ICES Journal of Marine Science: Journal du Conseil Advance Access originally published online on August 17, 2007
ICES Journal of Marine Science: Journal du Conseil 2007 64(8):1503-1511; doi:10.1093/icesjms/fsm106
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Anglerfish catchability for swept-area abundance estimates in a new survey trawl
1 Fisheries Research Services Marine Laboratory, PO Box 101, 375 Victoria Road, Torry, Aberdeen AB11 9DB, UK
2 Department of Geography, Kings College, University of London, The Strand, London WC2R 2LS, UK
Correspondence to D. G. Reid: tel: +44 1224 876544; fax: +44 1224 295511; e-mail: reiddg{at}marlab.ac.uk
Reid, D. G., Allen, V. J., Bova, D. J., Jones, E. G., Kynoch, R. J., Peach, K. J., Fernandes, P. G., and Turrell, W. R. 2007. Anglerfish catchability for swept-area abundance estimates in a new survey trawl. – ICES Journal of Marine Science, 64: 1503–1511.In 2005, a new trawl survey was launched in Scotland to estimate anglerfish (Lophius spp.) abundance using swept-area estimates. This required an understanding of the herding of the fish by the gear, particularly in the zone between the doors and wing ends. TV observations at the wing ends and along the sweeps were used to quantify the behavioural reactions of anglerfish. These observations were analysed to develop a gear efficiency estimate. This paper details the construction of the net and the procedures for data collection on the survey. In all, 54 reliable observations of anglerfish were recorded at the groundgear, the wing ends, and along the sweep/bridle combination. Detailed analysis of the recordings showed that all fish in the path of the net were captured, whereas more than half of the fish between the wings and the doors were not. The fish did not appear to herd and many of the encounters with the wires were passive. An individual-based particle-tracking model was constructed to use the behavioural observations to simulate the capture process and generate an efficiency factor. The calculated efficiency factor, based on the behavioural observations, was 1.04, indicating that almost all fish encountering the sweeps and bridles were lost. The implications and suggestions for development of this work are discussed.
Keywords: anglerfish, catchability, survey trawl
Received 31 August 2006; accepted 7 June 2007; advance access publication 17 August 2007.
 |
Introduction |
|---|
 Top Introduction Material and methodsResultsDiscussionReferences |
|---|
The northern shelf and North Sea anglerfish (Lophius piscatorius and L. budegassa) fisheries are of considerable commercial importance to Scotland. Current quotas are set at
9800 t, worth around 
24.5 million. By weight, it is the second most abundant demersal species landed in Scotland. Until recently, the assessment for these species has depended on information on effort and landings from the fishery and on fishery-independent data from bottom-trawl surveys. However, effort and landings data were considered unreliable, and the ICES Working Group on Northern Shelf Demersal Species (ICES, 2005, 2006) also examined the available survey data (from the ICES International Bottom Trawl Surveys in ICES Areas IV and VIa) and found conflicting signals. As a result, the ICES Advisory Committee on Fisheries Management recommended the establishment of a suitable survey that could provide reliable abundance estimates for these species to be used as tuning factors in subsequent assessments. The aim was to use the surveys to establish absolute abundance, using swept-area estimates of fish density and raising these to strata (Sparholt, 1990). Converting trawl cpue to biomass estimates requires an estimate of the swept area covered by the trawl and an estimate of the survey trawl efficiency. Somerton et al. (1999) divided catchability into three components: vertical herding, horizontal herding, and escapement under the footrope. In a species such as anglerfish, the type of vertical herding described by Godø and Totland (1996) is unlikely to occur, because these fish tend to stay on the seabed. Escapement under the footrope (Engås and Godø, 1989; Walsh, 1992) may occur, but is not considered here. The present study was designed to investigate the degree of horizontal herding in the survey net, particularly by the bridles, and to use this to provide a survey trawl catchability estimate and a corrected swept-area abundance estimate. The aim was to use direct underwater observations of anglerfish behaviour in the path of the survey trawl to parameterize an individual-based model of the capture process and use the model to explore what effect different behaviour parameters had on estimated efficiency.
