Diamond: Estimation of shrimp trawl bycatch 



495 



Table 6 



Mean percent bias of each of the estimators with normal 

 and delta lognormal distributions offish (F) and shrimp (S) 

 from simulated data. Percent bias (Eq. 17) was calculated 

 separately for each simulated fleet. N = the number of 

 fleets in each analysis. The * indicates that the mean 

 estimated bycatch is significantly different than the mean 

 actual bycatch in a paired sample t-test (P<0.05). The 

 CPUE-mean-per-unit estimator (CPUE=catch per unit 

 of effort) uses unit effort as a proxy for sample size. The 

 basic F;S ratio estimator is the mean of the individual fish 

 to shrimp ratios, and the grand F:S ratio estimator is the 

 ratio of the mean catch offish to the mean catch of shrimp. 

 The basic CPE ratio estimator is the mean of the ratios of 

 catch per effort, where effort is a variable measure such as 

 hours fished, and the grand CPE estimator is the ratio of 

 the mean catch offish to the mean estimate of effort. 



Estimator 



CPUE mean-per-unit 

 Basic F;S ratio 

 Grand F:S ratio 

 Basic CPE ratio 

 Grand CPE ratio 



F:S estimator, and showed significant interactions with 

 other parameters in both the grand F:S and grand CPE 

 ratio estimators. 



For the delta lognormally distributed data, the CPUE- 

 mean-per-unit estimator was the only estimator whose 

 estimated bycatch was not significantly different than 

 the actual simulated bycatch (% bias=0.087%, P=0,64), 

 All four of the ratio estimators significantly overestimated 

 bycatch (Table 6), with estimates ranging from a less than 

 1% overestimate using the grand CPE ratio estimator to a 

 30% overestimate with the basic CPE ratio estimator (Table 

 6). Using all 2-way and 3-way interactions in the ANOVA, 

 I found that significant main effects for both the basic and 

 grand F:S ratio estimators were the probability of catching 

 shrimp and the CV of the shrimp catch. The CV of the fish 

 catch and observer coverage were also main effects in the 

 grand F:S ratio method. The probability of catching fish was 

 an additional main effect in the basic F:S ratio method. The 

 only significant main effect in both CPE ratio estimators was 

 the CV of effort, and the only significant main effect in the 

 CPUE-mean-per-unit method was the CV of the fish catch. 

 All five methods exhibited several statistically significant 

 2-way and 3-way interactions (Table 7), 



Discussion 



The differences in bycatch estimates generated from the 

 field data show how confusing bycatch estimation can be 



SV\n(^P 



\ 400 

 %' 300 

 <S'/^ 200 



^H -0.2 



^H 0.0 



E=3 0.2 



I 1 1.5 



^H 2.0 



Figure 2 



An example of a significant 3-way interaction among 

 parameters in the bycatch simulations. This interaction is 

 between shrimp CV and observer coverage (n ) for the grand 

 F:S ratio estimator with a correlation coefficient (r) of 0.5 

 between the catches offish and shrimp. The simulated fish 

 and shrimp catches were normally distributed. The lack of 

 pattern in the 9r bias due to the interactions among various 

 parameters was a common feature of all the estimators. 



and how difficult it is to compare bycatch estimates from 

 past and recent studies. There was tremendous variability 

 in the bycatch estimates generated by the different meth- 

 ods in the field study; for example, the estimate of Atlantic 

 croaker bycatch in the northern region of North Carolina 

 was either 28 million (±24 million), 84 million (±264 mil- 

 lion), 144 million (±8.4 million), or 186 million (±8.4 mil- 

 lion) fish depending on the estimator used. In addition, 

 each method could give estimates that were higher or lower 

 than the other methods without any consistent pattern by 

 region or species; for example, in the northern region, the 

 basic F:S ratio estimate was the highest estimate for spot, 

 but it was one of the lowest for weakfish. 



Based on the simulation results, the CPUE-mean-per- 

 unit estimator was the best estimator, both because it 

 showed less bias than the ratio estimators and because it 

 was less influenced than the ratio estimators by param- 

 eters such as the mean or variance of the catch, observer 

 coverage, or the correlation between the catches offish and 



