Comparing the catch composition, profitability and discard survival from different trammel net designs targeting common spiny lobster (Palinurus elephas) in a Mediterranean fishery

Sampling

A total of 35 fishing trips on three, small, commercial boats with lobster trammel nets were surveyed in Majorcan waters off the Port d’Andratx and Portopetro harbors from April to August in 2015 and 2016. Scientific sampling was carried out on fishing boats that conducted their normal commercial activity while allowing for onboard observers under the condition of minimal disruption to their activity. Sampling protocols were therefore adopted to maximizing data collection while mitigating disturbance of fishing procedures. Each boat was rigged with two sets of trammel nets. A trammel net set consisted of several (between 10 and 30) netting walls (or “netting panels”), each approximately 100 m long and with a mesh size of 160 mm (full mesh knot to knot). Two manufacturing materials are currently used by fishermen in the Balearic Islands. Netting walls studied were either composed of standard PMF or MMF. A total of 981 PMF and 499 MMF netting walls were sampled. Trammel nets were either composed of one or a combination of the two materials.

Additionally, conventional MMF nets were compared to MMF nets with an additional modification called greca (Fig. 1), which is considered to reduce the capture of unwanted benthic organisms and reduce negative impacts on demersal communities. A greca is a piece of net approximately 20 cm high made of a thicker nylon material with a mesh size of 45 mm that is sown to the bottom of the main net along its entire length. Trammel nets to test the effect of a greca were composed of 10 netting walls alternating between five MMF with a greca and five MMF without a greca. Data from a total of 14 trammel net deployments were recorded. In total, 70 netting walls with greca and 70 without were evaluated. Apart from the netting materials and the greca, all trammel nets were comparable in their setup, i.e. mesh size, net length, floats and lead weighted rope used.

The date and time of deployment, location, soaking time, depth and type of net (PMF, MMF or MMF with greca) were recorded by an observer on board. The organisms caught were categorized into wanted (“marketable”) and unwanted catches (“non-marketable”) by the observer following the decisions made by the fishermen about the catch. All the marketable species and most of the large discarded species can be assigned to a specific netting wall because they remain entangled in the net. Thus, the netting wall was considered as the sample unit for those animals, which were immediately identified to the lowest taxonomic level possible, which was usually the species level, and measured with a ruler or (if not possible) visually inspected and assigned into 10 cm length classes. Visual size assessments were done to allow rapid recordings of animals that were manually removed from the net and directly discarded by the fishermen. A total of 10 cm size bins were chosen to facilitate length class allocation within a logistically difficult sampling environment and to ensure consistencies between observers. Species were classified as belonging to either the commercialized or non-commercialized fractions. It needs to be noted that during the retrieval of the trammel nets part of the organisms that have been entangled in the net may become dislodged during the process falling back into the sea before reaching the boat. As the dislodgment of entangled individuals can happen below as well as above the water this fraction cannot be quantified during normal commercial activities. Quantifying this part of the catch would require dedicated complex sampling involving scuba divers and was outside the scope of the current study. The abundances and biomass of catches presented are therefore relative values of the quantities caught by the fishing gear used. These may underestimate the true absolute quantities of organisms that did get entangled (caught) by the net. This data limitation is an inherent attribute of commercial catch and bycatch studies. In addition, the survival status (i.e., alive or dead) was noted for each discarded animal as the gear was hauled on board; where “dead” animals were observed to have no reflexes or active movement following the protocol by Benoit, Hurlbut & Chasse (2010), described below in more detail.

During net retrieval, small invertebrates or bed-forming organisms (e.g., Posidonia oceanica) were passively disentangling and falling on the deck in a continuous pattern; therefore, it was not possible to assign this fraction of the catch to a specific netting wall. Instead, this fraction was recorded at the level of the whole net and quantified by determining its total volume in liters. A subsample of 20 L in volume was taken and kept for further analysis at the laboratory from which all items were identified to species level when possible and weighed. All data were organized into a hierarchical (fishing trip, net, netting wall, and item) database.

