Toward elimination of unwanted catches using a 100 mm T90 extension and codend in demersal mixed fisheries

Most European fishing fleets will need to drastically reduce their unwanted catches to comply with new rules of the common fisheries policy. A more practical way to avoid increasing on-board sorting time and issues linked to storage capacity is to prevent unwanted catches in the first place. We assessed the selectivity properties of an experimental fishing gear that combined a 100 mm T90 cylinder with 130 meshes in the extension and a 100 mm T90 codend of 33 meshes (experimental gear) compared to a 100 mm diamond mesh extension and codend (control gear) during commercial trips using twin trawls. Analysis of the relative size composition of catches indicated a significantly higher escapement of small fish of several target species (e.g. Lepidorhombus whiffiagonis, Melanogrammus aeglefinus, Raja spp, and Lophius spp) and non-target species (e.g. Capros aper and Gurnards spp) from the T90 experimental trawl compared to the control trawl (n = 49 hauls), resulting in a significant reduction of unwanted catches of Gadidae, Triglidae, and Caproidae. In contrast, non-negligible commercial losses of small grade target gadoid species were observed. Mixed general linear models showed that the proportion of ray, haddock and anglerfish retained per length class decreased with increased tow duration. The T90 experimental gear will perform at a commercial level when targeting monkfish, megrim, rays and large haddock, however fishers are not likely to use this gear when targeting smaller-bodied species such as cephalopods, small haddock, whiting (Merlangius merlangus) and hake (Merluccius merluccius), because the gear is likely to allow large numbers to escape. Selectivity studies often focus on a short list of target species; however, catches of non-target species under quota can be problematic for some fisheries. For example, under the implementation of the Landing Obligation catches of boarfish could choke the French whitefish demersal fisheries in the Celtic sea, as France has no national quota for that species. The device tested constitutes an efficient solution to mitigate catches for such non-target schooling fish.


Introduction
The European Union common fisheries policy, which regulates fishing activities in European waters, was reformed in 2013, resulting in a new regulation, including article 15 on the landing obligation [1]. It stipulates that all catch from the stocks under total allowable catch (TAC) or minimum conservation reference size (MCRS) regulations must be landed, prohibiting the common practice of throwing unwanted catches back into the sea. These discarding practices represent a substantial protein waste and may lead to unsustainable management of marine resources [2].
Spatio-temporal changes in fishing practices and improvement in selecting gear are the two main practices that fishers can implement to comply with the regulation. Numerous studies have focused on testing gear selectivity in recent decades (see reviews of [3,4]) by using netting with meshes of different sizes and shapes (e.g. diamond, square, T90) in a variety of places on the trawl (e.g. baiting, extension, codend) and/or by adding devices such as escape grids, dispersive floats, separator panels, lights and pingers (see review [5]). The combinations of gear modifications are almost infinite and warrant sea trials to assess their efficiency. Variability in fish behavior among areas, seasons [6,7] and in the net depending on gear configurations such as vertical openings [8,9], prevent straightforward extrapolation of results to other case studies.
There are many options that increase size selectivity without increasing mesh size, and one of the easiest to implement is to rotate diamond mesh by 90˚(i.e. "T90 mesh"). For a given nominal mesh size, T90 mesh remains more open than diamond mesh during towing [10]. First introduced in the 1990s, T90 mesh has regained interest in recent years. Trials using T90 mesh have been undertaken in various locations, including several European waters, the Mediterranean Sea [11,12], and in the United States [13], and on a diversity of fleets, from trawlers targeting fish [12,14], shrimp [11] or Nephrops norvegicus [15] to beam trawlers targeting flatfish [16]. Most experiments tested T90 mesh in the codend, except for Kopp el al. [14], who mounted a 100 mm cylinder of T90 mesh on a subsection of the extension section. Compared to diamond mesh, T90 mesh increases the size selectivity of roundfish species, shrimp and Norway lobster, but tends to decrease the selectivity of flatfish [16]. T90 mesh in the extension piece are mandatory as part of the multiannual management plan for stocks of cod, herring and sprat in the Baltic Sea to reduce the amounts of unwanted catches of cod [17].
In this study, we assessed the ability of an experimental fishing gear that combined a 100 mm T90 cylinder of 130 meshes in the extension and a 100 mm T90 codend of 33 meshes to reduce unwanted catch, for both target species below the MCRS and all sizes of fish of non-target species. Comparison of catches between the experimental and control gear were carried out for 6 species using data collected in commercial conditions over an entire year using a twin trawl (n = 49). We tested the effect of covariates on the proportion retained and discuss the results in terms of profitability for the fisheries.

