486 



Fishery Bulletin 103(3) 



abundance in a consistent way. For example, the NMFS 

 Bering Sea trawl survey includes measurements of depth 

 and surface and bottom temperatures at all trawl sites 

 (Goddard and Walters 4 ) that could be used for post- 

 stratification. Similarly, the Pacific West Coast trawl 

 survey includes measurements of surface and bottom 

 temperature and salinity at all stations (Lauth et al. 5 ) 

 that could be used. Poststratification allows for use of a 

 wide range of stratification variables, including tempo- 

 rally dependent variables that are not available before 

 sampling is complete, e.g., temperature and salinity. 



For surveys where habitat information is not collected 

 at trawl sites, habitat information from other sources 

 can be paired with fish distribution information after 

 collections have been made. For instance, when habitat 

 information is available, but has not been collected at 

 each site, spatial statistics can be used to krige the 

 habitat information over the study area and to predict 

 the specific habitat data value at the sampling sites. If 

 there is a consistent relationship between species abun- 

 dance and the habitat variable, the catch and habitat 

 data paired at sample sites can then be used to identify 

 areas of suitable habitat and areas of high fish density 

 within suitable habitat. How well habitat and HFD 

 areas are estimated will depend on the number and 

 distribution of habitat measurements, the contouring 

 algorithms used, and the estimates of areas within 

 contours. Even if species are not distributed in direct 

 response to particular environmental characteristics, 

 the characteristics may serve as proxies for effects that 

 are more difficult to measure (Perry and Smith, 1994). 

 Once habitat and HFD areas are identified, poststrati- 

 fication can be conducted for total abundance estimates, 

 and statistically significant changes between years can 

 be assessed with an index of relative abundance. These 

 methods could yield more accurate estimates of abun- 

 dance for use by managers. The goal of most sampling 

 plans is to provide statistical estimates with the small- 

 est possible confidence limits at the lowest cost (Krebs, 

 1989). Thus, being able to use data collected indepen- 

 dently of a survey should be appealing. 



The NRC (2000) recommends using data from com- 

 mercial or sportfishing vessels in scientific assessments 

 of abundance. A primary difficulty in using commercial 

 fisheries data for scientific estimates of abundance is 

 that the data do not represent random samples of the 

 fish population. As a result, commercial fisheries data 



4 Goddard, P., and G. Walters. 1998. 1995 bottom trawl 

 survey of the eastern Bering Sea continental shelf. AFSC 

 Processed Report 98-08, 170 p. Resource Assessment and 

 Conservation Engineering Division, Alaska Fisheries Science 

 Center, NMFS, NOAA, 7600 Sand Point Way N.E., Seattle, 

 Washington, 98115. 



5 Lauth, R. R., M. E. Wilkins, and P.A. Raymore Jr. 1997. Re- 

 sults of trawl surveys of groundfish resources of the West 

 Coast upper continental slope from 1989 to 1993. NOAA 

 Tech. Memo. NMFS-AFSC-79, 342 p. National Technical 

 Information Service, U.S. Department of Commerce, 5285 

 Port Royal Road, Springfield, Virginia 22161. 



present a biased perspective of the population that may 

 change over time and may not correlate well with ac- 

 tual fish abundance (NRC, 2000). Although commercial 

 fishery-dependent data may provide biased estimates of 

 abundance, fishery-dependent data also provide large 

 sample sizes and a wide range of information not avail- 

 able from other sources. For example, commercial and 

 sportfishing data often provide broader geographic and 

 temporal coverage. Poststratification of haphazard data 

 from commercial and sportfishing sources may be one 

 way to reduce inherent bias and provide useable scien- 

 tific information. For instance, Buckland and Anganuzzi 

 (1988) described how data collected on commercial tuna 

 fishing vessels can be used to estimate dolphin abun- 

 dance when survey data are not sufficient. The data 

 collection sites were not randomly selected. Instead, 

 the sampling sites were directly related to dolphin 

 sightings, because dolphins and tuna schools are often 

 closely associated. As a result, areas of high dolphin 

 density corresponded with areas of high fishing effort. 

 Poststratification was used to decrease the bias result- 

 ing from nonrandom distribution of both search effort 

 and dolphin schools. A second example is a retrospec- 

 tive study that combined survey and commercial fishing 

 data. In this study (Halliday 8 ), 1958-60 poststratified 

 survey data were used to develop a relationship between 

 the survey abundance of the 1954-1959 year classes 

 and their abundance estimates from commercial fishery 

 data. This relationship was then used, along with 1969 

 survey data, to predict the size of the 1966-68 year 

 classes. The same process was used to predict the size 

 of later year classes with later years of survey data. 



Poststratification also facilitates the use of a single 

 data set for multiple objectives. Collecting data is costly 

 and many data sets are collected and analyzed for a 

 single objective and then not used again. Although it 

 is preferable to use data for multiple objectives, it can 

 be difficult to meet statistical assumptions when the 

 data are re-used for a different purpose. For example, 

 a multispecies survey may be stratified according to the 

 distribution of one or more of the most commercially 

 valuable species collected. An example is the stratifica- 

 tion of Pacific west coast bottom trawl surveys in 1980, 

 1983, and 1986, which were focused to improve the 

 precision of canary and yellowtail rockfish abundance 

 estimates (Weinberg et al. 2 ). If the stratification used 

 was not effective for decreasing the variance of abun- 

 dance estimates for other species, treating the data as 

 if they were haphazardly collected, recognizing that 

 the estimator may be biased, and poststratifying the 

 data by habitat variables that are closely related to the 



6 Halliday, R. G. 1970. 4T-V-W haddock: recruitment 

 and stock abundance in 1970-72. ICNAF Res. Doc 

 70/75, 12 p. Approved for citation by Tissa Amaratunga, 

 Deputy Executive Secretary, Northwest Atlantic Fisher- 

 ies Organization. [Available from the Secretariat Library, 

 Northwest Atlantic Fisheries Organization, 2 Morris Drive, 

 Burnside Industrial Park, Dartmouth, Nova Scotia, Canada, 

 B3B 1K8.] 



