838 



Fishery Bulletin 100(4) 



NOSHDB contains data digitized from hydrographic sur- 

 veys completed from 1930 to 1965 and from survey data 

 acquired digitally on NOS survey vessels since 1965. The 

 total amount of habitat by depth and latitude, as well as 

 the location of the shelf break, was determined by analysis 

 of the NOSHDB. 



Because the NOSHDB contains a vast amount of data, 

 we computed depth profiles only at each 0.5 intei-val of 

 latitude. Profiles were obtained by contouring a narrow 

 (0.05°=5.56 km) swath of depth soundings by using in- 

 verse distance to a power for interpolations (Surfer, 1995). 

 After correcting for latitudinal differences in the relation- 

 ship between longitude and distance, the resulting depth 

 profiles were used to estimate the total amount of habitat 

 [km] in 50-m depth intervals. In addition, the distance 

 offshore and the depth of the continental shelf break were 

 estimated. The location of the shelf break was estimated 

 by using a three-parameter segmented linear model that 

 minimized the sums of squared differences between a con- 

 toured depth profile and points along two linear sections 

 of a segmented line. The fitted join point of the two seg- 

 ments was then used to estimate the location of the shelf 

 break. In the estimation procedure, the offshore end of the 

 offshore segment was fixed at the exact 500 m depth value 

 obtained from the computed profiles. 



Information on rockfish abundance was obtained from 

 trawl survey data collected by the Resource Assessment 

 and Conservation Engineering division of the AFSC. 

 Trawl samples were generally collected by using a sam- 

 pling design that was stratified by depth and latitude, and 

 where allocation of sample sizes was based on prior fish- 

 ery catches (Wilkins et al., 1998). From 1977 to 1998, trawl 

 samples in the Eureka, Monterey, and Conception INPFC 

 areas (Fig. 1) were typically taken between 50 and 500 m 

 from June to August with a standardized Nor'eastern 

 high-opening rockfish bottom trawl rigged with roller 

 gear. Measurements recorded for each trawl sample were 

 the following: trawl net width and height; time of the tow; 

 distance traveled; and the number and weight of species in 

 the catch (Wilkins et al., 1998). 



The fundamental objective of the AFSC triennial conti- 

 nental shelf trawl survey is to estimate the distribution and 

 abundance of fishes vulnerable to capture by bottom trawl 

 along the U. S. west coast. This basic goal has not changed 

 since the first year of the survey in 1977, although specific 

 objectives have changed over time, which has resulted 

 in alterations in the distribution of sampling effort. For 

 example, sampling effort in 1977 was stratified by depth 

 and latitude according to rockfish fishery information. 

 Sampling efforts in 1980, 1983, and 1986 were shifted to 

 improve biomass estimates of canary and yellowtail rock- 

 fish. However, the lack of any significant improvement in 

 the precision of the rockfish biomass estimates prompted a 

 shift in the 1989 and 1992 surveys to include all demersal 

 groundfish and to improve estimates of Pacific hake (Mer- 

 hicciiis prociitctus) and juvenile sablefish (Anoplopoma 

 fimbria) abundance. More recently in 1995 and 1998 the 

 survey was expanded to include slope rockfish found in 

 deeper waters (to 500 m) with an emphasis on obtaining a 

 uniform sampling density. These changes in the goals and 



objectives of the AFSC shelf survey significantly altered 

 the data on the spatial distribution of samples over time, 

 which, in turn, confounded interannual comparisons of the 

 spatial distributions of rockfish species. 



Due to changes in survey design detailed above, all 

 years of the survey were simply pooled into a single com- 

 posite data set, from which the starting position of each 

 haul, depth of haul (m|, net width (m), distance towed 

 (km), numbers of species collected, and species weights 

 (kg) were extracted for analysis. The distribution of all 

 hauls was analyzed by depth and latitude to reveal any 

 patterns that might affect inferences about rockfish dis- 

 tributions or co-occurrences. Because management is pri- 

 marily concerned with biomass estimates of abundance, 

 only species weights were used in our analysis. All trawl- 

 specific species weight measurements were converted to 

 a catch-per-unit-of-effort (CPUE) statistic by dividing 

 species catch weight by the product of the distance towed 

 and net width, i.e. the area swept (ha). An analysis of the 

 frequency of occurrence of each species in trawls was con- 

 ducted to obtain a subset of the most ubiquitous rockfishes 

 for use in all subsequent analyses. These species were se- 

 lected based on their occurrence in at least six of the eight 

 survey years, with the exception of halfbanded rockfish (S. 

 semtcinctus), which was included because it yielded mod- 

 erately frequent catches in five of eight years (Table 1). 



The data representing the selected subset of species 

 were plotted by depth and latitude to display distribu- 

 tional patterns of CPUE. Next, interspecific distributional 

 overlaps were computed by calculating the percentage of 

 joint occurrences with other species based on presence or 

 absence (Krebs, 1989). Joint occurrences were determined 

 both on a trawl-specific basis and after catches had been 

 aggregated into 50-m depth and 0.5°-latitude intervals. In 

 addition, Sefeas^t'.s diversity and species richness were com- 

 puted for each haul to summarize the overall distribution 

 of rockfishes captured in the survey For our analysis, di- 

 versity was computed by using the Shannon-Wiener index, 

 and the number of species was used to scale richness 

 (Krebs, 1989). Diversity and richness measures were then 

 spatially contoured over depth and latitude dimensions to 

 display spatial structure (Surfer, 1995). 



Distributional patterns and groupings of rockfish based 

 on the CPUE data were analyzed by indirect gradient 

 analysis by using multivariate ordination and partitioning 

 methods. Multivariate analyses are often strongly infiu- 

 enced by the choice of distance or (dis)similarity measure. 

 Members of the set of Minkowski distance measures (e.g. 

 Manhattan, Euclidean, maximum, etc.) tend to be strongly 

 affected by extreme values. Moreover, species composition 

 data from trawl surveys have a high proportion of zero 

 catches and a distance measure that is little affected by this 

 property is desirable. The Bray-Curtis index, also known 

 as Czekanowski's quantitative index, is a commonly used 

 statistic in other similar applications and is robust to the 

 presence of zero values (see Bloom, 1981; Field et al., 1982; 

 Krebs, 1989; Rogers and Pikitch, 1992; Weinberg, 1994; 

 Meuter, 1999). A fourth-root transformation of the data 

 was conducted before calculation of the Bray-Curtis index, 

 as suggested by Field et al. (1982) and as implemented by 