PLOS ONE
Selectivity of T90 extension and codend in demersal mixed fisheries PLOS ONE | https://doi.org/10.1371/journal.pone.0235368 July 8, 2020 2 / 12 data collection and data analysis. The funder Les Pêcheurs de Bretagne helped in finding the volunteer fishing vessel "An triskell" on which the study was performed and ensured the link between the scientific team at Ifremer, Oceanic Development (the private company which was contracted to collect the data on-board) and the fishers along the project. The specific roles of these authors are articulated in the 'author contributions' section.
Competing interests: We confirmed the PO Les pêcheurs de Bretgane agreed to made the manuscript and related data public. This does not alter our adherence to PLOS ONE policies on sharing data and materials as detailed online in our guide for authors http://journals.plos.org/plosone/ s/competing-interests.

Materials and methods
Experiments at sea were conducted on the commercial vessel An Triskell (23,95m -453 KW, 102.27 tx), which was rigged with two warps twin otter trawls, a 36 m ground rope and a 28 m headline. The control trawl corresponded to the gear usually used by the vessel. Its codend was 25 meshes deep, 100 free diamond meshes in circumference and made of 4 mm double twine polyethylene (TPE), with a nominal stretched-mesh size of 100 mm. Its extension was 100 diamond meshes deep and made of 5 mm single TPE, with a nominal stretched-mesh size of 100 mm. The experimental trawl was equipped with T90 netting with a nominal stretched-mesh size of 100 mm, a cylinder of 130 meshes mounted in the extension (single 5 mm TPE) and 33 meshes in the codend (double 4 mm TPE) with a 66-mesh circumference (Fig 1). The rate of assembly between standard diamond mesh and T90 mesh equals three diamond meshes to two T90 meshes. The extensions and codends of both trawls had the same lengths (ca. 10 and 2.5 m, respectively) since 13 meshes in T90 equal 10 diamond meshes in length. Trawl geometry was monitored using the on-board trawl monitoring systems of the vessel to ensure proper functioning of the gear. We estimated the vertical opening height at 2.5-3 m and horizontal wingspread between the doors at 79m. However, as the data were not stored, effect of variation in trawl geometry on the results is unknown. Five fishing trips were conducted between September 2014 and October 2015 in the southern part of the Celtic sea (Fig 2). Mean (± standard deviation) tow duration was 3.4 (±0.2) h,

PLOS ONE
and mean fishing depth was 219 (±108 m (S1 Table)). For 49 hauls, wanted catches and unwanted catches (as usually sorted by the crew) were weighed (kg) per species, and the length of individual fish was measured (to the nearest 1 cm). When the total catch was too large to allow measurement of every individual, random sub-sampling was performed and the weight ratio between total catch and the subsample was used to raise the data. Due to identification issues, data for certain individual species were combined into taxonomic groups: Lophius spp (Lophius piscatorius and L budegassa), Loligo spp (Loligo forbesii and L. vulgaris), rays, gurnard and Trisopterus spp (Trisopterus luscus, T. minutus and T. esmarkii).
Length-dependent relative size selectivity between the experimental and control trawls was analyzed. Logistic regression in the mixed generalized linear model framework was fitted for 6 species according to the methods used by Holst and Revill [19] and Vogel et al. [20]. Between haul variability was incorporated by testing the significance of fixed effects: depth, tow duration and catch. Including haul as random effect, allowed accounting for an additional between haul variability, not explained by the fixed effects, and which can be interpreted as the effect of environmental and/or external conditions (e.g. current, time of day, fish behavior and physiology, fish species composition, and season) on catches. The within haul variability was taken into account at the individual level, with the effect of size on retention probability. For each species, observations outside the 0.025 and 0.975 quantiles of length distribution (which represents 5% of the data set) were excluded to reduce the influence of outliers on the fits. The proportion of fish retained per length class, P(l), were

PLOS ONE
calculated and modeled as: where N t,h (l) is the number of fish of length l in the test trawl; N c,h (l) is the number of fish of length l in the control trawl; m is the degree of polynomial function that models the effect of size (m = 0-3); W h total catch weigh in kg, TD h the tow duration in hour, D h the fishing depth in meter, S h the subsampling ratio and δ h is the random effect on the haul. The subsampling ratio is modeled as an offset [19]: with q t (h) and q c (h) the proportions taken out for measurements from the catch bulk of the experimental and control gear, respectively. The minimal dataset is provided as a supporting information file. For each species, the best-fitted model was selected based on the Akaike information criterion (AIC) considering improvement between two models to be significant when differences between AIC values exceeded 5. When two models had similarly "good" AIC values (i.e. Δ AIC < 5) the most parsimonious one was selected. The 32 models tested are summarized in Table 1. Statistical analyses were performed using R 3.5.3 [21].

