W

Ka

b

a

ARRA

KWSDFM

1

WsrIifiibomiohtri

T

c

0h

Fisheries Research 143 (2013) 109– 118

Contents lists available at SciVerse ScienceDirect

Fisheries Research

jo u r n al homep age: www.elsev ier .com/ locate / f i shres

alleye pollock (Theragra chalcogramma) behavior in midwater trawls

resimir Williamsa,b,∗, Chris D. Wilsonb, John K. Hornea

School of Fisheries and Aquatic Sciences, University of Washington, Seattle, WA, United StatesAlaska Fisheries Science Center, National Marine Fisheries Service, NOAA, Seattle, WA, United States

r t i c l e i n f o

rticle history:eceived 6 March 2012eceived in revised form 23 January 2013ccepted 28 January 2013

eywords:alleye pollock

tereo-camerasIDSONish behavior

a b s t r a c t

Trawls are standard tools for surveying fisheries resources, yet they are selective in what they retain,and thus provide potentially misleading information about fish populations. In order to evaluate thepotential for selective retention in a midwater survey trawl used in conjunction with acoustic surveysof walleye pollock, fish behavior was examined using an integrated approach of optical, acoustic andrecapture net methods. A stereo-camera system was used to provide length, position and orientationinformation, and a dual-frequency identification sonar was used to track fish targets in the trawl. Fishescaping the trawl were sampled using recapture, or pocket, nets mounted to the outside of the trawl.Most fish were found to be oriented along the main trawl axis, facing the forward trawl opening. Nearestdistance to the trawl panel did not appear to be length-dependent, however, at night when ambient light

idwater trawl levels were lower, fish maintained less distance to the trawl panel compared to daytime observations.Consequently, significantly more fish escapes occurred at lower light levels. Trajectories of fish escapingthe trawl were highly variable compared with fish that herded into the net, or those whose retentionstate was unknown. Greatest escapement into pocket nets was observed from the bottom panel of thetrawl at night. These findings suggest that survey trawl samples will be less biased due to selectivitywhen trawls are conducted during the day.

. Introduction

Trawls are species and size selective (Wileman et al., 1996).hen trawls are used as sampling tools during fisheries abundance

urveys, selectivity is a source of error, as the catch may not be rep-esentative of the size and species composition in the environment.n acoustic fishery abundance surveys, pelagic trawls are used todentify the length and species composition of acoustically detectedsh aggregations (Simmonds and MacLennan, 2005). Selectivity-

nduced sampling error in both species and length composition canias abundance estimates and misrepresent the length compositionf fished populations, which negatively affect fisheries manage-ent efforts (Godo et al., 1998). Trawl selectivity is also important

n the context of the commercial fishing industry, where regulationf selectivity is a potent tool for minimizing bycatch and controllingarvest size of fish (MacLennan, 1992). Consequently, it is impor-

ant to understand fish behavioral mechanisms that cause selectiveetention in trawls, and how these behaviors may be utilized tomprove both bycatch reduction and survey accuracy.

∗ Corresponding author at: 7600 Sand Point Way NE, Seattle, WA, United States.el.: +1 206 526 4133.

E-mail addresses: kresimir.williams@noaa.gov (K. Williams),hris.wilson@noaa.gov (C.D. Wilson), jhorne@u.washington.edu (J.K. Horne).

165-7836/$ – see front matter Published by Elsevier B.V.ttp://dx.doi.org/10.1016/j.fishres.2013.01.016

Published by Elsevier B.V.

Trawls exploit fish behavior by utilizing predator avoidance andherd fish into the trawl opening (Wardle, 1993; Ryer, 2008). Pelagictrawls differ from demersal trawls in that the trawl mouth open-ing is typically larger, often tens of meters in both vertical andhorizontal dimensions. After entering the trawl mouth, fish arecontinually herded toward the central axis of the trawl as theymove towards the codend. This process effectively captures fish,which occur in low densities in the water column by filtering largevolumes of water. Mesh sizes in the forward portion of a midwa-ter trawl greatly exceed fish size. Retention in this portion of thetrawl exploits avoidance behavior and the reluctance of fish to passthrough open meshes (Valdemarsen, 2001; Glass et al., 1993). Meshsizes are gradually reduced toward the aft portions of the net, wherefish are retained when they are unable to pass the meshes.

Herding behavior is critical to the efficiency of midwater trawlsand is likely to be size-dependent due to the reduced sensory andswimming abilities of juvenile fish relative to adults (e.g. Arimotoet al., 1991; Zhang and Arimoto, 1993). Size-dependent herdingability may be the primary factor linking behavior and selectiv-ity. Despite their importance, stimuli that control fish herding andbehavior responses that result from them are not well understood

for pelagic trawls. In this study, herding behavior exhibited by wall-eye pollock (Theragra chalcogramma, hereafter pollock) is examinedin a midwater trawl used to sample fish during an acoustic abun-dance survey. A comprehensive view of fish behavior is achieved by

110 K. Williams et al. / Fisheries Resea

Fig. 1. Aleutian wing trawl (AWT) design. The AWT is used during acoustic sur-veys of walleye pollock to sample fish aggregations. Net panels with mesh size>10.2 cm are constructed with nylon twine, and 10.2 cm mesh panel is constructedw

catiagt

2

n(Df

2

tNtcio∼tStTfc

ments of the vertical and fore/aft movements of the fish targets as

ith polyethylene twine.

ombining physical sampling and measurements from optical andcoustic instruments, such as distance to the net and movementrajectories. The results provide a quantitative measure of behav-oral patterns underlying trawl selectivity for pollock, which can bepplied to other pelagic or semi-demersal species and future trawlear development for fisheries surveys and commercial applica-ions.

