Francis et aL: Quantifying annual variation in catchability 



295 



error. By setting a value for one of these CVs we are im- 

 plicitly making an assumption about the other component: 

 annual variation in catchability. We did not solicit informa- 

 tion on how these CVs were set for individual assessments. 

 However, the most common way is based on a subjective 

 assessment of the "reliability" of the associated biomass 

 indices: the less reliable the indices are judged to be, the 

 higher the assumed CV (by "reliability" we mean the com- 

 bination of two very different properties of an index, its 

 precision, and its "quality" — the extent to which it is likely 

 to be proportional to biomass). This is why almost half of 

 the CPUE series (13 out of 30) have assumed CVs of 0.35, 

 and many of the trawl survey series (13 of 18) have CVs 

 of 0.25. In most cases, the CVs assumed for trawl survey 

 indices differ from the observation error CVs calculated 

 from the trawl survey data (Fig. IB). 



Trawl survey data Data from all New Zealand random 

 trawl surveys were considered. The surveys were grouped 

 into series, each of which contained surveys covering 

 (approximately) the same area at about the same time of 

 year and using the same (or similar) vessel(s) and gear 

 Some series were split into two, by area, because they were 

 deemed to survey two distinct fish communities. Series 

 with fewer than four surveys were rejected. This left 17 

 series, with between four and 1 1 surveys per series. 



For each trawl survey series a list of "suitable" species 

 was generated by listing all species caught in the series and 

 then excluding species deemed to be "unsuitable" for any of 

 the following reasons: 



• species caught in only a small percentage of tows; 



• species caught in small quantities (low mean catch per 

 tow); 



• species not well caught by the net because they are too 

 small, too large, too close to the sea floor, or too high in 

 the water column; 



• species for which identification was poor, or inconsis- 

 tent over time; 



• species whose range was poorly covered by the surveys 

 (e.g. those occurring mostly on rough ground, or mostly 

 in water shallower or deeper than that covered by the 

 series). 



The idea was to include as many species as possible for 

 each series. In considering a particular species in a spe- 

 cific trawl survey series, the following question was useful; 

 "If this were a valuable commercial species would it be 

 appropriate to use this series of trawl surveys to generate 

 biomass indices to put into a stock assessment?" An answer 

 of "yes" (or even "maybe") was a good reason to include this 

 species. The CVs of biomass estimates were not considered 

 in making this decision. For each series the list of suitable 

 species was compiled by people with an intimate knowledge 

 of that series and the associated species. No attempt was 

 made to derive consistent objective criteria (e.g. exclude 

 all species that occurred in fewer than 30'7f of tows) for all 

 series. The exclusion of a species from one series was no 

 barrier to its inclusion in another The number of accept- 

 able species in a series varied between 4 and 25. 



40- 



y 30- 



20- 



10- 



> 



C = CPUE indices 



T = trawl survey indices 



C 

 C 

 C 



9; 



0.1 



0.2 



— I — 

 0.3 



— I — 

 0.4 



— 1 — 

 0.5 



— I — 

 0.6 



Assumed CV 



Figure 1 



Descriptive statistics for the assessment data sets: 

 assumed CV plotted against (A) the number of indi- 

 ces (all data sets) and (B) the median observation 

 error CV (data sets with trawl survey indices only). 

 Each point represents one data set: in A the points 

 are jittered slightly to separate coincident points; 

 where different indices in a data set had different 

 assumed CVs the median of these is plotted. 



Biomass indices, and CVs, were calculated for each suit- 

 able species in each survey. In all series but two, vulner- 

 ability, areal availability, and vertical availability were set 

 to 1. There was a wide range of estimated CVs. Even when 

 the CVs for each species in a series were averaged over all 

 surveys, these averaged values spanned an order of mag- 

 nitude, from 0.07 to 0.70 (Fig. 2). 



Analyses 



Our analyses addressed a series of questions, which are 

 given as subheadings in this section. 