|
Material and methods |
|---|
|
TopIntroductionMaterial and methodsResultsDiscussionReferences |
|---|
The survey
The trawl used in this study and for the subsequent stock estimation surveys was based on standard commercial gear used in the anglerfish fishery and designed in collaboration with industry representatives. A drawing of the gear is presented in Figure 1, the groundgear in Figure 2, and the towing rig in Figure 3. The gear was also fitted with a 19-mm tickler chain, mounted between the wings and rigged to run ahead of the groundgear.
|
|
|
The observations were carried out on the Scottish research vessel FRV "Scotia" in October 2005, immediately before the stock estimation survey and in the same area to ensure comparability of results. The trawl locations (Figure 4) were based on information from the commercial fishery and chosen to provide clear tows with good expected catches of anglerfish and over a range of depths. In all, 60 trawl stations were completed, totalling 63 h of trawling.
|
On 43 hauls, low-light cameras (ROS "Navigator") coupled with self-recording video tape units were mounted on one or both of the wing ends, looking down at the junction between the bridles and the net. When ambient light was insufficient, single 20 W artificial light units were necessary. In good light conditions, a camera was also mounted on the headline, 5–10 m away from the centre and looking down and across the net opening. In total, there were 85 net camera deployments.
Once the net had been deployed, and where depth and weather permitted (26 of the 43 hauls), a towed remote-control television vehicle (RCTV) with real-time monitoring and recording via a fibre-optic cable was deployed in addition to the net cameras (Wardle and Hall, 1993). A high-frequency sonar (SIMRAD SM 2000) was used to steer the RCTV to a point above one of the sweeps (Jones et al., 2001). These could then be monitored using the colour zoom camera (Kongsberg-Simrad OE14-366) mounted on a pan-and-tilt unit. The camera required illumination with artificial light which, depending on water clarity, gave a working visibility range of 3–4 m. The RCTV was maintained in this position for the duration of the tow using a Magnus rotor system, although it was not always possible to keep the sweeps in view.
Gear performance was monitored during the tows using SCANMAR trawl surveillance equipment to provide depth, headline height, door spread, wing spread, bridle angle, and average speed over the ground. These data were used to set up the gear geometry parameters for the model described later. The trawl catches were processed to provide catch numbers, length frequency, weight, and age information on anglerfish and other species.
Video data analysis
Following the survey, the video recordings from the net-mounted cameras and the RCTV were examined for sightings of anglerfish; 54 sightings were recorded. Each observation was scored for a suite of behaviours. These are summarized below.
- Fish is inactive and partly buried—run over by the sweep or herded farther into net path;
- Fish is active (not recessed) and rises off bottom, allowing the sweep to pass underneath;
- Fish burst-swims on contact with sweep—into or out of the path of the net;
- Fish swims upwards—into or out of the path of the net.
The fish were also scored for the direction and, where possible, the distance of any movement, as well as the position on the sweep or bridle. Fish seen on the headline camera were all observed entering the net. They were scored on whether they encountered the tickler chain and/or the groundgear and on how they behaved at that point.
Behavioural analysis and model of trawl efficiency
For the purposes of this analysis, anglerfish behaviour was divided into a number of simple classes:
- Observation platform—RCTV camera, wing-mounted, or headline-mounted camera;
- Initial "state" of the fish—partially buried in the sediment or actively moving above the seabed;
- Outcome—after the encounter, the fish escaped or moved into path of the net;
- Direction—for fish that moved into the path of the net, the angle of movement relative to the sweep/bridles (90° represents movement normal to the wires).
The gear was modelled in two parts, the groundgear and the sweep/bridle section. For simplicity, the model used only one side of the net, i.e. half the groundgear and one set of bridles and sweeps. The groundgear was positioned at 90° to the direction of tow. The sweep/bridle was moved at an angle representing the calculated mean bridle angle, but could be changed to represent a range of bridle angle options. The fish were represented as individual particles, initially distributed at random across a field representing the swept area of the full gear (i.e. door spread by tow length). The model then moved the gear through the field of the fish distribution at 1 s intervals and at a speed representing the observed speed in the field. At each step, the model would check for an encounter with a fish "particle".
Upon encounter, each particle was able to respond to the arrival of the gear following a set of simple rule-based decisions. The first decision was whether it had encountered the groundgear or the bridle/sweep. If it encountered the groundgear, it was scored as being captured. If it encountered the bridle/sweep, it was offered a range of possibilities:
- A proportion of the fish (initially 25%, based on video recordings) were considered as inactive and partially buried; these were treated as having been run over and escaped.