. Materials and methods

This study simultaneously deployed recapture nets, or pocketets, a still-image stereo-camera system, and an imaging sonardual-frequency identification sonar or DIDSON, Sound Metrics).ata were analyzed using a diversity of techniques providing dif-

erent insights into fish behavior in the trawl.

.1. Data collection

Pollock behavior observations were collected during six haulsaken on July 27, 2007 in the eastern Bering Sea (EBS) aboard theOAA ship Oscar Dyson. Hauls were made using an Aleutian wing

rawl (AWT; Fig. 1), which is a scaled-down version of a commer-ial pollock midwater trawl with a 1.3 cm mesh size liner placedn the codend. The AWT has a vertical opening of 25 m, horizontalpening of 35 m, and length from the wingtips to the codend of160 m. Average towing speeds during trawling ranged from 2.9

o 3.5 kn. The trawl was instrumented with a trawl sonar (FS70,imrad), a depth and temperature logger (SBE 39, Sea-Bird Elec-ronics), and a light meter (MK9 archival tag, Wildlife Computers).

he catch was sorted to species, and approximately 300 pollockrom the codend were measured for fork length to the nearestentimetre.

rch 143 (2013) 109– 118

2.2. Pocket nets

Fish escapement from the AWT was studied using small recap-ture bags, or pocket nets, mounted to the outside of the trawl.Previous experiments in the EBS (Williams et al., 2011) found thatthe majority of fish escapement occurred in the aft-most quarterof the trawl. To explore escapement behavior at higher resolu-tion, the same type of pocket nets were used in the current study,but were arranged only over the aft section (10.2 cm mesh size)where escapement was expected to be highest. A total of 12 pocketnets were attached to the trawl on all 4 trawl panels (top, bottom,and sides) in 3 locations along the trawl axis, with each pocketnet covering approximately 4 m2 area of the trawl surface duringfishing (Fig. 2). The catches were removed from the pocket netsand processed after each tow. The pocket nets were constructedof 1.9 cm stretch mesh clear monofilament gill net material, whichoffered low visibility and drag making it less likely to elicit avoid-ance reactions by pollock escaping the trawl. The results of the catchwere compared using a three-way ANOVA, with pocket net samplesgrouped by day versus night, trawl section, and trawl panel.

2.3. Stereo-camera

The stereo-camera system consisted of two Canon Rebel XT digi-tal cameras and a strobe in three separate underwater housings. Thehousings were fixed to a sled enabling the system to be calibratedfor stereo processing. The cameras were calibrated following thetechniques described by Williams et al. (2010), using the cameracalibration and analysis software package written in the Matlab®

computing language by Bouguet (2008). Images were taken at 5 sintervals to ensure adequate re-sampling of individual fish and tominimize effects of the strobe flash during deployment. The sledwas fastened inside the section of the trawl comprised of 40.6 cmmeshes (Fig. 3), with the cameras facing across the trawl, ortho-gonal to the main trawl axis. The cameras were pointed at thebottom panel, to overlap with the area insonified by the DIDSON.

Combining high resolution (8 Mp) images with high powerstrobe lighting allowed small fish (10–12 cm) to be identified tospecies and measured at distances up to 4.5 m from the camera.To determine fish length, tilt, yaw, and position in the trawl fromstereo-images, pixel points corresponding to the fish snout andtail were identified in the paired images. The pixel coordinatesof corresponding points were transformed into three-dimensionalcoordinates using stereo-triangulation and the stereo-geometrydetermined by calibration (Williams et al., 2010).

2.4. Dual frequency identification sonar (DIDSON)

The DIDSON is an acoustic device that uses multiple beams toform a near-video quality “image” at longer ranges than can beachieved by optical instruments (>8 m) without the use of artificiallighting. The beam array consists of 96 beams at the high frequencysetting, with equivalent beam angles of 0.3◦ by 12◦. This array ofbeams is assembled into a sector (“fan”) along the narrow dimen-sion of the beams to constitute a 29◦ × 12◦ field of view. Becausethe beams are arranged in a line, fish movement can be trackedacross the multi-beam fan within the distance range of the beams(3–11 m in this study), while movement orthogonal to that fan isundetectable.

The DIDSON was mounted to the outside of the trawl on thebottom 40.6 cm mesh size panel. The system faced aft with thenarrow dimension of beam oriented vertically, enabling measure-

well as their interactions with the bottom trawl panel (Fig. 3). Datawere recorded internally on a hard disk, and power was providedindependently using a separate battery housing. Observations were

K. Williams et al. / Fisheries Research 143 (2013) 109– 118 111

in the

cptt

tadirsifsstneta

Fig. 2. Pocket net placement used

ollected at a frequency of 1.8 MHz and a sample rate of eight frameser second. The DIDSON data provided a two-dimensional view ofhe bottom panel of the trawl, which was conceptually analogouso the view of the stereo-cameras.

