ICES Journal of Marine Science: Journal du Conseil Advance Access originally published online on July 23, 2007
ICES Journal of Marine Science: Journal du Conseil 2007 64(8):1558-1568; doi:10.1093/icesjms/fsm101
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
PRESEMO—a predictive model of codend selectivity—a tool for fishery managers
1 Fisheries Research Services, Marine Laboratory, 375 Victoria Road, Aberdeen AB9 11DB, UK
2 Danish Institute for Fisheries Research, North Sea Centre, DK-9850 Hirtshals, DK
Correspondence to F. G. O'Neill: tel: +44 1224 295343; fax: +44 1224 295511; e-mail: B.Oneill{at}marlab.ac.uk
O'Neill, F. G., and Herrmann, B. 2007. PRESEMO—a predictive model of codend selectivity—a tool for fishery managers. – ICES Journal of Marine Science, 64: 1558–1568.The codend selectivity simulation model PRESEMO is a predictive model based on an understanding of the physical, biological, and behavioural mechanisms that underpin codend selection. In this paper, PRESEMO is used to predict the selectivity of a large range of codends of varying design. In particular, the selectivity of codends with mesh sizes in the range 80–160 mm, number of meshes around in the range 60–140, and netting twine thickness in the range 3–6 mm are predicted and, where possible, the predictions are validated with experimental data. Using the simulated data, the codend selectivity parameters are expressed in terms of the gear design parameters and in terms of both catch size and gear design parameters. The potential use of these results in a management context and for the development of more selective gears is highlighted by plotting iso-l50 and iso-sr curves used to identify gear design parameters that give equal estimates of the 50% retention length and the selection range, respectively. It is emphasized that this approach can be extended to consider the influence of other design parameters and, if sufficient relevant quantitative information exists, biological and behavioural parameters. As such, the model presented here will provide a better understanding of the selection process, permit a more targeted approach to codend selectivity experiments, and assist fishery managers to assess the impact of proposed technical measures that are introduced to reduce the catch of undersized fish and unwanted bycatch.
Keywords: catch weight, codend selectivity, mesh size, number of meshes around, PRESEMO, stochastic simulation, twine thickness
Received 1 September 2006; accepted 15 April 2007; advance access publication 23 July 2007.
 |
Introduction |
|---|
 Top Introduction MethodsResultsDiscussionReferences |
|---|
Mesh size regulations for codends in trawls aim to reduce fishing mortality by allowing small fish to escape through the meshes. In recent years, the minimum mesh size of the codends fished in the North Sea has been increased many times. Restrictions also exist on the maximum number of meshes around and the maximum twine thickness. Many studies have investigated the influence of these parameters, and some have developed empirical models that relate these parameters to the selectivity of the codend (Reeves et al., 1992; Galbraith et al., 1994). These models are used by fishery managers to assess the effect on future stocks of proposed codend technical measure changes. We must be very careful, however, in applying these models to cases outside the range of data from which they have been parameterized. Indeed, these models should not be extrapolated. Nevertheless, it is still necessary to be able to respond quickly to perceived changes in the fish stocks and to be able to consider alternative gear designs without having to carry out new experimental trials for each proposed design.
In this paper, we use the codend selectivity model PRESEMO to predict the haddock selectivity of a large range of diamond mesh codends of varying design. In particular, the selectivity of codends with mesh sizes in the range 80–160 mm, number of meshes around in the range 60–140 meshes, and made from double braided polyethylene (PE) twines of thicknesses in the range 3–6 mm are predicted. Where possible, the predictions are validated with experimental data. Because PRESEMO is a predictive model, that is, based on an understanding of the physical, biological, and behavioural mechanisms that underpin codend selection, we are confident of its predictions outside the experimental data range.
We analyse the simulated selectivity data in two ways: first in terms of the mesh size, the number of meshes around, and the twine thickness, and then in terms of these design parameters and the catch size. We also plot iso-l50 and iso-sr curves, which identify the combination of design parameters that give equal estimates of the 50% retention length (l50) and the selection range (sr), respectively.
|
Methods |
|---|
|
TopIntroductionMethodsResultsDiscussionReferences |
|---|
As explained in Herrmann (2005), PRESEMO is an individual-based structural model of the selection process in the codend of a trawl fishing gear. It models different populations of fish entering the codend during a tow. Each fish is assigned a weight and a maximum width and height according to its length, and is assumed to be of elliptical cross section. Each is also allocated a travel time down the codend, a time it can swim in the codend without being exhausted, a time between escape attempts, and a packing density for swimming in front of the catch. An escape attempt is deemed successful if the fish can pass through the mesh opening at the point of the codend where the attempt takes place. The openness of a mesh is a function of the codend geometry and is calculated using the methods of O'Neill (1997, 1999). Fish that do not escape fall back and become part of the catch when their exhaustion time is reached. The codend shape is continually updated as the catch builds up during the tow. At the end of a simulation, a logistic function is automatically fitted to the simulated selection data to obtain estimates of l50 and sr.