- The remainder was programmed to make a burst-swimming response at an angle and distance determined from the range of behavioural observations.
The model allowed control of the angle of burst-swimming, how far the particle moved, how many times it could react before exhaustion, and how many fish were buried and run over. We could also vary gear parameters, particularly the bridle angle. The model was run through a number of repeated cycles for a given scenario of burst distance, burst angles, number of bursts, buried fish run over, bridle angle, etc.
Calculation of catch efficiency
Trawl efficiency was calculated following the equation defined by Somerton et al. (1999). This is based on an expression of trawl efficiency as a combination of the net efficiency (i.e. the proportion of the fish in the path of the net that are caught) and the bridle/sweep efficiency (i.e. the proportion of fish from the area between the net wings and the doors that is herded into the path of the net) such that
|
|
In this model, it is assumed that all fish in the path of the net will be caught, based on the headline camera observations. Accordingly, component e of the efficiency calculation will always be 1. The final trawl efficiency will be 1 or more depending on how many of the fish in the area between the wing ends and the doors are herded into the path of the net.
|
Results |
|---|
|
TopIntroductionMaterial and methodsResultsDiscussionReferences |
|---|
Gear monitoring
Anglerfish were identified in video footage from 17 hauls; Table 1 shows the average gear geometry results recorded. The mean door spread, wing spread, and net speed were used as fixed input parameters to the modelling exercise. The mean bridle angle was also used to set the default for the modelling, and the results from the individual hauls were used to provide a reasonable range across which to alter this parameter in the model.
|
Behavioural observations
Of the 43 hauls where footage was collected, 54 anglerfish were reliably identified in 17 hauls. Eleven fish were from the headline cameras, and all passed into the net; 29 were from the wing cameras, 14 moving into the path of the net and 15 escaping; and 14 were from the RCTV, of which 7 moved into the path of the net and 7 escaped. Given the relatively small number of observations available from the wing and RCTV cameras, it was decided to use these together to represent behaviour along the bridles and sweeps. The split by behaviour and camera is presented in Figure 5, including fish that did not respond upon encounter with the wires.
|
Catch rates of anglerfish were relatively low during the survey (Table 2), and this was reflected in the small number of observations. It should be noted that these catch rates were similar to those seen in the stock estimation surveys and by commercial fishers. In addition, the RCTV and wing cameras had limited fields of view, with footprints of
2–4 m2, so were only able to monitor a small proportion of possible fish/gear encounters.
|
The angle of movement was recorded for all fish that reacted by moving into the path of the net. Of these, 22% moved at 45°, i.e. in the same direction as the net movement, but inwards; 56% at 90°, i.e. directly in from the bridle sweep; and 22% at 135°, i.e. back into the net and against the direction of the net's travel. Data on the direction of movement are presented in a kite diagram in Figure 6.
|
It was not possible to be sure in all cases that a fish observed and recorded on the RCTV was not also recorded at one of the wing cameras, except when the RCTV was on the opposite side of the net. All observations were treated as independent, and the outcome of the efficiency modelling (see below) would suggest that very few fish seen at the wings would also have been seen along the wires. Also, it should be noted that anglerfish observations taken on a single tow should not be regarded as strictly independent: there may be interaction between individuals seen and recorded. Given the low numbers of fish present and the rarity of observations, we decided to treat each fish observation as independent.
Model runs
Following initial tuning of the model under fixed conditions in which all particles reacted to the bridles/sweep at a fixed angle and burst distance (Allen, 2006), the model was run for a series of scenarios to examine the effect of the different variable input parameters.
Default conditions for these parameters for the model runs were:
- Direction of burst—as in Figure 6
- Number of burst movements = 5
- Length of burst = 2 m (based on estimates from video observations)
- Percentage buried and run over = 25% (based on estimates from video observations)
- Bridle angle = 14° (from gear geometry measurements)
- Each parameter could then be varied, whereas the others retained the default settings.
- Number of burst movements = 5
Base case with all parameters set at default
For the base case, the default conditions were set and 20 runs carried out. The resulting mean efficiency Q was 1.04 (s.e. 0.01). Under this scenario, only a very small percentage of the fish that were initially distributed between the wing ends and doors were herded into the path of the net.
