208 
Fishery Bulletin 111(3) 
tom temperature was sampled, it was generally within 
0.7°C of sub-mixed-layer temperature (depth=20 m) 
for March, April, and May. We focused on data from 
these months because they include the time periods 
immediately before bottom trawl sampling and, there- 
fore, could best indicate changes in environmental 
conditions that might affect catch rates. Moore et al. 
(2008) demonstrated strong intra-annual coherence of 
oceanographic properties within Puget Sound basins; 
therefore these data are likely representative of intra- 
annual environmental conditions throughout the cen- 
tral Puget Sound basin. 
Analysis 
For most years, all 20 depthxtime combinations were 
successfully sampled, but gear malfunction and other 
events resulted in missing sets for some sampling 
sites. These missing sets constituted only 5% of the to- 
tal sample design, but we wanted to account for them 
in deriving annual catch levels. We first ascertained 
whether these differences can alter annual estimates 
of catch rates by fitting an analysis of variance (ANO- 
VA) for each of our study species with year, depth, 
time, and a depthxtime interaction term. All but one 
species, the Shiner Perch ( Cymatogaster aggregata), 
showed either a significant effect of depth, depth+time, 
or a depthxtime interaction term. 
We used a simple approach to account for the small 
numbers of missing sets. Rather than fitting general- 
ized linear models to calculate a statistical “year ef- 
fect,” we instead calculated an annual average catch 
anomaly for each year on the basis of expected catches 
for each timexdepth combination. This approach is 
equivalent to fitting a generalized linear model with a 
time+depth+timexdepth interaction term, but it has a 
straightforward interpretation and permitted a paral- 
lel calculation for the trawl and environmental data. 
We calculated the mean catch rate (number of fish/tow) 
for each depthxtime combination for each species with 
data from the entire sampling period. We then calcu- 
lated the catch anomaly as the difference between ob- 
served species-specific catch and the expected (mean) 
catch rate given the depth and time of sampling. The 
annual abundance index for each species was equal to 
the average catch anomaly over all samples conducted 
within a year. We used the same approach to gener- 
ate temperature and salinity anomalies for each year. 
For each month and monitoring site, we calculated the 
mean temperature and salinity values from all avail- 
able data, generated anomalies for each year, month, 
and site, and averaged these across months to derive a 
yearly anomaly value. 
We generally tracked abundances at the species 
level, but, in some cases, we aggregated closely relat- 
ed species. Rock soles were allocated to a single spe- 
cies when the survey began, but subsequent genetic 
work indicated that the rock sole genus ( Lepidopsetta ) 
consists of 3 species (Orr and Matarese, 2000), 2 of 
which occur in Puget Sound: Rock Sole ( Lepidopsetta 
bilineata ) and Northern Rock Sole ( Lepidopsetta po- 
ly xystra). We conducted our analysis at the scale of an 
aggregated species group because the 2 Puget Sound 
species are not readily distinguished in the field and 
we wanted to maintain consistency throughout the 
time series. Further, Speckled Sanddab (Citharichthys 
stigmaeus ) and Pacific Sanddab ( Citharichthys sordi- 
dus) are morphologically similar as juveniles; for this 
reason, species-level identifications were not reliable. 
We, therefore, combined all individuals identified as 
either species into a species group termed “sanddab; 
( Citha ri ch thys ) . ” 
We focused analysis on the most common species 
and species groups encountered with the sampling gear 
so that we had sufficient statistical power to detect 
changes in abundance through time. We set an arbi- 
trary threshold of 200 sampled individuals over the en- 
tire time period for species to be included in the analy- 
sis. This use of a threshold eliminated species so rarely 
encountered that trends would not be reliable, species 
for which the gear was not appropriate, and samples 
for which species identity could not be determined (e.g., 
samples in very early juvenile stages). For each species, 
we asked whether abundance changed through time, 
and, if so, whether it was best described by a continu- 
ous linear increase or decrease or a discontinuous shift 
in the mean catch rate. The latter is consistent with re- 
gime shifts as reflected by rapid and persistent chang- 
es in population densities (Rodionov and Overland, 
2005). For each time series, we used Akaike’s informa- 
tion criteria adjusted for a small sample size (AIC c ) to 
choose between 3 models: constant, linear, or change 
point. For each model, we assumed normally distrib- 
uted residuals. We used the changepoint package (vers. 
0.6; Killick and Eckley, 2011) in R software (vers. 2.13; 
R Development Core Team, 2011) to assess discontinu- 
ous shifts in the mean catch rate. We required that the 
best fitting change-point model consist of time periods 
spanning at least 4 years of data. In other words, es- 
timated change points that broke the time series into 
increments shorter than 4 years were discarded, thus 
preventing the model from placing change points at the 
beginning or end of time series. 
Because we found evidence of change points for 
many flatfishes, we explored the data for flatfish spe- 
cies in more detail. The gear captures individuals 
across a wide size range and range of life history stag- 
es; therefore we evaluated whether changes in catch 
rate could be attributed to changes in recruitment pat- 
terns. If changes in densities were driven by changes 
in recruitment, we would expect to see time trends 
of abundance for small size classes to lead trends for 
larger size classes. For each flatfish species, we calcu- 
lated catch anomalies separately for small and large 
size classes (individuals below the 33 rd percentile and 
above the 66 th percentile of the cumulative length- 
frequency distribution, respectively). Size-at-age data 
are not available for most species, but for English 
