Size Selectivity of a Diamond-Mesh Codend of Demersal Trawl for Largehead Hairtail (Trichiurus lepturus Linnaeus, 1758) in the Beibu Gulf, in the South China Sea

Abstract

Experiments were conducted to assess the selectivity of diamond mesh codends in capturing largehead hairtail (Trichiurus lepturus) in the demersal trawl fishery of the Beibu Gulf, located in the South China Sea. The selectivity experiments involved four codends with different mesh sizes (30, 35, 40, and 45 mm), and the covered codend method was employed. Catch data were analyzed using the maximum likelihood method, with both Logistic and Richards curves fitted to the data, and the abundance–biomass comparison (ABC) curve method was used to analyze the impact of codend on fishery resources. Model deviance was compared with the degree of freedom to choose the best fit curve. The Logistic curves gave a better fit for the codends with 30 and 40 mm mesh sizes, while the Richards curves had a better fit for the codends with 35 and 45 mm mesh sizes, respectively. The 50% retention lengths (L₅₀) and selection ranges (SR) were determined for 30 mm (L₅₀: 11.40 cm; SR: 3.81 cm), 35 mm (L₅₀: 11.65 cm; SR: 3.98 cm), 40 mm (L₅₀: 12.92 cm; SR: 5.30 cm), and 45 mm (L₅₀: 13.28 cm; SR: 4.14 cm) mesh codends. The results indicated that the present minimum mesh size, 40 mm diamond mesh, did not match with minimum landing size of largerhead hairtail, 190 mm anal. length. Based on the observed linear relationship between L₅₀ and mesh opening, it was determined that the diamond mesh codend should be adjusted to a larger mesh opening of approximately 72.47 mm in order to achieve an L₅₀ value of 190 mm. The ABC curve characteristics indicate that the fish community structure in the Beibu Gulf in the South China Sea is in a severely disturbed state. This study suggests that the square mesh codend and T90 codend (diamond mesh netting turned by 90 degrees) should be tested and compared with the diamond mesh codend, to obtain the optimal configuration and mesh size of the codend for demersal trawl fishery in the Beibu Gulf in the South China Sea.

1. Introduction

Trawl fishing is the most important fishing method in the South China Sea [1]. It contributes to the highest proportion of total fishing production. But like most trawl fisheries around the world, trawl fishery in the South China Sea is also characterized by high by-catch problems due to its poor selective properties [2,3,4,5]. The Beibu Gulf has long held its esteemed status as one of the South China Sea’s pivotal fishing grounds, boasting an abundance of groundfish and even setting records for the highest catch rates in bottom trawling operations. However, over the past decade, a concerning shift has unfolded, revealing a tale of overexploitation and the rampant depletion of fishery resources within the Beibu Gulf. The once-harmonious balance has been disrupted as fishing intensity among fishermen has surged beyond optimal thresholds, triggering an alarming decline in the region’s marine biodiversity [1,3,4,5].

Largerhead hairtail (Trichiurus lepturus) is the most important species for demersal trawl fishery in the South China Sea, along with other species such as jack mackerel (Trachurus japonicus), cardinal seabream (Evynnis cardinalis), and blue round scad (Decapterus maruadsi). The production of largerhead hairtail in the Guangdong, Guangxi, and Hainan provinces of the South China Sea reached peak annual production of 320,000 t in 2006, while it dropped to only 250,000 t in 2022 [6,7]. In the Beibu Gulf of the South China Sea, Largerhead hairtail is also the most important species for demersal trawl fishery, and its total landing was 72 991 t in 2007 [8]. But fisheries resources of largehead hairtail have declined due to overfishing, climate change, and environmental deterioration in China [9,10,11,12]. It has been suggested that overfishing has resulted in extreme changes to largehead hairtail populations, such as the miniaturization of individuals and early sexual maturation [13].

