Developing a semi-pelagic trawl to capture redfish in the Gulf of St. Lawrence, Canada

Model construction, flume tank test, and full-scale modifications

A 1:10 scale model of an existing groundfish balloon trawl was constructed using a combination of Froude and Newton scaling principles (Dickson, 1959; Fridman, 1973). Force and geometric modelling laws were used during the scaling process to approximate full-scale bottom trawl characteristics and performance (Araya-Schmidt et al., 2021) . The model was evaluated in a flume tank (Fig. 1) located at the Fisheries and Marine Institute of Memorial University of Newfoundland, Canada (Winger, Louche & Legge, 2006). The model was spread with a pair of Morgere PF doors and evaluated across a range of towing speeds and rigging configurations.

The full-scale net plan of the trawl is previously described in Cheng et al. (2020). It has a headline length of 40.2 m, and a fishing line of 44.5 m. The rockhopper groundgear consists of 40.6 cm ∅ rollers on average and the headline consists of 132 floats (20.3 cm ∅). The trawl belly sections were constructed with the same netting, 170 mm diamond PE twine with a 3.5 to 4.0 mm ∅ (Winger, Louche & Legge, 2006).

Off-seabed trawling is represented by seabed clearance, which was tested at different rigging scenarios. Seabed clearance is defined as the vertical distance between groundgear and the seabed (in this case the flume tank belt), calculated by subtracting headline height (the height of the headline to the seabed) and vertical opening (headline to groundgear). The headline height and vertical opening parameters were recorded using cameras (Cheng et al., 2022). A total of five rigging scenarios were tested and scaled to full-size terms, detailed in Table 1. The first rigging scenario was to connect the upper bridles to the warps at 30.5 m forward of the doors (i.e., fork connection forward of the door; Fig. 2) and tested at four flow velocities (i.e., 2.5, 2.8, 3.0, and 3.2 kt). Rigging scenarios 2 and 3 were to reduce the fork connection from 30.5 m to 20.42 and 15.24 m, respectively and tested at a flow velocity of 2.8 kt. Scenarios 4 and 5 tested the rigging scenario 3 with different warp-to-depth ratios, where the warp-to-depth ratio increased (i.e., from 2.6:1 for scenario 4 to 6.9:1 for scenario 5).

Field tests

Two field experiments were conducted off the west coast of Newfoundland in the GSL, CA, in April 2021 and April 2022 (Fig. 3) onboard the commercial fishing trawler F/V Lisa M (overall length 19.8 m; gross tonnage 122.5 t; engine power 700 hp; 1 hp = 746 W). The trawl used for field experiments was a groundfish balloon trawl, described in section 2.1 and Cheng et al. (2020). The trawl was spread with a pair of low-aspect trawl doors (Injector Door Limited, Søvik, Norway), which were 4 m² in area. Fishing locations were chosen in collaboration with fishers (i.e., captain and crews) who have experience fishing redfish; fishing was carried out 24 h a day (i.e., both day and night hauls). During field experiment No. 1, the codend had to remain open because we could not land redfish due to a combination of no available redfish quota, licensing constraints, and no local market for redfish at the time of fishing. For field experiment No. 2, the codend was closed. A T90 codend (nominal 90 mm mesh size) described in Cheng et al. (2020) was used. Mesh measurements (wet) were obtained with an Omega gauge (Fonteyne, 2005), n = 60 with a mean of 89.5 mm and a standard deviation (SD) of 3.3.

Field experiment No. 1

Experimental design and data collection

Trawl mensuration included door spread, headline height, and vertical opening (headline to groundgear) using Notus trawl mensuration sensors (Notus Electronics Ltd. St. John’s, Newfoundland and Labrador, CA). Cameras and lights were used to observe the interaction between groundgear and seabed, and fish behaviour at the trawl mouth. A set up with a camera (GoPro, Woodman Laboratories, Inc., Half Moon Bay, CA, USA) and two flashlights (DIV08W diving lights from Brinyte Technology Ltd., Guangdong, China) using red light placed within waterproof housings, and connected to a plastic panel was attached to the middle of the trawl mouth. The position of the camera system was just aft of the fishing line, looking forward and at a slight angle toward the port to observe more footage of fish interacting with the groundgear and clearer documentation of when the trawl was on or off the seabed (Fig. 4). Videos collected were analyzed using Adobe Premiere Pro (Adobe Systems, Inc., San Jose, CA, USA) by a single observer. Interaction between groundgear and seabed (trawl state) was determined at the start of the tow and defined as when the trawl was fishing off- (groundgear was off the seabed) or on-seabed (groundgear was slightly off-seabed, light on seabed or hard on the seabed). Trawl state was confirmed via video. The duration of each trawl state was counted every minute (min) from underwater videos.

