Selectivity of penaeid trap nets in south-eastern Australia

Two experiments were done to estimate the selectivity of commercial and modified trap-net configurations in New South Wales (NSW), southeastern Australia. In the first experiment, a commercial trap net made entirely from 25 mm mesh and designed for use in shallow water was alternatively fished with a fine-meshed (9.5 mm netting) trap net (used as a control). In the second experiment, two trap-net configurations designed for use in deeper water and comprising the same anterior section (made from 25 mm mesh), but with different bunts made from (i) the conventional 25 mm mesh and (ii) 31 mm mesh were alternately fished against the control. Both of the conventional trap nets (comprising 25 mm mesh throughout) had low amounts of bycatch and similarly selected eastern king Penaeus plebejus , greasyback Metapenaeus bennettae and school prawns Metapenaeus macleayi across narrow selection ranges (< 3.4 mm) and at 50% retention lengths (between 18.53 and 21.50 mm) that were larger than the average commercially-accepted sizes (15-17 mm CL). Analyses of the selectivities and relative efficiencies of the trap-net configurations comprising the 25 and 31 mm bunts showed no benefit, in terms of maintaining prawn catches and reducing unwanted bycatch, associated with increasing mesh size in these gears. The utility of trap nets for selectively harvesting penaeids is discussed. We conclude that this type of fishing gear appears to have few deleterious impacts.


INTRODUCTION
Penaeids form the basis of several important commercial fisheries in New South Wales (NSW), Australia that have a total value of more than $A26 M per annum. Catches include 6 species, although eastern king Penaeus plebejus, school Metapenaeus macleayi and greasyback prawns Metapenaeus bennettae account for more than 98% of the total annual production (approx. SUMMARY: Two experiments were done to estimate the selectivity of commercial and modified trap-net configurations in New South Wales (NSW), southeastern Australia. In the first experiment, a commercial trap net made entirely from 25 mm mesh and designed for use in shallow water was alternatively fished with a fine-meshed (9.5 mm netting) trap net (used as a control). In the second experiment, two trap-net configurations designed for use in deeper water and comprising the same anterior section (made from 25 mm mesh), but with different bunts made from (i) the conventional 25 mm mesh and (ii) 31 mm mesh were alternately fished against the control. Both of the conventional trap nets (comprising 25 mm mesh throughout) had low amounts of bycatch and similarly selected eastern king Penaeus plebejus, greasyback Metapenaeus bennettae and school prawns Metapenaeus macleayi across narrow selection ranges (< 3.4 mm) and at 50% retention lengths (between 18.53 and 21.50 mm) that were larger than the average commercially-accepted sizes (15-17 mm CL). Analyses of the selectivities and relative efficiencies of the trap-net configurations comprising the 25 and 31 mm bunts showed no benefit, in terms of maintaining prawn catches and reducing unwanted bycatch, associated with increasing mesh size in these gears. The utility of trap nets for selectively harvesting penaeids is discussed. We conclude that this type of fishing gear appears to have few deleterious impacts.
targeted throughout their distributions across nearshore and estuarine habitats (Coles and Greenwood, 1983) using a combination of static (stow and trap nets) and towed (otter trawls and seines) fishing gears.
All of these gears are managed by a range of input controls that include limits on their dimensions, effort, methods and areas of operation and minimum and maximum legal mesh openings (for a definition, see Ferro and Xu, 1996). These legal mesh sizes vary between 40 and 45 mm in the codends of otter trawls, 30 and 36 mm throughout seines and stow nets and 25 and 36 mm in trap nets. The use of these small-meshed gears throughout habitats that are typically characterised by diverse assemblages and abundances of small fauna (Bell et al., 1988;Gray et al., 1996) is of considerable concern and has resulted in several quantitative studies of catches (e.g. Gray et al., 1990;Andrew et al., 1995;Kennelly et al., 1998;Gray, 2001). These studies revealed that at some locations and times, penaeid fishing gears, and especially otter trawls, retain incidental catches (collectively termed 'bycatch' sensu Saila, 1983) that often comprise juveniles of commercially-important teleosts, molluscs and crustaceans, including small, unwanted conspecifics (< approximately 15 mm carapace length -CL; Broadhurst et al., 2004) of the targeted species. Concerns over the mortality of these organisms and the potential impacts on their stocks have resulted in successful attempts at improving gear selectivity. The majority of this work has concentrated on otter trawls used throughout various marine (e.g. Broadhurst and Kennelly, 1997) and estuarine (Broadhurst and Kennelly, 1994;Broadhurst et al., 2004) fisheries. Considerably less attention has been directed towards assessing static gears (but see Macbeth et al., 2004) and no work has been done on trap nets.
Trap netting in NSW involves up to 95 operators who are permitted to fish in 12 coastal lakes and lagoons, although more than 40% of the effort is concentrated at Tuggerah Lakes (33 o 19'S, 151 o 30' E). Fishing mostly occurs at night and between the last and first quarter phases of the moon. All of the trap-net configurations used in NSW are similar, and consist of a wall of 25 mm netting (i.e. the minimum legal mesh size for this gear) up to 140 m in length. The width of this wall of netting varies from approximately 1 to 6 m, depending on the depths fished. For example, trap nets used at sites that are shallow and with slow currents typically have a narrow (e.g. 1 m) anterior section made from stiff, positivelybuoyant polyethylene (PE) netting, which helps to maintain net distension during fishing (Figs 1A and 2B). In contrast, trap nets used in deeper (e.g. > 2 m), faster-flowing water have wide transverse sections (i.e. up to approx. 6 m) of negatively-buoyant polyamide (PA) netting throughout.
Trap nets are set by attaching one end to a vertical stanchion near the shore and the other end to the horizontal gunwale of a dory anchored on the lake ( Fig. 2A and B). Currents cause the anterior section of the netting to distend and assume a parabolic shape, effectively trapping migrating penaeids and directing them along the wall of netting towards the horizontally-orientated bunt (at the dory). Fishers facilitate this movement of catch by regularly lifting and hauling sections of the headline and footrope over a second dory so that it passes underneath the trap net and the catch is progressively rolled towards the bunt (Fig. 2C).
Anecdotal information from fishers suggests that trap nets have low amounts of bycatch and are more selective than other gears used to target penaeids in NSW. However, there are no formal estimates of the selectivity of the various configurations or of the effects of any modifications on their performance. Our aims in the present work were to address this lack of information by (i) quantifying the selectivity of the most common configurations used and (ii) examining the relative efficiency of a larger mesh size in the bunt of one of these gears.