To allow an estimate of the variance in the model outcomes, the original burst direction data (into and out of the path of the net) and numbers not reacting were bootstrapped to obtain the 5 and 95 percentiles of the probability distribution. The model was then re-run 50 times with these values and the same parameter settings as the base case. Mean efficiency Q for the lower boundary was again 1.03 (s.e. 0.003); for the higher boundary, it was 1.12 (s.e. 0.01).
Varying the proportion moving into the path of the trawl
In the base case, the observed percentage of fish reacting into the path of the net was 39%. Using a range of options, this proportion was varied from 50% to 100% (Table 3). In all cases, the balance between the three possible directions of burst movement into the path of the net was maintained, and the proportion escaping outwards was reduced. The results are presented in Figure 7. Q remained close to 1 for all cases up to 70%. At 80%, it rose to 1.2 (s.e. 0.03), and reached 1.36 (s.e. 0.02) when 100% of the fish reacted towards the path of the net. It seems likely that the dominant factor in this set of simulations is the number of burst movements a particle can carry out before it becomes exhausted. Even with most of the fish reacting into the path of the net, many escape subsequently through exhaustion.
|
|
Varying the number of possible burst movements
The model was tested, allowing the particles between 1 and 10 burst movements before exhaustion. Trawl efficiency remained close to 1 for all scenarios (Figure 8). The dominant factor in these scenarios is the high probability of escape upon each encounter. Effectively, given that approximately half of the particles escape on each encounter, very few of them actually take ten encounters before escaping.
|
Varying the burst distance
The number of times the particle could move was fixed at five, but the distance moved was varied between 1 and 10 m. As might be expected, trawl efficiency showed a steady but small increase over the range of burst distances. At 10 m, it was 1.27 (s.e. 0.02). The results are presented in Figure 9. In these scenarios, it seems likely that the longer burst distances mean that the particles reach the path of the net in fewer bursts, so have a reduced number of encounters and hence chance of escape.
|
Varying the bridle angle
During the trials at sea, the bridle angle varied with depth of tow between
12° and 15°. Angles of 5°, 8°, 17°, and 20° were also tested. The results (Figure 10) show that, for all angles greater than 10°, the efficiency was close to 1. However, for very shallow bridle angles, efficiency did increase slightly to 1.15 (s.e. 0.03) at 5°. It seems likely that, at such shallow angles, the number of encounters needed to reach the path of the net is reduced, so a small gain in efficiency is achieved.
|
Varying the number buried and run over
In all runs, a proportion of the fish (set at 25%, based on the video recordings) was considered as buried in the sediment and run over by the net. A range of options between 0% and 50% was tested, however, and in all cases, trawl efficiency was close to 1 (Figure 11). In these scenarios, there are still substantial numbers of particles able to escape on each encounter so, in a similar fashion to the lower percentage scenarios in the base case above, most particles will have sufficient encounters to escape.
|
|
Discussion |
|---|
|
TopIntroductionMaterial and methodsResultsDiscussionReferences |
|---|
The aim of this study was to observe anglerfish behaviour directly at the net and bridle/sweeps to develop an efficiency estimate for the net to be used in producing swept-area biomass estimates. There were two clear conclusions from the direct observations. First, all fish seen in the path of the net itself were caught. Second, large proportions of fish in the path of the bridle/sweep combination escaped the path of the gear and were not caught.
The most important finding from direct observation was that anglerfish do not appear to be herded by the gear, which is the more classic response seen in species like cod, haddock, and flatfish (Hemmings, 1973; Walsh and Godø, 2003). Most of the active responses observed were of fish making short bursts of movement on contact with the sweeps and bridles, then settling back to the seabed. However, many fish also simply rose off the seabed, allowing the bridle or sweep to pass under them. Some fish did not appear to react to the wires at all, and were simply run over by them. As a result, fish in both categories were not caught by the trawl. This passive response to the gear has also been reported for some flatfish species with a similar cryptic, buried habit (Bublitz, 1996; Ryer and Barnett, 2006). In addition, Winger et al. (1999) identified the behavioural responses of flatfish as being important in the capture process. A burst-and-glide behaviour was common in small fish and often resulted in them being overtaken by the sweeps on exhaustion.
On the basis of these behaviours, more than half the fish would escape on each encounter. The remainder would move a short distance, most likely encountering the wires again, with the same probability of escape. Thus, multiple encounters would allow a high probability of escape.
