FIELD EXPERIMENTS SHOW THAT ACOUSTIC PINGERS REDUCE MARINE MAMMAL

FIELD EXPERIMENTS SHOW THAT ACOUSTICPINGERS REDUCE MARINE MAMMAL BYCATCHIN THE CALIFORNIA DRIFT GILL NET FISHERY

JAY BARLOW

GRANT A. CAMERON1

Southwest Fisheries Science Center,National Marine Fisheries Service,

8604 La Jolla Shores Drive, La Jolla, California 92037, U.S.A.E-mail: [email protected]

ABSTRACT

A controlled experiment was carried out in 1996–1997 to determine whetheracoustic deterrent devices (pingers) reduce marine mammal bycatch in theCalifornia drift gill net fishery for swordfish and sharks. Using Fisher’s exact test,bycatch rates with pingers were significantly less for all cetacean species combined(P , 0.001) and for all pinniped species combined (P ¼ 0.003). For species testedseparately with this test, bycatch reduction was statistically significant for short-beaked common dolphins (P ¼ 0.001) and California sea lions (P ¼ 0.02). Bycatchreduction is not statistically significant for the other species tested separately, butsample sizes and statistical power were low, and bycatch rates were lower inpingered nets for six of the eight other cetacean and pinniped species. A log-linearmodel relating the mean rate of entanglement to the number of pingers deployedwas fit to the data for three groups: short-beaked common dolphins, other cetaceans,and pinnipeds. For a net with 40 pingers, the models predict approximately a 12-fold decrease in entanglement for short-beaked common dolphins, a 4-fold decreasefor other cetaceans, and a 3-fold decrease for pinnipeds. No other variables werefound that could explain this effect. The pinger experiment ended when regulationswere enacted to make pingers mandatory in this fishery.

Key words: bycatch, fishery, pinger, cetacean, dolphin, pinniped, Delphinus delphis,Zalophus californianus, short-beaked common dolphin, California sea lion.

Acoustic deterrent devices (pingers) reduced the bycatch of harbor porpoise(Phocoena phocoena) in bottom-set gill nets during controlled experiments: in theGulf of Maine (Kraus et al. 1997), in the Bay of Fundy (Trippel et al. 1999), alongthe Olympic Peninsula (Gearin et al. 2000), and in the North Sea.2 In all cases

1 Current address: Scripps Institution of Oceanography, UCSD, 9500 Gilman Drive, La Jolla,California 92093, U.S.A.

2 Larsen, F. 1997. Effekten af akustiske alarmer pa bifangst af marsvin i garn. Report number 44-97(unpublished). Available from the Danish Institute for Fisheries Research, Jægersborgvej 64-66,DK-2800 Kgs. Lyngby, Denmark.

265

MARINE MAMMAL SCIENCE, 19(2):265–283 (April 2003)� 2003 by the Society for Marine Mammalogy

a large (approximately 77%–90%) decrease in harbor porpoise mortality wasachieved in short-term experiments. The mechanisms are not well understood(Kraus et al. 1997), but in field trials and in captive studies, the sounds produced bypingers appear to be aversive to harbor porpoises (Kastelein et al. 1995, 2000; Laakeet al.;3 Culik et al. 2001). Another pinger experiment was conducted in 1994 ona drift gill net fishery for swordfish along the U.S. east coast whose bycatchincluded a wide variety of cetaceans. Results of that experiment were somewhatequivocal: in paired tests pingered nets had lower bycatch, but both pingered andunpingered nets in the experiment had higher bycatch than unpingered nets in therest of the fleet.4 Prior to these recent successes, the use of active or passive acousticdeterrents showed little or no effect on net entanglement of Dall’s porpoises(Phocoenoides dalli) (Hatakeyama et al. 1994), and there was little optimism in thescientific community that such approaches would work with other species (Dawson1994, Perrin et al. 1994, Jefferson and Curry 1996). The recent success of pingers inreducing harbor porpoise entanglements in bottom set gill nets prompted a re-evaluation of their potential to reduce mortality of other cetacean species in otherfisheries.5 In this paper we describe an experiment to evaluate the effectiveness ofpingers to reduce cetacean mortality in the drift gill net fishery for swordfish andsharks along the coasts of California and Oregon.

This drift gill net fishery typically operates 37–370 km offshore from southernCalifornia to northern California and, in some years, to Oregon (Fig. 1). Theprimary season for broadbill swordfish (Xiphias gladius) is between 15 August and31 January, but some vessels fish for sharks (primarily common thresher, Alopiusvulpinas, and shortfin mako, Isurus oxyrinchus) between 15 May and 15 August.There were approximately 130 vessels actively fishing in 1995.6 Vessels aretypically 9–23 m in length, and each vessel fishes at night with one multifilamentgill net (stretched mesh size of 43–56 cm) with a maximum length of 1,830 m.Nets are suspended completely below the surface by float lines which werea minimum of 11 m in length. Previous bycatch included a wide assortment ofcetacean species (Julian and Beeson 1998) including delphinids (common dolphins,Pacific white-sided dolphins, northern right whale dolphins, Risso’s dolphins, pilotwhales, bottlenose dolphins, and killer whales), beaked whales (Cuvier’s beakedwhales, Baird’s beaked whales, and Mesoplodon spp.), dwarf sperm whales, spermwhales, and humpback whales (see Table 2 for scientific names). Based on the

3 Laake, J., D. Rugh and L. Baraff. 1998. Observations of harbor porpoise in the vicinity of acousticalarms on a set gill net. U.S. Department of Commerce, NOAA Technical Memorandum NMFS-AFSC-84 (unpublished). 40 pp. Available from the National Marine Mammal Laboratory, 7600 Sand PointWay NE, Seattle, WA 98115, U.S.A.

