Hiellvik et al : Measurement error in marine survey catches 



725 



times the distance between the vessels, we believe this 

 factor to be of minor importance. 



Another problem is the determination of the towed 

 distance. The uncertainty connected with subjective 

 judgments and inaccuracies in the GPS should be included 

 in the measurement error because these factors are also 

 present at a standard survey haul. However, it is not 

 obvious to what extent the differences in the recorded 

 towed distances are due to differences in subjective 

 judgments or to differences in actual towed distances. In 

 our calculations, we used the recorded values from both 

 vessels for d,. . If there is no real difference in the towed 

 distances within a comparison, d~ is expected to decrease 

 by setting^, j=(i, 2 f"*" ^1' hauls, thus eliminating one factor 

 of uncertainty. At the other extreme, if the subjective 

 judgments are perfect, and the recorded differences in 

 towed distance are due to real differences, one would 

 expect (T^ to increase by setting (/, ,=c?, .2, because an extra 

 error then is added. Actually, by using the values from 

 vessel 2 only and by setting c^, i=(^, ■, the resulting estimate 

 of&l is a'i=0.061, which is a reduction by about 119< . Even 

 though this value is statistically insignificant, it indicates 

 that uncertainty connected to the measurement of towed 

 distance constitutes a part of the measurement error (see 

 also God0 et al., 1990). 



In the "Results" section, the null hypothesis of equal 

 efficiency for all the participating vessels was rejected. By 

 joining data from the groups where the same pair of vessels 

 participates, the ability to detect differences in efficiency 

 between the vessels increases through an increased 

 sample size and a smaller number of simultaneous tests 

 (with N tests and a nominal level a, the null hypothesis 

 of equal efficiency is rejected for P-values smaller than 

 the Bonferroni corrected level a/N). For the N=6 tests, we 

 obtained P-values 0.0005 for group 4, 0.009 for groups 1 

 and 2, 0.011 for groups 6 and 10, 0.034 for groups 5 and 

 9, 0.107 for groups 7 and 8, and 0.123 for group 3. With 

 a level a=0.05 we have a/6=0.0083, and for group 4 the 

 difference is clearly statistically significant. The higher 

 efficiency of Jan Mayen (JM) in this group was probably 

 due to her heavier trawl doors. At a 10*7^ level, the vessel 

 Army Kraemer (LIZY) was significantly more efficient than 

 Johan Hjort (JH) in groups 1 and 2, and Michael Sars 

 (MS) was significantly more efficient than JH in groups 

 6 and 10. The differences between G.O. Sars (OS) and JH 

 (groups 5 and 9), OS and JM (groups 7 and 8) and GS and 

 LIZY (group 3) were not significant. However, excluding 

 group 4, and ignoring statistical significance, the vessels 

 can, in fact, be ranged consistently after increasing 

 efficiency as 



higher for small fish. One explanation may be that the 

 interaction between small fish and the trawl gear is more 

 variable (God0 and Walsh, 1992); another possibility is 

 that small fish operate more in patches than do large fish. 

 If the last assumption is correct, a reduction in (T^ could be 

 expected with increasing tow length. However, for the .set 

 of tows considered, we found no significant difference in a^ 

 due to tow distance. 



Consistent with the length dependency of the measure- 

 ment error is the length dependency of the total variability 

 of the surveys. The average varly, ) for the winter surveys 

 1996-2000 for fish <31cm and >64cm, was 2.49 and 0.98, 

 respectively; whereas for the unstratified data it was 1.66. 

 The corresponding numbers for the autumn surveys 1996- 

 2000 were 3.03, 1.47. and 2.98 for small, large, and unstrati- 

 fied fish, respectively. All numbers are for nonzero catches. 



Trawl catches have been considered highly variable (see 

 e.g. Gulland, 1964; Doubleday and Rivard, 1981) and as a 

 result the reliability of trawl survey estimates have been 

 questioned. Abrupt changes in catch size and composition 

 over a short time in a limited area have demonstrated the 

 difficulties in using the information as a relative estimate 

 of density without an understanding of the nature and 

 causes of the variability (Godo, 1994). Unexpected an- 

 nual changes in survey indices may also be a problem for 

 a reliable evaluation of fish stocks and can be attributed 

 to a variable bias (changes in catchability) among years 

 (Pennington and Godo, 1995). Our analysis demonstrates 

 that for the bottom trawl survey in the Barents Sea, catch 

 rates and composition from the applied survey trawl are 

 repeatable up to a relatively small and constant measure- 

 ment error and are hence expected to give a reliable pic- 

 ture of the relative fish density at a given site and time. 

 Further, the measurement error of this sampling gear is 

 small compared with the total observed variability. For a 

 particular survey it appears that most of the sui-vey vari- 

 ance is caused by station-to-station differences in catches 

 rather than local conditions at a station. This may be taken 

 as an indication that shorter and more frequent tows may 

 be more efficient for monitoring this cod stock. Moreover, 

 when controlling trawl geometry (Godo and Engas, 1989) 

 and towed distance (Godo et al., 1990), it should be possible 

 to establish explanatory factors to be included in the sur- 

 vey assessment procedure. To the degree that one is able to 

 establish models to determine fish densities at any station, 

 the comparability of density measures throughout the dis- 

 tribution area will improve. The consequences will thus not 

 only be more reliable survey estimates, but we also expect 

 a better understanding of distributional patterns in rela- 

 tion to the physical and biological environment. 



JM- 



-^GS- 



15,9,31 



-^JH- 



' LIZY /MS. (4) 



Acknowledgments 



The numbers in parentheses refer to experiment groups, 

 and for each group one vessel to the left, and one to the 

 right of the corresponding arrow, are involved, the one to 

 the right being always the most efficient one. 



There also seems to be a significant difference in the 

 measurement error for small and large fish, it being 



We are grateful to Atle Totland for help with handling 

 the data and to Michael Pennington for comments on the 

 paper. We are also indebted to the scientific editor and three 

 anonymous referees for a number of useful comments that 

 improved the paper. The work was financially supported by 

 the Norwegian Research Council (127198/120). 