Redfish behaviour at the trawl mouth

Fish behaviour and its effects on trawl entry were analyzed based on observations and behaviours of individual redfish at the center of the trawl mouth (Figs. 4 and 5). Methods were derived from a similar behavioural study outlined by Bayse, Pol & He (2016). Variables were determined between the first detection on video until trawl entry that included trawl entrance (Entry; fish entered trawl above fishing line) or escape (Escaped; fish escaped under the fishing line) and observations and behaviours were then placed into eight categories detailed in Table 2. Fish that went out of the screen without a clear trawl entry were considered unknown and removed from analysis. Of the eight categories, fish position (Position) was considered at first detection and was split into “Above” or “Below” the fishing line. Orientation is noted by the direction of the fish head in relation to the towing direction and the middle roller of the groundgear as: the head that oriented with and against the towing direction was classified as “away” and “toward”, and the head oriented to the left and right of middle roller of the groundgear were classified as “left” and “right”, respectively. Swimming behaviour was classified into two categories, Swimming behaviour 1 and Swimming behaviour 2. Swimming behaviour 1 considered behaviours in the horizontal plane. This behaviour included swimming with (With; fish that were swimming in the direction of trawling), swimming against (Against; fish that were swimming in the opposite direction of trawling), and passive (Passive; fish that were not swimming—holding station—or lying on the seabed). Swimming behaviour 2 considered behaviours in the vertical plane. This swimming behaviour included swimming up (Up; fish that swam upward), swimming down (Down; fish that swam downward towards the seabed), and no change (NC; fish that had no changes in their swimming direction in the vertical plane).

The variable “Contact” considered any contact between any section of the groundgear and redfish (Table 2). Other variables considered in the analysis included, time (period from first detection to trawl entry), trawl state (trawl on or off the seabed), and period (trawling during day or nighttime).

The observed effects from variables listed in Table 2 on the trawl entry of redfish were analyzed using a binomial generalized linear mixed model (GLMM) in R (version 4.2.2) statistical software (R Development Core Team, 2020). The model included trawl entry as the dependent variable, and independent variables listed in Table 2. Each individual tow (Tow) was considered a random effect on the intercept to account for variations in observations among tows due to extrinsic factors (i.e., environmental conditions, fish density, etc.). Model diagnostics were considered by investigating the data and models with the DHARMa package (Hartig, 2021) and multicollinearity with the vif function from the car package (Fox et al., 2012). Model selection was evaluated by information criterion (IC), where both the Bayesian information criterion (BIC; Schwarz, 1978) and the Akaike information criterion (AIC; Akaike, 1974) values with a correction for small sample sizes (AICc) were investigated. Initially all model combinations (n = 256) with Tow as a random effect on the intercept were run and parameters were estimated by maximum likelihood estimation using the automated model selection package glmulti (Calcagno & De Mazancourt, 2010) to select the candidate models which included important variables using IC and the relative importance of model terms plot in the glmulti package. The final models were run using the glmer function in the lme4 package (Bates et al., 2015), and the best model was determined from the minimum IC calculated from the BICtab or AICctab function from the bbmle package (Bolker, 2020). A delta IC of 2 or less indicated that models were similar, and the lowest IC was considered the best model.

Redfish behaviour under the groundgear

Redfish behaviour under the groundgear was recorded when the camera (in the same location as described above and in Fig. 4) was pointed straight down observing the area of the groundgear just under and behind the fishing line (Fig. 6). Redfish were observed before and after interacting with the passing groundgear. Noted observations and behaviours include, position (left or right), orientation (left, right, toward, away in relation to head position to the trawl path), swimming behaviour, turning (turn or no turn), trawl state (on- or off-bottom), and contact (contact or no contact). Swimming behaviour was grouped into passive swimming (PS; i.e., fish were laying on their side on the seabed, or slightly laying on the seabed with no swimming in response to upcoming trawl’ components), active swimming (i.e., swimming with or against the trawling direction; holding station), and startled reaction (swam sideways in relations to groundgear after being startled). The number of fish that contacted the groundgear was also noted for several swimming behaviour categories. Fish that passed under the groundgear (i.e., passing between rollers and rolled over by the rollers) were counted.