4 DeAlteris, J., E. Williams and K. Castro. 1994. Results of an experiment using acoustic devices toreduce the incidental take of marine mammals in the swordfish drift gillnet fishery in the NorthwestAtlantic Ocean. Unpublished report. 10 pp. Available from the University of Rhode Island, Kingston,RI 02881, U.S.A.

5 Reeves, R. R., R. J. Hofman, G. K. Silber and D. Wilkinson. 1996. Acoustic deterrence of harmfulmarine mammal-fishery interactions. Proceedings of a workshop held in Seattle, Washington, 20–22March 1996. U.S. Department of Commerce, NOAA Technical Memorandum NMFS-OPR-10(unpublished). 70 pp. Available from the NMFS Office of Protected Resources, 1335 East/WestHighway, Silver Springs, MD 20910, U.S.A.

6 Barlow, J., K. A. Forney, P. S. Hill, R. L. Brownell, Jr., J. V. Carretta, D. P. DeMaster, F. Julian, M.S. Lowry, T. Ragen and R. R. Reeves. 1997. U.S. Pacific Marine Mammal Stock Assessments: 1996.NOAA Technical Memorandum NOAA-TM-NMFS-SWFSC-248. 223 pp.

266 MARINE MAMMAL SCIENCE, VOL. 19, NO. 2, 2003

management scheme used in the United States, the estimated bycatch in 1992–1996 exceeded the PBR (Potential Biological Removal) for some marine mammalspecies and may not be sustainable.6 Concern about these bycatch levels promptedthe formation of the Pacific Offshore Cetacean Take Reduction Team to identifypotential solutions to this problem. The experiment described here was amongtheir first recommendations.

METHODS

Experimental Design

The experiment was designed to maximize statistical power and minimize bias.Each set was assigned randomly as either an experimental set (with pingers) ora control set (without pingers). The experiment was carried out only on those 20%–25% of fishing trips that carried National Marine Fisheries Service bycatchobservers. Prior to a trip, observers were given packets of 10 sealed and numberedenvelopes. Prior to each set, observers would open the envelope with the numbercorresponding to the sequential set number for that trip and would read a cardwhich would indicate whether that set was to be ‘‘experimental’’ or ‘‘control.’’Randomized within each packet of ten envelopes were five cards labeled ‘‘pingers’’and five labeled ‘‘no pingers.’’ If the number of sets per trip exceeded 10, a newpacket of envelopes was used starting with set number 11. To minimize thepotential for experimental manipulation, the selection of experimental and controlsets was made after the skipper had identified a fishing location and immediatelyprior to setting the net. A double-blind experimental design (such as that used byKraus et al. 1997 and Larsen2) was logistically infeasible.

Figure 1. Geographic distribution of sets with pingers (left) and without pingers (right)that were included in analyses.

267BARLOW AND CAMERON: PINGERS REDUCE BYCATCH

Dukane NetMark 10007 pingers were used during this experiment. Thesecommercially produced pingers emit a tonal signal of 300 msec duration every 4 secwith a fundamental frequency of 10–12 kHz and with significant harmonics up to100 kHz. The manufacturer cites a source level of 132 dB (re: 1 lPa @ 1 m), butindependent calibration studies have shown considerable variation in source levelsbetween 120 and 146 dB (X ¼ 138 dB, n ¼ 35).8,9 At a source level of 132 dB,these pingers were estimated to be 15 dB above ambient noise levels at 100 mdistance in the near-bottom environment in the Gulf of Maine (Kraus et al. 1997).Fishermen were instructed to place one pinger at each end of the floatline and at91 m intervals along the floatline and one pinger every 91 m along the leadlineoffset midway between the pingers on the floatline. A typical net of 1,830 m wouldtherefore require 21 pingers along the floatline and 20 pingers along the leadline.The actual number and configuration of pingers varied due to differences in netlength, pinger failures, and other uncontrolled factors (see below).

The experiment started at the beginning of the swordfish season in August 1996and continued until the end of October 1997 when pingers became mandatory inthis fishery. Based on previously measured rates of cetacean entanglement in thisfishery, an a priori power analysis10 indicated that approximately 1,100 sets wouldbe needed (550 with pingers and 550 without) to obtain a 90% probability ofdetecting a 50% decline in overall cetacean mortality (based on a Fisher exact testwith a ¼ 0.10, 1-tailed). A multiyear experiment was anticipated, but with only420 observed sets in 1996, the overall change in cetacean entanglement (a 77%reduction) was statistically significant.11 Based on these preliminary results, pingerswere made mandatory on 28 October 1997 via Federal regulations under theauthority of the U.S. Marine Mammal Protection Act, effectively ending thecontrolled experiment.

Data Collection

Observers on fishing vessels collected data on net specification (includingnumber of pingers used), environmental conditions at the beginning and end of theset, vessel activities during the set, and location at the beginning of the set (Table1). During net retrieval, the observer was stationed in a good position to observethe retrieval and recorded numbers and species of marine mammals (Table 2), seabirds, turtles, and fish caught. Data were checked by observers in the field and whenthey entered their data using a range-checking data entry program. Computer fileswere also checked for outliers, missing fields, and inconsistencies using an edit

7 The use of brand names does not imply endorsement by the National Marine Fisheries Service.8 Unpublished data from K. C. Baldwin, C. Pacheco, and S. D. Kraus, Center for Ocean

Engineering, University of New Hampshire, Durham, NH 03824, U.S.A.9 Unpublished data from D. Norris, Biomon, 718 C West Victoria Street, Santa Barbara, CA 93101,

U.S.A.10 Barlow, J. 1996. Design of an experiment to test the effectiveness of ‘‘pingers’’ to reduce marine

mammal by-catch in the west-coast drift gillnet fishery for swordfish and sharks. Unpublished report. 8pp. Available from the Southwest Fisheries Science Center, 8604 La Jolla Shores Drive, La Jolla, CA92037, U.S.A.

