432 
Fishery Bulletin 110(4) 
Multivariate analyses were based on pairwise Bray- 
Curtis similarity coefficients calculated between hauls 
or cruises. Bray-Curtis similarity coefficients are widely 
used in ecological studies because they are unaffected 
by changes in scale (e.g., with percentage or propor- 
tions) or the number of variables (e.g., species or hauls) 
used and produce a value of zero when both values be- 
ing compared are zero (joint absence problem) (Clarke, 
1993; Legendre and Legendre, 1998). In this applica- 
tion, similarity coefficients ranged from 0 (no catches 
in common) to 1 (identical catches). The fish assem- 
blage used in multivariate analyses was restricted to 
13 species-and-age classes of fish that were effectively 
captured by the purse seine because they were pelagic. 
We deliberately excluded demersal species from our 
analyses (e.g., flat fishes [Pleuronectidae], gunnels 
[Pholidae], and sculpins [Cottidae]) because they were 
unlikely to be effectively sampled with a purse seine 
(Bottom and Jones, 1990). The 13 species included the 6 
species-and-age classes of juvenile salmon, plus Ameri- 
can shad (Alosa sapidissima), longfin smelt ( Spirinchus 
thaleichthys), northern anchovy ( Engraulis mordax). 
Pacific herring ( Clupea pallasii), shiner perch ( Cyma - 
togaster aggregated), surf smelt ( Hypomesus pretiosus), 
and threespine stickleback ( Gasterosteus aculeatus). 
In these analyses, abundances of each species-and-age 
class in the 210 round hauls were transformed by using 
log(x+l). All species are native to the Columbia River 
and Pacific Northwest, with the exception of American 
shad (Hart, 1973; Eschmeyer et al., 1983; Hasselman 
et ah, 2012a). 
We compared pairwise Bray-Curtis similarity coeffi- 
cients calculated among subsets of the 210 round hauls 
to explore fine-scale spatial and temporal variation in 
catches. Specifically, we examined spatial variation by 
comparing pairwise similarities among catches at the 2 
stations, correcting for tidal stage (i.e., hauls occurred 
within 1 h of each other, where time 0 is at low tide) 
within the same cruise; differences were tested with the 
Mann-Whitney (MW) test for difference in medians. We 
explored fine-scale temporal variation by testing Bray- 
Curtis similarity coefficients among hauls made at the 
same station and within the same cruise but grouped by 
tidal stage at 1-h increments (i.e., within 1 h of low tide, 
2 h, etc.). We used the Kruskal-Wallis (KW) one-way 
ANOVA on ranks followed by KW multiple-comparison 
test to determine which groups were different from the 
others (Zar, 1984). 
We also used the matrix of Bray-Curtis similarity 
coefficients calculated either among hauls or cruises 
to construct MDS plots to graphically explore varia- 
tion in fish assemblage structure at the 2 scales (hauls 
or cruises). The MDS ordination technique places all 
points in MDS space in relation to their similarity (i.e., 
points farther apart in MDS space are less similar 
than those points closer together). In all MDS analy- 
ses, random starting locations were used for each of 
25 iterations to find the best solution; minimum stress 
was attained in multiple iterations which suggest a true 
minimum solution. Stress values of <0.20 indicate that 
spatial representation of data by the MDS plot is consis- 
tent with the structure of the original data set (Clarke 
and Corley, 2006). Finally, we quantitatively evaluated 
temporal variation in assemblage composition by either 
haul or cruise, using year and biweek (where biweek 
l=April 15-30, 2= May 1-15, etc.) for both analyses 
and Julian date and time after low tide for the matrix 
based on hauls. For this analysis, we used ANOSIM, 
which produces Global R values that indicate the degree 
of separation of groups generated by a particular factor 
(or pair of factors). These Global R values range from 0 
(no separation) to 1 (complete separation); the program 
also generates statistical probabilities by permutation. 
Environmental variation We evaluated the response of 
the pelagic fish assemblage to environmental variation 
at 2 scales: haul and cruise. For the former (haul), we 
used both local (in situ temperature and salinity mea- 
sured at depths of 1 and 7 m, and tidal stage and height) 
and regional (river flow and temperature, SST, SLH, UI, 
PDO, and NPGO) environmental parameters, whereas 
for the latter (cruise) we used only regional environ- 
mental parameters. For both analyses, environmental 
parameters were normalized, and then Euclidean dis- 
tances between hauls or cruises were calculated in envi- 
ronmental multivariate space. The relative difference 
between environmental conditions for each haul or cruise 
was then compared to haul- or cruise-specific Bray- 
Curtis similarity coefficients for the fish community (see 
above) with the BEST function in PRIMER-E (Clarke 
and Gorley, 2006). This function creates matrices from 
different combinations of environmental variables and 
then compares the order (rank) among the fish assem- 
blage with the environmental matrices to determine 
the environmental matrix with the highest correlation; 
statistical significance is estimated by permutation. 
Size comparisons We compared size distributions of 
fishes for 2 reasons: 1) to determine which fish species 
were similar in size to juvenile salmon and, therefore, 
might serve as alternative prey for salmon predators, 
and 2) to examine seasonal and interannual variation 
in juvenile salmon size. We did not compare the sizes 
of nonsalmonids between years because measurements 
were made on fish of multiple, undetermined ages; there- 
fore, interpretation of potentially detectable size differ- 
ences in length would be complex (i.e., could result from 
changes in growth or age composition). 
We evaluated the length of each juvenile salmon spe- 
cies-and-age class in 4 ways: 1) among years, 2) across 
the season (by date), 3) among years with weight as a 
covariate, and 4) between marked (known hatchery) and 
unmarked (hatchery and wild) fish by cruise. Length 
and weight data were transformed with ln(x+l) in all 
comparisons. We used one-way analysis of variance 
(ANOVA) to evaluate interannual variability in mean 
length among years, and two-way ANOVA to exam- 
ine size differences of clipped (hatchery) or unclipped 
(hatchery+wild) fish by cruise (Zar, 1984). We used 
analysis of covariance (ANCOVA) to evaluate seasonal 