Enhancing the selectivity of fishing gear to target specific sizes and achieve selective fishing in fisheries stands as a paramount approach towards mitigating bycatch and discard issues, while safeguarding precious fisheries resources [14]. Amongst the myriad of factors at play, the size of the trawl mesh, particularly the mesh size of the net capsule, assumes a pivotal role in determining the captured fish’s individual size. By fine-tuning the mesh dimensions, we can significantly influence the outcome of the catch, aiming to optimize the desired size range and minimize unintended catches. The selectivity of fishing gear plays an important role in the exploitation of fish stocks and fisheries management. In order to prevent the largehead hairtail stocks from collapsing, it is necessary to improve management measures of the demersal trawl fishery. Though some fishing experiments on diamond mesh codends and square mesh codends have been conducted in the South China Sea, the selective properties of diamond mesh codend for largehead haritail are still unknown in the Beibu Gulf [15,16]. There have been very few studies concerning the selectivity of diamond (or square) mesh codend for largehead haritail in China [12,17].

Trawl selectivity research serves as a vital cornerstone for protecting valuable fisheries resources and achieving scientifically informed fisheries management. Recognizing its significance, trawl fisheries worldwide have embraced the mesh size of codend as a key component in their regulatory measures. In line with this pursuit, China has taken a proactive stance by enacting the “Ministry of Agriculture Notice (2013) No. 1”, which lays out the implementation of the minimum mesh size system for fishing gear in marine fisheries. As part of this transition, the notice designates the double-boat winged single-capsule trawl as transitional fishing gear, with its codend’s minimum mesh size provisionally set at 40 mm, drawing reference from bottom trawls. This meticulous approach highlights our commitment to sustainable practices and paves the way for a harmonious coexistence between thriving fisheries and responsible management. However, it remains to be seen whether this minimum mesh size is applicable to all bottom trawl fisheries. The main objective of this study was to compare the size selectivity of diamond mesh codends of different mesh sizes. In particular, we intended to test whether the current minimum mesh size of trawl codend, 40 mm diamond mesh, would be able to ensure the escape of juvenile largehead haritail.

2. Materials and Methods

2.1. Fishing Experiments

A series of pair trawl fishing experiments was conducted in the fishing grounds of the Beibu Gulf, in the northern South China Sea, from 6 to 18 April 2016. The main fishing areas were located at 19°17′ N–20°45′ N, 108°13′ E–108°55′ E (Figure 1), where the water depth ranged from 42 to 68 m, mainly on a muddy and sandy sea bottom. Fishing times and locations were determined by the captain, aiming to establish a more representative view of the actual situation of commercial trawling in the northern South China Sea.

A pair of commercial trawlers, named Guibeiyu 26066 and Guibeiyu 26065, were used in the fishing experiments. Their main dimensions were identical. The total length was 34.75 m, maximum width 6.80 m, and engine power 294 kW.

A conventional demersal trawl net, with a total stretch length of 168.6 m and a circumference of 52 meshes (with a mesh size of 10 m) in the net mouth, was used in the fishing trials. As the pair trawl fishing was carried out, two vessels dragged a trawl net simultaneously, so no trawl door was needed. Except for the codends, the trawl net was nearly the same as those used in commercial fishing.

Four diamond mesh codends, made of polyethylene (PE) double twine, were tested in the fishing experiments. The nominal mesh sizes of the codends were 30, 35, 40, and 45 mm. They were defined as the D30, D35, D40, and D45 codend, respectively, according to their nominal mesh sizes. Specifications of the tested codends are listed in

Table 1. Mesh sizes and specification of four codends and cover. SD represents standard deviation.

Codend	Nominal Mesh Size (mm)	Mesh Opening (mm) ± SD	Description
D30	30	23.38 ± 1.08	660 × 450
D35	35	27.03 ± 1.40	560 × 385
D40	40	30.06 ± 0.83	495 × 340
D45	45	35.46 ± 1.50	440 × 300
Cover	15	12.45 ± 0.98	2000 × 1400

.

The fishing experiments were conducted with the covered codend method without hoops. The cover net was made of single PE twine netting with a nominal mesh size of 15 mm (Table 1). A total of 40 valid hauls (10 for each codend) were conducted. Warp length was roughly between 300 and 450 m, depending on the water depth. After the warp length was let out by the two vessels, the trawl net was towed for 3 h at a speed of approximately 3.2 knots, as is the normal practice in commercial fishing. For each haul, all catches for both codend and cover were handled separately, and classified into species levels. Each species was weighed and counted, and the standard length was measured.

