Refining a Nordmøre-grid bycatch reduction device for the Spencer Gulf penaeid-trawl fishery

Incremental refinements were made to a generic Nordmøre-grid to minimise bycatches of blue swimmer crabs Portunus armatus and giant cuttlefish Sepia apama, while maintaining catches of western king prawns Melicertus latisulcatus in the Spencer Gulf penaeid-trawl fishery. These refinements involved varying bar spaces, escape-exit areas and guiding-panel lengths, and were compared against a conventional trawl. Catches of teleosts and M. latisulcatus largely remained unaffected by the changes. Maximum reductions in P. armatus and S. apama bycatches (both ~90%) were achieved with a Nordmøre-grid comprising 38-mm bar spaces, 0.81- or 1.05-m2 escape exits and a 2.7-m guiding panel. Catching fewer P. armatus should reduce abrasion and crushing of M. latisulcatus in the codend and so increase the value of this targeted species. While noting some unresolved operational concerns, these results demonstrate the potential improvements in selectivity in this fishery using a Nordmøre-grid, primarily by mechanical separation.


Introduction
Relative to other fishing methods, penaeid trawling is generally considered to be poorly selective and can result in the discarding of large quantities of bycatch, which sometimes includes threatened or endangered species [1][2][3]. The most common technique for mitigating penaeidtrawl bycatch is to install physical modifications in the posterior trawl, termed bycatch-reduction devices (BRDs), which are designed to exclude organisms mainly based on differences in behaviour ('behavioural separators') or size ('mechanical separators') [4]. The latter category has been particularly successful in several fisheries, with designs like the 'Nordmøre-grid' consistently excluding up to 90% of total bycatch while maintaining catches of penaeids [5][6][7][8].
As for many penaeid-trawl fisheries over the past 20 years, those in Australia have made progressive attempts to refine mechanical separators to reduce the bycatch of various species of concern, including elasmobranchs and turtles [5,9]. Currently, many Australian penaeidtrawl fisheries use mechanical separators and virtually all require behavioural separators [ the performance of similar designs elsewhere [8,35]. Specifically, work was done to identify an optimal bar spacing for the grid (experiment 1) and then assess the utility of increasing the escape-exit area (experiments 2 and 3) and shortening the guiding panel (experiment 3). Performance of these refinements was assessed relative to three criteria of: (i) reducing total bycatch, with particular focus on maximising the escape of P. armatus and S. apama; (ii) maintaining and improving the quality of M. latisulcatus catches; and (iii) minimising technical handling issues.

Materials and methods
Three . All experiments were done within traditional trawl grounds for M. latisulcatus in depths of 10-30 m across sandy substrata, specifically in regions where P. armatus and/or S. apama are known to occur (see S1 File for geographic coordinates of trawl paths). The trawler was rigged with two identical 'Gundry' trawl nets, each with a headline length of 14.63 m spread by flat-rectangular otter boards (1.7 × 1.1 m) and towed at~1.9 m s -1 . Both posterior trawl bodies were fitted with zippers (Buraschi S146R, 2.0 m long) to facilitate changing the extensions/codends described below.

