270 
Fishery Bulletin 109(3) 
Interannual hydrographic pattern The ordination of 
hydrographic variables averaged broadly by region, year, 
and season (Fig. 2B) showed clear separation of samples 
by region, but failed to show evidence of other groupings 
due to years or seasons at this scale of resolution. Unlike 
the corresponding MDS community plot, the years 2004 
and 2005 did not occupy a distinct quadrant of the PCA 
environmental plot. 
Regional hydrographic pattern Stations in summer 
cruises did not form clearly distinct regional groups 
based on PCA (Fig. 3E). Rather, GF stations overlapped 
broadly with those of the NC, and weak regional separa- 
tion was apparent on a gradient described mostly by PCI 
axis. Most GF stations fell on the right of the first axis 
(characterized by higher TMP, lower SAL, and lower 
DEN) and most NC stations fell on the left (character- 
ized by the opposite conditions). The trio of eastern 
nearshore GF stations (23, 24, and 25) was again placed 
together and was characterized by higher CHL and lower 
DIS, DEP, and PAR than other summer stations. Sta- 
tions in fall cruises (Fig. 3F) showed greater regional 
separation than summer stations. This division also 
fell along an environmental gradient captured mostly 
by the first axis, with the GF characterized by higher 
TMP, higher CHL, lower DEN, and lower SAL, and the 
NC characterized by the opposite conditions. The finding 
that water properties within the two regions were more 
dissimilar in fall than in summer was consistent with 
PERMANOVA pairwise tests. One GF station (number 
24) and two NC stations were excluded from the fall 
analysis because of incomplete CTD data sets. 
Seasonal hydrographic pattern Summer and fall sam- 
ples along the NC overlapped broadly but on average 
occupied mostly different sides of their PCA plot (Fig. 
3G) and separated primarily along a gradient described 
by PCI; most summer samples were characterized by 
higher SAL and DEN and lower TMP than fall samples. 
Summer and fall samples in the GF (Fig. 3H) showed 
greater separation on the PCI gradient, composed of 
roughly equal parts DEN, SAL, and TMP and capturing 
most of the seasonal variance. Summer samples in both 
regions were more variable than fall samples. 
Univariate pattern ANOVA tests showed that water 
properties varied significantly between regions for six 
of the seven variables and between seasons for four 
variables (Fig. 4). TMP increased from north to south 
and from summer to fall, indicating stronger coastal 
upwelling in summer and north of Pt. Reyes. CHL also 
increased significantly from north to south, but the 
seasonal pattern was not consistent in the two regions 
and summer and fall differences in CHL were not signifi- 
cant. SAL and DEN were positively correlated (r 2 =0.97, 
PcO.001) and showed similar patterns, with higher 
values along the NC and a significant decrease from 
summer to fall in both regions. PAR was the only vari- 
able not significantly different between regions, but it 
was significantly higher in summer than in fall. DEP 
and DIS varied regionally but not seasonally. On aver- 
age, GF stations were shallower and farther from shore, 
reflecting the broader shelf in the GF. The interaction 
between region and season was not significant for any 
of the variables, although it was close for TMP (P=0.06) 
and PAR (P=0.08). 
Relationship of environment to community structure 
The ordination of samples based on species similarity 
(Fig. 2 A) was structurally related to the ordination of 
samples based on environmental similarity (Fig. 2B), 
as determined by direct comparison of their underlying 
similarity matrices with the BIO-ENY routine. Using 
first the full set of seven environmental variables with 
the BIO-ENV protocol, we found significant similarity 
of multivariate pattern between the biotic and the envi- 
ronmental data (p s =0.439, P= 0.001). However, a reduced 
subset of these seven variables generated an improved 
match with the community pattern. The solution that 
maximized the Spearman rank correlation between the 
two resemblance matrices was a four-variable combina- 
tion of DEP, DIS, TMP, and SAL (p s =0.471, P=0.001) 
that performed slightly better than the full seven-vari- 
able comparison. Closely following this four-variable 
solution were three three-variable combinations (DEP, 
DIS, DEN; DEP, DIS, TMP; DEP, DIS, SAL) that gave 
p s -values only slightly smaller than the four-variable 
combination. These three are arguably the best solu- 
tions because they achieve essentially the same level of 
correlation with one less variable. 
Discussion 
Abundance data from a six-year survey of coastal marine 
fishes captured in surface trawls revealed significant dif- 
ferences in community structure based on region, year, 
and season. These patterns were mirrored by differences 
in a small suite of mostly physical oceanographic vari- 
ables collected along with the biotic samples, indicat- 
ing that epipelagic fish communities were responding 
to interannual, seasonal, and relatively small-scale 
spatial variability in oceanography. Multivariate ordi- 
nation placed samples with similar fish communities 
in arrangements that corresponded with a boundary 
somewhere in the vicinity of the headland at Pt. Reyes. 
GF stations were more dispersed and variable in ordina- 
tions than NC stations, and the plotted arrangement of 
several GF stations appeared to be influenced by their 
common proximity to features such as the San Fran- 
cisco Bay outflow and possibly the Farallon Islands. 
Differences in community structure were due primarily 
to differences in relative abundance and occurrence of 
about 6-12 ubiquitous species, most of which were regu- 
larly caught in both regions and during both seasons of 
the study. Thus, community patterns were not driven 
by abrupt turnover of dominant taxa across regional 
boundaries or between seasons, but rather by gradients 
in local abundance. Most of the dominant species identi- 