In the ‘Description’ column, the multiplication expressions represent the size of the mesh around the circumference and the size of the mesh in vertical direction of the codend and cover.

2.2. Data Analysis

2.2.1. The Selectivity Curve and the Parameters Estimation

As there were insufficient data in each haul to estimate the between-haul variability of codends, our study focused on the mean selectivity curves by pooling data of all hauls. The most commonly used selectivity curves, the Logistic curve and the Richards curve, were applied to estimate the size selectivity of codends to the Largehead hairtail. The logistic curve is symmetric, and is specified by two parameters, a and b, as follows:

$$r(l_i)=\frac{\exp (a+b{l}_i)}{1+\exp (a+b{l}_i)}$$

(1)

where r(l_i) is the probability that a fish of size class l_i would be retained. With the parameters a and b, the 50% retention length, L₅₀, and selection range, SR, can be obtained as:

$$L_{50}=-a/b$$

(2)

$$SR=(2\ln 3)/b$$

(3)

The Richards curve was applied to analyze the fishing data to compare with the Logistic curve. The Richards curve is asymmetric according to an additional asymmetry parameter $\delta$ in the following form:

$$r(l_i)={[\frac{\exp (a+b{l}_i)}{1+\exp (a+b{l}_i)}]}^{1/\delta }$$

(4)

when $\delta$ > 1, the curve has a longer tail to the left of L₅₀, and when 0 < $\delta$ < 1 it has longer tail to the right. When $\delta$ = 1 it changes to the (symmetric) Logistic curve [18]. In the Richards curve, L₅₀ and SR can be expressed as follows:

$$L_{50}=\frac{-\ln (2^{\delta }-1)-a}{b}$$

(5)

$$SR=\frac{\delta \ln 3+\ln (1-{0.25}^{\delta })-\ln (1-{0.75}^{\delta })}{b}$$

(6)

All parameters, including a, b, and $\delta$ in both the Logistic curve and the Richards curve, can be estimated by maximizing the following log-likelihood function [18,19,20,21]:

$$l(\theta )=\ln (L)={\displaystyle \sum _i\{N_{n_i}\cdot \ln [r(l_i)]+N_{c_i}\cdot \ln [1-r(l_i)]\}}$$

(7)

where N_ni is the catch number of largehead hairtail in the tested codends and N_ni is the catch number in the relative covers.

To maximize Equation (7), we used SOLVER in Microsoft Excel-2003 [19,20]. In both the Logistic curve and the Richards curve, the standard errors of parameters including a, b,$\delta$, L₅₀ and SR could be obtained by the estimated method [18,22].

To assess the models’ ability to accurately represent the experimental data, we conducted hypothesis testing by calculating the p-value. Our first hypothesis, H₀, aimed to determine if there was a lack of fit in each model. This was achieved by comparing the deviance with the number of degrees of freedom (DOF). Our second hypothesis, H₁, aimed to evaluate whether δ = 1 was accepted or rejected, thereby helping us select the most appropriate model between the Logistic curve and the Richards curve [18,21].

2.2.2. The Abundance–Biomass Comparison Curve

The abundance–biomass comparison (ABC) curve method was proposed by Warwick [23] in 1986 to analyze the characteristics of communities under different disturbance conditions by comparing quantitative dominance curves and biomass dominance curves in the same coordinate system [23,24]. The ABC curve method reflects the theoretical background of the traditional evolution of r selection and k selection. In an undisturbed state, the biomass dominance curve lies above the data dominance curve; when the two curves intersect, it is in a moderately disturbed state; and when the biomass dominance curve is below the number dominance curve, it indicates that the community is in a severely disturbed state [24].

The relative relationship statistics between biomass and abundance in the ABC curve were expressed in terms of W:

$$W={\displaystyle \sum _{i=1}^S\frac{(B_i-A_i)}{50(S-1)}}$$

(8)

where $B_i$ and $A_i$ are the cumulative percentages of biomass and abundance corresponding to the species ordinal numbers in the ABC curve, and S is the number of species occurring.