Nordmøre-grids and codends
One control (conventional) codend and six Nordmøre-grids and their extension sections/ codends were constructed (Table 1; Fig 2). The control comprised 41-mm (stretched mesh opening-SMO) mesh (2.2-mm diameter-Ø braided, green polyethylene-PE twine) measuring 105 meshes in the normal (N) direction and 150 meshes in the transverse (T) direction. Following conventional configurations, a cylindrical panel (termed a 'crab bag') of 150-mm SMO mesh (4.0-mm Ø braided green twine) and measuring 30 T × 6.5 N was inserted 32 N anterior to the last row of meshes in the codend to separate the larger P. armatus from M. latisulcatus (Fig 2A). All grids were flat, rectangular (1978 × 1000 mm) and constructed from solid aluminium rod (frame: 20-mm Ø; bars: 16-mm Ø) ( Table 1; Fig 2B-2D). The key differences among grids were the spaces-dictated by the number of bars that could fit into the frame-and the location of a horizontal support (16-mm Ø rod) ( Table 1; Fig 2B-2D). The first two grids (experiment 1) had 35.37-(termed the '35-mm grid') and 45.00-mm ('45-mm grid') bar spaces and a centre horizontal rod (Table 1; Fig 2B and 2C). Based on their testing (see Results), the remaining four grids all had 38.22-mm bar spaces and their horizontal support located closer to the top (Table 1; Fig 2D).
Each grid was inserted at~30˚into an extension section made from 41-mm SMO mesh (3.0-mm Ø braided black PE twine) measuring 150 T × 116 N and with an anterior guiding panel (37.5 T wide; Table 1; Figs 2A and 3A-3C). The 35-and 45-mm grids both had 75-N (~3.3 m) guiding panels terminating at the grid base and triangular escape exits measuring 37 bars (B) on each side, providing an opening of 0.43 m 2 (Table 1; Figs 2A and  3A and 3B).
The four 38-mm grids differed in their escape-exit areas and/or guiding-panel lengths. The first and second 38-mm grids had a 75-N guiding panel (as above); but while the first had a triangular escape exit (termed the 'AB-exit 75N-panel grid'), the second had an escape exit made by cutting forward 1N2B either side, to provide an opening of 0.81 m 2 ('lN2B-exit 75N-panel grid') ( Table 1; Fig 3A). The third 38-mm grid also had a 1N2B escape exit, but with a shorter  Fig 3A and 3C). Each Nordmøre-grid extension had lifting straps laterally attached 30 N anterior to the escape exit, sufficient 200-mm Ø polyurethane buoys (behind the top of the grid) to achieve neutral buoyancy, and a posterior codend (and crab bag) identical to the control, except it measured 62 N (Fig 2A). Zippers were attached to all grid extensions and the control codend to facilitate attachment and removal from the two trawl bodies (Fig 2A).

Experimental design
The control was tested with the (i) 35-and 45-mm grids in experiment 1, (ii) AB-and 1N2Bexit 75N-panel grids in experiment 2, and (iii) 1N2B-exit 75N-panel and 1N2B-and 1N1B-exit 62N-panel grids in experiment 3 (Table 1). During each experiment, the control and grids being tested were alternately zippered to the nets on each side of the vessel and towed for 30 min. On each night, at least three replicates of each configuration were attempted through all possible combinations of pairings, although in experiment 2, some deployments were done where a treatment or control was not paired (i.e. another grid configuration was attached but was later omitted from analysis due to its non-identical codend).

Data collected and statistical analyses
At the end of each deployment, the following data were collected from each codend: (i) total weights of M. latisulcatus by industry categories (i.e. sorted using a Haldane PTY grading machine), including 'U8', 'U10', '10-15', '16-20' and '21-30' individuals per pound (454 g) and 'soft and broken' (i.e. post-moult and/or damaged M. latisulcatus); (ii) total weights of S. australis and Balmain bug I. peronii; (iii) total numbers, weights and, as an index of size in the absence of individual measurements, mean individual weights (total weight � total number) [36] for P. armatus, S. apama and S. novaehollandiae; (iv) total weight of the remaining miscellaneous bycatch and its main components (elasmobranchs, porifera, algae/seagrasses and teleosts); and (v) numbers of individual teleost species. Representative subsampling was required to estimate the weights and numbers of P. armatus and teleosts in the miscellaneous bycatch, and to measure total lengths (TL, rounded down to the nearest 0.5 cm) of teleosts. In addition to the above quantitative data, at the end of each deployment any debris that collected in the extensions or at the escape exits was noted and cleared and information describing on-board handling of the grids was collected. In the event of an unsuccessful deployment, the likely technical reason was noted (where possible) and the deployment was repeated.
Catch variables within each experiment were separately assessed in linear mixed models (LMMs) with all log-transformed, except the mean individual weights of key species, so that the predicted effects would be multiplicative. The LMMs included 'configuration' as a fixed effect, while 'side' of the vessel, 'night' and the interaction between 'night' and 'deployment' were included as random terms. Models were fitted using the ASReml-R package [37,38] with the significance of configuration determined using a Wald F-statistic. Significant effects of configuration were subsequently explored using the Benjamini-Hochberg-Yekutieli procedure to control the false-discovery rate (FDR) for multiple pairwise comparisons [39,40]. Predicted means were back-transformed (where required from the log scale) to the original scale using the bias-correction formula of Sprugel [41].
To facilitate interpreting results between sequential trials, the hypothesis of no inter-experimental (n = 3) differences in the mean individual weights of P. armatus, S. apama and S. novaehollandiae was also tested. In these analyses, LMMs were fitted to data from the control only, with 'experiment' fixed, and night and side used as random terms. Significant differences were separated as above.