11 Julian, F. 1997. Cetacean mortality in California gill net fisheries: preliminary estimates for 1996.Paper SC/49/SM2 (unpublished). 13 pp. Available from the International Whaling Commission, TheRed House, Station Road, Histon, Cambridge CB4 4NP, United Kingdom.


Table1.

Des

crip

tion

sof

vari

able

suse

din

anal

yses

.V

aria

ble

types

coded

:ca

t¼

cate

gor

ical

,cn

t¼

con

tin

uou

s,or

d¼

ord

inal

,in

v¼

inte

rval

.R

ange

ofco

nti

nuou

sva

riab

leor

cate

gor

ies

ofca

tegor

ical

vari

able

giv

enunder

‘‘Val

ues

.’’M

ean

stat

isti

csco

nsi

stof

arit

hm

etic

mea

nfo

rco

nti

nuou

san

din

terv

alva

riab

les

and

odds

of‘‘1

’’fo

rbin

ary

cate

gor

ical

vari

able

s.‘‘3

’’in

dic

ates

test

sper

form

edon

each

vari

able

.‘‘E

nta

ngle

’’in

dic

ates

Wil

coxo

nte

sts

ofva

riab

lefo

rdif

fere

nce

inen

tangle

men

tra

te.‘

‘Pin

gs’’i

ndic

ates

Wil

coxo

nte

stfo

rdif

fere

nce

sin

vari

able

bet

wee

nse

tsw

ith

and

wit

hou

tpin

ger

s(c

hec

kon

random

izat

ion).

‘‘GLM

’’in

dic

ate

vari

able

incl

uded

inth

eG

ener

aliz

edLin

ear

Mod

elan

alys

es.

Mea

nst

atis

tics

Tes

ts

Var

iab

len

ame

Des

crip

tion

Typ

eV

alu

esA

llse

tsW

ith

pin

ger

sW

ith

out

pin

ger

sE

nta

ng

leP

ing

sG

LM

Controllable

mechanical

variables

dli

gh

td

eck

lig

hts

onat

nig

ht?

(1¼

on)

ord

f0,

1g

0.7

40

.75

0.7

63

33

eng

ine

eng

ine

onat

nig

ht?

(1¼

on)

ord

f0,

1g

0.0

80

.04

0.1

23

33

gen

erg

ener

ator

onat

nig

ht?

(1¼

on)

ord

f0,

1g

0.8

30

.83

0.8

23

33

stic

ks

nu

mb

erof

lig

ht

stic

ks

dep

loye

dcn

t[0

,4

0]

4.9

4.4

5.4

33

stic

ks

pre

sen

tli

gh

tst

ick

sd

eplo

yed

?(1

¼d

eplo

yed

)or

df0

,1g

0.4

20

.38

0.4

63

33

pin

gs

nu

mb

erof

pin

ger

sd

eplo

yed

cnt

[0,

45

]1

73

20

3

pin

gs

pre

sen

tp

ing

ers

dep

loye

d?

(1¼

dep

loye

d)

ord

f0,

1g

0.4

81

.00

.03

3

soak

nu

mb

erof

hou

rsn

etsu

bm

erg

edcn

t[0

,6

2]

12

.51

2.7

12

.33

3

soak

lo/h

i0¼

(soa

k<

12

h)

ord

f0,

1g

0.5

10

.53

0.4

83

33

1¼

(soa

k.

12

h)

son

arso

nar

onat

nig

ht?

(1¼

on)

ord

f0,

1g

0.1

40

.13

0.1

53

33


Table1.

Con

tin

ued

.

Mea

nst

atis

tics

Tes

ts

Var

iab

len

ame

Des

crip

tion

Typ

eV

alu

esA

llse

tsW

ith

pin

ger

sW

ith

out

pin

ger

sE

nta

ngle

Pin

gs

GL

M

Environmentvariables

bcl

dcl

oud

cove

rat

star

tof

set:

inv

f0,

1,...

,9g

3.7

33

.61

3.8

53

lin

ear

scal

e0¼

0%

,8¼

10

0%

,9¼

too

dar

kb

cld

lo/h

i0¼

clea

r(b

cld,

5)

ord

f0,

1g

0.3

60

.35

0.3

63

33

1¼

clou

dy

(bcl

d>

5)

ecld

clou

dco

ver

aten

dof

set:

inv

f0,...

,9g

5.2

5.1

5.3

3li

nea

rsc

ale

0¼

0%

,8¼

10

0%

,9¼

too

dar

kec

ldlo

/hi

0¼

clea

r(e

cld,

5)

ord

f0,

1g

0.3

00

.31

0.3

03

33

1¼

clou

dy

(ecl

d>

5)

bb

eau

lo/h

iB

eau

fort

sea

stat

eat

star

tof

set

ord

f0,

1g

0.4

90

.49

0.4

83

33

0¼

calm

(,3

),1¼

rou

gh

(>3

)eb

eau

lo/h

iB

eau

fort

sea

stat

eat

end

ofse

tor

df0

,1g

0.4

40

.45

0.4

33

33

0¼

calm

(,3

),1¼

rou

gh

(>3

)se

ason

0¼

May

–O

ct,

1¼

Nov

–A

pr

cat

f0,

1g

0.5

60

.55

0.5

63

33

dep

thW

ater

dep

that

tim

eof

net

retr

ieva

l(f

ath

oms)

cnt

[0,

2,7

00

]1

,15

01

,16

71

,13

53

3

dep

thlo

/hi

0¼

shal

low

(,1

,00

0fa

thom

s)or

df0

,1g

0.4

60

.48

0.4

43

33

1¼

dee

p(.

1,0

00

fath

oms)

Net

variables

exte

nd

lo/h

i0¼

(ext

end,

37

ft)

cat

f0,

1g

0.2

50

.27

0.2

43

33

1¼

(ext

end>

37

ft)


Table1.

Con

tinued

.

