302 



Fishery Bulletin 101(2) 



results of Table 4 suggest that 0.2 is a reasonable default 

 CV for annual variability in survey catchability. This CV 

 should be "added" to the observation error CVs to obtain a 

 CV for use in stock assessments. 



The blanket use of default CVs is clearly undesirable. It 

 is obviously wrong to assume that all CPUE indices have 

 the same CV, regardless of which species or fishery they 

 describe, or the quality and quantity of data from which 

 they are calculated. Similarly, we should expect that an- 

 nual variability in trawl survey catchability will vary from 

 stock to stock. However, we have little choice in this matter. 

 In most stock assessments we do not have the information 

 to depart from a default value (although there is some- 

 times evidence that CPUE data sets were unusually weak, 

 Doonan et al.-^). The above default values imply smaller 

 CVs for CPUE than for trawl surveys. This is surprising 

 and contrary to the prevailing view that trawl surveys are 

 more "reliable" than the CPUE (in the sense defined in the 

 above section on the assessment data). Nevertheless, it is 

 clearly indicated by the data sets examined here. 



Can we detect years of extreme trawl survey 

 catchability? 



There is clear evidence of extreme years in New Zealand 

 trawl surveys, i.e. years in which the biomass indices for 

 many species are extreme (all low, or all high). However, 

 can we be confident that these extreme years are caused by 

 extremes in catchability? There are two other factors that 

 could cause these extremes. 



The first is sampling error, which is associated with the 

 element of chance involved in whether there happen to be 

 many fish at a randomly chosen location at the time it is 

 sampled by the trawl. Because some pairs of species co-oc- 

 cur, we can expect that if we are "lucky" with one species 

 (i.e. we happen to hit dense concentrations of it), then we 

 will tend to be "lucky" with its co-occurring species. Thus, 

 the sampling errors of co-occurring species will be cor- 

 related. It seems unlikely that the extreme mean ranks 

 shown in Figure 5 (or the biomass ratios in Table 5) were 

 caused solely by correlated sampling errors. In principle, 

 we should be able to quantify this likelihood. From the sur- 

 vey tow-by-tow data we could infer the extent of between- 

 species correlations at the level of individual stations, from 

 which we could calculate correlations for whole sui^veys (we 

 would expect more correlations in surveys covering a wider 

 range of species). This information could then be used to 

 calculate the probability of generating biomass ratios as 

 large as those in Table 5. However, to do so would be a 

 major multilevel simulation exercise which is beyond the 

 scope of the present work. What we do know, from other 

 studies, is that between-species correlations, when they 

 exist, are not large. Values of 0.2 to 0.4 seem to be typi- 

 cal (for .square-root-transformed catch rates in the same 



depth stratum. Bull"*). It does not seem at all likely that 

 such small correlations would cause the very substantial 

 synchronous fluctuations we see in Figure 5 and Table 5. 



A second interpretation of the extreme years is that 

 they occur because changes in abundance of co-occurring 

 species are correlated (because fishing that reduces the 

 abundance of one species is likely to do the same for co-oc- 

 curring species). Table 5 allows a subjective evaluation of 

 the likelihood that biomasses in the extreme years changed 

 by as much as the survey biomass indices suggest. This 

 evaluation is complex because the likelihood depends on 

 the magnitude of the ratios, the number of years between 

 surveys, the number of species involved, and any "adja- 

 cent" changes (e.g. for series 9, a large drop in biomass in 

 1989 is less plausible because it appears to be followed by 

 a large rise in the next year). Thus it is not easy to provide 

 a threshold and say that some changes are plausible but 

 others are not; but there is a clear range of plausibility. At 

 one extreme are the changes associated with 1980 in series 

 1 and 1989 in series 9; we have argued above that these 

 changes are clearly implausible. At the other extreme the 

 changes for 1995 in series 13 are not as implausible, but it 

 is a matter of judgment as to whether one could call them 

 plausible. 



Another point to bear in mind is that if we use observa- 

 tion error CVs (as routinely calculated from trawl survey 

 data) in stock assessments, we obtain residuals that are, 

 more often than not, larger than they ought to be. 



We are left with the conclusion that the trawl survey 

 data contain clear evidence that research-vessel catch- 

 ability does vary significantly from year to year In most, if 

 not all, of the circled years in Figure 5 the catchability of 

 many species appears to have been either much higher or 

 much lower than normal. This finding is consistent with 

 those of Myers and Cadigan (1995), who expressed this 

 variation in terms of between-age within-year correlations 

 in trawl-survey estimates of numbers at age. Also, Millar 

 and Methot (2002) found evidence of significant departures 

 from mean catchability in four of eight years in the trien- 

 nial series of trawl surveys carried out on the Pacific coast 

 of the United States. This variation in catchability may be 

 environmentally driven. It would not be difficult to find 

 plausible environmental variables that were extreme in 

 the right years. However, because most of our trawl-survey 

 time series were short we could have little confidence that 

 this correlation was indicative of causation. Another pos- 

 sible cause of variation in catchability is between-survey 

 changes in gear and fishing practice (although care is taken 

 to avoid such changes). 



Are there consistencies between data sets? 



Our only important result under this heading is that some 

 estimates of trawl survey catchability are not credible. For 

 two species, we found that some estimates were implausi- 



■^ Doonan, I. J., P. J. McMillan. R. P. Coburn, and A. C. Hart. 

 1999. A.ssessmont of OKO .'iA black orco for 1999-2000. N.Z. 

 Fish. A.ssoss. Res. Doc. 99/52. .30 p. National In.stitute of Water 

 and Atmo,sphi'ric Research, P.O. Box 14901, Wellington. New 

 Zealand. 



Bull. B. 2000. Personal commun. National histitute of 

 Water and Atmospheric Research. P.O. Box 14901. Wellington, 

 New Zealand. 