PRESEMO requires information on codend geometry, fish behaviour, the escape process, the fish population structure, and the fish morphology. The following parameter settings/descriptions were used.
Codend geometry
The codends examined were made of 3-, 4-, 5-, and 6-mm double braided PE netting, had 60, 80, 100, 120, and 140 open meshes around, and had mean inside mesh sizes of 80, 100, 120, 140, and 160 mm. All combinations of these design parameters were investigated, giving 100 different codends that were used in the simulations. The following relationships were used to relate the knot centre to knot-centre mesh size, mkc, and the twine bending stiffness, EI, with the twine thickness, t, and the inside mesh size, mi,
|
|
Simulation of selection
Herrmann and O'Neill (2005) outlined a protocol for using PRESEMO that takes between-haul variability into account. In their study, only catch weights up to 500 kg were considered. Here, we want to consider catch sizes up to 1200 kg and, therefore, have had to change some of the input parameter values. In particular, the number of fish in the target and bycatch populations entering the codends during a simulated tow was increased. The towing time also was increased to prevent the density of fish in the codend from becoming too large, and the population size structure was changed to ensure that codends with large mesh sizes retained a sufficient number of both target and bycatch fish. Table 1 summarizes the fish data used in the simulations. Herrmann and O'Neill (2005) also described a model of fish escape during the early part of a tow, when the tension in the mesh bars is low. In a subsequent paper (Herrmann and O'Neill, 2006), they relate their fish escape model to twine thickness, and it is this model that is applied here.
|
Between-haul variability is modelled by varying the size, spatial distribution, and structure of the target and bycatch population between each simulation (Herrmann and O'Neill, 2005). For each codend design, 1000 such simulations are made, from which 1000 estimates of l50, sr, and total catch weight are calculated. As there are 100 codend designs, 100 000 hauls are simulated.
Experimental data
For the comparison of simulated results with similar experimental ones, we have chosen published haddock selectivity data from Galbraith et al. (1994), Lowry and Robertson (1996), O'Neill and Kynoch (1996), Kynoch et al. (1999, 2004), and Dahm et al. (2002). These data are summarized in Table 2, and represent results from a number of different cruises and a number of different codend designs. To facilitate sufficient comparisons, the experimental data are from codends whose mesh size, number of meshes around, and twine thickness differ from the simulated codends by as much as ± 5 mm, ± 5 meshes, and ± 1 mm, respectively. These studies provided selectivity estimates for 152 individual, covered codend hauls.
|
Analysis of simulated data
In order to use and interpret the simulated data in a relatively straightforward way, two types of cubic polynomials were fitted to the estimates of l50 and sr. The first of these (Model 1) used mesh size, m, number of open meshes around, n, and twine thickness, t. The total catch weight at the end of the haul, w, is treated as a random effect, and l50 and sr are expressed as follows:
|
| (1) |
In the second (Model 2), w is considered to be a fixed effect, and l50 and sr are expressed as:
|
| (2) |
|
Results |
|---|
|
TopIntroductionMethodsResultsDiscussionReferences |
|---|
Simulated data
Figure 1 plots the distributions of l50 and sr for a selection of the simulated data. In Figure 1a, the number of meshes around is 100 open meshes, twine thickness is 4 mm, and the mesh size is 80, 100, and 120 mm, respectively. While it is clear that the mean l50 and sr increase with mesh size, there is considerable overlap (especially for sr) of the frequency distributions. In Figure 1b, the mesh size is 100 mm, the twine thickness is 4 mm, and the number of meshes around is 80, 100, and 120. Here we see, as expected, a decrease in l50 with increasing number of meshes around and a greater degree of overlap. In Figure 1c, the number of meshes around is 100 open meshes, the mesh size is 100 mm, and the twine thickness is 3, 4, and 5 mm. Although there is a considerable degree of overlap between the distributions, we can still identify that l50 decreases with increasing twine thickness, whereas there is no real difference in the sr distributions.
|
These plots not only highlight the extent of between-haul variability and the relative influence of the gear design parameters on codend selection, but also emphasize the difficulties that may be encountered in designing experiments to test a given hypothesis.