Mea

nst

atis

tics

Tes

ts

Var

iab

len

ame

Des

crip

tion

Typ

eV

alu

esA

llse

tsW

ith

pin

ger

sW

ith

out

pin

ger

sE

nta

ng

leP

ing

sG

LM

exte

nd

dis

tan

ceb

etw

een

cork

lin

ean

dsu

rfac

efl

oats

(ft)

cnt

[12

,7

8]

38

.23

7.3

38

.23

3

mes

hst

retc

hed

mes

hsi

ze(i

n.)

cnt

[15

,2

2]

20

.32

0.4

20

.33

3m

esh

lo/h

i0¼

(mes

h<

20

)or

df0

,1g

0.5

10

.51

0.5

03

33

1¼

(mes

h.

20

)n

tcol

orco

lor

ofn

etca

tfg

reen

,re

d,

blu

e,b

row

n,

oth

erg

22

2,

24

,4

,6

4,

30

10

9,

13

,2

,3

1,

11

11

3,

11

,2

,3

3,

19

3

net

dp

thn

um

ber

ofm

esh

es,

cork

lin

eto

lead

lin

ecn

t[3

6,

1,0

50

]1

28

.51

28

.21

28

.83

3

net

len

len

gth

ofn

et(f

ath

oms)

cnt

[22

2,

1,0

00

]9

50

.79

49

.39

51

.73

3sl

ack

per

cen

tsl

ack

:ca

lcu

late

dfr

omn

um

ber

mes

hes

han

gin

gan

dh

ang

ing

len

gth

cnt

[0,

50

]4

2.1

42

.24

23

3

Location/seasonvariables

reg

ion

0¼

sou

thof

34

.58N

,ca

tf0

,1g

0.4

50

.46

0.4

43

33

1¼

nor

thof

34

.58N

lat

lati

tud

eat

star

tof

set

cnt

[30

,4

7]

34

.79

34

.87

34

.71

3lo

ng

lon

git

ud

eat

star

tof

set

cnt

[11

7,

12

6]

12

0.6

12

0.9

12

0.4

3


Table1.

Con

tinued

.

Mea

nst

atis

tics

Tes

ts

Var

iab

len

ame

Des

crip

tion

Typ

eV

alu

esA

llse

tsW

ith

pin

ger

sW

ith

out

pin

ger

sE

nta

ng

leP

ing

sG

LM

area

Fiv

efi

shin

gre

gio

ns.

Reg

ion

s1

,3

,4

,5

sep

arat

edb

yla

titu

des

33

.838,

34

.338,

and

42

.008N

.R

egio

n2

com

pos

edof

smal

ldis

join

tar

eas

surr

oundin

gC

han

nel

Isla

nd

s.

cat

f1,...

,5g

31

5,

13

,7

,2

53

,2

1

16

1,

10

,5

,1

28

,1

0

15

4,

3,

2,

12

5,

11

mon

thm

onth

ofse

tca

tf1

,...

,1

2g

69

,0

,0

,0

,0

,0

,0

,5

4,

13

3,

15

7,

10

3,

93

36

,0

,0

,0

,0

,0

,0

,2

6,

66

,9

0,

48

,4

8

33

,0

,0

,0

,0

,0

,0

,2

8,

67

,6

7,

55

,4

5

3


Table2.

Fre

qu

ency

dis

trib

uti

onof

net

enta

ng

lem

ents

by

spec

ies

for

all

sets

,fo

rse

tsw

ith

pin

ger

s,an

dfo

rse

tsw

ith

out

pin

ger

s.

Sets

wit

hp

ing

ers

(n¼

29

5)

Sets

wit

hou

tp

ing

ers

(n¼

31

4)

All

#of

Enta

ngle

men

tsper

set

#of

Enta

ngle

men

tsper

set

Spec

ies

sets

12

3T

otal

12

3T

otal

Com

mon

dol

phin

,sh

ort-

bea

ked

24

33

17

22

1Delphinus

delphis

Com

mon

dol

phin

,lo

ng-b

eaked

10

11

Delphinus

capensis

Nor

ther

nri

gh

tw

hal

ed

olp

hin

81

35

5Lissodelphisborealis

Pac

ific

whit

e-si

ded

dol

phin

41

11

13

Lagenorhynchusobliquidens

Ris

so’s

dol

ph

in1

11

0Grampusgriseus

Dal

l’sp

orp

oise

31

12

2Phocoenoidesdalli

Shor

t-finned

pil

otw

hal

e1

01

1Globicephalamacrorhynchus

Sper

mw

hal

e1

11

0Physetermacrocephalus

‘‘Oth

erce

tace

ans’

’1

94

01

71

01

01

2(e

xclu

din

gsh

ort-

bea

ked

com

mon

dol

ph

in)

All

ceta

cean

s4

37

01

10

27

30

33

Nor

ther

nel

eph

ant

seal

13

33

10

10

Miroungaangustirostris

Cal

ifor

nia

sea

lion

18

44

14

14

Zalophuscalifornianus

All

pin

nip

eds

31

70

07

24

00

24


program. Observers opportunistically recorded data on marine mammal sightingsduring the day as the vessel traveled from one location to another.

Data Selection

Experimental protocols were not followed on every set. Sometimes skippers chosenot to employ pingers in rough seas (18 cases), during the first set of a season or thefirst set with an inexperienced crew (7 cases), when pingers were causing problems (2cases), or for other reasons (20 cases). Occasionally, skippers chose to employ pingerseven when the protocol called for none (because marine mammals were known to bepresent, 5 cases). For analyses presented here, we excluded every set which did notfollow the experimental protocols. To prevent experimental manipulation of results,we also excluded all the sets from trips that were judged to be substantially out ofcompliance with experimental protocols (more than one-third of sets not followingprotocols). Of the 713 sets that were observed during the experiment, 104 wereexcluded, resulting in 609 sets that we included in our analyses.