Representativeness of the sampled boats

Due to logistical constrains, only three vessels were sampled. Therefore, to ensure the representativeness of this sample, the commercialized catches of the three surveyed boats were compared to catches made by the remaining fleet targeting lobster. Weight (kg) and first sale price (Euros/kg) for any sale of the entire Majorcan fleet are easily available because all sales of fish are carried out in the single fishing wharf of the island. The landings data can be disaggregated by commercial category, day and boat and thus the commercialized catch is known for every fishing trip (Palmer et al., 2017). Fishing trips with trammel nets targeting lobster were filtered out using the algorithm described by Palmer et al. (2017). The data corresponding to 2015 and 2016 (the same period as this study sampling) were used in a principal component analysis (PCA) using the rda function of the vegan library (Oksanen et al., 2015) in the R package (R Core Team, 2008) on Hellinger-transformed data (Borcard, Gillet & Legendre, 2011; decostand function of vegan library). The similarity patterns in landings at the between-boat and within-boat level were depicted and assessed using a plot of the scores in the first two dimensions of the PCA space.

After compiling all the fishing trips of the lobster trammel net fleet, the within-boat variability in terms of landings composition (i.e., landed biomass of 170 commercial categories; Palmer et al., 2017) was similar between the sampled boats and the rest of the fleet (Fig. 2). The convex hull polygons that include all the fishing trips of the three sampled boats covered the same spatial area in the graph as the average scores of the remaining boats in the fleet. The average scores of the three commercial boats sampled were found well within the space occupied by the average scores of the remaining fleet. This indicates that within-boat variability with respect to the season is comparable or even greater than between-boat variability. Therefore, the three sampled boats could be considered representative of the small-scale Majorcan lobster fishing fleet.

Revenue patterns at the netting wall level: effects of manufacturing material and the use of greca on commercial catches

Revenues from the commercialized fraction (i.e., gross revenue corresponding to the first sale in Euros) for any single individual were estimated from the weight and average price of the corresponding commercial category. Sale prices of a given category were estimated after averaging the sales prices of this category over all the trammel net fishing trips corresponding to the 2015 and 2016 spiny lobster fishing seasons.

After preliminary inspection of the cumulative (across species) revenue per netting wall, the observed distribution seemed to follow the same pattern as the biomass amount. Therefore, the same statistical model and Bayesian fitting approach described in the next section were adopted.

Patterns of discards at the trammel and netting wall levels: effects of manufacturing material and the use of greca

Between-species comparability in terms of abundance was problematic because discards ranged from large fish (e.g., rays) to fragments of reef-forming algae that cannot be enumerated as individual organisms. Due to the logistical constraint of sampling on board of commercial vessels weight measurements could not be attained for fish thus total length measurements to the nearest centimeter were preferred for most individuals. To circumvent this problem, fish catches were converted to biomass (kg). Published length/weight relationships were used for this conversion (Morey et al., 2003; Ilkyaz et al., 2008; Bilge et al., 2014). However, when precise length measurements were not practical the organisms were assigned into 10 cm length classes (see justification above), except for P. elephas which were assigned into 5 cm CL classes. For the calculation the maximum of each length class category was used. In the absence of a size record for an individual item within a catch, the average size of the sampled conspecifics was used to estimate its weight. Due to the low precision of visual length estimates and the absence of size data for occasional individuals as aforementioned the calculated biomass values need to be viewed with this limitation in mind. For the invertebrate bycatch fraction weight was determined via a subsample in the laboratory on land (see details above).

After preliminary inspection of the cumulative biomass (i.e., pooling weight across species) per netting wall, the distribution of the data for netting walls in which some discard was reported seemed to follow a negative exponential distribution, but the number of netting walls without discards largely exceeded the expected value. Therefore, the following model was considered: $$\begin{array}{l}{x}_i\text{dexp}\left(\frac{1}{{\text{mean}}_i}\right)\\ {\text{mean}}_i=(\text{mean}.{\text{wall}}_{\text{type}}+\text{mean}.{\text{rnd}}_{\text{boat}}+\text{mean}.{\text{rnd}}_{\text{net}})w_i\\ w_i\text{dbinom}\left(p_{\text{type}}\right)\end{array}$$where x_i is the discarded biomass of the netting wall i; w_i is an indicator variable (zero in case of no catch, one otherwise); mean_i is the inverse of the characteristic parameter of the exponential distribution for netting wall i, which results from combining mean.wall_type, a fixed effect related to the type of manufacturing material (or the use of greca); and mean.rnd_boat and mean.rnd_net are random effects that accounted for the fact that all netting walls from a given boat or net were structurally linked. The strings dexp and dbinom indicate the assumed underlying distributions for the exponential and binomial distributions, respectively. The parameters mean.rnd_boat and mean.rnd_net were assumed to be normally distributed with zero mean across boats or nets. Two separate sets of analyses were completed: (i) PFM vs. MMF and (ii) MMF vs. MMF + greca.