The modelling work allowed us to calculate the efficiency of the trawl and to explore how the different behaviours had an impact on it. The efficiency factors possible from this model ranged from 1, where all fish in the path of the net were caught, and all fish in the path of the sweeps and bridles were not, to 3.5 where all fish in the path of the trawl were caught.
In the base case scenario, the efficiency of the net was close to 1, and therefore almost all fish (98.5%) in the path of the sweeps/bridles would escape. The proportions buried and run over, burst-movement direction, distance, and bridle angle were all based on the observations taken during the study. The only factor that could not be informed by direct observation was the number of burst movements a fish could make before exhaustion. Therefore, it was encouraging that the scenarios where the number of possible burst movements was varied showed little deviation from the base case, suggesting that this was not the critical factor.
The behavioural factors that the model suggested could increase efficiency were distance of burst movement and, critically, the number of fish reacting away from the trawl on encounter with the wires. Distance moved could only be estimated from the direct observations within a limited field of view. Most were in the order of the 2 m used in the base case, although a small number of fish were seen making longer excursions. The number of fish moving into the trawl path was believed to be reliably estimated from the observations at 40%. The model scenarios suggest that 80% or more of the fish would have to react into the net path to make any impact on efficiency.
Even with extreme changes in all input parameters (e.g. all fish that react do so into the path of the net), the efficiency was only raised to 1.36, meaning that 85% of the fish encountering the wires would escape.
It is important to note that the behavioural observations were not obtained under perfect experimental conditions. Visibility was often poor, and decreased light levels are known to affect the response behaviour of fish (Glass and Wardle, 1989; Walsh and Hickey, 1993; Engås, 1994). Artificial light (used in some of the observations) may have modified the behaviour, although Weinberg and Munro (1999) found only one species of flatfish that was affected by artificial light in the capture process. It was only possible to observe fish at a single point and time, so it was impossible to track each fish and follow, say, the number of times it encountered gear components. The observations of the fish at the bridles and sweeps were taken from two different platforms, with the RCTV tending to observe an area farther forward along the wires than the wing-mounted cameras. It was not possible to be sure in all cases that a fish observed and recorded on the RCTV was not also recorded at one of the wing cameras, except when the RCTV was on the opposite side of the net. All observations were treated as independent, and the outcomes of the efficiency modelling would suggest that very few fish seen at the wings would also have been seen along the wires. It should also be noted that anglerfish observations taken on a single tow should not be regarded as strictly independent, so there may be interaction between individuals seen and recorded. However, given the low numbers of fish present and the scarcity of observations, we believed it was viable to treat each fish observation as independent.
A critical issue still to be resolved is the efficiency of the net itself. In this study, we assumed 100% catch rates, based on the camera observations. However, escape under the groundgear is highly likely, given previous studies of both flatfish and gadoids (Somerton et al., 1999; Munro and Somerton, 2002). Future trials will make use of auxiliary nets behind the groundgear to test for this possibility (Engås and Godø, 1989; Munro and Somerton, 2002; Ingólfsson and Jørgensen, 2006).
Although the model proved very useful in understanding the link between behaviour and net efficiency, there are a range of improvements possible. Most important would be to include differences between first and subsequent encounters. It might be assumed that most fish would be buried in the substratum on first encounter and would be swimming more actively thereafter. These encounters could be parameterized and modelled separately. Differences between the observations farther along the sweep (RCTV) and those at the wing end could be useful in this respect. The efficiency of the net component should also be varied, based on groundgear net trials.
This study was carried out using swept-area estimates from trawl surveys for stock abundance estimation. Essentially, the requirement was to learn how efficient the survey net was and to use this to correct the catch data based on gear spread and tow distance. When applying such correction factors, it is important to know if the correction actually improves the estimate. Munro (1998) suggested applying a correction only when it reduces mean square error. In this case, the correction factor would be applied globally across the surveys and would aim to reduce bias rather than variance.
In conclusion, the present study showed that it is possible to observe and quantify the behaviour of anglerfish in a survey trawl and to use this information to derive a usable net efficiency factor for use in abundance estimation.
|
References |
|---|
|
TopIntroductionMaterial and methodsResultsDiscussionReferences |
|---|
-
Allen V. J. Using an individual based model to investigate the influence of gear parameters and fish behaviour on the efficiency of an anglerfish (Lophius spp.) trawl survey. (2006) London: Kings College. MSc thesis.
