New approach for modelling size selectivity in shrimp trawl fisheries

New approach for modelling size selectivity in shrimp trawl fisheries Abstract In the deep sea trawl fishery targeting shrimp (Pandalus borealis) and other cold-water shrimp species, fishers often use a Nordmöre sorting grid ahead of a small mesh codend to avoid bycatch. However, small fish can pass through the grid and are subsequently retained in the codend. This makes shrimp size selection complex and the size-dependent curve for both the shrimp and the bycatch species often exhibits a bell-shaped signature. In this study we developed a new model and method to estimate size selection in this fishery, conducted fishing trials in the Northeast Barents Sea, and applied the new method to quantify the individual and combined size selection of the Nordmöre grid and codend for deep water shrimp and two bycatch species. The size selectivity for both bycatch species showed the expected bell-shaped signature with low retention probability of very small and larger fish. The Nordmöre grid had high passage probability for all sizes, although it decreased slightly for the largest shrimps. The smallest shrimps were released by the codend. Introduction Shrimps are a commercially important species fished over the world. Although the species and sizes of targeted shrimp vary, in many shrimp fisheries the selectivity of the gear is based upon a grid to exclude non-shrimp bycatch followed by a size selective codend. This is the case for the deep water shrimp (Pandalus borealis) fisheries in the Northeast Atlantic shrimp fishery, which have used such selectivity devices since the early 1990s. The Nordmöre grid, first developed based on a device used to exclude jellyfish, proved to be efficient in excluding fish bycatch during shrimp trawling (Isaksen et al., 1992). The grid typically consists of a guiding funnel, a 30–50° sloped grid, and a triangular fish outlet in the upper panel just in front of the grid. This bycatch reducing device was introduced in the Norwegian inshore shrimp fishery in 1990. The use of this sorting grid became compulsory for the international deep sea shrimp trawl fishery in the Northeast Atlantic (i.e. the Norwegian Sea and the Barents Sea) in 1993 and the technique is today used in several other shrimp fisheries around the world (Eayrs, 2007; He and Balzano, 2007; García et al., 2007; Suuronen and Sardà, 2007; Frimodig, 2008). All vessels targeting shrimp in Norwegian waters are obliged to use a Nordmöre grid with a 19-mm bar spacing followed by a codend with a minimum mesh size of 35 mm (Norwegian Directorate of Fisheries, 2011). Thomassen and Ulltang (1975) tested several codend mesh sizes for the northern shrimp fisheries at the end of the 1960s and found acceptable retention lengths for deep water shrimps with the 35-mm mesh size. Despite the many changes in the northern shrimp fishery that have occurred since this investigation, the minimum codend mesh size remains at 35 mm. The introduction of the Nordmöre grid eliminated the issue of bycatch of larger sizes of fish because they are unable to pass through the grid into the trawl codend (Grimaldo and Larsen, 2005; Grimaldo, 2006). However, small-sized fish such as juveniles of various species are able to pass through the grid and enter the codend together with the targeted shrimps (He and Balzano, 2007, 2013). The current regulations of the Northeast Atlantic shrimp fishery allow retention of low numbers of juvenile fish from regulated species. For example, the fishing areas are closed if a catch sample exceeds three redfish (Sebastes spp.), and three Greenland halibut (Reinhardtius hippoglossoides) per 10 kg shrimps. Additionally, the shrimp catch can contain no more than 10% by weight of undersized (i.e. < 15 mm carapace length) shrimps (Norwegian Directorate of Fisheries, 2011). These strict bycatch rules have led to frequent temporary closures of several large shrimp fishing grounds in the Northeast Atlantic over the last 20 years (Gullestad et al., 2015), often lasting for weeks or months until bycatch levels fall below the threshold. Bycatches of juvenile fish and undersized shrimps also cause practical problems when sorting the catch on-board the fishing vessels. The use of a Nordmöre sorting grid ahead of a small mesh codend makes the size selection processes complex, and the size-dependent curve for both shrimp and bycatch often exhibits a bell-shaped signature. However, a selection model that can properly describe and estimate these signature curves is not available, which challenges an ability to assess size selectivity in these fisheries. Considering the challenges described above, the aim of the current study was to: Develop a new model to estimate size selection in shrimp trawls using a Nordmöre grid and a small mesh codend. Quantify the size selectivity of deep water shrimp using a Nordmöre grid with a 19-mm bar spacing followed by a diamond mesh codend with 35-mm mesh size. Quantify the size selectivity of juveniles of some of the most important bycatch species using the Nordmöre grid and diamond mesh codend Material and methods Research vessel, study area, and experimental design The fishing trials were conducted using a commercial shrimp trawl with Nordmöre grid followed by a size selective codend. Using a procedure described in Wileman et al. (1996), the bar spacing in the Nordmöre grid was 18.8 ± 0.4 mm (mean ± SD) and the meshes in the codend were 33.8 ± 1 mm (mean ± SD). In order to independently assess the contribution of the grid and the codend to the overall selectivity of the trawl, shrimp and fish released by the grid and the codend could be collected separately using two independent covers: one over the opening of the grid and the other surrounding the entire codend. Such double cover setups have been used previously to collect selectivity data in finfish fishery (Sistiaga et al., 2010). However, in a shrimp fishery, the mesh size of the cover is very small and may affect selectivity by masking codend meshes or affecting water flow adjacent codend meshes. In addition, using two covers significantly complicates the selectivity study because the covers can become entangled, they can modify trawl performance or alter shrimp and fish behaviour, and they create operational challenges on-board. Therefore, to avoid these challenges, we used two different experimental setups during the sea trials to evaluate size selectivity of shrimp and bycatch. In the test haul setup, we fished with a standard Nordmöre grid and 35 mm codend with a small-mesh cover [mesh size 16.4 ± 0.5 mm (mean ± SD)] collecting fish and shrimps escaping from the opening in front of the Nordmöre grid (Figure 1). In the control haul setup, the codend contained a small-meshed inner net [mesh size 18.5 ± 0.9 mm (mean ± SD)] installed with a low hanging ratio to prevent the escape of fish and shrimp. In this setup, fish and shrimps that escaped in front of the Nordmöre grid were collected in a small-meshed cover [mesh size 18.9 ± 1.2 mm (mean ± SD)] (Figure 1). Test and control hauls were deployed in the same fishing area during the same cruise. Figure 1. View largeDownload slide Experimental gear configuration: with separate group of hauls with test gear (top) and control gear (bottom). The gear configuration consisting of a forward Nordmöre grid and a small mesh codend is compulsory in the Northeast Atlantic trawl fishery targeting deep-water shrimps. Figure 1. View largeDownload slide Experimental gear configuration: with separate group of hauls with test gear (top) and control gear (bottom). The gear configuration consisting of a forward Nordmöre grid and a small mesh codend is compulsory in the Northeast Atlantic trawl fishery targeting deep-water shrimps. For the test hauls, the catch was collected in the test grid cover (GT) and in the test codend (CT), whereas for control hauls, the catch was collected in the control grid cover (GC) and in the blinded codend (CC). For each haul, the catch was sorted by species, length measured, and sorted into 1-cm wide length groups for fish and 1-mm wide length groups for shrimp. Thus, the catch data consisted of count numbers (n) representing the number of individuals of the different species collected in each of the compartments. The total length of the fish was measured using a measuring board, and the carapace length of the shrimps was measured using a calliper. The fishing trials were performed on board the research trawler (R/V) “Helmer Hanssen” (63.8 m LOA and 4080 HP) from 16 to 28 February, 2016 at the fishing grounds located at the Central bank, east of Hopen Island in the north of the Barents Sea). The trials were carried out using two identical Campelen 1800# trawls built entirely of 80–40 mm meshes in the wings and belly [2 mm polyethylene (PE) twine]. Thyboron T2 (6.5 m2 and 2200 kg) trawl doors were used, and an 8-m long rope was linked between the warps 80 m in front of the doors, which kept the distance between the doors at 48–52 m during the tows. The Campelen trawl has a 19.2 m fishing line and is believed to work at its optimal wingspread (ca. 15 m) and height (ca. 6.5 m) when the door distance is kept in this range. We used 40 m double sweeps and a 19.2 m long rock hopper gear built of three sections with 46 cm rubber discs. Both trawls were equipped with 4-panel Nordmöre grid sections that are equivalent in dimensions and construction to the 2-panel standard Nordmöre grid section used by the Norwegian coastal fleet targeting shrimp. Each Nordmöre grid is made of stainless steel and measures 1510 mm high and 1330 mm wide. The grid in both trawls used was mounted so that it would maintain an angle of 45 ± 2.5° while fishing. Model for size selection The size selection system consists of two main parts. The first part is a Nordmöre grid, which the fish and shrimps must pass through to enter the codend. If they do not pass through this grid they are excluded during this first part of the selection process. To pass through the grid, two conditions need to be fulfilled: (a) they need to contact the grid and (b) morphologically they must be able to pass through the grid, which is dependent on their size and orientation when they reach with the grid. The second part is a codend that collects fish and shrimp that pass through the grid. The codend is also size selective, and its size selection is the second part of the combined selection process. Thus, for a fish or deep water shrimp to be retained by the gear [rcombined(l)], it must be retained by selectivity process of the Nordmöre grid [rgrid(l)] and the codend [rcodend(l)]:   rcombinedl=rgridl×rcodendl (1) where l denotes the length of the fish or the length of the shrimp carapace. This system can be defined as a sequential dual selection system. It is a dual system because it consists of two processes and it is sequential because the second process follows the first. The next step is to model each of the two size selection processes individually. For the grid selectivity process we need to consider that some fish or shrimps might not contact the Nordmöre grid at all [see Larsen et al. (2016) for the explanation of contact in this context] or that they might do so orientated in a way that independent of length they have no chance of passing though the grid. The probability of length-independent exclusion from the Normöre grid, Cgrid, is specified as a parameter with a range of 0.0 to1.0, where 1.0 means that every individual of the species contacts the grid in a way that gives it a length-dependent chance of passing through the grid. For an individual contacting the grid with sufficiently good orientation to give it a length-dependent chance of passing through grid (rcontactgrid(l)), the following logit model was used [see Wileman et al. (1996) for further information about the logit model]:   