DIDSON observations were used to estimate fish movement byracking the trajectories of individual fish targets. Tracking wasccomplished using a customized DIDSON tracking applicationescribed by Handegard and Williams (2008). Before the track-

ng algorithm was applied, data were pre-processed to remove theesidual image of the trawl in the DIDSON through backgroundubtraction, based on a moving average of 40 frames. The track-ng process associated individual targets into tracks by predictinguture target positions based on past target velocities. Tracks con-isting of associated individual target positions across frames weremoothed using spline fitting to reduce errors due to target posi-ion estimation. Target positions from the smoothed tracks wereormalized relative to the estimate of the trawl panel location from

ach data frame. An estimate of the angle of the trawl mesh relativeo the central trawl axis (i.e., the attack angle) was 10.8◦ based on

computer model of the trawl geometry. Track data were rotated

Fig. 3. The placement of the stereo-camera and DIDSON imaging sona

study of pollock behavior in AWT.

by this angle such that a trajectory of 0◦ was parallel to the maintrawl axis headed toward the codend.

The velocity for each individual target detection along a trackwas estimated by taking a derivative of the fitted spline function ateach time-step along the track. The velocity vectors were decom-posed into polar equivalents, resulting in the target direction d andrange r for each tracked target and frame. The direction d rep-resents target movement in a two-dimensional plane verticallyplaced in the trawl such that d = 0 indicates movement along themain trawl axis into the codend, and d = 90 indicates vertical move-ment upwards orthogonal to the main trawl axis. Target speed s foreach time step was calculated as the change in range divided by theframe interval (1/8 s).

Two analytical approaches were used for the DIDSON track data.The first approach used values of s and d independent of the trackassociations, so that the velocity for each target detection couldbe grouped based on other criteria, such as distance to the trawl

panel. In the second approach, values were averaged by track torepresent mean target direction (d̄) and mean target speed (s̄) ofindividual fish. For targets that passed throught the trawl panel,

r within a midwater trawl to observe walleye pollock behavior.

112 K. Williams et al. / Fisheries Research 143 (2013) 109– 118

Table 1Results of six hauls taken to evaluate pollock behavior in midwater trawls.

Haul parameters Units 1 2 3 4 5 6

Start tow time GMT 00:02:44 03:06:11 05:58:54 09:22:06 12:01:45 14:02:31Duration min 12.57 10.53 21 7.97 6.48 10.07Distance fished nmi 0.66 0.51 1.01 0.45 0.38 0.54Bottom depth m 139 140 140 140 140 140Footrope depth m 136 135.4 134.5 129.8 128.9 131.4Surface temperature oC 10.9 10.5 10.9 10.6 10.5 10.4Gear temperature oC 2.2 1.9 1.9 1.9 1.9 1.9Catch weight kg 435.5 868.6 379.6 547.4 248 376.6Average length cm 31.14 27.68 26.33 21.8 18.72 22.66Pocket catch num. 6 31 65 120 248 105Pocket avg. length cm 14.0 22.2 14.2 15.3 13.9 14.1

-02

otOfs

�

atbfimrbgtctc“t

btcm

Felo

DIDSON targets num. 117 24

Stereo camera targets num. NA NA

Light level �E m−2 s−1 4.71E-02 2.64E

nly the portion of the track within the net was averaged, becausehe behavior of fish outside of the trawl was not of primary interest.nly tracks consisting of four or more target detections were used

or this analysis. The change in direction (�d) from previous time-tep was computed as

d = dt+1 − dt

nd averaged by track (�d̄), which was used as an indication of tor-uosity, or how often the fish changed its swimming direction whileeing observed. Trawl velocity was not subtracted from estimatedsh speed, therefore these values do not represent actual swim-ing speed of the fish. Tracks were classified into three types of

esponses based on their final trajectories, which were determinedy averaging d for the last three detections in the track (Fig. 4). Tar-ets passing through the panel were classified as “escaping” whileargets whose final trajectory indicated they would not come inontact with the trawl netting were classed as “herding”. Targetshat were not observed to have escaped, but whose trajectory indi-ated them would encounter the trawl panel were categorized asundetermined”, since the outcome of their future encounter withhe panel could result in either herding or escapement.

A secondary objective of the study was to investigate the possi-ility of matching DIDSON and stereo-camera targets. A small metal

arget intended to be visible by the DIDSON and in the stereo-amera images was attached to the trawl to facilitate the targetatching.

ig. 4. Classification of DIDSON tracks into three response types. Fish classed asscaping have been observed to pass through the trawl panel, herding fish have aast trajectory greater than the angle of attack, and undetermined targets are notbserved to react to the panel during the observation period.

NA 80 124 63NA 52 209 996.62E-03 5.68E-06 2.24E-06 5.30E-06

3. Results

3.1. Trawl samples

Haul data are given in Table 1. Hauls were conducted in suc-cession, taking place over an 18 h period. Trawling durations werevaried in an attempt to standardize the amount of catch betweenhauls and ranged from 8 to 20 min, yielding catches of 248–869 kg.Trawling durations and the amount of catch reflect standard oper-ating procedures during surveys, where the usual maximum catchtarget is ∼800 kg. On average, pollock made up 99.5% of the catchby weight. Hauls 1 and 2 occurred during the day, haul 3 at dusk,and hauls 4–6 at night. The camera system did not operate correctlyfor hauls 1–3, while the DIDSON failed to collect data on haul 3.

3.2. Pocket net results

Pocket net catch data provide the relative distribution of escape-ment from the trawl, and a method to identify escaping targetsobserved as trajectories in the DIDSON data. The pocket net catcheswere dominated by pollock (88.5%), the majority of which wereage-1 (92.1%) ranging from 10 to 20 cm. In contrast, age-1 pollockmade up 41.1% of the codend catch by number.

