+, vers. 1.0.1 (PRIMER-E Ltd., Ivybridge, 
UK). 
The effect of environmental and spatial factors on the fish 
assemblage structure We assessed the relationship be- 
tween the variability in the fish assemblage structure 
and the oceanographic and spatial variables (Table 2). 
This analysis was based on 2 matrices. The first matrix 
