Diamond: Estimation of shrimp trawl bycatch 



499 



doreplication (Hurlbert, 1984) and could lead to erroneous 

 variance estimates if the tows in a trip are not independent 

 samples (Cochran, 1977). The choice of whether to use trips 

 or tows as the unit of effort is dependent on two factors: 1) 

 whether there is a high degree of covariance among tows 

 in a trip, and 2) whether there is an independent estimate 

 of the average number of tows per trip to use as an expan- 

 sion factor. 



Confidence intervals around the bycatch estimates are 

 not symmetrical, although they are shown in Table 4 as 

 symmetrical to allow for easier comparisons between the 

 methods in the field study. Because of the small numbers 

 of observations, most of the confidence intervals in the 

 field study were larger than the means, with the general 

 exceptions of the grand ratio estimators for all species in 

 the northern region, which were surprisingly small. Most 

 grand ratio estimators underestimate the true catch rate 

 and are positively skewed unless the sample size is greater 

 than 30 and the CVs of both the observed fish catch and 

 the auxiliary variable are less than 10% (Cochran, 1977). 

 As seen in Table 5, CVs of the observed fish catch from field 

 data are rarely as low as 10%, and many are over 100%. 

 The very small confidence intervals for all species in the 

 northern region, and the very large confidence intervals for 

 all species in the southern region generated by the grand 

 ratio estimators are due to the nonrandom sampling of 

 boats for the average catch of shrimp in both areas. This 

 nonrandom sampling affects the confidence intervals be- 

 cause the average catch per day is a term in the denomi- 

 nator of the equation used to estimate the variance of the 

 grand ratio estimator (Eq. 2.45 in Cochran, 1977). A very 

 large value for the average catch per day from trip tickets 

 compared to the value from observations as in the north- 

 ern region causes an underestimate in the variance and 

 reduces the confidence intervals, whereas a small value 

 for the average catch per day from trip tickets compared 

 to the value from observations as in the southern region 

 causes an overestimate in the variance and increases the 

 confidence intervals. 



The field data shown here indicate some of the problems 

 that are peculiar to observing and estimating bycatch in 

 shrimp trawl fisheries in comparison to other fisheries. 

 First, there are several hierarchical levels of variability 

 that are ignored because of the logistical difficulties of sam- 

 pling shrimp trawls. If the National Marine Fisheries Ser- 

 vice (NMFS) protocol for shrimp trawl bycatch is followed, 

 only one sample of the catch is taken from a net because 

 of the large numbers of organisms caught in a typical tow. 

 The NMFS protocol depends on the observer thoroughly 

 mixing the catch so that a single sample characterizes 

 the entire catch without variance, but mixing the catch to 

 obtain a random sample is sometimes difficult because of 

 the weight of the catch, the position of the culling tray, the 

 size of the boat, or weather conditions. In addition, some 

 species such as crabs may redistribute themselves after 

 the catch is mixed by simply walking away. Stender and 

 Barans (19941 found differences in fish-to-shrimp ratios 

 when sampling the net compared with enumerating ev- 

 erything in the net. However this source of variability is 

 not measured when following the NMFS protocol and not 



included in the bycatch estimates. Second, only one net 

 is generally sampled per tow, although the boat may tow 

 two, four, or more nets. There is therefore an expansion 

 from the sampled net to the number of nets per tow so that 

 variance among nets is ignored, and this process also adds 

 error. Third, the expansion term, regardless of whether the 

 total shrimp landings or the total shrimp effort is used, is 

 assumed known without error. To include the error in the 

 expansion term further widens the confidence intervals 

 around the final estimates (Diamond and Hanan'"). 



One of the most interesting findings from the simulations 

 was that all the methods tended to overestimate bycatch. 

 None of the overall bycatch estimates, and relatively few of 

 the individual fleet simulations, generated underestimates 

 of the actual values. Although the mean-per-unit and grand 

 ratio estimates overestimated bycatch by less than 19f , if 

 the bycatch is large enough, these estimators could errone- 

 ously add hundreds of thousands offish to the catch-at-age 

 matrices used in stock assessments. Inaccurate stock as- 

 sessments could have consequences for the management 

 of fisheries, particularly for species like red snapper that 

 are managed by quotas on the directed fisheries that are 

 based on the level of bycatch or that have target levels set 

 for rebuilding fish stocks. One method that might be used 

 to "correct" bycatch estimates for the mean-per-unit esti- 

 mator would be to use the estimator to calculate the catch 

 of the target species from the observations, and then to 

 compare the estimated target species catch with the total 

 landings. Although this correction method assumes that 

 the total landings of the target species are accurate (which 

 is rarely a valid assumption), comparison of the estimated 

 total catch of the target species to the actual landed catch 

 might help to pinpoint biases and to adjust the estimated 

 bycatch. 



Because of the differences in estimates generated by 

 the different methods of estimating bycatch, interpreta- 

 tions of bycatch estimates and comparisons of bycatch 

 studies should be made very cautiously. It is often difficult 

 to tell in past studies whether estimates were generated 

 by basic F:S or grand F:S methods, but basic F:S methods 

 overestimate bycatch to a much greater degree. Because of 

 the statistically significant 2-way and 3-way interactions 

 among parameters, it is unlikely that estimates generated 

 by one method can be converted or corrected to other meth- 

 ods, so bycatch estimates made over time using different 

 methods should not be directly compared. In addition, 

 any bycatch estimate should include some indication of 

 the variance around either the estimate or the catch rate, 

 although variance estimates can be misleading if samples 

 are not random. Finally, estimates of the weight or num- 

 ber of species taken as bycatch, no matter how large or 

 small, are meaningless without an estimate of population 

 abundance. Small populations could be harmed by rela- 

 tively small amounts of bycatch, whereas large populations 



'6 Diamond, S. L., and D. Hanan. 1986. An estimate of harbor 

 porpoise mortality in California set net fisheries April 1, 1983 

 through March 31, 1984. National Marine Fisheries Service 

 Admin. Report SWR-86-15, 40 p. 