For the comparative analysis, pocket net catches were expressedas a percentage relative to the number of age-1 fish caught inthe codend for each haul sample, and normalized by the cover-age area of each pocket net (m2). For the ANOVA, haul 2 wasexcluded because the size distribution of fish in the catch wasmarkedly different than those of the other five hauls, with predom-inant component of the catch being age-2 pollock. Due to a prioriexpectations of lower escapement of age-2 relative to age-1 pollockas observed in previous pocket net experiments (Williams et al.,2011), this haul sample was not directly comparable to the other

five samples in terms of escapement rate or mean length. Signifi-cantly more fish escaped from trawls conducted al lower ambientlight levels (Table 2). Escapement rate differed significantly amongpanel (p < 0.01), bud did not differ between sections of the trawl.

Table 2Three-way ANOVA table for mean pocket net escapement rate, expressed as # ofescaping fish per m2 per h. A total of six hauls were taken with pocket nets mountedon three sections (forward, mid, and aft), and on four panels (top, port, stbd, andbottom).

Source Sum sq. d.f. Mean sq. F Prob > F

Light level 0.16 1 0.16 4.71 0.034Section 0.17 2 0.09 2.56 0.087Panel 0.48 3 0.16 4.78 0.005Error 1.77 53 0.03

Total 2.58 59

K. Williams et al. / Fisheries Research 143 (2013) 109– 118 113

Fig. 5. Escape pattern observed from the pocket net data. Escapement rate is number of fish caught in pocket nets expressed as a percentage of same size fish caught in thec . Highs epresd he bot

Csddmtmo

3

ftnr(pt(b

TTwp

The median tilt value did not appear to change significantly withreduced distance to the trawl, but the proportion of fish orientedtoward the codend increased.

odend. Vertical lines represent the standard deviation across pocket net groupingshown by plotting escape rate as a function of ambient light levels (b). Dotted line rifferent trawl sections (c), while pollock appeared to preferentially escape out of t

omparisons of mean length of fish caught in pocket nets detectedignificant differences among trawl sections (Table 3), while theifferences in mean length between light levels and between theifferent panels were not significant. A multiple comparison ofean length across sections showed larger fish (Table 4) tended

o escape from the forward-most section of the trawl, and escape-ent rates through the bottom panel differed significantly from the

ther panels, which had similar rates.

.3. Stereo-camera observations

The camera system provided adequate lighting and resolutionor age-1 pollock to be identified to species and measured at dis-ances up to 4.5 m, but the long interval between images (5 s)ecessary for reducing artificial light-induced behavioral artifactsesulted in a relatively low number of targets for the analysisTable 1). Based on image-derived estimates of fish length and

osition within the trawl, the distributions of individual distanceo the trawl panel did not differ significantly between juvenile<20 cm) and larger fish (Kolmogorov–Smirnov test, p = 0.11), andoth groups had maximum densities at a distance of 0.5–1.5 m

able 3hree-way ANOVA table for mean pocket net fish lengths, in cm. A total of five haulsith pocket nets mounted on three sections (forward, mid, and aft), and on fouranels (top, port, stbd, and bottom) were compared.

Source Sum sq. d.f. Mean sq. F Prob > F

Light level 0.09 1 0.09 0.045 0.833Section 20.14 2 10.07 4.796 0.014Panel 7.06 3 2.35 1.12 0.354Error 75.59 36 2.1

Total 102.88 42

est escapement of walleye pollock was observed in hauls taken at night (a, 4–6), asents a linear fit of escapement rate to light level. Escapement was similar along thetom panel (d).

(Fig. 6). A comparison of tilt and yaw estimates yielded similarpatterns for the two size groups. The tilt distribution was wide,ranging from −50◦ to 50◦, with the 25th and 75th percentile ofobservations (middle 50%) falling on −4.4◦ and 15.4◦. Tilt mea-surements reflect an angular tilt relative to the camera platformrather than absolute estimate, although the camera mounting wasparallel to the main trawl axis, and therefore close to horizontal.Yaw estimates showed a bimodal pattern with most (76.7%) fishfacing toward the trawl mouth and the remainder heading intothe codend. No fish were observed in orientations parallel to thecamera view axis (90◦ and 270◦, i.e., orthogonal to the main trawlaxis).

Classifying the data into range bins revealed a pattern of increas-ing diversity of fish lengths closer to the trawl netting (Fig. 7).

Table 4Comparison of mean lengths of escaping pollock among three sections and meanescape rate among four sides of a midwater trawl. Variables with different letters inthe third column are significantly different from each other.

Comparison of fish lengths (cm)

Section Mean length Standard error Significancecomparison

Forward 16.1 0.60 aMid 13.7 0.44 bAft 14.0 0.41 bBottom 0.27 0.296 aStarboard 0.12 0.174 bPort 0.07 0.084 bTop 0.08 0.134 b

114 K. Williams et al. / Fisheries Research 143 (2013) 109– 118

Fig. 6. Position and orientation of walleye pollock in the midwater trawl as determined from stereo-camera images. Tilt angles indicate heading in the vertical plane with0◦ parallel to the main trawl axis. Yaw angles indicate the heading in the horizontal plane with 0◦ facing the codend, 180◦ facing the trawl mouth.