The parameters of the model were fitted using a Bayesian approach. Samples from the joint posterior distribution of parameters given the data were obtained using JAGS (Plummer, 2003; http://mcmc-jags.sourceforge.net/). Virtually non-informative priors were set. A custom script calling JAGS via R2jags library from the R package (R Core Team, 2008) was used. Three Markov chain Monte Carlo were run and 30,000 posterior samples were retained after proper burning (10,000) and thinning (1/10). Convergence was assessed by visual inspection of the chains and tested using the Gelman–Rubin statistic (Plummer et al., 2006). Thereafter, according to a Bayesian framework, effects attributable to a given treatment were labelled statistically not relevant or statistically relevant (corresponding to “statistically non-significant”/“statistically significant” of classic inferential statistics) depending on whether their corresponding 95% Bayesian credibility intervals overlapped or not with a targeted reference value.

In addition to the univariate analyses described above, differences in species composition attributable to the manufacturing material or greca use were tested using a redundancy analysis (RDA; Borcard, Gillet & Legendre, 2011). The analyses were completed using the rda function of the vegan library (Oksanen et al., 2015) in the R package on the Hellinger-transformed data (Borcard, Gillet & Legendre, 2011; decostand function of the vegan library). Between-category differences were assessed using the permutation procedure implemented in the anova function from the vegan library; thus, net level random effects were ignored.

Note that all the analyses above refer to the netting wall level, but some discards, i.e., invertebrate fraction could not be assigned to a given netting wall but only at the entire trammel net level (see above). Thus, both fractions (i.e., items attributable to netting wall level and those attributable to the net level) were pooled at the net level for a second set of analyses. After preliminary inspection of the distribution of log-transformed cumulative (across species) biomass (kg) per trammel net, the data appeared to be normally distributed, which allowed a simpler analysis in comparison with the analysis at the netting wall level. After removing all mixed nets (i.e., those with netting walls composed of more than one fiber type), the differences attributable to fiber type in the log-transformed biomass were estimated using the function lm in R. Note that only MMF vs. PMF was compared because the greca was always set in mixed nets.

In the analytical setup described above, using netting wall as a statistical unit, it can be deduced that consecutive netting walls cannot be considered as statistically independent. To address this concern the data were tested for the existence of spatial (i.e., between consecutive netting walls) autocorrelation which was rejected justifying the use of the presented analyses at netting wall level (For the test and its results see Fig. S1).

Survival assessment

Survival was assessed at two temporal scales: immediate and short-term (seven days). The aim of the first method was to determine the immediate survival of the animals to be discarded when the gear arrived on deck. The time that each animal had been retained in the net was unknown but was less than the maximum soaking time (48 h). The retrieval of the trammel net is relatively quick, and the air exposure of the animals to be discarded during processing is usually less than 2 min. Animals to be discarded were identified and measured, and their survival status was assessed (see survival assessment method below). The following species were caught in sufficient numbers (>30 individuals) to permit a formal analysis: spiny lobster (P. elephas), cuckoo ray (Leucoraja naevus), unidentified species of skates (Raja sp.), thornback ray (Raja clavata), lesser spotted dogfish (Scyliorhinus canicula), red scorpionfish (Scorpaena scrofa), angler fish (Lophius piscatorius) and royale cucumber (Parastichopus regalis). The proportion of living animals at deck arrival was estimated from the ratio of living to total observed animals in the unwanted catch, and the Bayesian 95% credibility interval around that proportion was calculated using an ad hoc R script (Table 1).

Additionally, short-term survival (seven days) was assessed for the target species only. A total of 16 discarded specimens of undersized lobsters collected in Port d’Andratx between 26/5/16 and 16/8/16 were transferred into small, aerated and refrigerated tanks and transported to an onshore laboratory (LIMIA facilities). The collection of juvenile lobster specimens for survival test was carried out under specific permission from the Balearic Government. In the laboratory lobsters were placed in 5,000 L fiberglass tanks and supplied with a continuous flow of 100 μm filtered and refrigerated water; the temperature was maintained below 18 °C. The tank was also equipped with plastic tub shelters. Tanks had a maximum occupancy of five individuals to avoid stress and aggressive behavior among individuals. The individuals were fed once per day with fish. The vitality status of the specimens was determined using a four-point vitality scale (Benoit, Hurlbut & Chasse, 2010; Table S1) which is based on an evaluation of damage and reflex impairment closely related to the reflex action mortality predictor proposed by Davis & Ottmar (2006) on days zero, one, two, four and seven. At the end of the seven days observation period, all surviving individuals were then marked with “T” marks and returned to the sea in a marine protected area. The small lobsters were submerged by scuba divers at a depth of 40 m and released at natural shelters to avoid predation.