MATERIAL AND METHODS
This study was done at commercial trap net sites in Tuggerah Lakes between the last and first quarter moon phases of January and February 2003. All sites ranged in depth from 0.5 to 3 m and encompassed a combination of sloping sand and mud bottoms with patches of seagrass.

Trap-net configurations examined
Four trap-net configurations were used at these sites. All trap nets were made from dark netting that had a maximum stretched depth of approx. 6 m and hung at 50% on buoyed headlines and weighted footropes, 140 m in length (Fig. 1). The first configuration, termed the 25 mm PE/PA trap net, represented those commercial designs that are typically fished at shallow sites (e.g. < 1 m deep-see discussion above) and comprised two sections that were each 70 m in length and made from 100 meshes (normal direction -N) of 25 mm knotted PE (approx. 1.2 mm diameter -Ø, 3-strand twisted twine) and 250 N of 25 mm knotted PA (approx. 0.4 mm Ø, 3-strand twisted twine) netting respectively (Fig. 1A). The second and third trap-net configurations had the same anterior section (300 N of 25 mm knotted, PA netting-i.e. the same material as that used above-120 m in length), but different bunts. Both bunts were 20 m long, approx. 6 meters wide and made from 0.4 mm Ø, 3-strand twisted PA twine, but with mesh sizes that were 25 and 31 mm respectively (Fig. 1B). These bunts and the anterior section of the trap net were rigged with zippers (Buraschi S146R, 6 m in length) to facilitate their attachment. The 25 mm bunt attached to the anterior section described above represented the majority of the commercial trap-net configurations used throughout NSW (Fig. 1B). The larger-meshed, 31 mm bunt attached to the anterior section represented a modified and previously untested trap-net configuration. We hypothesised that this larger-meshed bunt would increase the size selection of the trap net for penaeids and small individuals comprising the bycatch. The fourth trap net was termed the control, and comprised 600 N (i.e. approximately 6 m stretched depth) of 9.5 mm knotless PA netting (0.7 mm Ø, braided twine) throughout (Fig. 1C).