Fig. 7. Patterns of walleye pollock orientation and length relative to proximity to the midwater trawl netting panel, determined from stereo-camera images.

Fig. 8. Changes in walleye pollock trajectory (d) and speed relative to the trawl (s) as they get closer to the netting panel of a midwater trawl. Values <0 in the vertical axisindicate the position is outside of the trawl. Vertical line in left panel indicates the attach angle of the trawl panel relative to the main trawl axis, and the vertical line in theright panel indicated mean speed of the trawl the water. Results are based on DIDSON target tracking analysis.

K. Williams et al. / Fisheries Research 143 (2013) 109– 118 115

Table 5Frequency of pollock behavior response classes in a midwater trawl determined byanalysis of DIDSON tracks.

Class Haul

1 2 4 5 6

Escaping 1 2 11 9 9

3

aywttptttbMdtitta

mBttamwe

(Tdc

Fp

Fig. 10. A comparison of mean direction, mean change in direction, and mean speedof individual tracks of walleye pollock in a midwater trawl derived from DIDSON

Herding 17 6 19 25 9Undetermined 99 16 50 90 45

.4. Fish tracking using the DIDSON

DIDSON data were analyzed both as single target detectionsnd by using metrics from entire tracks. In the single-target anal-sis, speed (s) and direction (d) estimates for each target detectionere summarized by range intervals (Fig. 8). Targets furthest from

he trawl panel had a fairly uniform speed of ∼1.3 m s−1, slowerhan the mean towing speed of 1.8 m s−1. As fish got closer to theanel, median values of d transitioned from moving parallel to therawl axis (i.e., straight towards the codend) to nearly parallel tohe estimated angle of the trawl panel. This value reflects expectedrajectory for individual fish herding back into the net, which woulde equal to or greater than the trawl panel angle estimated at 10.8◦.ost animals did not appear to avoid the panel by changing their

irection until they were within 0.5 m of the trawl meshes. This pat-ern was accompanied by a decrease in median value of s and anncrease in variability closer to the panel. Pollock appeared unableo maintain stationary position relative to the trawl. As fish passedhrough the panel, their direction was highly variable, but on aver-ge indicated movement away from the trawl.

For the track-based analysis, 408 tracks were analyzed, with aean track length of 9.01 detections and a mean duration of 1.125 s.

ased on target classification, the majority of tracks (73.5%) were inhe “undetermined” category, providing no insight into the reten-ion of the target (Table 5). For many of these tracks, the end pointppeared to be very near to the trawl panel. It is possible that fishove in a lateral direction after striking the trawl netting, i.e. cross-ise to the main trawl axis, at which point the target would likely

xit the horizontally compressed acoustic beam of the DIDSON.More fish were observed to escape in the three nighttime hauls

Table 5), which corresponded to the higher pocket net catch rates.

argets differed in their nearest distance to the trawl panel betweenay and night hauls (Fig. 9), suggesting that increased escapementoincided with greater densities of fish closer to the trawl panels.

ig. 9. A comparison of day and night distributions of nearest distances of walleyeollock targets to trawl panel netting observed by the DIDSON.

data. Only portions of the track inside the trawl were used.

When comparing track metrics among the three differentclasses, ‘herding’ fish averaged a positive trajectory (away fromthe trawl panel), escaping fish displayed more variability but werenegative on average, and the “undetermined” tracks showed inter-mediate values opposite of the trajectory of the trawl, as wouldbe expected of non-mobile “passive” targets (Fig. 10). Averagedchange in direction (�d̄) showed a narrow range in values for“undetermined” tracks (i.e., approximate straight-line movement),contrasted by variable responses in the herding fish and to an evengreater degree those of escaping fish. Relative target speed of herd-ing and escaping fish was, on average, lower than “undetermined”targets, suggesting that fish interacting with the panel exhibit aneffort to maintain position relative to the trawl.

The common sampling volume where a target could be seen byboth the cameras and the DIDSON was small (∼2.95 m3), resultingin few targets that could be matched. This was partially due to thelow image acquisition rate of the cameras. The trawl portion viewedby the cameras ranged from 5 to 7 m in the DIDSON data, meaningonly larger (>25 cm) pollock remained visible. Overall, 19 pollocktargets could be reliably matched between the two devices, all ofwhich exceeded 25 cm in length. The small number of targets and

the possibility of length bias precluded the use of the matchingtargets for further analysis.

116 K. Williams et al. / Fisheries Research 143 (2013) 109– 118

Fig. 11. Exploring the potential effect of artificial strobing light on walleye pollock behavior inside a midwater trawl. Panel (a) represents the average change in directionin fish targets from the previous DIDSON time frame, derived from individual fish tracks. A spectral analysis of the mean absolute change in direction shows no pattern att lated

s

omctaf�facuesa

4

etTtcatvtpbb

ihcesmtr

he strobing frequency (b, 0.2 Hz), and the next harmonic frequency (0.4 Hz). A simuhows detectable peaks at 0.2 and 0.4 Hz (d).

The potential influence of the stereo-camera strobe lightingn DIDSON track data was investigated using spectral analysisethods. An assumption was made that fish reactions to the lights

ould be observed in track data as changes in �d corresponding tohe frequency of strobe lights (Fig. 11(a)). Values of �d were aver-ged in each frame to produce a time series of movement changeor the duration of the trawl event (�d̄). The spectral analysis of the

d̄f time series did not show any increased power at or near therequencies of the strobe (0.2 Hz) or the nearest harmonic (0.4 Hz),s might be expected if the strobe caused the fish to suddenlyhange their trajectory in response to the light stimulus. A sim-lated test data set (Fig. 11(b)) into which a periodic random error

= N(0,2) was added to �d̄f at time periods corresponding to thetrobe cycle shows the expected result of a frequency analysis given

measureable response to the lighting.