Field experiment No. 2

A small experiment compared both the semi-pelagic trawl described above at full-scale, as well as the trawl rigged as a conventional trawl, which simply involved the removal of the extended bridle cables and attached the warp/upper bridle back to the door. The conventional trawl setup is described in Cheng et al. (2020) and Nguyen et al. (2023). The goal of this work was to test both trawls ability to capture redfish and bycatch species. Catches were transferred from the codend to a hopper that fed a conveyer system. Redfish went directly to the fish hold and redfish total catch was estimated by the fisheries observer which is standard practice in the fishery. Bycatch was sorted, counted, and weighed to the nearest 0.1 kg by Marine Institute scientists. Large Atlantic halibut weights were visually estimated since we did not have large enough equipment to weigh them. The gear mensuration setup matched that described in Experiment 1.

The experimental fishing license of this study granted by Fisheries and Oceans Canada (NL-6020-20). The license required that all redfish catches be landed.

Flume tank test and full-scale modifications

Overall, the flume tank test showed that the semi-pelagic trawl was effective at fishing off the seabed and was dynamic in fishing between off- and on- the seabed. The first scenario indicated that the seabed clearance was between 0.8 and 2.3 m when connecting the upper bridles to the warps at 30.5 m forward of the doors, and the seabed clearance increased with increasing flow velocities. Reduction in bridle extension length reduced the seabed clearance (i.e., rigging scenarios 2 and 3; Table 1). Further, increasing the warp-to-depth ratios from 2.6:1 to 6.9:1 changed the trawl from being off-seabed to on-seabed (i.e., rigging scenarios 4 and 5; Table 1). Rigging scenario 1 was adapted to the existing groundfish trawl for the subsequent field experiments by extending the upper bridle with a 38.1 m cable (1.27 cm Ø) to the warp with a G-hook. A 30.5 m cable (1.27 cm Ø) was attached between previously described connections to the door. An additional 4.27 m chain was added to the aft of the lower bridles (Fig. 2).

Field experiment No. 1

Fish behaviour analysis

Fish behaviour was analyzed using 22 of 27 valid tows where video recordings were collected during the experiment, 19 tows (∼42 h) focused on redfish behaviour in the trawl mouth and three tows (∼6.5 h) with the camera pointed down towards the seabed.

Fish behaviour at the trawl mouth

A total of 2,196 redfish were observed, including 2099 individuals with a known trawl entry and 97 with an unknown trawl entry. Thus, redfish with an unknown entry outcome were removed from the analysis. Of the 2,099 redfish, 1,168 were observed to enter the trawl, and 931 escaped under the fishing line. The majority of redfish were first detected under the fishing line (73.1%), and 26.9% were observed above the fishing line of the trawl (Table 4). Redfish that were detected above the fishing line had a lower escape percentage (26.6%) than those seen on the bottom (50.9%). The most frequent swimming behaviour observed was fish that were swimming against the trawling direction (62.3%), second was swimming with the trawl (22.0%), followed by passive (15.8%). Escape rates for redfish swimming with the trawl (77.7%) were higher than those either passive or swimming with the trawl (62.2% and 28.1%, respectively) (Table 4). In relation to the vertical plane, many (60%) of redfish had no change in their swimming direction, 34% swimming upward, and 6.2% swimming downward. Redfish that were swimming downward had a higher escape rate (94.6%) than those which had no change in their swimming direction (63.9%); a few(0.8%) redfish that were swimming upward escaped under the trawl.

The vast majority (81.7%) of redfish did not contact the groundgear, while 18.3% had contact (Table 4). The mean time of redfish that entered the trawl was 1.05 s (±0.07 SEM (standard error of the mean)) vs. 0.99 s (±0.02 SEM) for those that escaped underneath the fishing line. Many (77.6%) of redfish were observed when the trawl was on-seabed, compared to 22.4% when the trawl was off-seabed. Tables 5 and 6 showed the fish behaviour when the trawl was on- and off-seabed, respectively. Additionally, more redfish (62.5%) were observed during the day than the night (37.5%, Table 4).

The automated model selection process showed convergence issues which were improved by removing the variable orientation. Thus, the total number of models ran was 128. AICc had nine models within ∼2 delta AICc and BIC had only two. AICc had lower IC values for more complicated models in comparison to BIC. The relative importance plots were similar between the different ICs. Both considered the three most important variables to be position, swimming behaviour 1, and Swimming behaviour 2. However, AICc valued swimming behaviour 1 equal to position and swimming behaviour 2 where BIC had swimming behaviour 1 at ∼20% lower than the other two variables. Thus, following the principal of parsimony and a clear model preference, BIC was used to determine that the best model included position and swimming behaviour 2 (Table 7).