A Kaplan–Meier survival function was calculated for the spiny lobster by combining immediate and short-term survival data. For estimating survival probability and 95% confidence intervals, the surviving animals were assumed to be censored at day zero (i.e., when they came on board). The analysis was conducted using the survival package (Therneau & Grambsch, 2000) in R.

General catch patterns

A total of 1,550 netting walls corresponding to 70 trammel nets and 35 fishing trips were surveyed. The fishing depth ranged from 63 to 130 m (mean depth ± S.D.: 94.4 ± 19.3 m). The average soaking time was 45 h. (2.3 ± S.D). The handling of the catch was generally fast, and the disentangling of items from the net took 2 min maximum.

A total of 706 animals, from 19 species, comprised the wanted/marketable catch (Fig. 3A). As expected, the most frequently recorded species was lobster, but other highly marketable species such as S. scrofa (17.68 Euros/kg) or Zeus faber (20.94 Euros/kg) were also frequently caught. R. clavata was the third most common marketable species.

The average number of commercialized items per netting wall was low. The 95% quantiles of the number of animals per netting wall ranged from 0 to 3 with a median value of 0. The biomass (kg) ranged from 0 to 4.4 kg with a median of 0 and a mean of 0.61 kg per netting wall. Finally, the 95% quantiles for revenues (Euros) ranged from 0.0 to 103.92 Euros with median of 0.0 and a mean value of 12.75 Euros per netting wall.

The total number of discarded species was 47. The number of discarded animals recorded at the netting wall level was 1,465. The most frequent discards were elasmobranchs (Fig. 3A). Specifically, the three most frequently discarded species were rays (Rajidae). It is also noteworthy that the 5th most commonly discarded species was the target lobster, where the discarded specimens were undersized. This species also was the main discard species with 6.35% following the EU landings obligation criteria. The “other” group included 24 different species, each one with less than five items per species. The number of discarded items per netting wall ranged from 0 to 4 (95% quantiles), with a median of 1. The biomass (kg) per netting wall ranged from 0 to 4.6 kg, with a median of 0.03 kg and a mean of 0.68 kg per netting wall.

The assemblage of non-marketable/discarded fraction at the net level (i.e., pooling animals at the netting wall level with the fraction falling on deck) differed from that seen at the net panel level (Fig. 3B). At the pooled net level, the most common species were the reef-forming, calcareous algae of the genus Lithothamnion ssp. (maerl) followed by four elasmobranchs. The overall discard faction by weight amounted to 69.74%. Note, that the units in Fig. 3B are kg and not items in order to facilitate between species comparison, i.e., inclusion of invertebrate and plant fraction. Thus Lithothamnion spp. have a proportionally larger weight compared to other organisms causing a bias in a direct comparison. Therefore, excluding maerl the discard faction by weight was 52.71%. Besides calcareous algae the discards also contained other habitat-forming organisms such as Seagrass (P. oceanica), Ascidians (Botryllus schlosseri), Kelp (Laminaria spp.), Sponges (Porifera) and hard corals (Dendrophyllia cornigera), indicating that the fishing gear interacted with the benthos.

Effects of manufacturing material and the use of greca on marketable catches

Regarding the MMF vs. PMF comparison (Fig. 4A), the proportions of netting walls/panels with some marketable catch were similar: PMF = 0.31 (95% Bayesian credibility, interval between 0.28 and 0.34) vs. MMF = 0.28 (95% Bayesian credibility, interval between 0.24 and 0.32). The estimated mean revenue when marketable items were reported in the netting wall was also similar: PMF = 41.5 Euros (32.9–50.3) vs. MMF = 42.3 Euros (31.5–52.6). The mean revenue for an average net (22 netting walls), after combining the two different estimations to calculate the expected mean commercialized catch per net, was PMF = 262.1 Euros/net (63.5–504.2) and MMF = 242.5 Euros/net (44.9–563.8). In all three cases, the differences between PMF and MMF were not considered statistically relevant.