Bublitz C. G. Quantitative evaluation of flatfish behaviour during capture by trawl gear. Fisheries Research (1996) 25:293–304.[CrossRef][Web of Science]
Engås A. The effects of trawl performance and fish behaviour on the catching efficiency of demersal sampling trawls. In: Marine Fish Behaviour in Capture and Abundance Estimations—Fernø A., Olsen S., eds. (1994) Oxford: Fishing News Books. 45–65.
Engås A., Godø O. R. Escape of fish under the fishing line of a Norwegian sampling trawl and its influence on survey results. Journal du Conseil International pour l'Exploration de la Mer (1989) 45:269–276.
Glass C. W., Wardle C. S. Comparison of the reactions of fish to a trawl gear, at high and low light intensities. Fisheries Research (1989) 7:249–266.[CrossRef][Web of Science]
Godø O. R., Totland A. A stationary acoustic system for monitoring undisturbed and vessel affected fish behaviour. (1996) 11. ICES Document CM 1996/B: 12.
Hemmings C. C. Direct observation of the behaviour of fish in relation to fishing gear. Helgoländer Wissenschaftliche Meeresuntersuchungen (1973) 24:348–360.[CrossRef][Web of Science]
ICES. Report of the Working Group on the Assessment of Northern Shelf Demersal Stocks (2005) 5. Copenhagen. 722. ICES Document CM 2005/ACFM: 01 2004 4–13.
ICES. Report of the Working Group on the Assessment of Northern Shelf Demersal Stocks (2006) 5. Murmansk, Russia. 803. ICES Document CM 2006/ACFM: 13 2005 10–19.
Ingólfsson Ó. A., Jørgensen T. Escapement of gadoid fish beneath a commercial bottom trawl: relevance to the overall trawl selectivity. Fisheries Research (2006) 79:303–312.[CrossRef][Web of Science]
Jones E. G., Copland P. J., Reid D. G. Combined acoustic and video observations of fish behaviour in a survey trawl. In. Report of the Joint Session of the Working Group on Fisheries Acoustics Science (WGFAST) and Technology (WGFAST) and Fishing Technology and Fish Behaviour (WGFTFB) (2001) 4–5. ICES Document CM 2001/B: 04.
Munro P. T. A decision rule based on the mean square error for correcting relative fishing power differences in trawl survey data. Fishery Bulletin US (1998) 96:538–546.
Munro P. T., Somerton D. A. Estimating net efficiency of a survey trawl for flatfishes. Fisheries Research (2002) 55:267–279.[CrossRef][Web of Science]
Ryer C. H., Barnett L. A. K. Influence of illumination and temperature upon flatfish reactivity and herding behavior: potential implications for trawl capture efficiency. Fisheries Research (2006) 81:242–250.[CrossRef][Web of Science]
Somerton D., Ianelli J., Walsh S., Smith S., Godø O. R., Ramm D. Incorporating experimentally derived estimates of survey trawl efficiency into the stock assessment process: a discussion. ICES Journal of Marine Science (1999) 56:299–302.
Sparholt H. An estimate of the total biomass of fish in the North Sea. Journal du Conseil International pour l'Exploration de la Mer (1990) 46:200–210.
Walsh S. J. Size-dependent selection at the footgear of a groundfish survey trawl. North American Journal of Fisheries Management (1992) 12:625–633.[CrossRef]
Walsh S. J., Godø O. R. Quantitative analysis of fish reaction to towed fishing gears—what responses are important? Fisheries Research (2003) 63:289–292.[CrossRef][Web of Science]
Walsh S. J., Hickey W. M. Behavioural reactions of demersal fish to bottom trawls at various light conditions. ICES Marine Science Symposium (1993) 196:68–76.
Wardle C. S., Hall C. D. Marine video. In Video Techniques in Animal Ecology and Behaviour. Wratten S. D., ed. (1993) London: Chapman and Hall. 90–111.
Weinberg K. L., Munro P. T. The effect of artificial light on escapement beneath a survey trawl. ICES Journal of Marine Science (1999) 56:266–274.
Winger P. D., He P., Walsh S. J. Swimming endurance of American plaice (Hippoglossoides platessoides) and its role in fish capture. ICES Journal of Marine Science (1999) 56:252–265.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||