Fishing conditions, catches and variability in mean sizes among experiments
Weather conditions during the three experiments (12-34 km h -1 winds and <1.3 m seas) and catches were typical of those experienced during conventional fishing. When pooled across experiments (147 replicate tows), the partitioned catches comprised 7152, 330 and 7377 kg of M. latisulcatus, by-product (S. australis and I. peronii combined) and bycatch, respectively (supplementary data, S1 Table). Portunus armatus, S. apama and S. novaehollandiae comprised 32.5, 2.1 and 0.9% of the total bycatch, respectively. While P. armatus and S. novaehollandiae maintained consistent trends between experiments, S. apama were more abundant during experiments 1 and 3.

Experiment 1: Assessing bar spaces (control vs 35-mm vs 45-mm grids)
Despite the large size of the grids, few technical problems were encountered, with two deployments (one for each grid) repeated after the extension twisted during setting. However, debris was occasionally observed wedged between the top of the grid and the sides of the escape exit, which required cleaning. Where possible, this debris was retrieved on deck and included in catches of the main taxonomic groups. Although on-board handling of the grids was mostly straightforward, in one deployment a rock became dislodged and fell through the escape exit while the codend was being emptied.
No significant differences were detected among configurations for any of the industry categories or total weights of M. latisulcatus, although slightly lower total catches were predicted with the 35-mm grid (by 4%) than either the 45-mm grid or control (LMM, p > 0.05; Table 2; Fig 5A). These results precipitated a decision to test the 38-mm bar space in subsequent experiments (Table 1). By comparison, significant configuration effects were detected for the numbers and weights of P. armatus and elasmobranchs, the number and weight of S. apama, and their mean individual weight, and weights of I. peronii, porifera and total bycatch (LMMs, p < 0.01; Table 2). No other variables (including all teleosts) were significantly affected by configuration (LMM, p > 0.05).

Experiment 2: Assessing escape-exit area (control vs AB-exit 75N-panel vs 1N2B-exit 75N-panel grids)
To alleviate the risk of injury from falling debris during retrieval, the crew routinely checked the grid extensions and, as required, extracted debris and brushed the grid clean as the extension was hauled aloft. There were no deployment or on-board handling issues, and clearly less clogging around the frame line of the 1N2B-exit grid than that with the smaller AB-exit. No weed/debris accumulated near the base of the grid with the horizontal support at the top third. Refining a Nordmøre-grid bycatch reduction device As in experiment 1, LMMs failed to detect significant differences among configurations for the industry categories and total weights of M. latisulcatus (p > 0.05), except for those that were soft and broken, with a 44% lower predicted mean catch by the AB-exit 75N-panel grid than the control (FDR, p < 0.05; Fig 4B). Notwithstanding the above, there also was a trend of smaller total catches of M. latisulcatus with the 1N2B-exit 75N-panel grid than the control (by 14%; Fig 5A).