rcontactgridl,L50grid,SRgrid=1.0-logitl,L50grid,SRgrid=1.01.0+expln9SRgrid×l-L50grid (2) Model (2) considers that the probability of being able to pass through the grid is length dependent and will decrease for larger individuals. L50grid denotes the length of fish or shrimp with 50% probability of being retained, and SRgrid (selection range) describes the difference in length between fish or shrimp with 75 and 25% probability of being retained, respectively. Based on the above, the following model was used for the size selection in the first process [rgrid(l)]:   rgridl,Cgrid,L50grid,SRgrid=Cgrid1.0+expln9SRgrid×l-L50grid (3) The escape probability through the outlet in front of the Nordmöre grid was therefore modelled by:   egridl,Cgrid,L50grid,SRgrid=1.0- rgridl,Cgrid,L50grid,SRgrid=1.0- Cgrid1.0+expln9SRgrid×l-L50grid (4) For the small mesh codend selectivity process, we assumed that the retention probability can be modelled by a logit model:   rcodendl,L50codend,SRcodend=logitl,L50codend,SRcodend=expln9SRcodend×l-L50codend1.0+expln9SRcodend×l-L50codend (5) Thus, by inserting (3) and (5) into (1), we arrived at the following combined size selection model:   rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodend=Cgrid1.0+expln9SRgrid×l-L50grid×expln9SRcodend×l-L50codend1.0+expln9SRcodend×l-L50codend (6) Model (6) is a so-called structural model because it is based on modelling the individual processes expected to be involved in the combined size selection in the system. One advantage of applying a structural model compared to an empirical-based model is that once the values of the parameters in the model are estimated, they can be applied to investigate not only the combined processes in the system but also the individual processes. In this context, each of the model parameter values obtained can be directly interpreted. In the case of model (6), five parameters need to be estimated to be able to describe the size selection in the system: Cgrid, L50grid, SRgrid, L50codend, and SRcodend. If all individuals contact the grid with a mode of contact that provide a length dependent probability to pass through, then the value for Cgrid should be 1.0. However, this is not necessarily the case, as some individuals may escape through the escape outlet in front of the Nordmöre grid (Figure 1) without contacting the grid first. Other individuals may be so poorly orientated when they meet the grid that the probability of them passing through will be similar to those not contacting the grid, which will also be reflected in the value of Cgrid. Thus, L50grid and SRgrid are respectively the L50 and SR values for individuals contacting the grid with a reasonable mode of orientation. L50codend and SRcodend are the L50 and SR values for the codend selection conditional upon organisms contacting and passing through the Nordmöre grid (Figure 1). As different species have different morphology and behaviour, values of the parameters Cgrid, L50grid, SRgrid, L50codend, and SRcodend for the same combined system will be species specific. Therefore, the analysis was applied separately for the different fish species and for the deep water shrimp. Data analysis and parameter estimation Catch data were collected in two groups. One of the groups consisted of control hauls obtained by summing compartments GC and CC (Figure 1). Together, they sampled the size and species composition of fish entering the selective parts of the trawl (section with the Nordmöre grid and codend), and in this respect the control hauls can be paired with the test hauls so that a paired-gear estimation method can be used (Wileman et al., 1996). However, compared to the standard paired-gear method in which none of the selective parts of the system uses covers to collect fish or shrimps escapees, our test hauls are special because they use a cover (GT) to collect fish and shrimp escaping ahead of the Nordmöre grid. Therefore, our experimental data collection design represents a combination of the paired and covered data collection and estimation methods (Wileman et al., 1996). To estimate the average size selection of the Nordmöre grid and codend in the test trawl , we paired the pooled catch data with the pooled catch data from the control hauls. Based on this approach, the experimental data in the analysis were treated like three compartment data. Fish or shrimp caught were observed in GT, CT, or (GC + CC). For the estimation based on the size selection model established in the “Model for size selection” section, we needed to express the probabilities that fish or shrimp of a specific length l would be observed in each of these three compartments conditioned they were caught. The probability that the fish or shrimp would enter the selection section in one of the test hauls and in one of the control hauls was modelled by the split factor, SP, as is traditionally done for paired-gear data analysis (Wileman et al., 1996). This means that the probability that a fish or shrimp will enter the test haul is SP, whereas the probability of them entering the control haul is 1.0-SP. All fish or shrimp entering the control haul are retained because both the cover over the grid outlet and the blinded codend retain all potential escapees. For a fish or shrimp entering one of the hauls included in the analysis (test or control), the probability that it will be retained in the cover in front of the Nordmöre grid in one of the test hauls would, based on Equation (4), be SP×egridl,Cgrid,L50grid,SRgrid. For a fish or shrimp entering one of the hauls included in the analysis, the probability that it will be retained in the codend of a test haul would, based on Equation (6), be SP×rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodend. Considering this, the probability γ that a fish or shrimp entering one of the test or control hauls will be observed in one of the three compartments (GT, CT, or GC + CC) can be expressed as:   γl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=SP×egridl,Cgrid,L50grid,SRgrid+SP×rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodend+1.0-SP (7) Based on Equation (7) and the considerations above, the probabilities pGT, pCT, and pGC+CC that a fish or shrimp observed in the catch will be found in compartment GT, CT, or GC + CC, respectively, can be expressed by:   pGTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=SP×egridl,Cgrid,L50grid,SRgridγl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SPpCTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=SP×rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodendγl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SPpGC+CCl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=1.0-SPγl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP (8) Using Equation (8), the values for the parameters in the selection model (6) can be estimated from the collected experimental data by minimizing the following function with respect to Cgrid,L50grid,SRgrid,L50codend,SRcodend,and SP:   -∑l∑i=1anGTliqGTi×lnpGTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP+∑i=1anCTliqCTi×lnpCTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP+∑j=1bnGCljqGCj+nCCljqCCj×lnpGC+CCl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP} (9) where the outer summation is over length classes l in the experimental data and the inner summation is over experimental fishing hauls i (from 1 to a) and j (from 1 to b) with, respectively, the test and control setup. nGTli, nCTli,  nGClj, and nCClj are the number of fish or shrimp length measured of length class l in haul i and j in the respective compartment, with qGTi, qCTi, qGCj, and qCCj being the corresponding sampling factors (i.e. the fraction of the catch that was length measured). The estimation with (9) is based on the expected total catches and the sampling factors are therefore used to expand the subset of length-specific counts based upon the fractions of the haul measured for lengths. Minimizing (9) with respect to the parameters in it is the same as maximizing the likelihood for the observed experimental data based on a multinomial model, assuming that the formulated model (8) describes the experimental data sufficiently well. The observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC, which model (8) is expected to describe, are given by:   pGTl^=nGTlqGTnGTlqGT+nCTlqCT+nGClqGC+nCClqCCpCTl^=nCTlqCTnGTlqGT+nCTlqCT+nGClqGC+nCClqCCpGC+CCl^=nGClqGC+nCClqCCnGTlqGT+nCTlqCT+nGClqGC+nCClqCC. (10) Due to the experimental procedure followed, there was no obvious way to pair the data from the individual test and control hauls. Hence, to estimate the mean selectivity parameters for the experimental gear, the length dependent expected total catches for the test hauls were combined and compared with the combined expected total catches for the control hauls as formulated in function (9). The confidence limits for the parameters and curves for the size selection model were estimated using a double bootstrapping method that accounts for the uncertainty resulting from this unpaired nature of the data collection. For this, we adopted and further generalized the method for estimating uncertainty in size selectivity based on unpaired trawl data described by Sistiaga et al. (2016). This procedure accounts for uncertainty caused by between-haul variation in size selection processes (Fryer, 1991) and by the unpaired data collection method with groups of test and control hauls by selecting independently a hauls with replacement from the test hauls and b hauls with replacement from the control hauls during each bootstrap loop. Uncertainty caused by finite sample sizes on haul level (within-haul variability) is accounted for by randomly selecting fish with replacement from each of the selected hauls for each compartment separately, where the number selected from each compartment in each haul is the same as the number sampled in that compartment in that haul. These data are then raised and combined as described above, and the selectivity parameters are again estimated. The additional uncertainty in the estimation caused by subsampling is automatically accounted for by raising the data after the re-sampling (Eigaard et al., 2012). We performed 1000 bootstrap repetitions to calculate the 95% percentile confidence limits (Efron, 1982; Chernick, 2007) for the selection parameters and curves. The model’s ability to describe the experimental data sufficiently well was evaluated based on the p-value, model deviance versus degrees of freedom (DOF) and inspection of how the model curve reflects the length-based trend in the data (Wileman et al., 1996). The p-value expresses the likelihood to obtain at least as big a discrepancy between the fitted model and the observed experimental data by coincidence. In case of a poor fit statistics (p-value being < 0.05; deviance being ≫DOF), the model curve plots were inspected to determine whether the poor result was due to structural problems when describing the experimental data using the model or if it was due to over-dispersion in the data (Wileman et al., 1996). The analysis was carried out using the software SELNET (Herrmann et al., 2012, 2013a,b), which implements the models and the bootstrap method described above. Results Collected data We conducted eight control hauls (Table 1) and eight test hauls (Table 2). To ensure that the group of test and control hauls were fishing a similar size and species compositions all hauls were conducted within 5 days in February 2016 at same fish ground within a small area, at similar depth and with similar towing time (Tables 1 and 2). In all hauls, a subsample of the shrimp catch was length measured whereas catches of every American plaice (Hippoglossoides platessoides) and redfish (Sebastes spp.) were length measured. The total number of shrimp, American plaice and redfish length measured during the data collection were respectively 4405, 8773, and 4439 individuals. Table 1. Overview of the fish and shrimp length measured in the control hauls carried out during the trials. Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GC (% measured)  CC (% measured)  GC  CC  GC  CC  1  60  268  20  7604.9 N  03526.9 E  123 (72.31%)  160 (1.63%)  208  177  56  36  2  61  257  20  7605.4 N  03517.8 E  120 (58.14%)  153 (1.95%)  238  182  143  37  3  60  278  21  7605.3 N  03511.1 E  163 (7.47%)  173 (1.16%)  438  187  404  169  4  60  271  21  7605.9 N  03533.8 E  108 (9.60%)  171 (1.20%)  265  156  184  86  5  63  266  21  7605.9 N  03521.9 E  144 (40.54)  160 (1.91%)  321  121  108  20  6  61  271  22  7606.5 N  03531.9 E  169 (100%)  175 (2.02%)  206  150  68  34  7  60  271  22  7606.6 N  03521.9 E  208 (22.74)  169 (1.02%)  391  287  187  94  8  63  272  22  7606.5 N  03531.9 E  189 (21.12)  190 (0.73%)  327  301  164  120  Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GC (% measured)  CC (% measured)  GC  CC  GC  CC  1  60  268  20  7604.9 N  03526.9 E  123 (72.31%)  160 (1.63%)  208  177  56  36  2  61  257  20  7605.4 N  03517.8 E  120 (58.14%)  153 (1.95%)  238  182  143  37  3  60  278  21  7605.3 N  03511.1 E  163 (7.47%)  173 (1.16%)  438  187  404  169  4  60  271  21  7605.9 N  03533.8 E  108 (9.60%)  171 (1.20%)  265  156  184  86  5  63  266  21  7605.9 N  03521.9 E  144 (40.54)  160 (1.91%)  321  121  108  20  6  61  271  22  7606.5 N  03531.9 E  169 (100%)  175 (2.02%)  206  150  68  34  7  60  271  22  7606.6 N  03521.9 E  208 (22.74)  169 (1.02%)  391  287  187  94  8  63  272  22  7606.5 N  03531.9 E  189 (21.12)  190 (0.73%)  327  301  164  120  For fish species, all specimens were measured for length. For shrimp, values in brackets represent the proportion of individuals subsampled for length measurements. The catch was collected in the control gear grid cover (GT) and in the gear control gear codend (CT). Table 2. Overview of the fish and shrimp length measured in the test hauls carried out during the trials. Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GT (% measured)  CT (% measured)  GT  CT  GT  CT  9  60  268  22  7606.1 N  03522.3 E  150 (63.13%)  150 (1.34%)  391  283  211  42  10  62  265  23  7605.4 N  03523.2 E  123 (31.72%)  146 (0.94%)  444  347  392  65  11  64  268  23  7605.8 N  03525.1 E  98 (66.77%)  134 (1.05%)  482  402  494  108  12  62  265  23  7605.7 N  03522.1 E  7 (100%)  121 (2.10%)  283  309  211  47  13  63  274  23  7605.9 N  03523.4 E  21 (100%)  141 (1.76%)  239  212  354  91  14  60  256  24  7604.7 N  03516.8 E  50 (100%)  161 (2.67%)  256  202  98  33  15  63  252  24  7604.0 N  03512.9 E  75 (80.61%)  146 (1.08%)  230  320  135  82  16  66  269  24  7606.1 N  03517.2 E  140 (8.18%)  167 (1.78%)  298  120  142  24  Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GT (% measured)  CT (% measured)  GT  CT  GT  CT  9  60  268  22  7606.1 N  03522.3 E  150 (63.13%)  150 (1.34%)  391  283  211  42  10  62  265  23  7605.4 N  03523.2 E  123 (31.72%)  146 (0.94%)  444  347  392  65  11  64  268  23  7605.8 N  03525.1 E  98 (66.77%)  134 (1.05%)  482  402  494  108  12  62  265  23  7605.7 N  03522.1 E  7 (100%)  121 (2.10%)  283  309  211  47  13  63  274  23  7605.9 N  03523.4 E  21 (100%)  141 (1.76%)  239  212  354  91  14  60  256  24  7604.7 N  03516.8 E  50 (100%)  161 (2.67%)  256  202  98  33  15  63  252  24  7604.0 N  03512.9 E  75 (80.61%)  146 (1.08%)  230  320  135  82  16  66  269  24  7606.1 N  03517.2 E  140 (8.18%)  167 (1.78%)  298  120  142  24  For fish species, all specimens were measured for length. For shrimp, values in brackets represent the proportion of individuals subsampled for length measurements. The catch was collected in the test grid cover (GT) and in the test codend (CT). Size selectivity for shrimp The model used reflected the pattern observed in the experimental data well (Figure 2a–c). Thus, although the p-value (Table 3) observed in the fit statistics was low, we are confident that the model used represents the data adequately. We considered the low p-value to be a consequence of over-dispersion in the experimental catch portioning data that resulted from working with pooled and subsampled data with low sampling rates (Tables 1 and 2). Such cases have led to low p-values and high dispersion (Brčić et al., 2015; Alzorriz et al., 2016; Notti et al., 2016) especially for length classes with relative low frequency, such as our shrimp data for individuals with carapace length above 25 mm (Figure 2a–c). All shrimp were estimated to make contact with the Nordmöre grid, and most of them passed through it. However, the grid passage probability was estimated to decrease slightly with increasing shrimp size. The codend selectivity showed size-dependent release for shrimp with carapace length < 23 mm, with only about 20% of the shrimp with carapace length of 15 mm being retained by the codend. L50codend was estimated to be 17.72 mm, and SRcodend was estimated to be 3.63 mm (Table 3). L50grid was 49.2 mm, which at first glance could seem meaningless because it is above the size range for this species of shrimp (Table 3). However, this value is expected to be above a biologically meaningful value and confirms that most of the shrimp can pass through the grid except for the slight decrease for large shrimp. The combined selectivity for the grid and codend exhibited a slightly bell-shaped signature, with few shrimp < 15 mm being retained, a maximum retention rate for shrimp with carapace length of 25 mm, and a slight decrease for shrimp above this size. Table 3. Size selectivity parameters and fit statistics results for shrimps, American plaice and redfish based on fitting the model (8) to the experimental data.   Shrimps  American Plaice  Redfish  Cgrid  1.00 (0.98–1.00)  1.00 (0.97–1.00)  0.90 (0.75–0.99)  L50grid  49.17 (37.16–68.57)  19.40 (18.41–20.20)  13.61 (13.06–14.28)  SRgrid  16.52 (8.02–27.82)  7.47 (6.44–8.61)  3.46 (2.93–3.97)  L50codend  17.72 (16.10–22.59)  6.84 (5.46–7.68)  9.78 (8.85–10.45)  SRcodend  3.63 (1.79–9.45)  1.66 (0.10–2.66)  1.74 (1.33–2.60)  SP  0.51 (0.42–0.70)  0.55 (0.49–0.61)  0.63 (0.51–0.74)  DOF  34  90  54  Deviance  175.66  118.38  101.91  p-value  <0.0001  0.0241  0.0001    Shrimps  American Plaice  Redfish  Cgrid  1.00 (0.98–1.00)  1.00 (0.97–1.00)  0.90 (0.75–0.99)  L50grid  49.17 (37.16–68.57)  19.40 (18.41–20.20)  13.61 (13.06–14.28)  SRgrid  16.52 (8.02–27.82)  7.47 (6.44–8.61)  3.46 (2.93–3.97)  L50codend  17.72 (16.10–22.59)  6.84 (5.46–7.68)  9.78 (8.85–10.45)  SRcodend  3.63 (1.79–9.45)  1.66 (0.10–2.66)  1.74 (1.33–2.60)  SP  0.51 (0.42–0.70)  0.55 (0.49–0.61)  0.63 (0.51–0.74)  DOF  34  90  54  Deviance  175.66  118.38  101.91  p-value  <0.0001  0.0241  0.0001  L50 and SR values are given in mm for shrimps and in cm for American plaice and redfish. Values in brackets are 95% confidence limits. Figure 2. View largeDownload slide Size selectivity plots for shrimps. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of shrimps found in the grid cover of the test gear, plot “b” shows the length dependent portioning of shrimps observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of shrimps found in the control gear. The points in plots “a”-”c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Figure 2. View largeDownload slide Size selectivity plots for shrimps. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of shrimps found in the grid cover of the test gear, plot “b” shows the length dependent portioning of shrimps observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of shrimps found in the control gear. The points in plots “a”-”c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Size selectivity for American plaice The model used reflected the pattern observed in the experimental data well (Figure 3a–c). Despite the p-value being < 0.05 (Table 3), the model represented the data adequately and we are confident about the performance of the model. All American plaice were estimated to make contact with the Nordmöre grid, and most of them passed through it. The grid passage probability was very high for American plaice up to 12 cm long, followed by a decrease and then very low passage probability for fish > 30 cm long. The codend only showed low size selectivity for American plaice with an L50codend value of 6.84 cm, thus all American plaice > 10 cm long that entered the codend were retained (Table 3). The combined selectivity for the grid and codend showed a clear bell-shaped signature, with a high retention probability for American plaice ∼10 cm long (ca. 90% retention). Retention of individuals < 5 cm long was practically 0 and retention of fish in the range of 10–30 cm decreased, with really low retention rates for fish > 30 cm long. In the range of 6–23 cm, retention probability for American plaice for the gear was >25%, meaning that this gear would not be adequate in areas where the numbers of American plaice within this range are high. Figure 3. View largeDownload slide Size selectivity plots for American plaice. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of American plaice found in the grid cover of the test gear, plot “b” shows the length dependent portioning of American plaice observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of American plaice found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Figure 3. View largeDownload slide Size selectivity plots for American plaice. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of American plaice found in the grid cover of the test gear, plot “b” shows the length dependent portioning of American plaice observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of American plaice found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Size selectivity for redfish For redfish, the model used represented the experimental data well up to l = 22 cm (Figure 4a–c). Because fish above this size are outside the selective range of the gear, the model adequately describes the size selection process in the gear. The discrepancy between model and experimental data above the selective range of the gear is regarded as a consequence of unequal entry of bigger redfish into the test and control gears. The combined size selection showed a clear bell-shaped signature, with >60% of redfish around 12 cm long being retained by the gear but <25% of redfish <9 cm and >15 cm long being retained. The grid passage probability was high (>80%) for redfish <12 cm long, and it decreased monotonously, with no redfish >20 cm entering the codend. The codend size selection showed that none of the redfish >14 cm would be released by the codend and that L502 and SR2 were 9.8 and 1.7 cm, respectively (Table 3). Figure 4. View largeDownload slide Size selectivity plots for redfish. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of redfish found in the grid cover of the test gear, plot “b” shows the length dependent portioning of redfish observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of redfish found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC (Equation 10). The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Figure 4. View largeDownload slide Size selectivity plots for redfish. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of redfish found in the grid cover of the test gear, plot “b” shows the length dependent portioning of redfish observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of redfish found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC (Equation 10). The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Discussion The size selection data resulting from the grid and codend configuration used in this study were based on a new model and estimation method that is an extension of the unpaired method described in Sistiaga et al. (2016). This new approach models the observed data summed over hauls for a group of test and controls hauls, and it combines a structural dual sequenced size selection model with unpaired data collection for groups of test and control hauls. This model effectively described the length-dependent portioning of the observed catch between the test codend, the test grid cover, and the control gear for all species investigated. In addition to enabling estimation of the combined size selection for the Nordmöre grid followed by the diamond mesh codend, this method enabled estimation of the size selection for each of the selection devices individually because the structural model explicitly described the selectivity processes in each of the devices. Structural size selection models have previously been developed and applied to describe size selection in other trawl fisheries. This includes models for fish sorting grids in combination with codends in finfish fisheries (Sistiaga et al., 2010; Herrmann et al., 2013a), square mesh panels in combination with selective codends (Zuur et al., 2001; O’Neill et al., 2006; Alzorriz et al., 2016), double grid sorting devices (Larsen et al., 2016: Sistiaga et al., 2016), and excluding grids combined with a selective codend (Brčić et al., 2015; Stepputtis et al., 2015; Lövgren et al., 2016). Excluding grids combined with a selective codend result in the same bell-shaped selection pattern as the Nordmöre grid followed by a size selective codend. However, our study is the first time such a modelling process has been applied to a shrimp trawl fishery and the first time that a sequential model with two compartment data collection in test and control gears have been used. Our gear configuration was more complex than the methods previously applied, but it is necessary due to the practical problems that potentially could have resulted from using a small mesh cover over the test codend. The method and model presented herein offer new possibilities for studying size selectivity in other shrimp fisheries. In particular, our approach enables detailed mapping of which sizes of bycatch species would have especially high risk of being caught if they are abundant in the shrimp fishing grounds. In this study, we demonstrated the ability of the new model to represent bell-shaped selectivity data in detail for shrimp and two fish bycatch species: American plaice and redfish. For the juvenile bycatch species, our results demonstrated very high and length-dependent grid passage probability. Thus, in conjunction with the small-meshed diamond mesh codend used in the shrimp fishery, the gear has high catch risk for certain size ranges of these bycatch species. The use of the combined bycatch reducing and size selective system consisting of the Nordmöre grid and 35 mm codend mesh is well established in the Northeast Atlantic shrimp fishery. However, the data from our study clearly show that fish within a limited size range and undersized shrimps retained in the 35-mm codend will continue to be a problem for the northern shrimp fleet. Some precaution is required since our fishing trial was based on only 16 hauls and the number of shrimp, American Plaice and redfish length measured was respectively limited to 4405, 8773, and 4439 individuals (Tables 1 and 2). This leads to a degree of uncertainty in the estimated size selection curves (Figures 2–4) and needs to be considered when interpreting the results. However, this uncertainty is reflected in the selectivity parameters (Table 3) and confidence bands around the size selection curves. A recent study by Herrmann et al. (2016) has shown that estimating trawl size selectivity to a specific level of precision based on a paired gear method will require around 10 times as many fish as needs to be measured using a covered method. Based on that claim we had concerns whether it would be possible to estimate size selectivity with an acceptable level of precision as our gear configuration uses a combination of a paired and covered sampling design. However, the relative narrow confidence bands around the size selection curves in our study demonstrates the practicality and appropriateness of our method. Acknowledgements We thank the crew of RV “Helmer Hanssen” and assistants Ingeborg Sætra, Hermann Pettersen, Tomas Araya, and postdoctoral researcher André Frainer for valuable assistance on board. We appreciate the efforts of the editor and the three anonymous reviewers, which we feel improved our manuscript significantly. We are also grateful to the Arctic University of Norway UIT in Tromsø and the Norwegian Seafood Research Fund for funding the experiments carried out in this study. References Alzorriz N., Arrregi L., Herrmann B., Sistiaga M., Casey J., Poos J. J. 2016. Questioning the effectiveness of implemented technical measures under the EU landings obligation: the Basque Otter Bottom Trawl fishery case study. Fisheries Research , 175: 116– 126. Google Scholar CrossRef Search ADS   Brčić J., Herrmann B., Sala A. 2015. Selective characteristics of a shark-excluding grid device in a Mediterranean trawl. Fisheries Research , 172: 352– 360. Google Scholar CrossRef Search ADS   Chernick M. R. 2007. Bootstrap Methods: A Guide for Practitioners and Researchers, 2nd edn. Wiley Series in Probability and Statistics . Wiley, New York. Eayrs S. 2007. A Guide to Bycatch Reduction in Tropical Shrimp-Trawl Fisheries  [ Revised edition]. FAO, Rome. 108 pp. Efron B. 1982. The jackknife, the bootstrap and other resampling plans. SIAM Monograph No 38, CBSM-NSF. Eigaard O. R., Herrmann B., Nielsen J. R. 2012. Influence of grid orientation and time of day on grid sorting in a small-meshed trawl fishery for Norway pout (Trisopterus esmarkii). Aquatic Living Resources , 25: 15– 26. Google Scholar CrossRef Search ADS   Frimodig A. J. 2008. Bycatch Reduction Devices Used in the Pink Shrimp Trawl Fishery. Report to the California Fish and Game Commission. California Department of Fish and Game Marine Region State Fisheries Evaluation Project (4/14/2008). 12 pp. Fryer R. J. 1991. A model of between-haul variation in selectivity. ICES Journal of Marine Science , 48: 281– 290. Google Scholar CrossRef Search ADS   García E.G. (ed), Ragnarsson S. Á., Steingrimsson S. A. 2007. Bottom Trawling and Scallop Dredging in the Arctic: Impacts of Fishing on Non-Target Species, Vulnerable Habitats and Cultural Heritage . Nordic Council of Ministers, Copenhagen. 374 pp. Grimaldo E. 2006. The effects of grid angle on a modified Nordmöre-grid in the Nordic Shrimp Fishery. Fisheries Research , 77: 53– 59. Google Scholar CrossRef Search ADS   Grimaldo E., Larsen R. B. 2005. The cosmos grid: a new design for reducing by-catch in the Nordic shrimp fishery. Fisheries Research , 76: 187– 197. Google Scholar CrossRef Search ADS   Gullestad P., Blom G., Bakke G., Bogstad B. 2015. The “Discard Ban Package”: Experiences in efforts to improve the exploitation patters in Norwegian fisheries. Marine Policy , 54: 1– 9. Google Scholar CrossRef Search ADS   He P., Balzano V. 2007. Reducing the catch of small shrimps in the Gulf of Maine pink shrimp fishery with a size-sorting grid device. ICES Journal of Marine Science , 64: 1551– 1557. Google Scholar CrossRef Search ADS   He P., Balzano V. 2013. A new shrimp trawl combination grid system that reduces small shrimp and finfish bycatch. Fisheries Research , 140: 20– 27. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M. B., Nielsen K. N., Larsen R. B. 2012. Understanding the size selectivity of redfish (Sebastes spp.) in North Atlantic trawl codends. Journal of Northwest Atlantic Fishery Science , 44: 1– 13. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M., Larsen R. B., Nielsen K. N. 2013a. Size selectivity of redfish (Sebastes spp.) in the Northeast Atlantic using grid-based selection systems for trawls. Aquatic Living Resources , 26: 109– 120. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M., Larsen R. B., Nielsen K. N., Grimaldo E. 2013b. Understanding sorting grid and codend size selectivity of Greenland halibut (Reinhardtius hippoglossoides). Fisheries Research , 146: 59– 73. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M., Santos J., Sala A. 2016. How many fish need to be measured to effectively evaluate trawl selectivity? PLoS One , 11: e0161512. Google Scholar CrossRef Search ADS PubMed  Isaksen B., Valdemarsen J. W., Larsen R. B., Karlsen L. 1992. Reduction of fish by-catch in shrimp trawl using a rigid separator grid in the aft belly. Fisheries Research , 13: 335– 352. Google Scholar CrossRef Search ADS   Larsen R. B., Herrmann B., Sistiaga M., Grimaldo E., Tatone I., Onandia I. 2016. Size selection of redfish (Sebastes spp.) in a double grid system: quantifying escapement through individual grids and comparison to former grid trials. Fisheries Research , 183: 385– 395. Google Scholar CrossRef Search ADS   Lövgren J., Herrmann B., Feekings J. 2016. Bell-shaped size selection in a bottom trawl: a case study for Nephrops directed fishery with reduced catches of cod. Fisheries Research , 184: 26– 35. Google Scholar CrossRef Search ADS   Norwegian Directorate of Fisheries. 2011. J-209-2011: Forskrift om maskevidde, bifangst og minstemål m.m. ved fiske i fiskevernsonen ved Svalbard (In Norwegian). Notti E., Brčić J., De Carlo F., Herrmann B., Lucchetti A., Virgili M., Sala A. 2016. Assessment of the relative catch performance of a surrounding net without the purse line as an alternative to a traditional boat seine in small scale fisheries. Marine and Coastal Fisheries , 8: 81– 91. Google Scholar CrossRef Search ADS   O’Neill F. G., Kynoch R. J., Fryer R. J. 2006. Square mesh panel in North Sea demersal trawls: separate estimates of panel and codend selectivity. Fisheries Research , 78: 333– 341. Google Scholar CrossRef Search ADS   Thomassen T., Ulltang Ø. 1975. Report from mesh selection experiments on Pandalus borealis in Norwegian waters. ICES CM 1975/K:51. Sistiaga M., Herrmann B., Grimaldo E., Larsen R. B. 2010. Assessment of dual selection in grid based selectivity systems. Fisheries Research , 105: 187– 199. Google Scholar CrossRef Search ADS   Sistiaga M., Brinkhof J., Herrmann B., Grimaldo, Langård L., Lilleng D. 2016. Size selection performance of two flexible sorting grid section designs in the Northeast Arctic cod (Gadus morhua) and haddock (Melanogrammus aeglefinnus) fishery. Fisheries Research , 183: 340– 351. Google Scholar CrossRef Search ADS   Stepputtis D., Santos J., Herrmann B., Mieske B. 2015. Broadening the horizon of size selectivity in trawl gears. Fisheries Research , 184: 18– 25. Google Scholar CrossRef Search ADS   Suuronen P., Sardà F., 2007. By-catch reduction techniques in European fisheries traditional methods and potential innovations. In Bycatch Reduction in the World’s Fisheries. Reviews: Methods and Technologies in Fish Biology and Fisheries  7, pp. 37– 74. Ed. by Kennelly S. J.. Springer, The Netherlands. 288 pp. Wileman D. A., Ferro R. S. T., Fonteyne R., Millar R. B. (Eds.). 1996. Manual of methods of measuring the selectivity of towed fishing gears. ICES Cooperative Research Report No. 215. Zuur G., Fryer R. J., Ferro R. S. T., Tokay T. 2001. Modelling the size selectivities of a trawl codend and an associated square mesh panel. ICES Journal of Marine Science , 58: 657– 671. Google Scholar CrossRef Search ADS   © International Council for the Exploration of the Sea 2017. All rights reserved. For Permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png ICES Journal of Marine Science Oxford University Press

New approach for modelling size selectivity in shrimp trawl fisheries

Loading next page...
 
/lp/ou_press/new-approach-for-modelling-size-selectivity-in-shrimp-trawl-fisheries-OLEFQG3RFc
Publisher
Oxford University Press
Copyright
© International Council for the Exploration of the Sea 2017. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
1054-3139
eISSN
1095-9289
D.O.I.
