As the catches in gillnets depend on the match between fish girth and mesh sizes it is generally recommended to collect information on the relation between length and various girth measurements. Similarly, for longline one may consider collecting information on length and mouth size. This sort of information is easy and cheap to acquire, as it only requires samples of fish over a large length range.

There exist simple methods to estimate gillnet selection from morphometric information. The estimates from these methods are not based directly on selection taking place, and so are not reliable. The methods may, however, be very useful for experimental planning, when deciding on which mesh size should be included in a selection study or for stock assessment surveys.

Girth may be measured by using a simple tape measure. More precise and quicker measurements may be obtained by using simple devices designated for girth measurements (e.g. see EU 1997). The string used for the measurements should be inelastic. In some studies the measurements have been standardised by applying a constant tension as, for example, measured by using an angling spring balance (e.g. EU 1997). Applying a constant tension is reasonable from a standardisation point of view, but leads to measurements that are influenced by both the tension and the compressibility of the fish body. If this is undesirable, measurements may be made by applying a gentle pull not allowing the string to cut into the skin of the fish.

McCombie and Berst (1969) suggested a plot for mapping the shape of fish useful for interpreting gillnet selection. Fish girth was measured at various parts of the body including those positions considered most important for the catch. The distance from the girth measuring positions to the snout was similarly measured, as was the total fish length. All distances and girth measures were normalised with respect to the fish length, i.e. normalised measure = (measure / fish length). Plotting the normalised girth values against the normalised distance measures provides simple girth profiles of various species. McCombie and Berst (1969) used these size profiles to interpret selection differences between species or between different maturity stages of the same species.

A simpler approach is to relate size selection directly to girth measures. Hovgård
(1996b) and Pet *et al.* (1995) note that species which differ with respect to selection
measured against length, become more similar when the selection is expressed as
relative to girth.

**Figure 4.1 The morphometric considerations used for estimating selection by
Sechin's method. The first and second panel show the fraction of fish not able
to swim through a mesh-size with a perimeter of 8 cm. The third and fourth
panel show the fraction of fish able to press the opercula through the mesh.
The fifth panel, which is the product of the second and fourth panel data,
provides the fraction of fish caught.**

Sechin (1969) and Kawamura (1972) suggested simple and conceptually appealing methods where length-girth information is used to derive selection curves for gillnets. The two approaches are almost identical. The method below is based on Sechin's formulation, and only accounts for fish being gilled or wedged.

The basic idea behind Sechin's method is to assume selection is a result of two size dependent processes - a) the fish must be small enough to get its head into the mesh and b) large enough to be retained by the mesh. The girth dimensions determine both processes. However, as there is variability in the girth for a given length, a knife edge selection by girth is not equivalent to a knife edge selection by length. Figure 4.1 first panel shows this feature, as all fish with a maximal girth above a mesh perimeter of 8 cm is retained (marked by squares) whereas all fish below this size may swim through the mesh (marked by crosses). Figure 4.1 second panel shows the proportion retained for each length class. Figure 4.1 third and fourth panels show a similar process with regard to being able to get the head into the mesh where all fish with an opercular girth below 8 cm are allowed to pass. The length composition caught is dependent on both processes and is derived as: Proportion retained * proportion passing (Fig. 4.1, fifth panel).

The actual formulation of Sechin's method is more complex than that shown in Figure
4.1. First of all twine is elastic and the fish tissue is compressible. These effects are
accounted for by adding correction factors (“K-factors”) to the length-girth
relationship. The compressibility of the tissue is expected to be larger at the position
of the maximum girth (due to the soft tissue of the body cavity) compared to the gill
region (associated with the fish skull). Therefore, two K-factors are needed: K_{max} and
K_{gill}. The K-factors may be derived from the ratio: mesh perimeter/measured girth,
where the measured girth is made along meshmarks found close to the gill, or close
to the site of maximal girth, respectively. Secondly, Sechin notes that there is also
variability in the mesh perimeter (σ). Thirdly, Sechin does not use the observed
length-girth data directly, but assumes a normal distribution of the variability (σ^{gill},
σ^{max}) around the expected value (*K ^{gill} G^{gill}*,

The percent retained and the percent passing the gill are derived by using the cumulative normal distribution (Φ). The probability of being retained is expressed as:

and similarly that the probability of a passing by the opercula is expressed as:

The selection curve is finally derived as the product of the probabilities of retaining and passing, i.e.