Fig 6. Predicted mean individual weights (± SE) of (a) Portunus armatus, (b) Sepia apama and (c) S. novaehollandiae by configuration and experiment.
Where there was a significant configuration effect (p < 0.05) among a species within an experiment, letters above bars denote the false-discovery-rate (FDR) adjusted pairwise comparisons (p < 0.05). https://doi.org/10.1371/journal.pone.0207117.g006 Refining a Nordmøre-grid bycatch reduction device Significant configuration effects were detected for a similar set of bycatch species/groups to those identified in experiment 1, including the numbers and weights of P. armatus and elasmobranchs, and weights of porifera and total bycatch, but also the mean individual weight of P. armatus and number of P. wrighti (LMMs, p < 0.05; Table 2). No other bycatch species were affected by configuration (LMMs, p > 0.05; Table 2). Unlike in experiment 1, there were also no significant effects of configuration on the weights of I. peronii or the number, weight and mean individual weight of S. apama (LMMs, p > 0.05; Table 2). However, both species were retained in very low densities (i.e. <1 per deployment) across the trawled area.
In terms of the directions of significant differences, compared to the control, the AB-and 1N2B-exit 75N-panel grids caught significantly lower numbers and weights of P. armatus (by up to 67 and 63%, respectively) and elasmobranchs (up to 98% for both grids), and weights of porifera (91 and 93%) and total bycatch (54 and 62%) (FDRs, p < 0.05; Fig 5C and 5E-5G). The only variables not similarly affected were the number of P. wrighti, with significantly fewer (by 59%) caught by the 1N2B-exit 75N-panel grid than the AB-exit 75N-panel grid only (FDR, p < 0.05) and the mean individual weight of P. armatus, which could not be separated, although the 1N2B-exit 75N-panel grid caught smaller P. armatus (115 ± 20 g) than the ABexit 75N-panel (120 ± 21 g) or control (143 ± 20 g) (FDR, p > 0.05; Fig 6A).

Experiment 3: Assessing escape area and guiding panel length (control vs 1N2B-exit 75N-panel vs 1N2B-exit 62N-panel vs 1N1B-exit 62N-panel grids)
No technical issues were experienced with the three grids, except for one deployment needing to be repeated when the 1N2B-exit 72N-panel grid codend was empty (possibly owing to the trawl being twisted during deployment and not achieving bottom contact). There were also few issues with clogging among the grids, particularly for the 1N2B-and 1N1B-exit 62N-panel grids.
Similar to experiment 2, a significant configuration effect was detected for soft and broken M. latisulcatus (LMM, p < 0.01; Table 2), where lower weights were predicted with the 1N2Bexit and 1N1B-exit 62N-panel grids (34 and 33%, respectively) than the control (FDR, p < 0.05; Fig 4C). The only other M. latisulcatus category significantly affected by configuration was the U10 (LMM, p < 0.05; Table 2); however, this was not separated by the FDRs (p > 0.05; Fig 4C). By comparison, significant configuration effects were detected for the numbers, weights and mean individual weights of P. armatus and S. apama, number and weight of elasmobranchs, and weights of I. peronii, porifera and total bycatch (LMMs, p < 0.05; Table 2).

Discussion
The results from this study demonstrate the utility of cumulatively assessing technical changes to a generic Nordmøre-grid to approach a desired selectivity. The total bycatch reduction by the optimal design (~80%) was within the upper range of those observed for mechanical separators tested in other crustacean-trawl fisheries throughout the world (typically 50-90%) [5-8, 42, 43]. Further, the relative numbers of P. armatus and S. apama excluded by Kennelly and Broadhurst [7] were more than doubled in the present study from~35 to >80%, with no loss of M. latisulcatus. These findings can be discussed in terms of species-specific morphologies and possible behavioural/mechanical responses to the various refinements.