10.1093/icesjms/fsx117
Publisher site
See Article on Publisher Site

Abstract

Abstract In the deep sea trawl fishery targeting shrimp (Pandalus borealis) and other cold-water shrimp species, fishers often use a Nordmöre sorting grid ahead of a small mesh codend to avoid bycatch. However, small fish can pass through the grid and are subsequently retained in the codend. This makes shrimp size selection complex and the size-dependent curve for both the shrimp and the bycatch species often exhibits a bell-shaped signature. In this study we developed a new model and method to estimate size selection in this fishery, conducted fishing trials in the Northeast Barents Sea, and applied the new method to quantify the individual and combined size selection of the Nordmöre grid and codend for deep water shrimp and two bycatch species. The size selectivity for both bycatch species showed the expected bell-shaped signature with low retention probability of very small and larger fish. The Nordmöre grid had high passage probability for all sizes, although it decreased slightly for the largest shrimps. The smallest shrimps were released by the codend. Introduction Shrimps are a commercially important species fished over the world. Although the species and sizes of targeted shrimp vary, in many shrimp fisheries the selectivity of the gear is based upon a grid to exclude non-shrimp bycatch followed by a size selective codend. This is the case for the deep water shrimp (Pandalus borealis) fisheries in the Northeast Atlantic shrimp fishery, which have used such selectivity devices since the early 1990s. The Nordmöre grid, first developed based on a device used to exclude jellyfish, proved to be efficient in excluding fish bycatch during shrimp trawling (Isaksen et al., 1992). The grid typically consists of a guiding funnel, a 30–50° sloped grid, and a triangular fish outlet in the upper panel just in front of the grid. This bycatch reducing device was introduced in the Norwegian inshore shrimp fishery in 1990. The use of this sorting grid became compulsory for the international deep sea shrimp trawl fishery in the Northeast Atlantic (i.e. the Norwegian Sea and the Barents Sea) in 1993 and the technique is today used in several other shrimp fisheries around the world (Eayrs, 2007; He and Balzano, 2007; García et al., 2007; Suuronen and Sardà, 2007; Frimodig, 2008). All vessels targeting shrimp in Norwegian waters are obliged to use a Nordmöre grid with a 19-mm bar spacing followed by a codend with a minimum mesh size of 35 mm (Norwegian Directorate of Fisheries, 2011). Thomassen and Ulltang (1975) tested several codend mesh sizes for the northern shrimp fisheries at the end of the 1960s and found acceptable retention lengths for deep water shrimps with the 35-mm mesh size. Despite the many changes in the northern shrimp fishery that have occurred since this investigation, the minimum codend mesh size remains at 35 mm. The introduction of the Nordmöre grid eliminated the issue of bycatch of larger sizes of fish because they are unable to pass through the grid into the trawl codend (Grimaldo and Larsen, 2005; Grimaldo, 2006). However, small-sized fish such as juveniles of various species are able to pass through the grid and enter the codend together with the targeted shrimps (He and Balzano, 2007, 2013). The current regulations of the Northeast Atlantic shrimp fishery allow retention of low numbers of juvenile fish from regulated species. For example, the fishing areas are closed if a catch sample exceeds three redfish (Sebastes spp.), and three Greenland halibut (Reinhardtius hippoglossoides) per 10 kg shrimps. Additionally, the shrimp catch can contain no more than 10% by weight of undersized (i.e. < 15 mm carapace length) shrimps (Norwegian Directorate of Fisheries, 2011). These strict bycatch rules have led to frequent temporary closures of several large shrimp fishing grounds in the Northeast Atlantic over the last 20 years (Gullestad et al., 2015), often lasting for weeks or months until bycatch levels fall below the threshold. Bycatches of juvenile fish and undersized shrimps also cause practical problems when sorting the catch on-board the fishing vessels. The use of a Nordmöre sorting grid ahead of a small mesh codend makes the size selection processes complex, and the size-dependent curve for both shrimp and bycatch often exhibits a bell-shaped signature. However, a selection model that can properly describe and estimate these signature curves is not available, which challenges an ability to assess size selectivity in these fisheries. Considering the challenges described above, the aim of the current study was to: Develop a new model to estimate size selection in shrimp trawls using a Nordmöre grid and a small mesh codend. Quantify the size selectivity of deep water shrimp using a Nordmöre grid with a 19-mm bar spacing followed by a diamond mesh codend with 35-mm mesh size. Quantify the size selectivity of juveniles of some of the most important bycatch species using the Nordmöre grid and diamond mesh codend Material and methods Research vessel, study area, and experimental design The fishing trials were conducted using a commercial shrimp trawl with Nordmöre grid followed by a size selective codend. Using a procedure described in Wileman et al. (1996), the bar spacing in the Nordmöre grid was 18.8 ± 0.4 mm (mean ± SD) and the meshes in the codend were 33.8 ± 1 mm (mean ± SD). In order to independently assess the contribution of the grid and the codend to the overall selectivity of the trawl, shrimp and fish released by the grid and the codend could be collected separately using two independent covers: one over the opening of the grid and the other surrounding the entire codend. Such double cover setups have been used previously to collect selectivity data in finfish fishery (Sistiaga et al., 2010). However, in a shrimp fishery, the mesh size of the cover is very small and may affect selectivity by masking codend meshes or affecting water flow adjacent codend meshes. In addition, using two covers significantly complicates the selectivity study because the covers can become entangled, they can modify trawl performance or alter shrimp and fish behaviour, and they create operational challenges on-board. Therefore, to avoid these challenges, we used two different experimental setups during the sea trials to evaluate size selectivity of shrimp and bycatch. In the test haul setup, we fished with a standard Nordmöre grid and 35 mm codend with a small-mesh cover [mesh size 16.4 ± 0.5 mm (mean ± SD)] collecting fish and shrimps escaping from the opening in front of the Nordmöre grid (Figure 1). In the control haul setup, the codend contained a small-meshed inner net [mesh size 18.5 ± 0.9 mm (mean ± SD)] installed with a low hanging ratio to prevent the escape of fish and shrimp. In this setup, fish and shrimps that escaped in front of the Nordmöre grid were collected in a small-meshed cover [mesh size 18.9 ± 1.2 mm (mean ± SD)] (Figure 1). Test and control hauls were deployed in the same fishing area during the same cruise. Figure 1. View largeDownload slide Experimental gear configuration: with separate group of hauls with test gear (top) and control gear (bottom). The gear configuration consisting of a forward Nordmöre grid and a small mesh codend is compulsory in the Northeast Atlantic trawl fishery targeting deep-water shrimps. Figure 1. View largeDownload slide Experimental gear configuration: with separate group of hauls with test gear (top) and control gear (bottom). The gear configuration consisting of a forward Nordmöre grid and a small mesh codend is compulsory in the Northeast Atlantic trawl fishery targeting deep-water shrimps. For the test hauls, the catch was collected in the test grid cover (GT) and in the test codend (CT), whereas for control hauls, the catch was collected in the control grid cover (GC) and in the blinded codend (CC). For each haul, the catch was sorted by species, length measured, and sorted into 1-cm wide length groups for fish and 1-mm wide length groups for shrimp. Thus, the catch data consisted of count numbers (n) representing the number of individuals of the different species collected in each of the compartments. The total length of the fish was measured using a measuring board, and the carapace length of the shrimps was measured using a calliper. The fishing trials were performed on board the research trawler (R/V) “Helmer Hanssen” (63.8 m LOA and 4080 HP) from 16 to 28 February, 2016 at the fishing grounds located at the Central bank, east of Hopen Island in the north of the Barents Sea). The trials were carried out using two identical Campelen 1800# trawls built entirely of 80–40 mm meshes in the wings and belly [2 mm polyethylene (PE) twine]. Thyboron T2 (6.5 m2 and 2200 kg) trawl doors were used, and an 8-m long rope was linked between the warps 80 m in front of the doors, which kept the distance between the doors at 48–52 m during the tows. The Campelen trawl has a 19.2 m fishing line and is believed to work at its optimal wingspread (ca. 15 m) and height (ca. 6.5 m) when the door distance is kept in this range. We used 40 m double sweeps and a 19.2 m long rock hopper gear built of three sections with 46 cm rubber discs. Both trawls were equipped with 4-panel Nordmöre grid sections that are equivalent in dimensions and construction to the 2-panel standard Nordmöre grid section used by the Norwegian coastal fleet targeting shrimp. Each Nordmöre grid is made of stainless steel and measures 1510 mm high and 1330 mm wide. The grid in both trawls used was mounted so that it would maintain an angle of 45 ± 2.5° while fishing. Model for size selection The size selection system consists of two main parts. The first part is a Nordmöre grid, which the fish and shrimps must pass through to enter the codend. If they do not pass through this grid they are excluded during this first part of the selection process. To pass through the grid, two conditions need to be fulfilled: (a) they need to contact the grid and (b) morphologically they must be able to pass through the grid, which is dependent on their size and orientation when they reach with the grid. The second part is a codend that collects fish and shrimp that pass through the grid. The codend is also size selective, and its size selection is the second part of the combined selection process. Thus, for a fish or deep water shrimp to be retained by the gear [rcombined(l)], it must be retained by selectivity process of the Nordmöre grid [rgrid(l)] and the codend [rcodend(l)]:   rcombinedl=rgridl×rcodendl (1) where l denotes the length of the fish or the length of the shrimp carapace. This system can be defined as a sequential dual selection system. It is a dual system because it consists of two processes and it is sequential because the second process follows the first. The next step is to model each of the two size selection processes individually. For the grid selectivity process we need to consider that some fish or shrimps might not contact the Nordmöre grid at all [see Larsen et al. (2016) for the explanation of contact in this context] or that they might do so orientated in a way that independent of length they have no chance of passing though the grid. The probability of length-independent exclusion from the Normöre grid, Cgrid, is specified as a parameter with a range of 0.0 to1.0, where 1.0 means that every individual of the species contacts the grid in a way that gives it a length-dependent chance of passing through the grid. For an individual contacting the grid with sufficiently good orientation to give it a length-dependent chance of passing through grid (rcontactgrid(l)), the following logit model was used [see Wileman et al. (1996) for further information about the logit model]:   rcontactgridl,L50grid,SRgrid=1.0-logitl,L50grid,SRgrid=1.01.0+expln9SRgrid×l-L50grid (2) Model (2) considers that the probability of being able to pass through the grid is length dependent and will decrease