Concerning the standard MMF vs. MMF + greca comparison (Fig. 4B), the differences in the probability of obtaining some commercial catch were statistically not relevant: MMF = 0.33 (0.23–0.44) vs. MMF + greca = 0.44 (0.33–0.56). Similarly, the estimated mean revenue when marketable items were reported in the netting wall did not to differ: MMF = 40.5 Euros (27.5–62.5) vs. MMF + greca = 39.5 Euros (27.9–57.7). After combining these estimates, the differences in the expected (mean) revenues from the pooled net (22 netting walls) was −90.0 Euros/net favored the MMF + greca, but this difference was not statistically relevant (95% quantiles ranging between −540.9 and 331.7 Euros/net).

Effects of manufacturing material and the use of greca on non-marketable/discarded catch

Concerning the MMF vs. PMF comparison at the netting panel level (Fig. 4C), the proportions of panels with unwanted catch were similar: PMF = 0.50 (0.47 and 0.51) vs. MMF = 0.49 (0.44 and 0.53). The estimated mean weight of discarded animals was also similar: PMF = 1.38 kg (0.70–1.47) vs. MMF = 1.34 kg (1.02–1.71). The differences in these two variables attributable to net type were not relevant. The expected (mean) weight of discards at the net level (i.e., 22 netting walls pooled; the average number of netting walls per net) after combining the two model parameters were PMF = 14.0 kg/net (5.7–27.6) vs. MMF = 13.5 kg/net (4.9–26.9). These differences were not statistically relevant.

Comparing the standard MMF vs. MMF + greca at the netting wall level showed relevant differences in the probability of netting walls with unwanted catches (Fig. 4D): MMF = 0.52 (0.41–0.63) vs. MMF + greca = 0.70 (0.58–0.79). The estimated mean discarded weight was statistically relevant, but in the opposite direction: MMF = 1.30 kg (0.90–1.86) vs. MMF + greca = 0.62 kg (0.43–0.91). Thus, MMF + greca netting walls tended to retain some discarded fauna more frequently, but the overall mean weight of discards was smaller. At the net level, the estimated amount of discard per net was smaller for MMF + greca, but the difference was not statistically relevant (prob. = 0.20; mean difference: 5.43 kg; 95% CI between 18.73 and −5.6 kg).

The RDA results for the differences in species composition are shown in Fig. 5. Differences between net types were statistically significant (variance explained: 0.50, residual variance = 21.49; F = 9.05 df = 2,767; probability after 1,000 random permutations < 0.001). The main differences in species composition between MMF/PMF vs. MMF + greca was due to the relative abundance of S. canicula (which has a smaller body size and was more frequently caught with MMF + greca) (Fig. 6A) and Rajidae species (which have a larger body size and were more frequently caught with MMF/PMF) (Fig. 6B).

The second set of analyses including all discards were pooled at the trammel net level, standardized by the number of netting walls. In this case, the sample size considered (i.e., number of nets) was smaller (44 nets) because nets with more than one type of manufacturing material were removed, which is the case for all MMF + greca netting walls. The estimated mean biomass of discards for PMF was 1.19 kg/netting wall (95% CI [0.94–1.51]). The corresponding figures for MMF were mean = 0.92 kg/netting wall (0.63 and 1.35). In that case, the differences between MMF and PMF in discards weight (kg) per trammel net, after accounting for differences in the number of netting walls per net, were not significant (effect of the net type: df = 1, SS = 0.57; residual: df = 41, SS = 17.67; F = 1.34, p = 0.25) (Fig. S2).

Immediate and short-term survival assessment

A total of 1,216 animals from eight species were examined at arrival on board, and 353 were alive (Table 1). The immediate survival probabilities of the most abundant discarded species (n > 30; P. elephas, L. naevus and P. regalis) were all greater than 0.6. In contrast, the survival probability of other commonly discarded species was less than 0.2 (Fig. 7).

A sample (n = 16) of living, undersized spiny lobsters (mean CL = 7.1 ± 0.9 cm) was kept in captivity for seven days. From that sample, only one lobster died that already had a “poor” vitality status on day zero. In contrast, the animals that were classified with a “good” vitality status on day zero quickly improved vitality status to “excellent” (Fig. 8).

The data from immediate and short-term survival was combined using the Kaplan–Meier survival function (Fig. S3). The survival curve reached an apparent asymptotic value after one day in captivity, which provided an asymptotic estimate of survival probability (0.64; 95% CI between 0.54 and 0.76).