Analysis of simulated data and comparison with experimental data
The coefficient estimates of Models 1 and 2 are shown in Tables 3 and 4, respectively. The validation of these models is a two-stage procedure. First, we must demonstrate that the cubic expressions are a good fit to the simulated data and second, that their predictions are a good fit to the experimental data. Table 5 shows some comparisons of the predictions of Model 1 with the average simulated values for all the codends made from 4-mm double PE. These show that the cubic model is a good fit to the catch-averaged simulated data and that the low r2 value for sr (Table 3) is caused by the large variation of this parameter. A similar analysis for Model 2 identifies a small but systematic lack of fit for the extreme values of the design parameters, indicating that the cubic model may not model the simulated data accurately enough when catch is a fixed effect.
|
|
|
l50 = l50–pl50, and Some comparisons of predictions of these models with the experimental data of Table 2 are presented in Figure 2. The dashed lines are 95% confidence limits (i.e. ± 2 s.d.) of the simulated average parameter estimates for each codend, examples of which are found in Table 5. The data points are the relevant individual experimental haul estimates of Table 2, and the unbroken lines are the relevant empirical model predictions of the studies of Galbraith et al. (1994), Lowry and Robertson (1996), and Kynoch et al. (1999). Triangles are the predictions using Model 1, where catch weight is assumed to be a random effect, and the small square symbols are the predictions using Model 2, where catch is a fixed effect and given a value of 442 kg (the average catch weight of the experimental hauls, excluding catches > 1300 kg).
|
In general, the predictions of Models 1 and 2 provide a reasonable estimate of the experimental data and the empirical models in the literature. In particular, the l50 predictions are a very good reflection of the experimental results. Most experimental data points are within the 95% range of the simulated data, indicating that the simulations can explain a large portion of the variation found in the experimental data, although the plots of sr suggest that the model predictions may overestimate the between-haul variability.
A few experimental results are outside the predicted confidence bands. In Figure 2a, two l50 values are very small. Examining the experimental data in greater detail reveals that the catch weights in these hauls were 1390 and 1770 kg, while the cover contained only 100 and 140 kg. We are unable to explain or reproduce the large sr value of Figure 2d.
In Figures 3 and 4, we use Models 1 and 2 to generate iso-l50 and iso-sr curves. These are curves of constant l50 and sr and can be used to identify the set of parameter pairs to achieve a given selective performance. Figure 3 plots the iso-l50 and iso-sr curves in terms of the mesh size and the number of meshes around for codends made from 4-mm (Figure 3a) and 6-mm (Figure 3b) double-braided PE. This figure is very informative and, by looking along any horizontal or vertical line, we can see how the selectivity parameters will vary with either mesh size or number of meshes around. By comparing Figures 3a and 3b, we can further see the effect of increasing twine diameter. These plots are based on Model 1, which assumes that catch is a random effect.
|
|
The contours drawn in Figure 4 are based on Model 2, which considers catch to be a fixed effect. This model permits an explicit examination of the effect of catch size, and we can see from the figure how the dependence of the selectivity parameters on mesh size and number of meshes around varies as the catch size increases from 400 to 800 kg.
|
Discussion |
|---|
|
TopIntroductionMethodsResultsDiscussionReferences |
|---|
The aim of this paper is to provide a means of predicting codend selectivity outside the range of available experimental data. To do this, we employed the codend selectivity model PRESEMO to simulate the haddock selectivity of a large range of diamond-mesh codends. We then fitted a cubic model to this simulated dataset and expressed the selectivity parameters in terms of the gear design parameters and, in the case of Model 2, the catch size. We validated these models, as much as we could, with available experimental data. Because PRESEMO is a structural model, based on an understanding of the underlying physical, biological, and behavioural mechanisms that govern codend selection, we believe that its predictions outside the range of available experimental data are reasonably accurate. At the very least, we are more confident of these predictions than we would be of extrapolating the empirical models that have been developed by other authors. The obvious question then is, how far beyond the experimental data range can we accept the predictions of PRESEMO? Although there is no easy answer, we would be reasonably confident of the results presented here, at least up to mesh sizes of 140 mm and 140 meshes around. The issue of twine thickness is more difficult because twine bending stiffness also depends on the material type, twine construction, and any subsequent treatments. Furthermore, our model of twine deformation assumes a linear relationship between bending moment and twine curvature, which may not hold for composite twines, where there will be friction between the component fibres, each of which is likely to bend non-linearly [see O'Neill (2002) for a full discussion].
As a general rule, however, if this approach is to be applied to a given fishery, it will be necessary to ensure that the PRESEMO input parameters pertain to the fishery in question and that there is as much validation with experimental data as possible. Therefore, it should be possible for fishery managers to make reliable assessments of the effect of proposed codend technical measure changes, to respond quickly to changes in the fish stocks, and to be able to consider alternative gear designs without having to carry out new experimental trials for each proposed design.