. Discussion

Trawls have often been described as crude tools for collectingcological data and conducting fisheries assessments. This is par-ially due to the mechanism by which fish are captured in trawls.rawls function by evoking instinctual predator avoidance behavioro herd fish into the codend (Fernö, 1993; Ryer, 2008). This pro-ess is dependent on behavioral characteristics of different species,s well as environmental influences on behavior such as light andemperature (Glass and Wardle, 1989; Ryer, 2008), resulting inariable efficiency of capture (Suuronen et al., 1997). The abilityo accurately reconstruct species and length compositions of fishopulations in the environment from survey trawl catches maye limited without knowledge of how species or length-specificehaviors influence trawl retention.

The main objective of this study was to quantitatively exam-ne pollock behavior in a midwater survey trawl to understandow selective retention occurs. Studies of pollock escapement fromommercial tows indicate a dependence on fish density (Ericksont al., 1996), as in high densities fish may not encounter the mesh

urface, removing the opportunity for escape. Survey trawls haveuch smaller trawl openings compared with commercial pollock

rawls, resulting in less density within the trawl and thereforeeduced influence of density on escapement behavior. The mesh

data set with a periodic random error e = N(0,2) added at the strobing frequency (c)

size in the portion of the trawl observed by the DIDSON and cam-eras was sufficiently large (40.6 cm) for even adult pollock to passthrough. Fish were only retained in the net because of their reluc-tance to pass through meshes. This behavioral retention has beenshown to be length-dependent (Williams et al., 2011), suggestingthat there are length-dependent or ontogenetic factors that controlhow fish react to trawl netting.

Quantifying fish reactions to the trawl net requires selection ofmetrics that adequately describe behavioral responses, which are inturn limited by the choice of instruments used to observe behavior.An ideal observation tool would unobtrusively provide informationon target species, length, orientation, and movement relative to thetrawl. Optical instruments often require artificial light at depthswhere fish are found, potentially biasing observations of behavior(Stoner et al., 2008). High resolution acoustic cameras, such as theDIDSON, can provide unobtrusive observation without lights, butgenerally do not provide adequate resolution to identify targets tospecies and size in the case where both targets and the instrumentplatform are not stationary. This shortcoming reduces the valueof the DIDSON in selectivity studies where fish length is a criti-cal measure. In this study, attempts were made to directly matchtargets between DIDSON and stereo-camera systems, providinglength specific trajectory and escapement data. Matching targetswas problematic due to the small zone of overlap between the twoinstruments, and might have been more useful in higher fish den-sities or with a larger number of deployments. Given the potentialvalue of corresponding DIDSON and optical targets in providinginsightful information in trawl selectivity studies, efforts should becontinued to develop this combined approach.

Results of the current study were based on integrating multipledifferent sources of data to reconstruct behavioral patterns. Theresults reveal several interesting patterns of pollock behavior in amidwater trawl. Fish observed by the DIDSON could not be mea-sured for length, but pocket net catches indicated that fish seenescaping in the DIDSON were likely to be age-1 individuals. Simi-larly, the majority of DIDSON tracks (73.5%) indicated straight-line

movement along the main trawl axis at approximately the speed ofthe apparent water flow (relative to the trawl), and >75% of the fishseen in the cameras had tilt angles within 20◦ of horizontal and fac-ing the forward trawl opening. Combined, these patterns suggest

s Resea

tsrec

hdlmEt

waAfijiedvdLmaditttgNhpi

ptmttonmhmplGjfita

lelwtatgse

K. Williams et al. / Fisherie

hat the typical behavior of pollock inside a midwater trawl is towim in the same direction as the trawl, likely due to an optomotoreaction (e.g. Kim and Wardle, 2003) or vibration of netting, andventually be overtaken by the trawl and gradually fall back to theodend.

A comparison of escapement rates showed variability betweenauls (Fig. 5(a)), which could be attributed to the different lengthistributions encountered among trawls and different ambient

ight levels. Light was found to be an important factor in escape-ent response in pollock in previous pocket net experiments in the

BS (Williams et al., 2011), with higher catch rates correspondingo lower light levels (Fig. 5(b)).

A plausible expectation of length-dependent herding behaviorould be that larger pollock would detect the netting panel from

greater distance due to greater light gathering ability (Zhang andrimoto, 1993), and therefore could maintain a greater separation

rom the panel (i.e., a stronger herding reaction). This explanations consistent with the greater retention of adults compared withuveniles by the trawl, based on the pocket net data showing annverse relationship between light intensity and numbers of fishscaping the trawl. Visual acuity has been shown to be length-ependent for pollock, but nighttime light levels may be below theisual threshold for all sizes of pollock consistent with no significantifferences in panel distance between adult and juvenile pollock.aboratory studies using walleye pollock reported that juvenile fishove closer to the trawl panel as light levels were reduced (Ryer

nd Olla, 2000), similar with DIDSON tracking observations. Trackata showed increased density of targets closer to the trawl panel