Selection = P ^{retained} P ^{passing}

Example 4.1 provides an estimate of gillnet selection from morphological measurements using Sechin's approach.

**Example 4.1 Erhardt and Die (1988) used Sechin's approach to estimate the
gillnet selectivity of Spanish mackerel ( Scomberomorus maculatus).**

The length-girth relations were described by linear regression: Girth = a + b*length.
The estimated standard deviations (σ^{max}, σ^{gill}) did not show significant differences
between length classes and were hence pooled over all sizes. The variation in mesh
sizes (σ) were found to be insignificant and set at zero. The input parameters are
provided in Table 4.1.

**Table 4.1 The input parameters estimated by Erhardt and Die (1988)**

Max | Gill | |

a | -2.51 | 0.21 |

b | 0.51 | 0.38 |

k | 0.975 | 0.977 |

σ^{2} | 1.173 | 0.609 |

perimeter | 17.8 | 17.8 |

Erhardt and Die calculated the selection for each of five mesh-sizes. Calculations are easily performed in Excel spreadsheets where the cumulative normal distribution (Φ) is called NORMSDIST. The calculations are shown for the mesh size 8.9 cm (=> perimeter of 17.8 cm) in Table 4.2. The selection is plotted in Figure 4.2.

**Table 4.2 The estimated size selection using the input data given in Table 4.1
for a 8.9 mm mesh size.**

Length | Retainedby max girth | Passing Gill cover | Product passed and retained |

35 | 0.004 | 1.000 | 0.00 |

36 | 0.015 | 1.000 | 0.02 |

37 | 0.044 | 1.000 | 0.04 |

38 | 0.106 | 1.000 | 0.11 |

39 | 0.215 | 1.000 | 0.22 |

40 | 0.371 | 1.000 | 0.37 |

41 | 0.551 | 0.999 | 0.55 |

42 | 0.722 | 0.995 | 0.72 |

43 | 0.853 | 0.982 | 0.84 |

44 | 0.934 | 0.947 | 0.88 |

45 | 0.975 | 0.872 | 0.85 |

46 | 0.992 | 0.746 | 0.74 |

47 | 0.998 | 0.574 | 0.57 |

48 | 1.000 | 0.386 | 0.39 |

49 | 1.000 | 0.222 | 0.22 |

50 | 1.000 | 0.107 | 0.11 |

51 | 1.000 | 0.043 | 0.04 |

52 | 1.000 | 0.014 | 0.01 |

53 | 1.000 | 0.004 | 0.00 |

54 | 1.000 | 0.001 | 0.00 |

55 | 1.000 | 0.000 | 0.00 |

**Figure 4.2 Plot of the Sechin selection curve based on the data given in Table
4.2**

Despite the ease of collecting the necessary data, Sechin's method has been rarely
used in selection studies. In most studies, the estimated selection curves are
evaluated by comparing them to the observed length frequency distributions of the
catches (Kawamura 1972, Clarke and King 1986, Erhardt and Die 1988, Reis and
Pawson 1992, Pet *et al.* 1995). These comparisons give fits of varying qualities.
Comparisons between selection curves and length frequencies of the catch should
be interpreted cautiously, as the length frequency distribution is only a rough proxy of
selection (see Section 5.1).

From a theoretical viewpoint, the morphometric methods are useful as they provide attractive tools for establishing hypotheses regarding the expected shapes of selection curves. Two examples are provided below.

If the fish growth can be described as isometric, so the shape of the fish remains the
same irrespective of fish size, the selection curves derived from using Sechin's
approach follow Baranov's principle of geometric similarity. Isometric growth implies
that a) girth is proportional to length, and that b) σ^{max} and σ^{gill} are both proportional to
fish length (girth variability is characterised by a constant coefficient of variation).
Most fish species more or less grow in an isometric manner. Gillnet selectivity curves
should therefore follow Baranov's principle.

**Figure 4.3 The relationship between the body form and the Sechin selection
curve for two species of ‘Hypothesidae’.**

The shape of Sechin's selection curves is dependent on the difference between *K ^{gill}
G^{gill}* and