Unaccounted mortality in salmon ﬁsheries:
non-retention in gillnets and effects on estimates of
Matthew R. Baker*
and Daniel E. Schindler
School of Aquatic and Fishery Sciences, University of Washington, Box 355020, Seattle, WA 98195-5020, USA; and
Department of Biology, University of Washington, Seattle, WA 98195-1800, USA
1. Eﬀective and sustainable natural resource management is enhanced when the consequences of
exploitative practices are fully understood and acknowledged. Commercial ﬁsheries devote consid-
erable resources to maximize the harvest of target species and minimize interference with non-target
stocks. Appropriately, bycatch and discard of non-target stocks are recognized as critical economic
and conservation concerns. Few studies, however, have examined non-retention mortality in target
stocks. Non-retention, where ﬁsh are engaged by ﬁshing gear but not landed, is rarely quantiﬁed
and the eﬀects on stocks are unknown. Mortality due to non-retention may have important eﬀects
on the dynamics of exploited populations.
2. We surveyed spawning populations of sockeye salmon Oncorhynchus nerka that had traversed
commercial ﬁsheries in Bristol Bay, Alaska, to estimate the incidence of non-retention in gillnets
and the severity of injuries associated with entanglement. To better understand how gillnet injury
aﬀects spawning success, we tagged and monitored stream-spawning ﬁsh and applied a maximum
likelihood model to mark–recapture data.
3. A substantial portion (11–29%) of spawning sockeye salmon exhibited clear signs of past entan-
glement with commercial gillnets. Survival among such ﬁsh was signiﬁcantly reduced. More than
half of the ﬁsh that reach natal spawning grounds with ﬁshery-related injuries fail to reproduce. This
suggests that estimates of spawning stocks are inﬂated by 5–15% at minimum.
4. Synthesis and applications. Our analyses indicate that non-retention in gillnet ﬁsheries is an
important and under-appreciated consequence of the exploitation of salmon. Stock estimates for
exploited populations that do not account for non-retention mortality overestimate the number of
reproductively viable ﬁsh. Unaccounted mortality and interannual variation in the magnitude
of this mortality may prevent accurate estimates of viable spawners, confound our understanding
of the relationship between stock size and recruitment, impede optimal management and obscure
the ecosystem impacts of migratory stocks in coastal watersheds. Given the magnitude of non-
retention in this ﬁshery, explicit consideration of non-retention mortality may be warranted across
a wide range of exploited populations.
Key-words: delayed mortality, ecosystem engineers, ﬁshery-induced injury, mark–recapture
analysis, natural resource management, Paciﬁc salmon, population dynamics, stock-recruit-
Fishery-related injury in target stocks is rarely quantiﬁed but
may be an important source of mortality in heavily exploited
populations (Alverson 1997; Hall, Alverson & Metuzals 2000).
Both immediate and delayed mortality caused by encounters
with commercial gear is often high (Chopin & Arimoto 1995).
While bycatch, discard and release of non-target species is
often considered (Harrington, Myers & Rosenberg 2005),
damage sustained by target stocks is often ignored. Certain
gear types have low retention rates, enabling a portion of ﬁsh
that encounter gear to disentangle or escape, often leading to
delayed mortality. Such delayed mortality may have important
consequences for ﬁsheries management and the sustainability
*Correspondence author. E-mail: email@example.com
Journal of Applied Ecology 2009, 46, 752–761 doi: 10.1111/j.1365-2664.2009.01673.x
2009 The Authors. Journal compilation 2009 British Ecological Society
of exploited populations, especially where these stocks are
managed for explicit targets and ﬁshing eﬀort relative to stock
size is variable.
Many Paciﬁc salmon gillnet ﬁsheries are managed according
to escapement targets. These are terminal ﬁsheries, which har-
vest salmon on their return migration to freshwater and are
regulated to ensure that suﬃcient numbers of adults evade the
ﬁshery and spawn. While most ﬁsh intercepted by the ﬁshery
are harvested, many disentangle from nets and continue their
migration to natal spawning areas. Many of these ﬁsh sustain
serious injuries. Although counted as part of the aggregate
escapement of viable spawners, ﬁsh damaged in the ﬁshery
experience physical trauma, physiological stress, exhaustion
and increased susceptibility to disease (Ricker 1976; Davis
2002). These ﬁsh may die prior to spawning or have reduced
spawning success. Such losses have a direct bearing on esti-
mates of spawning adults. If a signiﬁcant portion of the enu-
merated escapement fails to spawn, escapement estimates will
not accurately reﬂect the eﬀective population of viable spaw-
ners. This will also confound analyses of the relationship
between spawning stock size and future recruitment to the
population. Where delayed mortality aﬀects a constant per-
centage of escaped stocks, this loss may be implicit in the
stock-recruit function. In most ﬁsheries, however, ﬁshing eﬀort
is variable between years, dependent on the size and timing of
the salmon run. The failure to account for inter-annual vari-
ability in ﬁshery-related injury to spawning stocks may gener-
ate signiﬁcant errors in stock assessment.