Statistical Analyses

Descriptions and summary statistics for variables that are likely to affect marinemammal entanglement are given in Table 1. We use abbreviated variable names(Table 1) throughout this report. Some continuous variables and categoricalvariables with multiple states were collapsed to two-state categorical variables forsome analyses; for example, the number of chemical light sticks (‘‘sticks’’) wasincluded as a continuous variable and as the categorical variable ‘‘sticks present.’’

The random distribution of net and set variables in pingered and unpingeredsets was tested using the two-sample Wilcoxon rank sum test (two-tailed). Thereduction in marine mammal bycatch when pingers were present was tested usinga one-tailed Fisher’s exact test using a 2 3 2 contingency table (no entanglementsvs. one or more entanglements per set). Reduction in the number of entanglementsper set was tested with a non-parametric Wilcoxon rank sum test (one-tailed test).The distributions of fish catch were far from Poisson or normal; therefore, thereduction in the number of target and non-target fish caught was tested only withthe Wilcoxon rank sum test (one-tailed).

Multivariate tests of the effect of pingers and other variables on marine mammalentanglement were conducted using a Generalized Linear Modelling (GLM)framework (McCullagh and Nelder 1989). A logarithmic link function was used toapproximate a Poisson error structure:

lnðE½Yi�Þ ¼ b0 þX

Xijbj

where Yi is the number of entanglements for observation i, (for a species or speciesgroup); Xij is the value of predictor variable j for observation i, which may includemain effects and interaction terms; bj is the model coefficient for predictor variablej; and b0 is the coefficient for a constant term. The error structure was actuallyallowed to vary as

varðYiÞ ¼ r2 E½Yi�where the dispersion parameter, r2, can be estimated from the residuals toaccommodate deviations from Poisson expectations (r2 ¼ 1.0). Maximum


likelihood estimates of the coefficients, bj, were computed using iterativelyreweighted least squares using SPLUS software. According to likelihood theory,these parameters are asymptotically normal for known variance, hence, a t-test wasused to determine whether an estimated coefficient is significantly different fromzero.

Three pinger response variables (entanglements of ‘‘short-beaked commondolphin,’’ ‘‘other cetaceans,’’ and ‘‘pinnipeds’’) were modeled as linear functions ofpredictor variables including the number of pingers (‘‘pings’’), the number ofpingers squared (‘‘pings squared’’), and each variable indicated under the ‘‘GLM’’column of Table 1. A ‘‘net volume’’ term, the product of soak time, net length,and net depth, was included by adding soak time, net length, and net depthsimultaneously in a single model. Preliminary multivariate models were built usingan approximate stepwise approach implemented in SPLUS. These models were thenpruned by sequentially removing the least significant variable until all remainingvariables were statistically significant using a test for a reduction in overall deviance(a ¼ 0.05). For Poisson-distributed entanglements, a chi-square test was used formodel selection, and for over-dispersed models, an F-test was used (McCullagh andNelder 1989).

RESULTS

Entanglements

A total of 74 marine mammals (43 cetaceans and 31 pinnipeds) was entangled inthe 609 sets during the experiment (Table 2). Short-beaked common dolphins werethe most common species and accounted for over half of the cetaceanentanglements. Pinniped entanglements included northern elephant seals (Mir-ounga angustirostris) and California sea lions (Zalophus californianus) in roughly equalnumbers. For both cetaceans and pinnipeds, entanglement rates in nets withpingers were approximately one-third the rates in nets without pingers (Table 3).

Most marine mammal entanglements consisted of single individuals; however,three northern right whale dolphins (Lissodelphis borealis) were found entangled ina single net (with 24 pingers). The empirical distributions of the number ofentanglements per set for ‘‘short-beaked common dolphins,’’ ‘‘other cetaceans,’’ and‘‘pinnipeds’’ did not differ significantly from the Poisson distribution (chi-squaregoodness of fit, a ¼ 0.05).

Possible Confounding Factors

There were no significant differences between pingered and unpingered nets forany of the variables tested except for the number of light sticks (‘‘sticks’’ and ‘‘stickspresent’’). Geographic distributions of sets showed no obvious differences betweenpingered and unpingered sets (Fig. 1). Only two variables other than the number ofpingers were related to entanglement rates. Entanglement of short-beaked commondolphins was significantly related to the number of common dolphins sightings onthat trip (‘‘cdsight,’’ Wilcoxon rank sum test, P ¼ 0.0008). Entanglement of ‘‘othercetaceans’’ was not significantly related to any other variables. Entanglement ofpinnipeds was significantly related to the cloud cover at the end of the set (‘‘ecld lo/hi,’’ Wilcoxon rank sum test, P ¼ 0.04). Using a Bonferroni correction for multiple


Table3.

Byc

atch

rate

san

don

e-ta

iled

stat

isti

cal

test

sof

dec

reas

esin

enta

ng

lem

ents

inse

tsw

ith

pin

ger

sco

mp

ared

tose

tsw

ith

out

pin

ger

s.

Byc

atch

rate

s(t

otal

byc

atch

/tot

alse

ts)

Stat

isti

cal

test

resu

lts

(P-v

alu

e)

Spec

ies

Sets

wit

hp

ing

ers

Sets

wit

hou

tp

ing

ers

Wil

coxo

nra

nk

sum

test

Fis

her

’sex

act

test

Com

mon

dol

phin

,sh

ort-

bea

ked

0.0

10

0.0

67

0.0

01

0.0

01

Delphinus

delphis

Com

mon

dol

phin

,lo

ng-b

eaked

0.0

00

0.0

06

0.2

27

0.2

58

Delphinus

capensis

Nor

ther

nri

gh

tw

hal

ed

olp

hin

0.0

10

0.0

16

0.0

70

0.1

24

Lissodelphisborealis

Pac

ific

whit

e-si

ded

dol

phin

0.0

03

0.0

10

0.3

17

0.3

29

Lagenorhynchusobliquidens

Ris

so’s

dol

ph

in0

.00

30

.00

00

.78

90

.48

5Grampusgriseus

Dal

l’sp

orp

oise

0.0

03

0.0

06

0.3

18

0.3

30

Phocoenoidesdalli

Shor

t-finned

pil

otw

hal

e0.0

00

0.0

03

0.2

27

0.2

58

Globicephalamacrorhynchus

Sper

mw

hal

e0

.00

30

.00

00

.22

70

.48

5Physetermacrocephalus

‘‘Oth

erce

tace

ans’