Ethics statement
Sea trials were carried out on board the vessel An Triskell in accordance with the European scientific fishing authorization granted by the French maritime fisheries and aquaculture directorate (n.o 2014/730412/SELECMCTrawl/0006 and 2015/730412/SELECMCTrawl/0001). Animals were not exposed to additional stress other than that involved in commercial fishing practice. Thus, no additional authorization or ethics approval was required to perform the study. This study did not involve endangered or protected species.

Target species
Lepidorhombus whiffiagonis. The best-fitted model (Table 2) indicated that megrim shorter than the MCRS (20 cm) were nearly absent from both trawls, those 20-30 cm long were retained more by the control than the test trawl and those longer than 30 cm had an equal retention probability for the two trawls (Fig 3). None of the covariables tested were retained during the model selection procedure (Table 2).
Lophius spp. Analysis of length showed that equal retention probability was reached at ca. 40 cm (Fig 3) and that the difference in the retention probability between the two trawls was small for shorter fish. Tow duration had significant effect ( Table 2, p-value <0.001), with in the proportion of anglerfish retained per length class decreasing with increasing tow duration.
Melanogrammus aeglefinus. The best-fitted model ( Table 2) indicated that haddock shorter than 30-35 cm, while present in the control trawl, were nearly absent from the test trawl, and the size at which retention probability was equal was ca. 40-45 cm (Fig 3). This size is much higher than the MCRS of 30 cm. Fishing depth was retained in model selection procedure ( Table 2, p-value<0.001) indicating that proportion of haddock retained per length class increased with increasing depth.

Rays
Despite high variability in the dataset, the best-fitted model ( Table 2) tended to indicate a slightly lower retention probability for test trawl than the control trawl for rays shorter than 45 cm (Fig 3). The proportion of rays retained per length class decreased with increasing tow duration ( Table 2, p-value<0.001).

Non-target species
Capros aper. Boarfish 11-16 cm long, abundant in the control trawl, were nearly absent in the test trawl (Fig 3). None of the covariables measured affected the retention probability of boarfish (Table 2).
Gurnard. Length analysis showed a higher retention probability for individuals shorter than 25-27 cm for the control trawl. The best-fitted model ( All model included haul as random effect and subsampling as an offset. The superscripts refer to the degree of polynomial function that models the effect of size. https://doi.org/10.1371/journal.pone.0235368.t001 the size at which retention probability was equal since the control trawl seemed to have systematically more gurnard 30-40 cm long (Fig 3). Tow duration and total catch weight were retained in model selection procedure (p-value = 0.002 and p-value = 0.003, respectively) indicating that the proportion of gurnard retained per length class decreased with increasing tow duration and catch weight.