3. Results

3.1. Overview of the Catch Species in the Sea Trials

During the fishing trials, a diverse range of species were captured, totaling 39 different species, comprising 38 fish species and 1 cephalopod species. The target species, largehead hairtail, and three bycatch species, jack mackerel, Spanish mackerel (Scomberomorus niphonius), and Dorab (Chirocentrus dorab), were the most dominant in terms of catch weight, accounting for 70.21% of the total catch. And the percentages of largehead hairtail were 42.46%, 53.69%, 69.06%, and 43.60% for D30, D35, D40, and D45 codend, respectively. Additionally, a large volume of jellyfish was caught in the sea trials. As these jellyfish had no commercial value, they were handled and discarded by the crew as soon as they were onboard. Though these jellyfish were not classified and quantified, they were conservatively estimated to be as much as 150 kg per haul. All of the jellyfish were caught by the codends. Among all species, only the catch data of largehead hairtail and jack mackerel were sufficient enough for further analysis.

A total of 2356 largehead hairtail and 1702 jack mackerel were subsampled and measured. The length distributions of largehead hairtail and jack mackerel of each codend are shown in Figure 2 and Figure 3, respectively. For all codends, the anal length of largehead hairtail ranged from 3.5 cm to 43.5 cm, while in the cover net the distribution ranged from 3.5 cm to 19.5 cm. For jack mackerel, a fork length less than 10.0 cm accounted for 99.06% of the total catch, both in the codends and the cover nets.

3.2. Size Selectivity of Codends with Different Mesh Sizes

Table 2. Estimated parameters and the testing of hypotheses of the Logistic and Richards curves for the largehead hairtail.

Parameters	D30				D35				D40				D45
	Logistic a		Richards		Logistic		Richards a		Logistic a		Richards		Logistic		Richards a
a	−6.583	(0.892)	−6.496	(3.597)	−6.330	(0.468)	−1.944	(3.575)	−5.363	(0.695)	−5.234	(3.313)	−7.611	(0.618)	−3.259	(2.329)
b	0.577	(0.073)	0.557	(0.191)	0.524	(0.036)	0.407	(0.059)	0.415	(0.055)	0.427	(0.156)	0.561	(0.046)	0.403	(0.066)
δ			1.119	(1.049)			0.085	(0.238)			0.821	(0.817)			0.169	(0.235)
L50	11.40	(0.27)	11.38	(0.37)	12.08	(0.16)	11.65	(0.21)	12.92	(0.25)	12.89	(0.26)	13.56	(0.19)	13.28	(0.22)
SR	3.81	(0.48)	4.10	(1.24)	4.19	(0.29)	3.98	(0.37)	5.30	(0.70)	4.86	(1.47)	3.91	(0.32)	4.14	(0.39)
H0: Model fit
Deviance	15.69		15.63		37.73		21.66		31.53		30.84		34.81		26.27
DOF	19		18		28		27		26		25		30		29
p-value	0.678		0.619		0.104		0.755		0.209		0.194		0.250		0.611
H1: δ = 1
Deviance			0.06				16.07				0.68				8.54
DOF			1				1				1				1
p-value			0.802				<0.05				0.408				<0.05

^a denotes that the curve will be applied; Values in parentheses are standard errors.

presents the estimated parameters, their corresponding standard errors, and the results of the hypothesis tests for both the Logistic and Richards curves applied to each codend in the case of the largehead hairtail. The values of L₅₀ and SR for each codend estimated from the Logistic curves were similar to those from the Richards curves. The hypotheses of model fit H₀ did not indicate any problems, as all p-values were larger than 0.05. The hypotheses of symmetry H₁: δ = 1 were rejected for the D35 codend and D45 codend. Tests indicated that the Richards curves provided a significantly better fit (p-value < 0.05) for the D35 codend and D45 codend, with a major reduction in the model deviances. As a result, we chose the Richards curves for the D35 codend and D45 codend, and the Logistic curves for the D30 codend and D40 codend, as the better fit curves. According to the chosen curves, the L₅₀ of the D30, D35, D40, and D45 codends for largehead hairtail were 11.40, 11.65, 12.92, and 13.28 cm, respectively, while the SR values were 3.81, 3.98, 5.30, and 4.14 cm, respectively. The selectivity curves of each codend for largehead hairtail are shown in Figure 4, and Figure 5 displays the deviance residuals obtained from the best-fit curves for largehead hairtail. The deviance residuals show that the model fits well.