’0

.02

40

.04

10

.08

70

.12

7(e

xclu

din

gsh

ort-

bea

ked

com

mon

dol

ph

in)

All

ceta

cean

s0

.03

40

.11

0,

0.0

01

,0

.00

1N

orth

ern

elep

han

tse

al0

.01

00

0.0

32

0.0

36

0.0

56

Miroungaangustirostris

Cal

ifor

nia

sea

lion

0.0

14

0.0

45

0.0

13

0.0

20

Zalophuscalifornianus

All

pin

nip

eds

0.0

22

0.0

76

0.0

03

0.0

03


testing (a ¼ 0.05/19 ¼ 0.002), only one variable (the number of common dolphinsightings) remained significantly related to entanglements.

Pinger Effects on Entanglements of Short-beaked Common Dolphins

The bycatch of short-beaked common dolphins was significantly lower in netswith pingers (P ¼ 0.001, for both the one-tailed Wilcoxon rank sum test and theFisher exact test, Table 3). The only other variable that appeared to be statisticallysignificant was the number of common dolphin sightings on a trip (P , 0.001).The only variable selected in the stepwise log-linear model was the number ofpingers squared (P ¼ 0.0001, Table 4, Fig. 2).

Pinger Effects on Entanglements of Other Cetaceans

The bycatch of ‘‘other cetaceans’’ (other than short-beaked common dolphins)was not significantly related to pinger use in univariate tests (P ¼ 0.08 and P ¼0.13 using the one-tailed Wilcoxon rank sum test and the Fisher exact test,respectively) (Table 3). However, when the number of pingers used was included ina GLM model (as number of pingers squared), the pinger effect was statisticallysignificant (P ¼ 0.03, Table 4, Fig. 3). The only other significant variable in theGLM model was the Beaufort sea state at the end of the set. Pingers were notsignificantly related to entanglement rates for any of the other species testedseparately, but sample sizes were low in all cases (only one to eight totalentanglements per species). Entanglement rates were lower in pingered nets for fiveout of the seven other cetacean species.

Pinger Effects on Entanglements of Pinnipeds

Pinniped bycatch was also significantly lower in pingered nets (P ¼ 0.003 or0.003, one-tailed Wilcoxon rank sum test or the Fisher exact test, respectively)(Table 3). For individual species tested alone, bycatch reduction was significant forCalifornia sea lions (P ¼ 0.01 or 0.02, respectively) and marginally significant fornorthern elephant seals (P ¼ 0.04 or 0.06, respectively). The number of pingers(‘‘pings’’) was one of four significant variables selected in the stepwise building ofa GLM model for pinniped entanglement (P ¼ 0.007, Table 4, Fig. 4). The othersignificant variables in the GLM model were water ‘‘depth,’’ ‘‘gener,’’ and ‘‘engine.’’In univariate tests the only significant variable in explaining pinnipedentanglement was cloud cover (‘‘ecldlohi’’). This variable is not correlated withpinger use and cannot be used to explain the effect of pingers on entanglement.

Pinger Effects on Catch

There were no significant differences in the catch rates for the three target fishspecies (broadbill swordfish, common thresher shark, and shortfin mako shark)(one-tailed Wilcoxon rank sum test, Table 5). The catch rates of the non-target fishspecies were also not significantly related to pinger use (Table 5).

DISCUSSION

Pingers significantly reduced total cetacean and pinniped entanglement in driftgill nets without significantly affecting swordfish or shark catch. Results also


Table

4.A

nal

ysis

ofdev

iance

table

sfo

rlo

g-l

inea

rfits

tom

arin

em

amm

alen

tangle

men

ts.

Init

ial

mod

els

buil

tusi

ng

appro

xim

ate

step

wis

eap

pro

ach

imp

lem

ente

din

SPL

US,

then

non

-sig

nifi

can

tva

riab

les

del

eted

(seq

uen

tial

lyre

mov

ing

leas

tsi

gn

ifica

nt)

un

til

rem

ain

ing

term

sal

lst

atis

tica

lly

sig

nifi

can

t(a

¼0

.05

).P

-val

ue

issi

gn

ifica

nce

leve

lfr

omei

ther

chi-

squ

are

test

or(f

or‘‘o

ther

ceta

cean

s’’)

app

roxi

mat

eF

-tes

tfo

rch

ang

ein

dev

ian

ce.

Mod

elR

esid

ual

dev

ian

ceD

egre

esof

free

dom

Ch

ang

ein

dev

ian

ceP

Est

imat

edco

effici

ent

Stan

dar

der

ror

ofco

effi

cien

t

Com

mon

dol

phin

(shor

t-bea

ked

)en

tangle

men

tm

odel

(est

imat

eddis

per

sion

¼1

.01

)G

ran

dm

ean

16

0.7

76

08

22

.72

10

.21

71

Pin

gs2

14

2.7

46

07

21

5.0

30

.00

01

20

.00

16

0.0

00

6

‘‘Oth

erce

tace

ans’’