Survival for ﬁsh entangled by gillnets is the lowest for all
gear types (ASFEC 1995). With regard to commercial salmon
ﬁsheries, there are no current estimates of gillnet-related injury
in exploited populations nor has there been extensive research
to determine the consequence of these injuries on spawning
success among escaped ﬁsh. Studies of mortality associated
with non-retention in salmonids have largely focused on catch-
and-release sport ﬁsheries (Vincent-Lang, Alexandersdottir &
McBride 1993; Booth et al. 1995) or commercial ﬁsheries using
troll and seine gear (Parker, Black & Larkin 1959; Thomas &
Associates Ltd 1997). The few existing studies that address
non-retention mortality in gillnet ﬁsheries either examine the
issue in an experimental context (Thompson, Hunter & Patten
1971; Thompson & Hunter 1973), document outdated harvest
regimessuchashighseasﬁsheries(Frenchet al. 1970; Ricker
1976), evaluate selective ﬁsheries practices where entangled ﬁsh
are deliberately released and revived (Buchanan et al. 2002;
Vander Haegen et al. 2004) or exclude severely damaged ﬁsh
from analysis (Thompson & Burgner 1952; Hartt 1963).
The Bristol Bay sockeye salmon Oncorhynchus nerka ﬁshery
is managed to achieve constant annual escapement. Our study
was designed to quantify the impact of gillnet injury on
escaped stocks, given the current operation of the ﬁshery. We
estimated the incidence and severity of injuries in ﬁsh returning
to natal streams and the eﬀect of such injuries on pre-spawning
mortality. The ﬁndings suggest that gillnet injuries are com-
mon and, in many cases, inhibit spawning. The eﬀects of such
unaccounted mortality may have important implications for
the designation of optimal escapement targets in exploited
populations, the estimation of spawner-recruit relationships,
the understanding of evolutionary processes driven by ﬁshery
selection and the characterization of the ecosystem eﬀects of
spawning activity in coastal watersheds.
Materials and Methods
ESTIMATION OF THE INCIDENCE OF GILLNET INJURY
Analyses were conducted in the Wood River system in south-west
Alaska (see Map Appendix S1, Supporting Information). The Wood
River system is the primary watershed in the Nushagak district of the
Bristol Bay ﬁshery, supporting one of the world’s largest stocks of
commercially exploited sockeye salmon (Hilborn et al. 2003).
Throughout the Wood River system, sockeye salmon gather within a
100 m range of their natal stream for a period of 1 month following
migration through the ﬁshery, entering spawning streams at matura-
tion (Hendry, Berg & Quinn 1999). This behaviour allowed us to sam-
ple discrete populations immediately prior to their entry to spawning
grounds. At Pick Creek, the site of our mark–recapture study, we
used beach seines to sample 200–500 ﬁsh each year for three consecu-
tive years (2005–2007) to determine the incidence and severity of gill-
net injuries in the pre-spawning population of sockeye salmon that
had successfully transited the ﬁshery. In 2006 and 2007, we expanded
sampling to include 10 populations throughout the Wood River sys-
tem. All sampling occurred within a 2-week period (12–24 July). We
sampled streams in accordance with historical peak spawning date
(University of Washington, unpublished data), immediately prior to
expected stream entry.
CLASSIFICATION OF GILLNET INJURY
All sockeye salmon were examined for ﬁshery-related injury. Clear
net marks, abrasions, contusions or scale loss spanningthe circumfer-
ence of the ﬁsh were considered evidence of gillnet entanglement. Gill-
net marked ﬁsh were grouped according to the severity of the injury:
(i) minor injuries included any evidence of gillnet entanglement,
including net marks and ⁄or scale loss; (ii) moderate injuries included
open wounds and ⁄or skin loss on 5–20% of the surface area of the
ﬁsh; and (iii) severe injuries included large open wounds, fractured
jaws or gill plates, and ⁄or skin loss on >20% of the surface area of
the ﬁsh (Fig. 1).