The current study reveals the relationship between the L₅₀ and the mesh opening (M) of the codend for largehead hairtail, as follows:

L₅₀ = 1.5854M + 75.095 (R² = 0.947)

(9)

For jack mackerel, the catch data were analyzed using both Logistic and Richards curves to determine the best fit. But the data were not converged to obtain available parameters from the maximum likelihood method.

3.3. ABC Curve Characteristics

The ABC curves and W-statistics of the catches from the different mesh sizes are shown in Figure 6. In undisturbed conditions, the general community structure is dominated by k-selected species (species with low reproductive rates, high survival rates, and longer lifespans), with the biomass dominance curve above the abundance dominance curve (w > 0). Conversely, under significant anthropogenic disturbances such as fishing, r-selected species (species with high reproductive rates, low survival rates, and shorter lifespans) rapidly increase, causing the abundance dominance curve to shift above the biomass dominance curve (w < 0). It can be seen from Figure 6 that the k-dominance curve of abundance intersected with the biomass curve between the first and third species, and the k-dominance curve of abundance was above the biomass curve. The W-statistic values of different mesh sizes were less than 0. However, the W values did not show a corresponding change pattern with the increase in the mesh size of the codend.

4. Discussion

4.1. The Size Selectivity of Codends with Different Mesh Sizes

Logistic curves, although symmetric in nature, have been observed to exhibit asymmetry in selectivity curves based on practical studies. This necessitates the adoption of Richards asymmetric curves for a more accurate comparative analysis. The Richards curves were confirmed to be effective to fit fishing data, with a relatively high retention of small fish [18,21]. The situation was almost the same in the present study. The retained number of small largehead hairtail, whose anal length was less than 11 cm, was 266 and 250 for the D35 and D45 codend, compared with only 80 and 113 for the D30 and D40 codend, respectively. Additionally, the Richards curves gave a better fit for the D35 and D45 codend.

For jack mackerel, the retention probabilities of small individuals were relatively high. Most of the jack mackerel, accounting for 99% of the total catch, were less than 10.0 cm. This resulted in the failure of the analysis of selective parameters, as the fishing data did not converge in the models.

According to the findings of Huang et al. [12], the L₅₀ value for largehead hairtail using a 50 mm diamond mesh codend in the East China Sea was reported to be 13.78 cm. In our current study, we obtained an L₅₀ value of 13.28 cm for the D45 codend, which closely aligns with their findings. Additionally, Huang et al. [12] also investigated the selectivity of diamond mesh codends with mesh sizes of 70 mm, 75 mm, and 80 mm for largehead hairtail. Their results indicated that the escape rate of largehead hairtail was >80% when the mesh size was larger than 70 mm, indicating a substantial reduction in the number of largehead hairtail when the mesh size of the codend was larger than 70 mm. Rajeswari et al. [25] tested the size selectivity of 40 mm mesh diamond codend for largehead hairtail along the Andhra Pradesh coast. But they used the total length data of largehead hairtail for selective analysis, making it impossible to compare their results with ours. At present, there is no minimum landing size (MLS) of largehead hairtail reported in the South China Sea. According to the estimation of Yan et al. [7], the 50% maturity length of the largehead hairtail is 190 mm anal length in the Beibu Gulf. We temporally used 190 mm anal length as the MLS of largehead hairtail to estimate the minimum mesh opening of the diamond mesh codend. The result indicated that the mesh opening of the diamond mesh codend should be enlarged to 72.47 mm to satisfy the L₅₀ equaling 190 mm. But as the largehead hairtail directed demersal trawl fishery is also a multispecies fishery, increasing the mesh size would affect the catch efficiency of other species.