enta

ng

lem

ent

mod

el(e

stim

ated

dis

per

sion

¼1

.26

)G

ran

dm

ean

14

1.1

26

08

23

.65

40

.29

61

Pin

gs2

13

5.4

36

07

25

.69

40

.03

20

.00

09

0.0

00

51

ebea

ulo

hi

13

0.7

46

06

24

.69

00

.05

1.0

23

0.5

55

Pin

nip

eden

tangle

men

tm

odel

(est

imat

eddis

per

sion

¼1

.01

)G

ran

dm

ean

18

7.4

06

07

22

.83

00

.20

61

Dep

th1

82

.46

60

62

4.9

40

.03

20

.00

06

50

.00

026

1P

ing

s1

75

.15

60

52

7.3

10

.00

72

0.0

31

0.0

11

1G

ener

17

0.7

06

04

24

.45

0.0

31

.09

00

.69

41

En

gin

e1

65

.87

60

32

4.8

30

.03

26

.73

01

0.3

5


indicate a greater reduction with a greater number of pingers. These results aresimilar to results of previous experiments that showed a significant reduction inharbor porpoise bycatch when pingers were used on set gill nets (Kraus et al. 1997,Larsen2, Trippel 1999, Gearin et al. 2000). Our experiment is, however, the firstunequivocal demonstration that pingers are correlated with a significant reductionin the bycatch for a delphinid cetacean (short-beaked common dolphin) and fora pinniped (California sea lion). The significant reduction in total cetacean bycatchwas largely driven by the reduction in bycatch of short-beaked common dolphins.Bycatch reduction was not statistically significant for any other cetacean species(although, bycatch was lower for most). An impractically large sample would berequired to find a statistically significant result for rare species, even if theirresponse was the same as for common dolphins.

Because of the potential importance of these results in reducing marine mammalbycatch worldwide, it is important to investigate potential spurious causes of thesepatterns. One potential concern is the lack of a true double-blind control in ourexperimental protocol. We cannot tell whether the observed pinger effect wascaused by the sound produced by the pingers or by the presence of something novelhanging from the net. We believe that the visual enhancement caused by thepresence of the pingers at night is trivial and that the sounds they emit almostcertainly caused the reduction in bycatch; however, our design does not allow us todistinguish between these hypotheses. A more serious concern is the possible director inadvertent manipulation of the results by the observers or the fishermen. Theobservers had no direct role in the design or analysis of the experiment and wouldnot directly benefit by manipulating the results (other than the common human

Figure 2. Predicted bycatch per set of short-beaked common dolphins as function ofnumber of pingers based on GLM. Dotted lines show approximate 95% confidence intervals.


desire for successful outcomes). Fishermen knew that their industry was undergrowing scrutiny and that, if bycatch were not reduced, they might face additionalregulations or even closure; therefore, fishermen had a strong incentive to show thatpingers worked. The ability for fishermen to manipulate results was limited becausethe fishermen had already chosen a location before a set was determined to be‘‘pingered’’ or ‘‘unpingered.’’ Sets were eliminated from analysis when this protocolwas not followed. Once a net is set in a given location, there is little that a fishermancan do to affect marine mammal bycatch. Of the variables that are under a captain’scontrol (‘‘dlight,’’ ‘‘engine,’’ ‘‘gener,’’ ‘‘sticks,’’ ‘‘soak,’’ and ‘‘sonar’’), only ‘‘sticks’’ wassignificantly correlated with pinger use, and none were significantly correlated withcetacean bycatch. The effect of pingers on bycatch was greater than the effects of anyother variables (except number of common dolphin sightings), and it would beimpossible to contrive such a strong pinger effect by subtle experimentalmanipulations. Additional analyses (including classification and regression trees,CART) were conducted to look for other variables that might explain patterns ofentanglements,12 and pingers also emerged as an important explanatory variable inthose studies.

Figure 3. Predicted bycatch per set of ‘‘other cetaceans’’ (other than short-beakedcommon dolphins) as function of number of pingers based on GLM. Dotted lines showapproximate 95% confidence intervals.

12 Cameron, G. 1999. Report on the effect of acoustic warning devices (pingers) on cetacean andpinniped bycatch in the California drift gillnet fishery. Administrative Report LJ-99-08C(unpublished). 71 pp. Available from the Southwest Fisheries Science Center, 8604 La Jolla ShoresDrive, La Jolla, CA 92037, U.S.A.


Table5.

Cat

ch(n

um

ber

offi

sh)

and

one-

tail

edst

atis

tica

lte

sts

for

dec

reas

esin

catc

hra

tes

for

sets

wit

han

dw

ith

out

pin

ger

s.

Tot

alca

tch

Sets

wit

hp

ing

ers

Sets

wit

hou

tp

ing

ers

Wil

coxo

nra

nk

sum

Spec

ies

(#of

fish

)C

atch

Cat

ch/s

etC

atch

Cat

ch/s

etP

-val

ue

Tar

get

Swor

dfi

sh,

bro

adb

ill

1,0

75

51

31

.74

56

21

.79

0.4

6Xiphias

gladius

Shar

k,

Com

mon

thre

sher

46

21

70

0.5

82

92

0.9

30

.24

Alopius

vulpinas

Shar

k,

mak

o8

15

41

81

.42

39

71

.26

0.5

3Isurus

oxyrinchus

Non

-tar

get

Mol

a,co

mm

on2

,16

21

,01

23

.43

1,1

50

3.6

60

.43

Molamola

Op

ah6

07

30

61

.04

30

10

.96

0.3

0Lam

prisguttatus

Shar

k,

big

eye

thre

sher

69

25

0.0

94

40

.14

0.3

2Alopius

supersiliosus

Shar

k,

blu

e2

,11

91

,06

63

.61

1,0

53

3.3

50

.71

Prionaceglauca

Tu

na,

alb

acor

e1

,11

76

96

2.3

64

21

1.3

40

.46

Thunnus

alalunga

Tu

na,

blu

efin

57

22

95

1.0

02

77

0.8

80

.37

Thunnus

thynnus

Tu

na,

skip

jack

58

02

74

0.9

33

06

0.9

70

.32

Katsuwonus

pelamis


Additional work is needed to determine the optimal number and placement ofpingers on drift gill nets. Log-linear models indicate that mortality rate is stilldecreasing with number of pingers within the range of 30–40 pingers (Fig. 2–4);however, there were few data during this experiment within the range of 1–20pingers, so there is considerable uncertainty about the shape of this response curvein that region. The GLM model identified Beaufort sea state, engine noise, andgenerator noise as possible explanatory variables in some analyses. All threevariables are sources of noise that might mask the sounds produced by pingers;however, engine and generator noise could also act to alert animals to the presenceof the net. Water depth is another explanatory variable for pinnipeds; this might beexpected because California sea lions forage only in the shallower, inshore portion ofthe operational range of drift gill net vessels.