for larger individuals. L50grid denotes the length of fish or shrimp with 50% probability of being retained, and SRgrid (selection range) describes the difference in length between fish or shrimp with 75 and 25% probability of being retained, respectively. Based on the above, the following model was used for the size selection in the first process [rgrid(l)]:   rgridl,Cgrid,L50grid,SRgrid=Cgrid1.0+expln9SRgrid×l-L50grid (3) The escape probability through the outlet in front of the Nordmöre grid was therefore modelled by:   egridl,Cgrid,L50grid,SRgrid=1.0- rgridl,Cgrid,L50grid,SRgrid=1.0- Cgrid1.0+expln9SRgrid×l-L50grid (4) For the small mesh codend selectivity process, we assumed that the retention probability can be modelled by a logit model:   rcodendl,L50codend,SRcodend=logitl,L50codend,SRcodend=expln9SRcodend×l-L50codend1.0+expln9SRcodend×l-L50codend (5) Thus, by inserting (3) and (5) into (1), we arrived at the following combined size selection model:   rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodend=Cgrid1.0+expln9SRgrid×l-L50grid×expln9SRcodend×l-L50codend1.0+expln9SRcodend×l-L50codend (6) Model (6) is a so-called structural model because it is based on modelling the individual processes expected to be involved in the combined size selection in the system. One advantage of applying a structural model compared to an empirical-based model is that once the values of the parameters in the model are estimated, they can be applied to investigate not only the combined processes in the system but also the individual processes. In this context, each of the model parameter values obtained can be directly interpreted. In the case of model (6), five parameters need to be estimated to be able to describe the size selection in the system: Cgrid, L50grid, SRgrid, L50codend, and SRcodend. If all individuals contact the grid with a mode of contact that provide a length dependent probability to pass through, then the value for Cgrid should be 1.0. However, this is not necessarily the case, as some individuals may escape through the escape outlet in front of the Nordmöre grid (Figure 1) without contacting the grid first. Other individuals may be so poorly orientated when they meet the grid that the probability of them passing through will be similar to those not contacting the grid, which will also be reflected in the value of Cgrid. Thus, L50grid and SRgrid are respectively the L50 and SR values for individuals contacting the grid with a reasonable mode of orientation. L50codend and SRcodend are the L50 and SR values for the codend selection conditional upon organisms contacting and passing through the Nordmöre grid (Figure 1). As different species have different morphology and behaviour, values of the parameters Cgrid, L50grid, SRgrid, L50codend, and SRcodend for the same combined system will be species specific. Therefore, the analysis was applied separately for the different fish species and for the deep water shrimp. Data analysis and parameter estimation Catch data were collected in two groups. One of the groups consisted of control hauls obtained by summing compartments GC and CC (Figure 1). Together, they sampled the size and species composition of fish entering the selective parts of the trawl (section with the Nordmöre grid and codend), and in this respect the control hauls can be paired with the test hauls so that a paired-gear estimation method can be used (Wileman et al., 1996). However, compared to the standard paired-gear method in which none of the selective parts of the system uses covers to collect fish or shrimps escapees, our test hauls are special because they use a cover (GT) to collect fish and shrimp escaping ahead of the Nordmöre grid. Therefore, our experimental data collection design represents a combination of the paired and covered data collection and estimation methods (Wileman et al., 1996). To estimate the average size selection of the Nordmöre grid and codend in the test trawl , we paired the pooled catch data with the pooled catch data from the control hauls. Based on this approach, the experimental data in the analysis were treated like three compartment data. Fish or shrimp caught were observed in GT, CT, or (GC + CC). For the estimation based on the size selection model established in the “Model for size selection” section, we needed to express the probabilities that fish or shrimp of a specific length l would be observed in each of these three compartments conditioned they were caught. The probability that the fish or shrimp would enter the selection section in one of the test hauls and in one of the control hauls was modelled by the split factor, SP, as is traditionally done for paired-gear data analysis (Wileman et al., 1996). This means that the probability that a fish or shrimp will enter the test haul is SP, whereas the probability of them entering the control haul is 1.0-SP. All fish or shrimp entering the control haul are retained because both the cover over the grid outlet and the blinded codend retain all potential escapees. For a fish or shrimp entering one of the hauls included in the analysis (test or control), the probability that it will be retained in the cover in front of the Nordmöre grid in one of the test hauls would, based on Equation (4), be SP×egridl,Cgrid,L50grid,SRgrid. For a fish or shrimp entering one of the hauls included in the analysis, the probability that it will be retained in the codend of a test haul would, based on Equation (6), be SP×rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodend. Considering this, the probability γ that a fish or shrimp entering one of the test or control hauls will be observed in one of the three compartments (GT, CT, or GC + CC) can be expressed as:   γl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=SP×egridl,Cgrid,L50grid,SRgrid+SP×rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodend+1.0-SP (7) Based on Equation (7) and the considerations above, the probabilities pGT, pCT, and pGC+CC that a fish or shrimp observed in the catch will be found in compartment GT, CT, or GC + CC, respectively, can be expressed by:   pGTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=SP×egridl,Cgrid,L50grid,SRgridγl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SPpCTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=SP×rcombinedl,Cgrid,L50grid,SRgrid,L50codend,SRcodendγl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SPpGC+CCl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP=1.0-SPγl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP (8) Using Equation (8), the values for the parameters in the selection model (6) can be estimated from the collected experimental data by minimizing the following function with respect to Cgrid,L50grid,SRgrid,L50codend,SRcodend,and SP:   -∑l∑i=1anGTliqGTi×lnpGTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP+∑i=1anCTliqCTi×lnpCTl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP+∑j=1bnGCljqGCj+nCCljqCCj×lnpGC+CCl,Cgrid,L50grid,SRgrid,L50codend,SRcodend, SP} (9) where the outer summation is over length classes l in the experimental data and the inner summation is over experimental fishing hauls i (from 1 to a) and j (from 1 to b) with, respectively, the test and control setup. nGTli, nCTli,  nGClj, and nCClj are the number of fish or shrimp length measured of length class l in haul i and j in the respective compartment, with qGTi, qCTi, qGCj, and qCCj being the corresponding sampling factors (i.e. the fraction of the catch that was length measured). The estimation with (9) is based on the expected total catches and the sampling factors are therefore used to expand the subset of length-specific counts based upon the fractions of the haul measured for lengths. Minimizing (9) with respect to the parameters in it is the same as maximizing the likelihood for the observed experimental data based on a multinomial model, assuming that the formulated model (8) describes the experimental data sufficiently well. The observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC, which model (8) is expected to describe, are given by:   pGTl^=nGTlqGTnGTlqGT+nCTlqCT+nGClqGC+nCClqCCpCTl^=nCTlqCTnGTlqGT+nCTlqCT+nGClqGC+nCClqCCpGC+CCl^=nGClqGC+nCClqCCnGTlqGT+nCTlqCT+nGClqGC+nCClqCC. (10) Due to the experimental procedure followed, there was no obvious way to pair the data from the individual test and control hauls. Hence, to estimate the mean selectivity parameters for the experimental gear, the length dependent expected total catches for the test hauls were combined and compared with the combined expected total catches for the control hauls as formulated in function (9). The confidence limits for the parameters and curves for the size selection model were estimated using a double bootstrapping method that accounts for the uncertainty resulting from this unpaired nature of the data collection. For this, we adopted and further generalized the method for estimating uncertainty in size selectivity based on unpaired trawl data described by Sistiaga et al. (2016). This procedure accounts for uncertainty caused by between-haul variation in size selection processes (Fryer, 1991) and by the unpaired data collection method with groups of test and control hauls by selecting independently a hauls with replacement from the test hauls and b hauls with replacement from the control hauls during each bootstrap loop. Uncertainty caused by finite sample sizes on haul level (within-haul variability) is accounted for by randomly selecting fish with replacement from each of the selected hauls for each compartment separately, where the number selected from each compartment in each haul is the same as the number sampled in that compartment in that haul. These data are then raised and combined as described above, and the selectivity parameters are again estimated. The additional uncertainty in the estimation caused by subsampling is automatically accounted for by raising the data after the re-sampling (Eigaard et al., 2012). We performed 1000 bootstrap repetitions to calculate the 95% percentile confidence limits (Efron, 1982; Chernick, 2007) for the selection parameters and curves. The model’s ability to describe the experimental data sufficiently well was evaluated based on the p-value, model deviance versus degrees of freedom (DOF) and inspection of how the model curve reflects the length-based trend in the data (Wileman et al., 1996). The p-value expresses the likelihood to obtain at least as big a discrepancy between the fitted model and the observed experimental data by coincidence. In case of a poor fit statistics (p-value being < 0.05; deviance being ≫DOF), the model curve plots were inspected to determine whether the poor result was due to structural problems when describing the experimental data using the model or if it was due to over-dispersion in the data (Wileman et al., 1996). The analysis was carried out using the software SELNET (Herrmann et al., 2012, 2013a,b), which implements the models and the bootstrap method described above. Results Collected data We conducted eight control hauls (Table 1) and eight test hauls (Table 2). To ensure that the group of test and control hauls were fishing a similar size and species compositions all hauls were conducted within 5 days in February 2016 at same fish ground within a small area, at similar depth and with similar towing time (Tables 1 and 2). In all hauls, a subsample of the shrimp catch was length measured whereas catches of every American plaice (Hippoglossoides platessoides) and redfish (Sebastes spp.) were length measured. The total number of shrimp, American plaice and redfish length measured during the data collection were respectively 4405, 8773, and 4439 individuals. Table 1. Overview of the fish and shrimp length measured in the control hauls carried out during the trials. Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GC (% measured)  CC (% measured)  GC  CC  GC  CC  1  60  268  20  7604.9 N  03526.9 E  123 (72.31%)  160 (1.63%)  208  177  56  36  2  61  257  20  7605.4 N  03517.8 E  120 (58.14%)  153 (1.95%)  238  182  143  37  3  60  278  21  7605.3 N  03511.1 E  163 (7.47%)  173 (1.16%)  438  187  404  169  4  60  271  21  7605.9 N  03533.8 E  108 (9.60%)  171 (1.20%)  265  156  184  86  5  63  266  21  7605.9 N  03521.9 E  144 (40.54)  160 (1.91%)  321  121  108  20  6  61  271  22  7606.5 N  03531.9 E  169 (100%)  175 (2.02%)  206  150  68  34  7  60  271  22  7606.6 N  03521.9 E  208 (22.74)  169 (1.02%)  391  287  187  94  8  63  272  22  7606.5 N  03531.9 E  189 (21.12)  190 (0.73%)  327  301  164  120  Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GC (% measured)  CC (% measured)  GC  CC  GC  CC  1  60  268  20  7604.9 N  03526.9 E  123 (72.31%)  160 (1.63%)  208  177  56  36  2  61  257  20  7605.4 N  03517.8 E  120 (58.14%)  153 (1.95%)  238  182  143  37  3  60  278  21  7605.3 N  03511.1 E  163 (7.47%)  173 (1.16%)  438  187  404  169  4  60  271  21  7605.9 N  03533.8 E  108 (9.60%)  171 (1.20%)  265  156  184  86  5  63  266  21  7605.9 N  03521.9 E  144 (40.54)  160 (1.91%)  321  121  108  20  6  61  271  22  7606.5 N  03531.9 E  169 (100%)  175 (2.02%)  206  150  68  34  7  60  271  22  7606.6 N  03521.9 E  208 (22.74)  169 (1.02%)  391  287  187  94  8  63  272  22  7606.5 N  03531.9 E  189 (21.12)  190 (0.73%)  327  301  164  120  For fish species, all specimens were measured for length. For shrimp, values in brackets represent the proportion of individuals subsampled for length measurements. The catch was collected in the control gear grid cover (GT) and in the gear control gear codend (CT). Table 2. Overview of the fish and shrimp length measured in the test hauls carried out during the trials. Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GT (% measured)  CT (% measured)  GT  CT  GT  CT  9  60  268  22  7606.1 N  03522.3 E  150 (63.13%)  150 (1.34%)  391  283  211  42  10  62  265  23  7605.4 N  03523.2 E  123 (31.72%)  146 (0.94%)  444  347  392  65  11  64  268  23  7605.8 N  03525.1 E  98 (66.77%)  134 (1.05%)  482  402  494  108  12  62  265  23  7605.7 N  03522.1 E  7 (100%)  121 (2.10%)  283  309  211  47  13  63  274  23  7605.9 N  03523.4 E  21 (100%)  141 (1.76%)  239  212  354  91  14  60  256  24  7604.7 N  03516.8 E  50 (100%)  161 (2.67%)  256  202  98  33  15  63  252  24  7604.0 N  03512.9 E  75 (80.61%)  146 (1.08%)  230  320  135  82  16  66  269  24  7606.1 N  03517.2 E  140 (8.18%)  167 (1.78%)  298  120  142  24  Haul Nr  Trawling time (min)  Depth (m)    Pos. start   Shrimp   American Plaice   Redfish   Day in February  Lat.  Long.  GT (% measured)  CT (% measured)  GT  CT  GT  CT  9  60  268  22  7606.1 N  03522.3 E  150 (63.13%)  150 (1.34%)  391  283  211  42  10  62  265  23  7605.4 N  03523.2 E  123 (31.72%)  146 (0.94%)  444  347  392  65  11  64  268  23  7605.8 N  03525.1 E  98 (66.77%)  134 (1.05%)  482  402  494  108  12  62  265  23  7605.7 N  03522.1 E  7 (100%)  121 (2.10%)  283  309  211  47  13  63  274  23  7605.9 N  03523.4 E  21 (100%)  141 (1.76%)  239  212  354  91  14  60  256  24  7604.7 N  03516.8 E  50 (100%)  161 (2.67%)  256  202  98  33  15  63  252  24  7604.0 N  03512.9 E  75 (80.61%)  146 (1.08%)  230  320  135  82  16  66  269  24  7606.1 N  03517.2 E  140 (8.18%)  167 (1.78%)  298  120  142  24  For fish species, all specimens were measured for length. For shrimp, values in brackets represent the proportion of individuals subsampled for length measurements. The catch was collected in the test grid cover (GT) and in the test codend (CT). Size selectivity for shrimp The model used reflected the pattern observed in the experimental data well (Figure 2a–c). Thus, although the p-value (Table 3) observed in the fit statistics was low, we are confident that the model used represents the data adequately. We considered the low p-value to be a consequence of over-dispersion in the experimental catch portioning data that resulted from working with pooled and subsampled data with low sampling rates (Tables 1 and 2). Such cases have led to low p-values and high dispersion (Brčić et al., 2015; Alzorriz et al., 2016; Notti et al., 2016) especially for length classes with relative low frequency, such as our shrimp data for individuals with carapace length above 25 mm (Figure 2a–c). All shrimp were estimated to make contact with the Nordmöre grid, and most of them passed through it. However, the grid passage probability was estimated to decrease slightly with increasing shrimp size. The codend selectivity showed size-dependent release for shrimp with carapace length < 23 mm, with only about 20% of the shrimp with carapace length of 15 mm being retained by the codend. L50codend was estimated to be 17.72 mm, and SRcodend was estimated to be 3.63 mm (Table 3). L50grid was 49.2 mm, which at first glance could seem meaningless because it is above the size range for this species of shrimp (Table 3). However, this value is expected to be above a biologically meaningful value and confirms that most of the shrimp can pass through the grid except for the slight decrease for large shrimp. The combined selectivity for the grid and codend exhibited a slightly bell-shaped signature, with few shrimp < 15 mm being retained, a maximum retention rate for shrimp with carapace length of 25 mm, and a slight decrease for shrimp above this size. Table 3. Size selectivity parameters and fit statistics results for shrimps, American plaice and redfish based on fitting the model (8) to the experimental data.   Shrimps  American Plaice  Redfish  Cgrid  1.00 (0.98–1.00)  1.00 (0.97–1.00)  0.90 (0.75–0.99)  L50grid  49.17 (37.16–68.57)  19.40 (18.41–20.20)  13.61 (13.06–14.28)  SRgrid  16.52 (8.02–27.82)  7.47 (6.44–8.61)  3.46 (2.93–3.97)  L50codend  17.72 (16.10–22.59)  6.84 (5.46–7.68)  9.78 (8.85–10.45)  SRcodend  3.63 (1.79–9.45)  1.66 (0.10–2.66)  1.74 (1.33–2.60)  SP  0.51 (0.42–0.70)  0.55 (0.49–0.61)  0.63 (0.51–0.74)  DOF  34  90  54  Deviance  175.66  118.38  101.91  p-value  <0.0001  0.0241  0.0001    Shrimps  American Plaice  Redfish  Cgrid  1.00 (0.98–1.00)  1.00 (0.97–1.00)  0.90 (0.75–0.99)  L50grid  49.17 (37.16–68.57)  19.40 (18.41–20.20)  13.61 (13.06–14.28)  SRgrid  16.52 (8.02–27.82)  7.47 (6.44–8.61)  3.46 (2.93–3.97)  L50codend  17.72 (16.10–22.59)  6.84 (5.46–7.68)  9.78 (8.85–10.45)  SRcodend  3.63 (1.79–9.45)  1.66 (0.10–2.66)  1.74 (1.33–2.60)  SP  0.51 (0.42–0.70)  0.55 (0.49–0.61)  0.63 (0.51–0.74)  DOF  34  90  54  Deviance  175.66  118.38  101.91  p-value  <0.0001  0.0241  0.0001  L50 and SR values are given in mm for shrimps and in cm for American plaice and redfish. Values in brackets are 95% confidence limits. Figure 2. View largeDownload slide Size selectivity plots for shrimps. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of shrimps found in the grid cover of the test gear, plot “b” shows the length dependent portioning of shrimps observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of shrimps found in the control gear. The points in plots “a”-”c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Figure 2. View largeDownload slide Size selectivity plots for shrimps. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of shrimps found in the grid cover of the test gear, plot “b” shows the length dependent portioning of shrimps observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of shrimps found in the control gear. The points in plots “a”-”c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Size selectivity for American plaice The model used reflected the pattern observed in the experimental data well (Figure 3a–c). Despite the p-value being < 0.05 (Table 3), the model represented the data adequately and we are confident about the performance of the model. All American plaice were estimated to make contact with the Nordmöre grid, and most of them passed through it. The grid passage probability was very high for American plaice up to 12 cm long, followed by a decrease and then very low passage probability for fish > 30 cm long. The codend only showed low size selectivity for American plaice with an L50codend value of 6.84 cm, thus all American plaice > 10 cm long that entered the codend were retained (Table 3). The combined selectivity for the grid and codend showed a clear bell-shaped signature, with a high retention probability for American plaice ∼10 cm long (ca. 90% retention). Retention of individuals < 5 cm long was practically 0 and retention of fish in the range of 10–30 cm decreased, with really low retention rates for fish > 30 cm long. In the range of 6–23 cm, retention probability for American plaice for the gear was >25%, meaning that this gear would not be adequate in areas where the numbers of American plaice within this range are high. Figure 3. View largeDownload slide Size selectivity plots for American plaice. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of American plaice found in the grid cover of the test gear, plot “b” shows the length dependent portioning of American plaice observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of American plaice found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Figure 3. View largeDownload slide Size selectivity plots for American plaice. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of American plaice found in the grid cover of the test gear, plot “b” shows the length dependent portioning of American plaice observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of American plaice found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC [Equation (10)]. The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Size selectivity for redfish For redfish, the model used represented the experimental data well up to l = 22 cm (Figure 4a–c). Because fish above this size are outside the selective range of the gear, the model adequately describes the size selection process in the gear. The discrepancy between model and experimental data above the selective range of the gear is regarded as a consequence of unequal entry of bigger redfish into the test and control gears. The combined size selection showed a clear bell-shaped signature, with >60% of redfish around 12 cm long being retained by the gear but <25% of redfish <9 cm and >15 cm long being retained. The grid passage probability was high (>80%) for redfish <12 cm long, and it decreased monotonously, with no redfish >20 cm entering the codend. The codend size selection showed that none of the redfish >14 cm would be released by the codend and that L502 and SR2 were 9.8 and 1.7 cm, respectively (Table 3). Figure 4. View largeDownload slide Size selectivity plots for redfish. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of redfish found in the grid cover of the test gear, plot “b” shows the length dependent portioning of redfish observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of redfish found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC (Equation 10). The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Figure 4. View largeDownload slide Size selectivity plots for redfish. The left column shows the fit of the selection model [Equation (8)] to the experimental catch portioning rates. Plot “a” shows the length dependent portioning of redfish found in the grid cover of the test gear, plot “b” shows the length dependent portioning of redfish observed in the codend of the test gear (left y-axis) and plot “c” shows the length dependent portioning of redfish found in the control gear. The points in plots “a”–“c” represent the observed experimental length dependent portioning of the catches between the three compartments GT, CT, and GC + CC (Equation 10). The grey curves in plots “a–c” represent the total catches in number of individuals (right y-axis) in the three compartments GT, CT, and GC + CC. The plots in the right column show the selectivity curves for the test gear with plot “d” showing the length dependent grid passage probability [Equation (4)], plot “e” showing the length dependent codend selectivity in the test gear [Equation (5)], and plot “f” showing the combined size selectivity of the Nordmöre grid and the codend for the test gear [Equation (6)]. The stippled curves in plots “d”–“f” represent 95% confidence bands. Discussion The size selection data resulting from the grid and codend configuration used in this study were based on a new model and estimation method that is an extension of the unpaired method described in Sistiaga et al. (2016). This new approach models the observed data summed over hauls for a group of test and controls hauls, and it combines a structural dual sequenced size selection model with unpaired data collection for groups of test and control hauls. This model effectively described the length-dependent portioning of the observed catch between the test codend, the test grid cover, and the control gear for all species investigated. In addition to enabling estimation of the combined size selection for the Nordmöre grid followed by the diamond mesh codend, this method enabled estimation of the size selection for each of the selection devices individually because the structural model explicitly described the selectivity processes in each of the devices. Structural size selection models have previously been developed and applied to describe size selection in other trawl fisheries. This includes models for fish sorting grids in combination with codends in finfish fisheries (Sistiaga et al., 2010; Herrmann et al., 2013a), square mesh panels in combination with selective codends (Zuur et al., 2001; O’Neill et al., 2006; Alzorriz et al., 2016), double grid sorting devices (Larsen et al., 2016: Sistiaga et al., 2016), and excluding grids combined with a selective codend (Brčić et al., 2015; Stepputtis et al., 2015; Lövgren et al., 2016). Excluding grids combined with a selective codend result in the same bell-shaped selection pattern as the Nordmöre grid followed by a size selective codend. However, our study is the first time such a modelling process has been applied to a shrimp trawl fishery and the first time that a sequential model with two compartment data collection in test and control gears have been used. Our gear configuration was more complex than the methods previously applied, but it is necessary due to the practical problems that potentially could have resulted from using a small mesh cover over the test codend. The method and model presented herein offer new possibilities for studying size selectivity in other shrimp fisheries. In particular, our approach enables detailed mapping of which sizes of bycatch species would have especially high risk of being caught if they are abundant in the shrimp fishing grounds. In this study, we demonstrated the ability of the new model to represent bell-shaped selectivity data in detail for shrimp and two fish bycatch species: American plaice and redfish. For the juvenile bycatch species, our results demonstrated very high and length-dependent grid passage probability. Thus, in conjunction with the small-meshed diamond mesh codend used in the shrimp fishery, the gear has high catch risk for certain size ranges of these bycatch species. The use of the combined bycatch reducing and size selective system consisting of the Nordmöre grid and 35 mm codend mesh is well established in the Northeast Atlantic shrimp fishery. However, the data from our study clearly show that fish within a limited size range and undersized shrimps retained in the 35-mm codend will continue to be a problem for the northern shrimp fleet. Some precaution is required since our fishing trial was based on only 16 hauls and the number of shrimp, American Plaice and redfish length measured was respectively limited to 4405, 8773, and 4439 individuals (Tables 1 and 2). This leads to a degree of uncertainty in the estimated size selection curves (Figures 2–4) and needs to be considered when interpreting the results. However, this uncertainty is reflected in the selectivity parameters (Table 3) and confidence bands around the size selection curves. A recent study by Herrmann et al. (2016) has shown that estimating trawl size selectivity to a specific level of precision based on a paired gear method will require around 10 times as many fish as needs to be measured using a covered method. Based on that claim we had concerns whether it would be possible to estimate size selectivity with an acceptable level of precision as our gear configuration uses a combination of a paired and covered sampling design. However, the relative narrow confidence bands around the size selection curves in our study demonstrates the practicality and appropriateness of our method. Acknowledgements We thank the crew of RV “Helmer Hanssen” and assistants Ingeborg Sætra, Hermann Pettersen, Tomas Araya, and postdoctoral researcher André Frainer for valuable assistance on board. We appreciate the efforts of the editor and the three anonymous reviewers, which we feel improved our manuscript significantly. We are also grateful to the Arctic University of Norway UIT in Tromsø and the Norwegian Seafood Research Fund for funding the experiments carried out in this study. References Alzorriz N., Arrregi L., Herrmann B., Sistiaga M., Casey J., Poos J. J. 2016. Questioning the effectiveness of implemented technical measures under the EU landings obligation: the Basque Otter Bottom Trawl fishery case study. Fisheries Research , 175: 116– 126. Google Scholar CrossRef Search ADS   Brčić J., Herrmann B., Sala A. 2015. Selective characteristics of a shark-excluding grid device in a Mediterranean trawl. Fisheries Research , 172: 352– 360. Google Scholar CrossRef Search ADS   Chernick M. R. 2007. Bootstrap Methods: A Guide for Practitioners and Researchers, 2nd edn. Wiley Series in Probability and Statistics . Wiley, New York. Eayrs S. 2007. A Guide to Bycatch Reduction in Tropical Shrimp-Trawl Fisheries  [ Revised edition]. FAO, Rome. 108 pp. Efron B. 1982. The jackknife, the bootstrap and other resampling plans. SIAM Monograph No 38, CBSM-NSF. Eigaard O. R., Herrmann B., Nielsen J. R. 2012. Influence of grid orientation and time of day on grid sorting in a small-meshed trawl fishery for Norway pout (Trisopterus esmarkii). Aquatic Living Resources , 25: 15– 26. Google Scholar CrossRef Search ADS   Frimodig A. J. 2008. Bycatch Reduction Devices Used in the Pink Shrimp Trawl Fishery. Report to the California Fish and Game Commission. California Department of Fish and Game Marine Region State Fisheries Evaluation Project (4/14/2008). 12 pp. Fryer R. J. 1991. A model of between-haul variation in selectivity. ICES Journal of Marine Science , 48: 281– 290. Google Scholar CrossRef Search ADS   García E.G. (ed), Ragnarsson S. Á., Steingrimsson S. A. 2007. Bottom Trawling and Scallop Dredging in the Arctic: Impacts of Fishing on Non-Target Species, Vulnerable Habitats and Cultural Heritage . Nordic Council of Ministers, Copenhagen. 374 pp. Grimaldo E. 2006. The effects of grid angle on a modified Nordmöre-grid in the Nordic Shrimp Fishery. Fisheries Research , 77: 53– 59. Google Scholar CrossRef Search ADS   Grimaldo E., Larsen R. B. 2005. The cosmos grid: a new design for reducing by-catch in the Nordic shrimp fishery. Fisheries Research , 76: 187– 197. Google Scholar CrossRef Search ADS   Gullestad P., Blom G., Bakke G., Bogstad B. 2015. The “Discard Ban Package”: Experiences in efforts to improve the exploitation patters in Norwegian fisheries. Marine Policy , 54: 1– 9. Google Scholar CrossRef Search ADS   He P., Balzano V. 2007. Reducing the catch of small shrimps in the Gulf of Maine pink shrimp fishery with a size-sorting grid device. ICES Journal of Marine Science , 64: 1551– 1557. Google Scholar CrossRef Search ADS   He P., Balzano V. 2013. A new shrimp trawl combination grid system that reduces small shrimp and finfish bycatch. Fisheries Research , 140: 20– 27. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M. B., Nielsen K. N., Larsen R. B. 2012. Understanding the size selectivity of redfish (Sebastes spp.) in North Atlantic trawl codends. Journal of Northwest Atlantic Fishery Science , 44: 1– 13. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M., Larsen R. B., Nielsen K. N. 2013a. Size selectivity of redfish (Sebastes spp.) in the Northeast Atlantic using grid-based selection systems for trawls. Aquatic Living Resources , 26: 109– 120. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M., Larsen R. B., Nielsen K. N., Grimaldo E. 2013b. Understanding sorting grid and codend size selectivity of Greenland halibut (Reinhardtius hippoglossoides). Fisheries Research , 146: 59– 73. Google Scholar CrossRef Search ADS   Herrmann B., Sistiaga M., Santos J., Sala A. 2016. How many fish need to be measured to effectively evaluate trawl selectivity? PLoS One , 11: e0161512. Google Scholar CrossRef Search ADS PubMed  Isaksen B., Valdemarsen J. W., Larsen R. B., Karlsen L. 1992. Reduction of fish by-catch in shrimp trawl using a rigid separator grid in the aft belly. Fisheries Research , 13: 335– 352. Google Scholar CrossRef Search ADS   Larsen R. B., Herrmann B., Sistiaga M., Grimaldo E., Tatone I., Onandia I. 2016. Size selection of redfish (Sebastes spp.) in a double grid system: quantifying escapement through individual grids and comparison to former grid trials. Fisheries Research , 183: 385– 395. Google Scholar CrossRef Search ADS   Lövgren J., Herrmann B., Feekings J. 2016. Bell-shaped size selection in a bottom trawl: a case study for Nephrops directed fishery with reduced catches of cod. Fisheries Research , 184: 26– 35. Google Scholar CrossRef Search ADS   Norwegian Directorate of Fisheries. 2011. J-209-2011: Forskrift om maskevidde, bifangst og minstemål m.m. ved fiske i fiskevernsonen ved Svalbard (In Norwegian). Notti E., Brčić J., De Carlo F., Herrmann B., Lucchetti A., Virgili M., Sala A. 2016. Assessment of the relative catch performance of a surrounding net without the purse line as an alternative to a traditional boat seine in small scale fisheries. Marine and Coastal Fisheries , 8: 81– 91. Google Scholar CrossRef Search ADS   O’Neill F. G., Kynoch R. J., Fryer R. J. 2006. Square mesh panel in North Sea demersal trawls: separate estimates of panel and codend selectivity. Fisheries Research , 78: 333– 341. Google Scholar CrossRef Search ADS   Thomassen T., Ulltang Ø. 1975. Report from mesh selection experiments on Pandalus borealis in Norwegian waters. ICES CM 1975/K:51. Sistiaga M., Herrmann B., Grimaldo E., Larsen R. B. 2010. Assessment of dual selection in grid based selectivity systems. Fisheries Research , 105: 187– 199. Google Scholar CrossRef Search ADS   Sistiaga M., Brinkhof J., Herrmann B., Grimaldo, Langård L., Lilleng D. 2016. Size selection performance of two flexible sorting grid section designs in the Northeast Arctic cod (Gadus morhua) and haddock (Melanogrammus aeglefinnus) fishery. Fisheries Research , 183: 340– 351. Google Scholar CrossRef Search ADS   Stepputtis D., Santos J., Herrmann B., Mieske B. 2015. Broadening the horizon of size selectivity in trawl gears. Fisheries Research , 184: 18– 25. Google Scholar CrossRef Search ADS   Suuronen P., Sardà F., 2007. By-catch reduction techniques in European fisheries traditional methods and potential innovations. In Bycatch Reduction in the World’s Fisheries. Reviews: Methods and Technologies in Fish Biology and Fisheries  7, pp. 37– 74. Ed. by Kennelly S. J.. Springer, The Netherlands. 288 pp. Wileman D. A., Ferro R. S. T., Fonteyne R., Millar R. B. (Eds.). 1996. Manual of methods of measuring the selectivity of towed fishing gears. ICES Cooperative Research Report No. 215. Zuur G., Fryer R. J., Ferro R. S. T., Tokay T. 2001. Modelling the size selectivities of a trawl codend and an associated square mesh panel. ICES Journal of Marine Science , 58: 657– 671. Google Scholar CrossRef Search ADS   © International Council for the Exploration of the Sea 2017. All rights reserved. For Permissions, please email: journals.permissions@oup.com

Journal

ICES Journal of Marine ScienceOxford University Press

Published: Jan 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off