Experimental design
All fishing was done at night (between 20:00 and 03:00) and according to normal commercial procedures. During all sets, the headline and footrope of the anterior end of the particular trap-net configuration being used (see below for details) were secured to a staked stanchion and the net set from a dory along the bottom of the lake in a straight line ( Fig anchored in position ( Fig. 2A and B). To mark the middle of the trap net, approx. 140 m of 4 mm Ø, PE rope was attached between the stanchion and the second dory, and a large float was clipped at 70 m ( Fig. 2B). Each trap net was left to soak for 25 minutes, after which the first dory returned to the middle of the gear and the headline and footrope were lifted onboard (Fig. 2C). Two fishers simultaneously hauled the headline and footrope so that the dory passed under the trap net and the catch was concentrated towards the bunt and then into the first dory (Fig. 2C). The trap net was then removed from the lake and the next configuration was set. Using this fishing method, two experiments were done during consecutive phases of the new moon. In the first experiment, the 25 mm PE/PA trap net was alternatively fished against the control at a shallow (< 1 m) commercial trap-net site. We attempted between 2 and 3 replicate, alternate 25 min sets of the treatment and control trap net on each night and completed a total of 15 balanced sets over 6 nights. In the second experiment, the 25 and 31 mm bunts were alternatively zippered to the anterior 25 mm PA trap-net section and fished against the control at two siteswhich were determined randomly and/or according to the prevailing weather conditions (i.e. the direction and strength of wind and waves) each night. Over 8 nights, we attempted two replicate nightly sets of each trap-net configuration and successfully completed a total of 14 balanced replicates.
Data collected from all trap-net sets included: the number and weight of total prawns; the numbers, weights and all carapace lengths (CL to the nearest 1 mm) of greasyback, eastern king and school prawns; the weight of total bycatch; the numbers of all fish and their fork lengths (FL to the nearest 5 mm); and the numbers of all other species. Where it was not possible to identify individual species, these were grouped at the levels of genus or family.

Statistical analyses
Data from the two experiments were analysed in different detail. Attempts were made at modelling and comparing the selectivity of all treatment trap nets for the key species encountered in both experi- ments. In addition, because the control was alternately fished against a commercial and modified, larger-meshed trap-net configuration during experiment 2, specific hypotheses concerning gear-related effects on catches were examined using multivariate and univariate analyses (detailed below).
For both experiments, the size-frequencies of individuals of species caught in sufficient quantities (at least 100 individuals from each trap net configuration) were combined across all tows for the control and each of the treatment trap nets. Parametric selection curves (logistic and Richards) were fitted to these data using maximum likelihood (Millar and Fryer, 1999). These fits used an estimated-split SELECT model (Millar and Walsh, 1992) and were assessed by visual examination of residual plots and by comparing model deviances and associated degrees of freedom against a χ 2 distribution. The standard errors of parameter estimates (i.e. 50% retention length -L 50 and difference in length between 25 and 75% retention lengths -selection range or SR) were adjusted according to an appropriately-derived replicate estimate of over-dispersion (Millar and Fryer, 1999). Pairwise bivariate Wald statistics were calculated using the estimated parameter vectors to test for differences between the selectivity curves for each of the treatment trap nets (Kotz et al., 1982). Using the full data set from experiment 2, nonmetric multivariate analyses were used to investigate differences in the structures of catches between the trap-net configurations, following the methodologies presented by Clarke and Warwick (2001). Abundances were √ transformed and used to develop Bray-Curtis similarity matrices. Ordination of the relationships among ranks of these similarities from individual sets of the three trap-net configurations was done by multi-dimensional scaling (MDS). Two-way crossed analyses of similarity (ANOSIM) were used to test for differences in catch assemblages from the 3 trap nets over the 8 nights fishing. Significant R values from these analyses were used to group the trap nets, which were subsequently explored using SIMPER (Clarke and Warwick, 2001).

SELECTIVITY OF PENAEID TRAP NETS 449
Parametric univariate analyses were used to examine differences in the catches of the key species and groups identified above among the trap-net configurations used in experiment 2. To provide balanced analyses, only nights with two replicate sets of each trap-net configuration were considered. Data were ln(x+1) transformed, tested for heterogeneous variances, and analysed by appropriate 2-factor (nights and trap nets as random and fixed factors respectively) orthogonal analyses of variance (ANOVA). To increase power for the main effects of trap nets, where the interaction term was non-significant at P < 0.25, it was pooled with the residual (Winer, 1971). All significant main effects of the trap net were investigated using Student-Newman-Keuls (SNK) multiple comparisons. The means for all significant interactions were graphed, but not investigated further owing to the low level of replication within nights (i.e. only two replicate sets of each trap net).