n the night tows relative to those that were conducted duringhe day. This trend was not confirmed by stereo-camera observa-ions, where the peak density was seen at approximately 1 m fromhe trawl. These different findings may result from differences ineometries of the imaging volumes between the two instruments.either device sampled the entire cross-sectional volume and bothad a disproportionate amount of sampling volume at the trawlanel (i.e., the farther away from the panel a fish went, less likely

t was to be sampled).Other laboratory studies have found that haddock preferentially

assed through meshes that were less visible or lower contrast withhe background (Glass et al., 1993), meaning that in reduced light

ore fish are expected to pass through the mesh simply as a func-ion of reduced visual stimulus. Based on pocket net catches fromhis and other pollock escapement studies (Williams et al., 2011),nly smaller fish (<20 cm) passed through the meshes in greaterumbers when light levels were low. While all sizes of pollockay not adequately detect meshes in low light, larger fish may be

erded by some other mechanism, such as physical contact with theesh. This concept would explain length-selective retention, as the

robability of smaller fish contacting the mesh would be less thanarger fish due to a lower ratio of pollock size and the mesh opening.iven the large mesh sizes in the trawl body (40–300 cm) relative to

uvenile pollock length (14–20 cm), it may be possible that smallersh partially pass through the mesh before contact with the meshwine, leading to a greater proportion of escapement post-contact,s opposed to a swim reaction back into the net.

The orientation of pollock and other gadoids in trawls at lowight levels has been described as chaotic or disorganized (Ollat al., 2000; Glass and Wardle, 1989) when observed by infraredight and strobe cameras. In this study, pollock observed at night

ere primarily oriented toward the trawl opening or, less often,he codend, but essentially never orthogonal to the main trawlxis in the horizontal plane. The tilt distribution was comparable

o pollock observations in captivity (Horne, 2003) and not muchreater than what might be expected based on other in situ or exitu observations of other gadoids (e.g. Pacific whiting, Hendersont al., 2008).

rch 143 (2013) 109– 118 117

Tracking data provided unique individual-based insight intofish movements with respect to the trawl panel. Escaping fishshowed a general trend for having the highest variability betweenindividuals across all tracking metrics, suggesting that escape-ment is not a characteristic response that could be described, butgenerally lacked predictable movement. The within-trawl trajec-tories of escaping fish were on average steeper relative to thetrawl panel than the main direction of the trawling. This find-ing suggests that rather than passively encountering the paneland allowing the netting panel to overtake them, escaping fishwere actively changing their positions relative to the netting panelas they moved to the outside of the trawl. Herding fish andthose whose response was classified as “undetermined” were lessvariable in their responses, but their mean track values differed(Fig. 10). Herding fish, by definition, headed away from the trawl,but resisted going into the codend by moving at a relative speedslower than that of the trawl, as expected of fish that perceive thetrawl panel and actively swim to maintain position (recalling thatmost fish were oriented facing forward). Undetermined-responsefish moved in a fairly straight line at a greater speed relativeto the trawl mounted DIDSON, directly toward the codend, andrepresented the dominant target category in the current set ofdata.

The pocket net data provided coarser information on pollockbehavior than the stereo-cameras and the DIDSON. The directiv-ity of escapement seen in the pocket net data suggests a tendencyfor pollock to move downward when disturbed, which was alsoseen in the Gulf of Alaska pocket net experiments with pollock(Williams et al., 2011). Gadoid schools have been observed toincrease in depth from when they are observed by the ship’sechosounder to when they encounter the trawl during fishing oper-ations (Handegard et al., 2003), and may be a common responseby other semi-demersal gadoids (Wardle, 1993). A major cause forthis reaction is probably radiated vessel noise, as fish diving after avessel passes has been demonstrated for pollock (De Robertis andWilson, 2006).

In conclusion, fish behavior in trawls is an important consider-ation in commercial trawling where bycatch reduction is important(MacLennan, 1992), and in fisheries surveys where selectivity mayproduce biased assessments of fish abundance (Nakashima, 1990;Godo and Sunnana, 1992). The results of this study may provideguidance for new designs or modifications of survey trawl gear,which may reduce trawl selectivity and thus make survey, trawlcatches more representative of sampled populations. For example,high visibility material may be used to enhance retention by pro-viding a stronger visual stimulus. Smaller mesh material could beused for the bottom panel of midwater trawls to reduce selectivitydue to the prevailing downward direction of escapement. Increasedescapement at night suggests that, in the context of acoustic-basedsurveys where trawls are used to identify backscatter, trawling atnight may result in more biased size compositions where smalland large pollock are encountered. Results from this and similarstudies of fish behavior in trawls can also be applied to commer-cial trawling operations, where reduced catches of juvenile fish aresought.

Acknowledgements

The authors wish to thank Rick Towler, David King, Darin Jones,and Abigail McCarthy for their assistance with data collections, aswell as the crew of the NOAA ship Oscar Dyson. This paper has

benefited from manuscript comments provided by Chris Rooper,Craig Rose, and Mike Guttormsen. Reference to trade names doesnot imply endorsement by the National Marine Fisheries Service,NOAA.