STREAM RESIDENCE AS AN INDICATOR OF
Our analysis sought to determine whether gillnetinjury resulting from
non-retention in commercial ﬁsheries prevents injured ﬁsh from
spawning. We examined a stream-spawning population of salmon,
using stream residence as a proxy for successful reproduction. Direct
observation of spawning activity was not possible given the spatial
extent of the survey. Egg retention estimates were compromised by
scavenging gulls Larus glaucescens. Therefore, a mark–recapture
study was conducted at Pick Creek (5933¢00¢¢N, 15904¢18¢¢W) to
determine relative diﬀerences in survival and stream residence
between ﬁsh with and without ﬁshery-related injuries. A second-order
stream, Pick Creek originates in a series of spring-fed ponds and ﬂows
2 km before entering Lake Nerka. The stream averages 33 cm deep
and 7Æ8 m wide (Hendry 1998) with high water clarity and relatively
constant discharge (Hendry et al. 1999). Spawning occurs at high
densities throughout the lower 2 km of the stream, with an average of
Unaccounted mortality in salmon ﬁsheries 753
2009 The Authors. Journal compilation 2009 British Ecological Society, Journal of Applied Ecology,46,752–761
8000–10 000 adult spawners (Rogers & Schindler 2008) and a spawn-
ing season of c. 40 days. Due to the presumption that mortality of
severely injured ﬁsh would increase as a function of distance travelled
from the ﬁshery, we sought to sample a population that represented
the average distance from commercial ﬁshery to natal stream for Bris-
tol Bay stocks. Throughout Alaska, sockeye salmon stocks migrate a
mean distance of 103 ± 70 km (n= 32) to an elevation
72 ± 104 m (n= 32). The average Bristol Bay sockeye migrates
94 km to 28 m (Burgner 1991). With a freshwatermigration of 98 km
to an elevation of22 m (Hendry & Berg 1999), the Pick Creekpopula-
tion is representative of the post-ﬁshery migration in Bristol Bay.
Pre-spawning mortality was assumed to occur where ﬁsh failed to
demonstrate suﬃcient stream residence to allow spawning opportuni-
ties. Although sockeye salmon enter spawning areas at reproductive
maturity, several days in-stream precede successful spawning at high
density sites. The reproductive lifespan (stream entry to senescence)
of Pick Creek ﬁsh is 17–20 days (Hendry et al. 1999). All sockeye sal-
mon perish soon after spawning. Typically ﬁsh hold in tight schools
during their ﬁrst days of stream residence and disperse to colonize
spawning habitat within a week of stream entry. Subsequent studies
in Pick Creek indicate females secure territory and spawn towards the
end of the ﬁrst week of in-stream residence (mean days post-
entry = 8Æ05±5Æ56) and defendtheir redd site until senescence, typ-
ically maintaining a consistent presence for a week or more (mean
days post-spawning = 6Æ93 ± 2Æ37; M. Baker, unpublished data).
While movement does not preclude reproductive success in males,
males establish dominance hierarchies on small spatial scales (Quinn,
Adkison, & Ward 1996) and typically demonstrate strong site ﬁdelity
following a period of initial exploration (Foote 1990; Rich et al.
2006). Competitive advantage among males is driven by prior resi-
dence (Foote 1990) and, as males remain reproductively active until
death, extended stream residence confers greater reproductive oppor-
tunities. Given these conditions, we determined any ﬁsh that failed to
demonstrate a minimum stream residence of 3 days failed to spawn
(sensitivityto this threshold value shown in Table 1).
MARK–RECAPTURE SAMPLING AND IN-STREAM
From 15 to 17 July 2005, we sampled and tagged pre-spawning adult
sockeye salmon at the mouth of Pick Creek. Fish were capturedusing
a beach seine.A sample of 100 gillnet-marked ﬁsh was tagged, includ-
ing 50 with minor injuries, 30 with moderate injuries and 20 with
severe injuries (42 males and 58 females). This distribution of severity
of injury reﬂects a representative sampling of the injured population
of ﬁsh at Pick Creek (n= 1863). A sample of 100 uninjured ﬁsh (50
males and 50 females) was also tagged as a control group. Each ﬁsh
was anaesthetized with tricaine methanesulphonate (MS-222; Wes-
tern Chemical, Inc., Ferndale, WA), tagged with an external uniquely
coded 3-cm Petersen disc tag (Floy Tag Co., Seattle, WA), rejuve-
nated in cold water and released (Fig. 2). This method of tagging cor-
responds to a well-established procedure that neither accelerates
mortality nor has lasting eﬀects on ﬁsh behaviour (Quinn & Foote
1994). Presence and severity of ﬁshery-related injury and presence of
fungal infection (Saprolegnia spp.) was assessed at this stage. Photo-
graphs of all injured ﬁsh were reviewed at the conclusion of sampling
to re-evaluate classiﬁcation and ensure standard ranking overtime.
Visual stream surveys of Pick Creek were conducted every other
day throughout the lifespan of all tagged ﬁsh (17 July to 25 August).
Surveys recorded the presence, absence and mortality of tagged ﬁsh.
For analysis,each 2 day period was considered a sampling event.
