208 
Fishery Bulletin 1 10(2) 
Recruitment versus PDO indices 
Stock assessments for 24 of the 62 species in the analyses 
have been published by the PFMC since 2005 (available 
at http://www.pcouncil.org/groundfish/stock-assess- 
ments/safe-documents/2011-safe-document/, accessed 
September 2011) and provide information on the number 
of recruits by year. We examined stock assessments for 
each species to determine whether strong recruitment 
events occurred during the period from the mid to late 
1990s through 2002. Annual recruitment strength is 
generally modeled in the assessments as random devia- 
tions about a stock- recruitment (S-R) relationship. These 
deviations and the central tendency of the S-R curve are 
informed by all other sources of available information 
(i.e., observed lengths, weights, age, and trend informa- 
tion from fishery-dependent and independent sources) 
and will reflect predation intensity, climate, and other 
influences (Methot, 2011). For this analysis, we defined 
strong recruitment as 1.7-5 times greater than the 
average recruitment during the 10 to 14 years before 
the most recent assessment. 
We subsequently subdivided the 62 species included 
in our study into three groups: those with strong re- 
cruitment during the late 1990s, those without strong 
recruitment during this period, and those with un- 
known recruitment levels. We summed the biomass 
indices for all species within each group and the overall 
biomass indices for all three groups and regressed these 
summed values versus year. We reasoned that declining 
biomass indices would be more tightly tied to time for 
those species with elevated recruitment as the resulting 
exceptionally strong cohorts declined due to natural and 
fishing-induced mortality in the early 2000s. 
Biomass indices for the aggregated subgroups and 
overall were also compared with the PDO, a widely 
used index of climate variability for the California Cur- 
rent system. The PDO is an index based on patterns 
of variation in sea surface temperature of the North 
Pacific from 1900 to the present (Mantua et al., 1997; 
Schwing et ah, 2009). Although derived from sea sur- 
face temperature data, the PDO index is well correlated 
with other environment factors, including sea level pres- 
sure, winter air temperature, wind shear, and precipi- 
tation, as well as other Pacific climate indices (ENSO 
[El Nino-Southern Oscillation] and MEI [multivariate 
ENSO index]). For comparison with the annual survey 
data, monthly PDO values were averaged annually 
(November to October) to include the survey period each 
year (Mantua 2 ). 
Species richness 
Coast-wide estimates of species richness were calcu- 
lated as area-weighted mean number of fish species 
taken per trawl sample. Estimates were stratified by 
survey year (2003-10), depth (55-183 m, 184-549 m. 
2 Mantua, N. 2010. Personal commun. Dep. Atmospheric 
Sciences, Univ. Washington, Seattle, WA 98195. 
and 550-1280 m), and geographic region (one degree 
latitudinal increments from 32° to 49°N) for all fish spe- 
cies. Estimates were built upon the number of distinct 
fish species reported for each trawl sample. Mean species 
counts were determined for each stratum and weighted 
by the proportion of stratum area within the total area. 
Annual species richness estimates were computed as 
the sum of these area-weighted species counts within 
the area of interest (per depth range or coast-wide) and 
survey year. Species richness variance within each area 
was similarly estimated as the sum of stratum vari- 
ances weighted by their associated squared proportion 
of stratum area within the total area. Standard errors of 
the mean were computed as the square root of the ratio 
of the variance estimate to the stratum count for each 
area (i.e., within a specific depth stratum or coast-wide). 
We compared species richness over time by regressing 
against year and also evaluated the relationship between 
species richness and the annual Pacific Decadal Oscilla- 
tion (PDO) index. In both cases, regression analyses by 
depth strata and overall depth were undertaken 
Statistical analyses 
For the biomass data we examined individual species, 
and present for comparison several aggregate groups 
formed by summing species coast-wide biomass indices 
(metric tons, t). For each species, regression analy- 
sis was used to initially investigate the relationship 
between annual biomass indices and year. To account 
for the large number of tests conducted, a sequential 
Bonferroni correction with a significance level of 0.05 
was applied to the data (Peres-Neto, 1999). Grouping 
data for later analyses (initially by depth or taxonomic 
group to examine trends over time for aggregated data 
and subsequently by the presence, absence, or unknown 
occurrence of exceptionally large year classes after 
recruitment) resulted in fewer tests and no Bonferroni 
correction was applied. Results for biomass and species 
richness were statistically compared with year and the 
PDO index by linear and multiple regression (GLM) 
by using SAS for Windows (SAS Institute, Inc., Cary, 
North Carolina). To stabilize the variance, the natural 
logarithm of the response variable was used in the 
regression models; however, even after the transforma- 
tion, annual variance estimates were highly variable 
for some species. Regressions weighted by the variance 
estimate of the annual values were therefore used to 
examine interactions between annual biomass indices 
and species richness versus year, PDO values, or both 
(Draper and Smith, 1981). 
The Akaike information criterion (AIC) was used to 
choose between competing models (i.e., recruitment, 
environmental variability or both) when comparing 
biomass values, summed by groups, versus year and the 
PDO index (Sakamoto et al., 1986). For each group, the 
best model was selected on the basis of the smallest AIC 
value ( AlC min ). A similar comparison was done between 
species richness versus year, the PDO index, and both 
year and the PDO index. To determine whether a model 