Grid evolution and possible escape mechanisms
Experiment 1 focused on identifying the optimal bar spacing between two grids: one with 35 mm, considered to be near the limit at which the largest M. latisulcatus (e.g. 39-49 mm carapace length with maximum carapace widths of 18-23 mm) [44] would be affected; and the other with 45 mm, tested by Kennelly and Broadhurst [7]. Neither grid significantly affected the catches of M. latisulcatus (although the 35-mm grid did retain slightly fewer in total), which largely can be attributed to the long, low-angled guiding panel directing the catch to the base of the similarly angled grid. This configuration provided minimal directional transition and clogging before sorting occurred across the entire surface of the grid [6][7][8]. Such a process clearly was sufficient for all M. latisulcatus to be more-or-less passively (considering their limited response to trawls) [45] orientated parallel to the bars and pass through. And this occurred despite a relatively (to many other penaeid fisheries) fast towing speed of~1.9 m s -1 .
Like M. latisulcatus, S. apama would have had considerable exposure to the grid surfaces and, considering their poor swimming speed, many also were probably orientated parallel to the bars. Consequently, their selection simply would have been a function of their mantle dimensions which, for experiment 1, meant only the 35-mm grid significantly reduced numbers. Portunus armatus exclusion would also be size-dependent, but their considerable morphological complexity means that spaces in the grids were likely to be less selective than for M. latisulcatus or S. apama.
In comparison, none of the teleosts caught were affected by the different bar spaces in experiment 1. Given that most teleosts were <14 cm TL and laterally compressed with widths considerably less than 35 mm, all simply passed through the grids. But, in terms of physical capabilities, many of these fish theoretically should have been able to avoid the grid if they had sufficient time to respond. Specifically, considering the horizontal distance of the grid (~1.7 m) and towing speed (~1.9 m s -1 ), organisms would need to swim a vertical distance of~1.0 m at a speed of at least~1.1 m s -1 to encounter the escape exit. Such swimming speeds should encompass the capabilities of most sizes caught [46]. Further, the proximity of the crab bag (~1 m) to the exit opening should have anteriorly displaced water (and caused a reduction in perceived flow), thereby assisting small teleosts to swim [47].
Considering the above, the lack of any behavioural escape by the teleosts implied insufficient visual cues and that bar spaces would have to be considerably narrower (or perhaps more visual) to affect their catches. Although not significant, narrower bars in the 35-mm grid promoted the escape of more S. apama, but the slight reduction in catches of M. latisulcatus implied concomitant losses with any further narrowing of spaces. For this reason, a bar spacing of~38 mm was chosen as a compromise for the subsequent experiments (i.e. those that compared different escape exits and guiding-panel lengths).
Within the 38-mm grids, increasing the escape-exit opening also failed to affect teleost catches, although the predicted means for some of the abundant species varied and, in some cases, with greater numbers caught using the grid(s) than the control. Inspection of LMM residuals did not reveal any explanation for these inconsistent results; nonetheless, it was difficult to reconcile whether the larger-exit grids facilitated the escape of more teleosts.
Escape-exit area similarly had no apparent effects on M. latisulcatus, except for the soft and broken category, where lower weights were predicted with the AB-and 1N2B-exit 75N-panel grids (although the latter was not significant). Such a result presumably reflects fewer damaged M. latisulcatus due to the reductions of P. armatus in catches [11].
Sepia apama were encountered at low densities during experiment 2, precluding any assessment of their responses to the revised grids. Further, few insights were gained from the more consistently abundant S. novaehollandiae owing to the large inter-specific size difference. Therefore, the 1N2B-exit opening grid was reassessed in experiment 3, but with a further increase in the exit opening and reduction in the guiding panel length. The rationale for these changes was to increase the opportunities for anterior escape by reducing contact distance with the grid [4].
Consistent with the earlier experiments, configuration effects were detected in experiment 3 for the same bycatch species/groups (i.e. numbers and weights of P. armatus, S. apama and elasmobranchs, and weights of porifera and total bycatch). The actual refinements appeared to be particularly beneficial in that all three Nordmøre-grids yielded the largest reductions in these categories for the whole study, while also maintaining M. latisulcatus catches. With no one particular Nordmøregrid consistently outperforming the others across all bycatch categories, the small variations manifested as slightly greater, but non-significant, reductions in total bycatch with each refinement.
An important consideration in interpreting the above observations is the potential for inter-experimental confounding of the sizes of organisms caught. Specifically, the mean individual weight of S. apama was larger in experiment 3 than in experiment 1; however, given that the difference was not significant, confounding effects might be small, particularly when compared to the disproportionately greater exclusion between configurations.