Model 1 may be considered more useful. Not only is it simpler and an accurate description of the simulated data, it also implicitly takes into account changes in catch size that may arise as a result of changes of codend selectivity. Thus, it may provide more accurate assessments of the effect on future stocks of technical measure changes.
Model 2 is also useful and could be used, for example, to distinguish between fisheries where catch sizes differ. However, there is some evidence that the predictions of Model 2 exhibit a small but systematic lack of fit to the simulated data at extreme values of the design parameters. This indicates that the form of the expressions of Equation (2) may not be sufficiently flexible, which probably can be remedied by using a higher order expansion or an appropriate surface fitting routine (e.g. cubic splines).
One concern regarding the simulated haul data is that the variance of sr is overestimated. It may be possible to address this by altering some of the PRESEMO input parameters (Herrmann and O'Neill, 2005). The contour plots and the iso-curves are a particularly useful way of presenting the results. They provide a quick and easy means of identifying the different combinations of gear design parameters that have a given selective performance and can be extended to consider the influence of other design parameters, such as mesh shape, twine material and—as discussed by Herrmann and O'Neill (2006), if sufficient quantitative information exists—to factors such as light levels, fish condition/swimming ability, and water turbidity.
|
Acknowledgements |
|---|
This work has been carried out with financial support from the Commission of the European Communities, on the specific RTD programme—Quality of Life and Management of Living Resources, "Development of predictive model of codend selectivity". It does not necessarily reflect the Commission's view and in no way anticipates its future policy in this area. The authors acknowledge the great debt they owe the partners in the EU projects PREMECS and PREMECS II. Financial support was also given from a project under the development programme for sustainable fishery financed by the Directorate for Food, Fisheries and Agri Business, Denmark.
|
References |
|---|
|
TopIntroductionMethodsResultsDiscussionReferences |
|---|
-
Dahm E., Wienbeck H., West C. W., Valdemarsen J. W., O'Neill F. G. On the influence of towing speed and gear size on the selective properties of bottom trawls. Fisheries Research (2002) 55:103–119.[CrossRef][Web of Science]
Galbraith R. D., Fryer R. J., Maitland K. M. S. Demersal pair trawl cod-end selectivity models. Fisheries Research (1994) 20:13–27.[CrossRef][Web of Science]
Herrmann B. Effect of catch size and shape on the selectivity of diamond mesh cod-ends. I. Model development. Fisheries Research (2005) 71:1–13.[CrossRef][Web of Science]
Herrmann B., O'Neill F. G. Theoretical study of the between haul variation of haddock selectivity in a diamond mesh cod-end. Fisheries Research (2005) 74:243–252.[CrossRef][Web of Science]
Herrmann B., O'Neill F. G. Theoretical study of the influence of twine thickness on haddock selectivity in diamond mesh cod-ends. Fisheries Research (2006) 80:221–229.[CrossRef][Web of Science]
Kynoch R. J., Ferro R. S. T., Zuur G. The effect on juvenile haddock by-catch of changing cod-end twine thickness in EU trawl fisheries. MTS Journal (1999) 33:61–72.
Kynoch R. J., O'Dea M. C., O'Neill F. G. The effect of strengthening bags on cod-end selectivity of a Scottish demersal trawl. Fisheries Research (2004) 68:249–257.[CrossRef][Web of Science]
Lowry N., Robertson J. H. B. The effect of twine thickness on cod-end selectivity of trawls for haddock in the North Sea. Fisheries Research (1996) 26:353–363.[CrossRef][Web of Science]
O'Neill F. G., Kynoch R. J. The effect of cover mesh size and cod-end catch size on cod-end selectivity. Fisheries Research (1996) 28:291–303.[CrossRef][Web of Science]
O'Neill F. G. Differential equations governing the geometry of a diamond mesh cod-end of a trawl net. Journal of Applied Mechanics. (1997) 64:1631–1648.
O'Neill F. G. Axissymmetrical trawl cod-ends made from netting of generalized mesh shape. IMA Journal of Applied Mathematics (1999) 62:245–262.
O'Neill F. G. Bending of twines and fibres under tension. Journal of the Textile Institute (2002) 93:1–10.
Reeves S. A., Armstrong D. W., Fryer R. J., Coull K. A. The effects of mesh size, cod-end extension length and cod-end diameter on the selectivity of Scottish trawls and seines. ICES Journal of Marine Science (1992) 49:279–288.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||