438 



Fishery Bulletin 97(3), 1999 



mercial tows of 45-min duration and randomly as- 

 signed after each tow (to eliminate biases between 

 different trawls and sides of the vessel). The loca- 

 tion of each tow was randomly selected from the avail- 

 able prawn-trawl locations that were possible under 

 the fishing conditions. Over a total period of five 

 nights, we initially completed 11 replicate tows of 

 the composite-square-2 codend against the control. 

 On the basis of results of this comparison and after 

 discussions with representatives of industry, we then 

 designed and constructed the composite-square-3 

 codend and immediately tested it against the con- 

 trol over another 11 replicate tows. 



After each tow, the two codends were emptied onto 

 a partitioned tray. Prawns and all individuals of com- 

 mercially important species comprising bycatch were 

 separated by species. The following categories of data 

 were selected for each tow: the total weight of prawns 

 and a subsample (100 prawns from each codend) of 

 their lengths (to the nearest 1-mm carapace length); 

 the total number of prawns (estimated from the 

 weight of the subsample); the weights of total dis- 

 carded bycatch, discarded noncommercial bycatch 

 (i.e. species of no commercial value), and retained 

 bycatch; the weights and numbers of commercially 

 or recreationally (or both ) important bycatch species; 

 and the sizes (to the nearest 0.5 cm) of commercially 

 or recreationally (or both) important fish. A random 

 sample of prawns (approx. 14 kg) from each codend 

 in each tow was separated and sent to A. Raptis & 

 Sons PTY LTD (seafood processing plant) in 

 Hindmarsh, Adelaide, where they were graded into 

 commercial categories (number per pound) by a lo- 

 cally built automated "dynamic grading machine." 



Several commercially important species were 

 caught in sufficient quantities to enable meaningful 

 comparisons. These were western king prawns, sand 

 trevally, Pseudocaranx wrighti, red mullet, Upen- 

 eichthys porosus, school whiting, Sillago bassensis, 

 Degen's leatherjacket, Thamnaconus degeni, south- 

 ern sand flathead, Platycephalus bassensis, cuttle- 

 fish. Sepia spp., small-toothed flounder. Pseudo- 

 rhombus jenynsii , and southern calamari. 



Analysis of catch data 



Data for all replicates that had sufficient numbers 

 of each variable (defined as >2 individuals in at least 

 8 replicates) were analyzed with paired /-tests 

 (P<0.05). Except for the weights and numbers of 

 prawns and small-toothed flounder, all variables 

 were analyzed by using one-tailed tests of the hy- 

 pothesis that the square-mesh codends caught less 

 than the control. Although the composite square- 

 mesh codends might have been expected to retain 



fewer prawns than the conventional diamond-mesh 

 control codend (due to the larger square-shaped 

 mesh), Broadhurst and Kennelly ( 1996; 1997) showed 

 that trawls fitted with composite square-mesh pan- 

 els caught more prawns than conventional trawls 

 (possibly owing to a reduction in drag and a corre- 

 sponding increase in swept area). Therefore, the 

 weight and number of prawns were analyzed by us- 

 ing two-tailed tests. Catches of small-toothed floun- 

 der, a species that cannot pass through the square 

 meshes, were also compared by using two-tailed tests. 

 With the exception of data for catches of prawns, 

 where analyses provided similar results for weights 

 and numbers of variables, only data concerning num- 

 bers were included in Figure 4 to conserve space. 



Size frequencies of commercially or recreationally 

 (or both) important fish were plotted and compared 

 with two-sample Kolmogorov-Smimov tests (P=0.05), 

 where there were sufficient data («>25 in each 

 codend, pooled across all tows). 



Analysis of size-selectivity of prawns 



The relatively large size of mesh used in the control 

 (45-mm diamond mesh), compared with the compos- 

 ite square-mesh codends (52-mm mesh on the bar), 

 meant that their selectivities overlapped. Insufficient 

 numbers of smaller prawns were retained among 

 individual hauls to enable analyses of between-haul 

 variation (see Fryer, 1991). To provide sufficient data 

 for analyses, size frequencies of prawns from each 

 codend were combined across all tows. An estimated 

 split model (Millar and Walsh, 1992) was used to fit 

 logistic selection curves to these data by maximum 

 likelihood method (Pope et al., 1975) by using the 

 program "CC Selectivity" (Wileman et al., 1996). 

 Logistic curve parameters, their standard errors, and 

 959^ confidence limits were calculated for each 

 codend. For each logistic curve, model deviance val- 

 ues were determined for a goodness-of-fit hypothesis 

 (i.e. to test H^: that the curves were logistic). 



Size categories of commercially graded prawns 

 from each codend were plotted and analyzed by us- 

 ing two-sample Kolmogorov-Smirnov tests (P=0.05 ). 

 In these analyses, the numbers of prawns pooled 

 across all samples from each codend were used for 

 the degrees of freedom. 



Results 



Analysis of catch data 



Compared with the control codend, both the compos- 

 ite-square-2 and composite-square-3 codends signifi- 