Economic and ecological benefits of using Nordmøre-grids
The catch dynamics associated with the tested Nordmøre-grids also revealed that an economic benefit should accrue to the SGPF as a result of retaining fewer P. armatus in trawls and presumed reductions in the associated damage to M. latisulcatus. Densities of both species are temporally and spatially variable, making extrapolations difficult. But, if the reduction in the soft and broken category of M. latisulcatus was proportionally allocated among all sizes, then based on recent landings and prices [48] Nordmøre-grids would increase value of the catch (not accounting for other costs associated with constructing and maintaining the grids) to the fishery by some A$0.44M per year. Such revenue would more than offset any concomitant loss of I. peronii catches (~A$0.04M per year).
The net impact of using the Nordmøre-grids on the P. armatus stock is less clear. While reductions in the bycatch of any species might normally be assumed to benefit subsequent stock(s), there is an absence of data on the fate of P. armatus escaping grids and some evidence of increased productivity of this scavenger species with trawl intensity [22,23,49]-the latter presumably due to higher volumes of discards being available as food, including its conspecifics via cannibalism. Notwithstanding the uncertainty of how the P. armatus stock might be impacted, the reduction in the remaining bycatch (i.e. excluding P. armatus) using the Nordmøre-grids-estimated by deduction to be in the order of 70-80%-is likely to have some positive ecological impacts.

Technical considerations and future research
Throughout the study, there were few issues (2% of replicate tows) with debris clogging the extensions or grids or repeated deployments due to twisting of the extension section. With more experience using grids, even fewer issues may arise. However, while a large grid appears to be a fundamental design component of a BRD for this fishery [7], the skipper noted this also made it cumbersome. Like many other fishers who are new to using such modifications, Spencer Gulf fishers are apprehensive of large mechanical-type separators due to operational and safety concerns. Acknowledging these concerns, a next phase in this work would be to undertake trials of the preferred grid system across broader spatio-temporal scales and with more fishers in the fishery. Also, refinements to the 62N-panel grid involving lighter materials, like polymers, may be worthwhile. By including operational data with the assessments of catches, it should be possible to objectively gauge any concerns fishers have with using the system and perhaps modify deployment and on-board handling procedures so that they are more acceptable to industry.

Conclusions
Combined with the previous work of Kennelly and Broadhurst [7], the present study highlights an incremental approach to the development of a Nordmøre-grid, whereby technical refinements were made over a series of experiments to identify the best grid configuration for the SGPF with respect to maximising reductions of P. armatus, S. apama and total bycatch, while maintaining M. latisulcatus catches. The greatest reductions of P. armatus, S. apama, elasmobranchs, porifera and total bycatch (with no loss of M. latisulcatus) were achieved by a large, low-angled Nordmøre-grid with 38-mm bar spaces, a support bar two thirds up the length, a guiding panel terminating~0.6 m anterior to the grid base, and a large escape exit. Further, probably due to fewer P. armatus being caught, damage to M. latisulcatus was reduced, resulting in a better quality and value of the retained target species. These results clearly demonstrate the potential for improved selectivity in this fishery using a Nordmøre-grid-primarily by mechanical exclusion of bycatch species from the target species largely owing to size and/or morphological differences.
Supporting information S1 Table. Scientific and common names and numbers (n) of bycatch species caught during experiments 1, 2 and 3, and, for subsampled teleosts, median and range of total lengths. (DOCX) S1 File. Numbers and/or weights of targeted, by-product and bycatch species/species groups caught during experiments 1, 2 and 3, and geographic coordinates of trawl paths. For columns J-AK, numbers of organisms are indicated by column headers labelled with the prefix 'num'; all other values are total weights in kg. (XLSX)