310 



Fishery Bulletin 100(2) 



Marine surveys typically yield data sets that are highly 

 variable and contain a substantial proportion of zero 

 catches, particularly when the data set is broken down 

 into age groups (Stefansson, 1996). Thus, in some cases 

 the data were analyzed in terms of presence and absence 

 of pollock (hereafter termed "occurrence"). Because non- 

 zero data often follow a lognormal distribution (Penning- 

 ton, 1996), the lognormal transformations (log(CPUE-i-l)) 

 of nonzero density data were used during our analyses. 



Changes in the depth distribution of pollock were an- 

 alyzed by lOG-m bottom depth intei-vals, corresponding 

 to the depth stratification used in the triennial surveys. 

 Changes in geogi-aphic distribution of CPUE were eval- 

 uated by GOA area boundaries, which again correspond- 

 ed to the stratification design used during the triennial 

 sui-veys. In our study, three areas were taken into con- 

 sideration: "Shumagin" ( 1.59°-170°W, area 610), "Chirikof" 

 (154°-159°W, area 620), and "Kodiak" ( 147°- 154° W, area 

 630). 



The distribution of pollock in relation to distance from 

 land was evaluated in 20 nautical mile (nnii) increments 

 by using GIS software (ESRI, Inc., 1996). However, the 

 20-nmi increments crossed the triennial sui-vey strata 

 boundaries, which may have resulted in the disruption 

 of the stratified random sampling scheme. By addressing 

 pollock occurrence or density at each station, instead of 

 biomass, we avoided the need for stratifying the data after 

 collection. Pollock usually reside at depths less than 300 m 

 within the shelf and slope regions of the GOA (NPFMC"'). 

 Thus, the relatively narrow shelf region would limit the 

 offshore distribution of pollock. 



Initial exploratory analyses involved three-dimensional 

 contingency tables to assess the relationship between the 

 occurrence of pollock and year, bottom depth, geographic 

 region, and distance from land. In cases where the three- 

 dimensional analyses rejected the null hypothesis of inde- 

 pendence, more detailed partial contingency tables were 

 constructed for the occurrence of pollock against each in- 

 dividual variable. Othei' studies suggested that bottom 

 depth and geographic location explain most of the vari- 

 ability in fish distribution data (e.g. Overholtz and Tyler, 

 1985; Jav, 1996). 



^ NPFMC (North Pacific Fishery Management Council). 1998. 

 Essential fish habitat assessment report for the groundfish 

 resources of the Gulf of Alaska region. NPFMC, 60.5 West 4"' 

 Avenue, Suite 306, Anchorage, AK 99.501, 117 p. 



Single-factor ANOVAs were performed on the nonzero 

 CPUE data to examine the sources of variation in the da- 

 ta. The In(CPUE-i-l) of the nonzero data was used as the 

 dependent variable among years. A separate ANOVA was 

 calculated for each category of bottom depth, geographic 

 region, and distance from land. 



Examination of the population density data revealed 

 that for most age groups, the number of stations with a 

 low density (<1000 fisli/km-) of pollock increased whereas 

 the number of high-density stations remained relatively 

 stable. This finding suggested that the distribution of 

 suitable habitat for the demersal fraction of the popula- 

 tion might have been expanding. Therefore, we examined 

 the distribution of low-density concentrations of pollock 

 separately in relation to the three physical character- 

 istics across the survey years. Although ocean bottom 

 temperature (OBT) was not measured at every survey 

 station, available data were used to help interpret our re- 

 sults. Mean bottom temperature was averaged over bot- 

 tom depth intervals for each year to provide a general 

 ovei'view (Fig. 3). Additional information on OBT has been 

 summarized by S. Hare (http://www.iphc.washington.edu/ 

 staff/hare/html/papers/OBT/obt.html ) from a variety of da- 

 ta sources and averaged over five-year periods. 



Interspecies associations were examined with Bray- 

 Curtis clustering techniques (Boesch, 1977; Walters and 

 McPhail, 1982). Such classification techniques are useful 

 in generating hypotheses about community structure, 

 which may then be used to aid management actions (Cor- 

 mack, 1971). Because the surveys used in the cluster 

 analyses were conducted during the summer months, the 

 results could not be extrapolated to other seasons. Trien- 

 nial survey data were log-transformed and clustered by 

 using the group average fusion strategy. For the species 

 clusters, the top thirty species by weight were chosen in 

 addition to the four age groups of pollock. After careful 

 evaluation of all the cluster dendrograms, a common dis- 

 similarity level (dissimilarity coefficient (A)=27) was cho- 

 sen for all years as the level at which the most clearly 

 defined clusters occurred throughout the triennial survey 

 years. 



Diversity indices were calculated for all five sui-vey 

 years. The top ten species by number of fish were com- 

 bined for all years to make one list of species for which 

 diversity was analyzed. Simpson's diversity index was cal- 

 culated both in terms of richness and evenness (Simpson, 

 1949; Tokeshi, 1993). Richness was interpreted to mean 

 "effective number of species," whereas evenness was un- 