The reduction we see in pinniped entanglements is particularly surprising becauseothers have predicted that pinnipeds might be attracted to nets to feed on thecaptured fish (the ‘‘dinner bell’’ effect). However, in an experimental study of theresponse of captive California sea lions to pingers, Anderson (2000) showed that theyinitially responded with a start followed by avoidance (five of six sea lions left thewater). This response helps explain the reduction we noted in sea lion entanglements.

Although pingers appear to reduce bycatch for a large range of marine mammalspecies, we echo the concerns that have been expressed by many other authors thatanimals may habituate to pingers. Given the relatively small number of nets andthe huge area fished, habituation may be less of a concern for the California driftgill net fishery than for intensive, localized set gill net fisheries in the Gulf of Maineand in the North Sea. We believe that pingers are unlikely to reduce the bycatch ofall cetacean species or all pinniped species.

Figure 4. Predicted bycatch per set of pinnipeds as function of number of pingers basedon GLM. Dotted lines show approximate 95% confidence intervals.


ACKNOWLEDGMENTS

We thank Tim Price and the hard-working fishery observers for collecting these data. Wethank the California drift gill net fishermen who have become actively involved in theprocess of improving their fishery. We are grateful to Rand Rasmussen for editing andmaintaining a reliable data base and Peter Perkins, Fred Julian and Cleridy Lennert for freelysharing their statistical expertise. Jim Carretta helped prepare Figure 1. Finally we wouldlike to thank all others who planned, funded, and helped carry out the pinger experimentwithin the California drift gill net monitoring program.

LITERATURE CITED

ANDERSON, R. C. 2000. Responses of captive California sea lions (Zalophus californianus) tonovel stimuli and the effects of motivational state. Master’s thesis, University of SanDiego, San Diego, CA. 192 pp.

CULIK, B. M., S. KOSCHINSKI, N. TREGENZA AND G. M. ELLIS. 2001. Reactions of harborporpoises Phocoena phocoena and herring Clupea harengus to acoustic alarms. MarineEcology Progress Series 211:255–260.

DAWSON, S. M. 1994. The potential for reducing entanglement of dolphins and porpoiseswith acoustic modifications to gillnets. Reports of the International WhalingCommission (Special Isssue 15):573–578.

GEARIN, P. J., M. E. GOSHO, J. L. LAAKE, L. COOKE, R. DELONG AND K. M. HUGHES. 2000.Experimental testing of acoustic alarms (pingers) to reduce bycatch of harbourporpoise, Phocoena phocoena, in the state of Washington. Journal Cetacean Research andManagement 2:1–9.

HATAKEYAMA, Y., K. ISHII, T. AKAMATSU, H. SOEDA, T. SHIMAMURA AND T. KOJIMA. 1994. Areview of studies on attempts to reduce the entanglement of the Dall’s porpoise,Phocoenoides dalli, in the Japanese salmon gillnet fishery. Reports of the InternationalWhaling Commission (Special Isssue 15):549–563.

JEFFERSON, T. A., AND B. E. CURRY. 1996. Acoustic methods of reducing or eliminatingmarine mammal-fishery interactions: do they work? Ocean and Coastal Management31:41–70.

JULIAN, F., AND M. BEESON. 1998. Estimates of marine mammal, turtle, and seabirdmortality for two California gillnet fisheries: 1990–95. Fishery Bulletin, U.S. 96:271–284.

KASTELEIN, R. A., D. DE HAAN, C. STAAL, S. H. NIEUWSTRATEN AND W. C. VERBOOM. 1995.Entanglement of harbour porpoises (Phocoena phocoena) in fishing nets. Pages 91–156 inP. E. Nachtigall, J. Lien, W. W. L. Au and A. J. Read, eds. Harbour porpoises—laboratory studies to reduce bycatch. De Spil Publishers, Woerden, The Netherlands.

KASTELEIN, R. A., H. T. RIPPE, N. VAUGHAN, N. M. SCHOONEMAN, W. C. VERBOOM AND

D. DE HAAN. 2000. The effects of acoustic alarms on the behavior of harbor porpoises(Phocoena phocoena) in a floating pen. Marine Mammal Science 16:46–64.

KRAUS, S., A. J. READ, A. SOLOW, K. BALDWIN, T. SPRADLIN, E. ANDERSON AND

J. WILLIAMSON. 1997. Acoustic alarms reduce porpoise mortality. Nature 388:525.MCCULLAGH, P., AND J. A. NELDER. 1989. Generalized linear models. Chapman and Hall,

New York, NY.PERRIN, W. F., G. P. DONOVAN AND J. BARLOW, EDS. 1994. Gillnets and cetaceans.Reports of

the International Whaling Commission Special Issue 15.TRIPPEL, E. A., M. B. STRONG, J. M. TERHUNE AND J. D. CONWAY. 1999. Mitigation of

harbour porpoise (Phocoena phocoena) by-catch in the gillnet fishery in the lower Bayof Fundy. Canadian Journal of Fisheries and Aquatic Science 56:113–123.

Received: 13 February 2002Accepted: 1 August 2002


Documents

FIELD EXPERIMENTS SHOW THAT ACOUSTIC PINGERS REDUCE MARINE MAMMAL