RESULTS
Thirty three families comprising more than 43 species were captured during this study (Table 1). It was not possible to distinguish between 2 and 3 species of glassy perchlets and gobies respectively, so these were grouped by genus and family.
Sufficient quantities and appropriate sizes of eastern king, greasyback and school prawns (Fig. 3) 450 M. BROADHURST et al. and southern herring (30-120 mm FL) were caught to enable attempts at modelling their selectivities for at least one of the treatment trap-net configurations in either experiment. A logistic model (Fig. 4, Table  2) was used in all cases because: (i) the null hypothesis for the goodness-of-fit test was not rejected (P > 0.05); (ii) the deviance residuals showed no clear structure; and (iii) there was no significant reduction in deviance associated with using a Richards curve. Pairwise bivariate Wald tests detected significant differences in parameter estimates between the 25 and 31 mm bunts for eastern king and greasyback prawns, with the larger-meshed bunt selecting individuals at larger L 50 s and across considerably greater SRs (P < 0.01, Fig. 4A and B, Table 2). No significant differences were detected in the estimated selection parameters for school prawns between these bunts (P > 0.05; Fig. 3C, Table 2). The 25 mm PE/PA trap net selected eastern king prawns at significantly greater and lower parameters than the 25 and 31 mm bunts respectively (Pairwise test P < 0.01, Fig. 3A, Table 2). The estimated selection parameters for southern herring were not signifi-cantly different between the 25 mm PE/PA trap net and the 31 mm bunt (P > 0.05, Table 2).
MDS of the abundance data from the control trap net and 25 and 31 mm bunts used in experiment 2 had a stress of 0.14 for the best two-dimensional ordination, indicating sufficient representation (Fig. 5)  nights (averaged across the three trap nets-ANOSIM Global R = 0.568, P < 0.01) and among trap-net configurations (averaged across the 8 nights-ANOSIM Global R = 0.68, P < 0.01). Pairwise tests revealed that the 25 and 31 mm bunts were significantly different to the control (R = 0.918, P < 0.01 and R = 0.923, P < 0.05 respectively) but not to each to other (R = 0.167, P > 0.05) (Fig. 5). SIMPER analyses of these two groups (i.e. control vs. treatment trap nets) showed that all species of penaeids, along with several species of small fish (including glassy perchlets, whitebait, southern herring, gobies, hardyhead and river garfish) were responsible for the differences between the control and treatment trap nets (Table 3). ANOVA of the appropriate univariate data from experiment 2 detected significant F ratios for the main effect of trap nets for the numbers of total, greasyback and eastern king prawns, hardyhead, river garfish and glassy perchlets (Table 4).
SNK tests of these means showed that the 31 mm bunt retained significantly fewer total and eastern king prawns than the commercially-used 25 mm bunt (mean reductions of 44 and 52% respectively) and the control ( Fig. 6A and C). Although not significant, the weights of total and eastern king prawns showed similar trends ( Fig. 6B and D). Similarly, the 31 mm bunt caught fewer greasyback prawns and river garfish than did the 25 mm bunt (by 50 and 73% respectively) ( Fig. 6G and J). Both treatment bunts retained comparable, and significantly fewer hardyhead and glassy perchlets than the control (Fig. 6I and K). Significant interactions were detected between nights and trap nets for the weight of greasyback prawns and the number of southern herring (Table 4). Like the results from above, the appropriate means of these differences revealed that the 31 mm bunt retained lower quantities of these species across most nights ( Fig.  7A and B).