Discussion
To the best of our knowledge, this was the first time a 100 mm T90 mesh configuration was tested simultaneously in the extension and codend, although T90 mesh has been individually tested in the codend [10,16,22] or the extension [14]. Given the previous results, the device tested in this study is interesting in terms of shape, length, mesh type and mesh size for the dermersal mixed fisheries in the Celtic Sea. The device aims to reduce, as much as possible, catches of target species below the MCRS and all sizes of non-target species. In practice, it should drastically decrease the overall amount of unwanted catches that fishers previously discarded and should now store on-board and land in port to comply with the new common fishery policy regulation. T90 mesh is known to increase the escape rate of a wide range of species, from flatfish to roundfish and from demersal to pelagic species [10][11][12][13][14][15][16], which is particularly useful for mixed fisheries. T90 mesh remains open while towing and, for a given stretched size, provides a larger space for fish to escape than diamond mesh. Compared to the mandatory upper panel designed to enhance escapement of species that exhibit "rising" escapement behavior, the use of a cylinder provides an escape area on the top, sides and the bottom of the gear. This should benefit a wider range of species that have other behaviors and escape strategies such as sinking or horizontal escapement behavior. At more than 160 mesh and 12.5 m long, the selective cylinder is longer than those used in other trials with a T90 extension and codend. Additionally, as explained by Kopp et al. [14] and Digre et al. [22], a selective panel mounted on the extension is designed to increase an escapee's survival potential by not damaging it and allowing it to avoid compression in the codend.
In this study, the use of T90 mesh increased the size selectivity of commonly encountered roundfish, as previously highlighted for beam trawling [16] and bottom trawling [13]. Our results confirm previous results for haddock and extend them to other species such as boarfish, gurnard, megrim, monkfish and rays. Catches of haddock below the MCRS were reduced to nearly zero (S3 Table). However, the size at which the retention probability was equal between the two gears was estimated to be much higher than the MCRS, which would increased nonnegligible commercial losses. By extension, even more individuals of other gadoid species, such as whiting and hake, with a thinner cross-section than haddock for a given length, would escape through this device. For fleets that depend on small-grade haddock and whiting, the 100 mm T90 extension and codend device could result in commercial losses that are too large to remain profitable. In such cases, decreasing the mesh size [23] or mounting the T90 only in the extension or codend could be useful alternatives.
Gurnard and boarfish, non-target species that could be caught in large quantities in a single haul, have coarse scales and spines that damage other species in the trawl. Fishers are thus greatly interested in decreasing gurnard and boarfish catches. More generally, the fishing master and crew report greatly improved quality in the catch when using T90 gear, which is an additional benefit [22,24]. Interestingly, even though the TAC of boarfish exceeds 100 000 t, France has zero quota for this species. Thus, the landing obligation will cause serious issues for French fishers, as this non-target species could choke the entire whitefish directed demersal fisheries. One important result is that the device tested almost eliminated capture of boarfish. The shape of T90 mesh perfectly suited the shape of boarfish, probably facilitating the its escape. Boarfish is usually captured in large schools that tend to obstruct the codend. The large surface area of the T90 mesh increased the probability that the entire school would escape from the extension and the codend. The few Atlantic horse mackerel (Trachurus trachurus) observed were not representative of large catches. However, our results are consistent with those of Kopp et al. [14], who reported the strong ability of T90 mesh to eliminate unwanted catches of small and medium pelagic species, for which demersal fleets have a low or zero quota at the vessel level.
Small megrim (smaller than MCRS of 20 cm) were rarely captured, even with the control trawl, but it was not possible to determine from our dataset whether small megrim escaped from the control trawl or whether they were not present in the environment. Only an absolute size selectivity experiment could address this issue. Our result showed a major reduction in marketable megrim of 20-30 cm long. Kopp et al. [14] observed increased selectivity for sole (Solea solea) using a 100 mm T90 extension. Mean target catch weights of other flatfish, such as M. kitt were similar between the two trawls, which differs from the results of Bayse et al. [16], who used beam trawls and found that a T90 codend decreased selectivity for flatfish. Similarly, Lomeli et al. [13] suggested that, compared to diamond mesh, T90 mesh decreased the 50% retention length for flatfish. Relative selectivity for monkfish and rays increased slightly, however, resulting in a significant decrease in unwanted catch weights for rays but not for monkfish. Our results rely on a large dataset collected during an entire year to assess impacts on the largest number of species possible. They highlight that the device should not be used during seasonal trips that target red mullet or cephalopods. Increasing size selectivity for red mullet in the Mediterranean Sea by using a smaller T90 mesh has been previously observed [12].
The effect of catch size on codend selectivity has been addressed in numerous studies, with conflicting results [25][26][27][28]. Indeed, the underlying process appears nonlinear, while moderate catch might increase size opening and enhance escapement in the codend, very large catch sizes can obstruct the codend meshes and thereby reduce the potential escape for fish. The effect of catch weight on catch at length was not retained by model selection for the 6 species studied. The only species for which catch weigh may have a small impact is gurnard, with a significant negative effect of total catch. Other environment conditions such as tow duration and fishing depth were incorporated in the models. The hypothesis that catch is proportional to tow duration has been tested in several studies [29,30]. Negative effect of tow duration on different phases of the haul (towing, hand up, surface) have already been highlighted for haddock, whiting and Norway lobster by Madsen et al. [31]. In this analysis, tow duration seems to decrease the retention probability at length for rays, monkfish and gurnards. Differences observed are often associated with size-related swimming capacities, with larger fish having a greater swimming capacity than small fish [32]. Time of the day (day/night), season or fish condition, has non negligible impacts on the effectiveness of the test gear. The sampling design did not allow us to formally assess their respective effect; however, this remaining variability was handled by implementing "haul" as a random term in the modeling process.
The experimental fishing gear evaluated in this study will perform well at a commercial level when targeting monkfish, megrim, rays and large haddock, however fishers are not likely to use this gear when targeting smaller-bodied species such as cephalopods, small haddock, whiting and hake. Selectivity studies often focus on a short list of target species; however, catches of non-target species under TAC and for which quota availability is small (or even null) can be problematic for some fisheries. The experimental fishing gear constitutes an efficient solution to mitigate catches non-target schooling fish such as boarfish.
Supporting information S1 Table. Haul characteristics in mean depth, duration, quarter and total catch weight.