Redfish behaviour under the groundgear

A total of 603 redfish were observed under the trawl, just behind the fishing line. Of those, 528 clearly escaped, while 75 had an unknown escape (swam out of view; though likely escaped) and were not further described. In general, redfish escape behaviour under the groundgear was observed in three ways: 291 (55%) were observed to show passive swimming (including 96 individuals laid on the seabed and 195 individuals were sitting or touching the seabed without swimming), 196 (37%) swam actively (i.e., swimming with or against the trawling direction), and 41 (7.7%) showed a startled response and swam sideways in relation to towing direction (Table 8). These different swimming behaviours led to differences in the ways that fish passed under the groundgear, where a total of 461 redfish passed between groundgear rollers versus 67 that were rolled over by the groundgear rollers. Most fish that swam against the upcoming trawl escaped through the escape opening between rollers, and these fish did not contact the groundgear. Fish that swam with the trawl direction kept the upcoming trawl at a short distance and mostly returned to escape between rollers until the trawl came closer. Though occasionally some of these fish were passed over or rolled over by the rollers as they swam slower than the upcoming trawl. Redfish that were startled in a sideways direction were observed to contact the front side or impinge with the inside of the rollers as they passed under the trawl. Most fish that have contact with the groundgear was observed when trawling on the seabed.

Field experiment No. 2

A total of 15 tows were completed during the experiment, including six hauls for the experimental trawl and nine hauls for the conventional trawl. Due to circumstances that included poor weather, vessel breakdowns, gear mensuration malfunctions, and lower than expected catch at the beginning of trials; comparative fishing was not attempted. Thus sampled hauls could not be paired for comparison and any day and nights effects could not be delineated. Collected data, including gear mensuration and catch of redfish and bycatch species were used for preliminary assessment of each trawl’s performance, ability to catch redfish off the seabed and potentially avoiding bycatch species in the catch regarding changing the warp length. Two different warp length ranges were conducted to improve redfish catch rates, including shorter warp length during first ten hauls (758.5 m mean in length, range 731.5–789.9 m) and longer warp length during last five hauls (849.3 m mean, range 823.0–890.6 m).

For the trawl’s performance, seabed clearance was only measured for the final two tows due to an equipment malfunction. For these tows, the trawl averaged 1.0 m (SD = 0.2) off of the seabed. The mean warp length was 791.0 m (range: 731.5 to 890.6 m). For the shorter warp length range, the average haul duration was 93.2 min (range: 55 to 148 min), the mean tow speed was 2.5 kt (range: 2.3 to 2.7 kt), the mean door spread was 61.1 m (range: 58.7 to 62.8 m), the mean depth of the fishing ground was 341.0 m (range: 330 to 354.7 m), and the average RPM(engine revolution per minute) was 1,405.6 (range: 1,356 to 1,459). For the longer warp length range, the average haul duration was 91.6 min (range: 55 to 125 min), the mean tow speed was 2.4 kt (range: 2.3 to 2.5 kt), the mean door spread was 60.7 m (range: 57.4 to 65.7), the mean depth of the fishing ground was 350.7 m (range: 334.6 to 361.1 m), and the RPM was 1441.8 on average (range: 1,419.5 to 1,464). The mean bottom water temperature was 7.3 °C (SD = 0.1).

For catches of redfish, a total of 26,341.0 kg (15 tows) was estimated to be caught during the experiment. The total catch of redfish caught during the shorter warp length tows estimated for the experimental trawl and the conventional trawl was 1,129.5 kg and 4,866.2 kg, respectively. The total catch of redfish caught during the longer warp length tows was 2,218.1 kg estimated for the experimental trawl and 18,143.7 kg estimated for the conventional trawl.

For bycatch species, a total of 15 bycatch species were observed during the experiment. Of those bycatch species, four species had capture totals above 50 kg, including Atlantic halibut (340.0 kg conventional trawl; 13.0 kg semi-pelagic trawl), white hake (175.2 kg conventional trawl; 13.0 kg semi-pelagic trawl), Atlantic cod (110.5 kg conventional trawl; 8.0 kg semi-pelagic trawl), and thorny skate (Amblyraja radiata; 75.0 kg conventional trawl; 3.0 kg semi-pelagic trawl). During the short warp length tows, the total catch of Atlantic halibut, white hake, Atlantic cod, and thorny skate were 31, 68.2, 40.5, and 17 kg, respectively. The total capture of these bycatch species during the long warp length tows (i.e., 322 kg, 132 kg, 78 kg, and 61 kg estimated for Atlantic halibut, white hake, Atlantic cod, and thorny skate, respectively).