DISCUSSION
This study showed that (i) the commercially-used trap-net configurations had comparable and appropriate selectivity parameters for the targeted sizes of all three species of penaeids, and (ii) these trap nets are considerably more selective than other larger-meshed, static and towed gears used to target penaeids in south eastern Australia and throughout many other temperate and tropical fisheries (Vendeville, 1990;Sobrino et al., 2000;Broadhurst et al., 2004;Macbeth et al., 2004). More specifically, the 25 mm bunt trap net selected all species at L 50 s between 18.53 and 19.42 mm across SRs that were less than 3.4 mm ( Table 2). This minimal inter-specific variability in selectivity parameters can be attributed to the considerable appendages and morphological discontinuities of penaeids, which strongly influence their selectivity irrespective of the species (Vendeville, 1990). These comparable L 50 s and narrow SRs mean that the majority of all individuals of the 3 species less than between approximately 18 and 20 mm CL escaped through the 25 mm mesh. These escapees were larger than the average industry-accepted commercial sizes (approx 15-17 mm CL; Broadhurst et al., 2004)  The selectivities of the commercially-used trap nets can be compared with other penaeid-fishing gears using a simple proportionality constant termed the 'selection factor' (SF) (Pope et al., 1975) or 'coefficient of selectivity' (Vendeville, 1990), calculated by dividing the size of mesh into the L 50 estimate. For the commercial trap nets examined here, all penaeid SFs ranged between 0.74 and 0.86. These values are considerably greater than those typically recorded for penaeid otter trawls, seines and stow nets, which frequently are less than 0.45 (e.g. Vendeville, 1990;Sobrino et al., 2000) and often lower than 0.30 (e.g. Broadhurst et al., 2004;Macbeth et al., 2004). For example, in a study examining the selectivity of conventional otter trawls used in the Clarence River, NSW, Broadhurst et al. (2004) demonstrated that codends made from 40 mm diamondshaped mesh had L 50 s of 8.6 and 10.3 mm (i.e. SFs of 0.21 and 0.26) and SRs of 3.9 and 3.5 mm for school and eastern king prawns respectively. Although constructed from meshes that were almost 40 % smaller than these trawl codends, the commercially-used trap nets selected individuals at L 50 s that were more than 2.5 times greater. Further, this selection occurred across a substantially narrower range of sizes.
Such a relatively more-defined selection by trap nets can be attributed to their design and method of operation. Hauling the headline and footrope of the entire posterior section (i.e. approx. 70 m) over the dory effectively spread large transverse sections of the netting (e.g. > 2 m) and maintained the maximum lateral mesh openings at an area where the catch was dispersed and being progressively rolled towards the bunt (Fig. 2C). This facilitated multiple contacts between all individuals in the catch and open meshes, providing numerous opportunities for selection to occur. In contrast, most of the selection processes in otter trawls, seines and stow nets occur in the codend. At this location, the mesh openings are mostly orientated parallel to the general movement of catch, frequently narrow in proportion to the mesh size, and often blocked by the distribution of the catch (Suuronen and Millar, 1992;Erickson et al., 1996). These characteristics limit the probability of small organisms contacting open meshes and escaping.
The selection mechanisms described above for trap nets provide one explanation for the relatively high SRs observed for all penaeids from the 31 mm bunt (Fig. 4, Table 2). Owing to an increase in L 50 proportional to the mesh size, this bunt maintained a SF of between 0.67 and 0.8 for all species. However, unlike the commercially-used 25 mm trap-net configurations, this selection occurred over a wide range of sizes (e.g. SRs ± SE between 4.36 ± 0.87 and 3.90 ± 1.55 mm, Table 2) more characteristic of towed gears. These wide SRs may be attributed to the comparatively short overall length (i.e. < 20 m) of larger-meshed netting available to select individuals in the 31 mm bunt section and their fewer contacts with open meshes as they were rolled towards the dory. Given the observations for the 25 mm trap nets, it is likely that the SRs of penaeids would be reduced if the entire posterior section (i.e. 70 m) was made from the 31 mm netting (and not just the 20 m bunt section).
Considering the results from the analyses of catch comparisons in experiment 2, it is apparent that using such a larger-meshed trap net would provide little benefit in terms of reducing unwanted bycatch while maintaining commercial catches. For example, MDS and the subsequent ANOSIM tests failed to detect any significant differences in overall catch structures between the 31 and the 25 mm bunts (Fig. 5). Compared to the control, both configurations were similarly effective in excluding large quantities of those small fish (such as whitebait, gobies, glassy perchlets and hardyhead) that typically inhabit coastal lakes and estuaries throughout NSW and are vulnerable to capture by penaeidcatching gears (e.g. Andrew et al., 1995;Gray, 2001) (Fig. 6, Tables 3 and 4). Further, there was no concomitant increase in the selection of southern herring (the most common larger-sized species), probably because this species has a high dorsal profile that limits their escape through diamond meshes, regardless of their size. The 31 mm bunt did exclude considerably greater numbers of small prawns, but a large proportion of commercial-sized individuals also escaped ( Fig. 6A and C).
The results presented here confirm anecdotal claims by fishers that the trap-net configurations used in NSW selectively harvest penaeids across well-defined and targeted sizes. Compared to most other penaeid-catching gears, trap nets intuitively are less likely to negatively impact estuarine habitats. Future research into ways of mitigating the perceived deleterious effects of commercial penaeid fishing gears in some areas may therefore benefit from considering the utility and benefits of this fishing method.