In the present study, the covered codend method was used in the fishing experiments. The cover net was 1.5 times larger than the codends, both in the fully extended length and at the circumference, as suggested by Wileman et al. [18]. But there was no hoop added to the cover net, as it was complicated to handle such a large hoop onboard in the fishing process. Of course, the presence of cover net would have a masking effect on the experiments, as reported by Madsen and Holst [26]. But it is hard to quantify these influences in our study.

4.2. Impact of Trawling on Fishery Resources

The ABC curve characteristics reflect the changes in the relative numbers of large and small species in the community and the changes in the size composition of individuals [23,24]. In this experiment, the ABC curve characteristics indicated that the fish community in the Beibu Gulf, located in the South China Sea, was in a severely disturbed state, and small fish or juvenile large fish were in absolute dominance. With the increase in mesh size, the W statistics did not show a corresponding trend, and the release effect of juvenile fish was not improved, indicating that the mesh size needs to be further enlarged. This is consistent with the previous selective results [24].

The jellyfish blooms and the effect on the fisheries have been reported in China and all around the world [27,28,29]. In our experiments, the by-catch of jellyfish did have some negative impact on the estimation of codends’ selectivity. Firstly, jellyfish would clog the mesh of the codends, as mentioned by Xian et al. [29], making it hard for the fish to escape. Secondly, a number of small fish, especially jack mackerel, were embedded in the belly of the jellyfish, and were caught by the codends with jellyfish. That is also one reason for the fishing data of jack mackerel failing to converge in the maximum likelihood method. Last but not least, a lot of jellyfish would significantly increase the codend catch weight, which was considered to affect the mesh opening of the diamond codends [30]. Furthermore, the crew had to spend more time sorting and discarding these jellyfish. Park et al. [31] tested a conical jellyfish exclusion device in a trawl net, and obtained a very good result. This jellyfish exclusion device might be a solution for largehead-hairtail-directed trawl fishery when jellyfish blooms are present in the Beibu Gulf.

4.3. Prospects for Trawl Fisheries

There are three simple ways to improve the size selectivity of codends, including increasing the diamond mesh size, reducing the circumference, and using square mesh codends [32]. In the present study, the codends were designed to have the same circumference, approximately 19 m, with mesh sizes gradually increasing. Our results indicated that it is hard to improve the size-selective properties of diamond codends by only increasing mesh sizes. A better way might be to dramatically improve the size selectivity of codends by reducing the mesh circumference, and meanwhile enlarging the mesh sizes. In particular, the square mesh codends should be tested in a largehead-hairtail-directed trawl fishery in the Beibu Gulf, because it has been reported that the square mesh codend has a better selective characteristic than the diamond mesh codend for largehead hairtail [25]. It has been reported that T90 codends, in which standard diamond mesh codends are turned 90 degrees, and square codends, can potentially improve size selectivity [32,33]. These codends should be tested in a largehead-hairtail-directed trawl fishery in the Beibu Gulf to find out whether they are able to address the poor selective properties of this multispecies demersal fishery.

During this survey, a notable observation was the abundance of largehead hairtail below MLS. This could be attributed to the fact that April–June is the spawning season for hairtail in the South China Sea, with April being the peak month [34,35]. Additionally, the transitional fishing of largehead hairtail resources might have contributed to the prevalence of undersized largehead hairtail in the survey. Previous research indicates a declining catch rate and catch ratio of largehead hairtail since 2007 [6,7], which further supports the phenomenon of undersized largehead hairtail observed in this survey. Furthermore, we need to consider other factors that influence the size selectivity, such as environmental factors and fish behavior. Environmental factors include water depth, substrate type, the structure of the marine ecosystem, and the geographical locations of the fishing grounds [1,23]. These factors can impact the distribution and behavior of different species, thereby affecting the selectivity of fishing gear. Additionally, fish behavior, such as evasion and aggregation, can also influence the encounter probability and capture efficiency of fishing gear. These aspects will be the focus in future research.