1 s Resea

R

A

B

D

E

F

G

G

G

G

H

H

H

H

K

M

18 K. Williams et al. / Fisherie

eferences

rimoto, T., Gang, X., Matsushita, Y., 1991. Muscle-contraction time of capturedwalleye pollock Theragra chalcogramma. Nippon Suisan Gakk. 57, 1225–1228.

ouguet, J.Y., 2008. Camera Calibration Toolbox for Matlab (online), Available from:http://vision.caltech.edu/bouguetj/calib doc/index.html (accessed 10.08).

e Robertis, A., Wilson, C.D., 2006. Walleye pollock respond to trawling vessels. ICESJ. Mar. Sci. 63, 514–522.

rickson, D.L., Perez-Comas, J.A., Pikitch, E.K., Wallace, J.R., 1996. Effects of catch sizeand codend type on the escapement of walleye pollock (Theragra chalcogramma)from pelagic trawls. Fish. Res. 28, 179–196.

ernö, A., 1993. Advances in understanding of basic behaviour: consequences forfish capture studies. ICES Mar. Sci. Symp. 196, 5–11.

lass, C.W., Wardle, C.S., 1989. Comparison of the reactions of fish to a trawl gear,at high and low light intensities. Fish. Res. 7, 249–266.

lass, C.W., Wardle, C.S., Gosden, S.J., 1993. A behavioural study of the principlesunderlying mesh penetration by fish. In: Wardle, C.S., Hollingworth, C. (Eds.),ICES Mar. Sci. Symp., vol. 196. , pp. 92–97.

odo, O.R., Sunnana, K., 1992. Size selection during trawl sampling of cod and had-dock and its effect on abundance indexes at age. Fish. Res. 13, 293–310.

odo, O.R., Karp, W.A., Totland, A., 1998. Effects of trawl sampling variability onprecision of acoustic abundance estimates of gadoids from the Barents Sea andthe Gulf of Alaska. ICES J. Mar. Sci. 55, 86–94.

andegard, N.O., Michalsen, K., Tjøstheim, D., 2003. Avoidance behaviour in cod(Gadus morhua) to a bottom-trawling vessel. Aquat. Liv. Resour. 16, 265–270.

andegard, N.O., Williams, K., 2008. Automated tracking of fish in trawls usingthe DIDSON (Dual frequency identification sonar). ICES J. Mar. Sci. 65,636–644.

enderson, M.J., Horne, J.K., Towler, R.H., 2008. The influence of beam position andswimming direction on fish target strength. ICES J. Mar. Sci. 65, 237–266.

orne, J.K., 2003. The influence of ontogeny, physiology, and behaviour on the tar-

get strength of walleye pollock (Theragra chalcogramma). ICES J. Mar. Sci. 60,1063–1074.

im, Y.H., Wardle, C.S., 2003. Optomotor response and erratic response: quantitativeanalysis of fish reaction to towed fishing gears. Fish. Res. 60, 455–470.

acLennan, D.N., 1992. Fishing gear selectivity – an overview. Fish. Res. 13, 201–204.

rch 143 (2013) 109– 118

Nakashima, B.S., 1990. Escapement from a diamond-ix midwater trawl during acous-tic surveys for capelin (Mallotus villosus) in the Northwest Atlantic. ICES J. Mar.Sci. 47, 76–82.

Olla, B.L., Davis, M.W., Rose, C., 2000. Differences in orientation and swimming ofwalleye pollock Theragra chalcogramma in a trawl net under light and dark con-ditions: concordance between field and laboratory observations. Fish. Res. 44,261–266.

Ryer, C.H., Olla, B.L., 2000. Avoidance of an approaching net by juvenile walleyepollock Theragra chalcogramma in the laboratory: the influence of light intensity.Fish. Res. 45, 195–199.

Ryer, C.H., 2008. A review of flatfish behavior relative to trawls. Fish. Res. 90,138–146.

Simmonds, J., MacLennan, D., 2005. Fisheries Acoustics: Theory and Practice. Black-well Publishing, Oxford, 437 pp.

Stoner, A.W., Ryer, C.H., Parker, S.J., Auster, P.J., Wakefield, W.W., 2008. Evaluatingthe role of fish behavior in surveys conducted with underwater vehicles. Can. J.Fish. Aquat. Sci. 65, 1230–1243.

Suuronen, P., Lehtonen, E., Wallace, J., 1997. Avoidance and escape behaviour byherring encountering midwater trawls. Fish. Res. 29, 13–24.

Valdemarsen, J.W., 2001. Technological trends in capture fisheries. Ocean Coast.Manage. 44, 635–651.

Wardle, C.S., 1993. Fish behavior and fishing gear. In: Pitcher, T.J. (Ed.), Behaviour ofTeleost Fishes. Chapmann & Hall, London, pp. 609–645.

Wileman, D.A., Ferro, R.S.T., Fonteyne, R., Millar, R.B., 1996. Manual of methods ofmeasuring the selectivity of towed fishing gears, ICES Coop. Res. Rep. No. 215.,Copenhagen.

Williams, K., Rooper, C.N., Towler, R.H., 2010. Use of stereo camera systems forassessment of rockfish abundance in untrawlable areas and for recording pollockbehavior during midwater trawls. Fish. Bull. 108, 352–362.

Williams, K., Punt, A.E., Wilson, C.D., Horne, J.K., 2011. Length-selective retention ofwalleye Pollock, Theragra chalcogramma, by midwater trawls. ICES J. Mar. Sci.

68, 119–129.

Zhang, X.M., Arimoto, T., 1993. Visual physiology of walleye pollock (Theragrachalcogramma) in relation to capture by trawl nets. In: Wardle, C.S., Holling-worth, C. (Eds.), ICES Marine Science Symposia. Fish Behavior in Relation toFishing Operations, vol. 196. , pp. 113–116.