NONPARAMETRIC ESTIMATOR FOR STREAM
Survival between sampling occasions and stream residence for each
category of gillnet injury were estimated through a nonparametric
estimator using a maximum likelihood approach. This allowed us to
separately estimate survival and account for failures to detect ﬁsh
during stream surveys. A model developed by Lady & Skalski (1998)
was adapted and used to estimate stream residence, following
approaches developed by Cormack (1964) and elaborated by Burn-
ham et al. (1987), whereby conditional survival probabilities are
estimated from one sampling event to the next based on release–
recapture data for marked individuals.
Maximum likelihood estimation (MLE) was used to derive survival
and detection probabilities,using the following function:
where Kis the number of sampling occasions; S
is the probabil-
ity that an individual alive at sampling occasion iwill be alive at
sampling occasion i+1; P
is the probability that an individual
alive at sampling occasion iwill be detected; kis the product of
ﬁnal survival and detection probabilities (S
is the num-
ber of marked individuals detected at sampling occasion i;c
Fig. 1. Photographs of relative severity of gillnet injury. Note that coloration is dark (red) and scales are absorbed in ﬁsh without injury. Fish have
less coloration (red ﬁblush ﬁsilver) and retain scales as severity of injury increases. Morphological traits associated with sexual maturity in
males (dorsal–ventral expansion and extended kype) are less pronounced among ﬁsh with injury. These trends suggest gillnet injury may retard
or inhibit sexual maturation.
754 M. R. Baker & D. E. Schindler
2009 The Authors. Journal compilation 2009 British Ecological Society, Journal of Applied Ecology,46,752–761
the number of marked individuals detected for the last time at
sampling occasion i;v
is the number of marked individuals
known to be alive at sampling occasion i+1;v
is the probabil-
ity that an individual alive at sampling occasion iwill not be
detected again (v
). Ris the number of individuals tagged
at the initial sampling occasion.
The maximum likelihood estimators for survival parameters (and
their variances and covariances) are derived by Burnham et al. (1987)
and reformulated by Lady & Skalski (1998):
Using these survival probabilities, Lady & Skalski (1998) devel-
oped the following estimator of stream residence time (T), operating
on assumptions that: (i) the distribution of deaths between sampling
periods is uniform and (ii) all individuals die prior to the ﬁnal
is the time of the ith sampling occasion relative to the
initial sampling occasion, t
Although technically developed to derive estimates of stream resi-
dence time, this model was applied to data on a beach spawning pop-
ulation (Quinn & Foote 1994), where ﬁsh were marked and
recaptured at the same location. In our study, ﬁsh were tagged at the
stream mouth and surveys were conducted within the main stem of
the stream. We therefore modiﬁed the model to estimate separate
probabilities for: (i) whether or not a ﬁsh entered the stream and (ii)
its survival and detection within the stream.
In our analysis, the ﬁrst period describes the probability of stream
entry or the interval between when a ﬁsh was marked (tagged at the
stream mouth) and its ﬁrst recapture (ﬁrst in-stream observation).
This is deﬁned as the joint probability of survival and stream entry.
The second period describes survival after stream entry, which we
characterize as stream residence. Stream residence was estimated only
for ﬁsh that were observed in the stream and initiated at the ﬁrst
in-stream observation. For integration with the model above, we
arranged the data such that the ﬁrst in-stream observation (stream
entry) for a given individual is considered the ﬁrst sampling occasion
(release) for that individual, regardless of calendar date. All sub-
sequent sampling occasions for that individual arerelative to that ﬁrst
in-stream observation, in eﬀect, modelling stream residence as a ﬁrst-
order approximation by entry date rather than calendar date. Calen-
dar date of spawning had no inﬂuence on the senescence schedule of
ﬁsh (Appendix S2).
Table 1. Estimated stream residence time and pre-spawning mortality according to severity of gillnet injury and presence of Saprolegnia spp.
Pre-spawning mortality Stream residence time (days)
Threshold for successful
3 days; range: 1–9 days)
in stream All ﬁsh
Uninjured 100 6% (2–25%) 10Æ78 11Æ01 14Æ4±8Æ314Æ7±8Æ1
Gillnet injured 100 51% (42–71%) 4Æ54 7Æ82 6Æ1±7Æ810Æ5±7Æ6
Minor 50 16% (8–44%) 8Æ14 8Æ85 10Æ9±8Æ011Æ9±7Æ6
Moderate 30 80% (67–93%) 1Æ37 4Æ11 1Æ7±3Æ75Æ2±4Æ8
Severe 20 95% (90–100%) – – 0Æ4±1Æ40Æ4±1Æ4
Fungal infection (Saprolegnia spp.)
No infection 157 11% (4–35%) 9Æ84 10Æ30 12Æ8±8Æ413Æ4±8Æ1
Fungal infection 43 93% (86–95%) 0Æ53 3Æ83 0Æ7±2Æ55Æ2±5Æ1
Stream residence was calculated for each category of gillnet injury through maximum likelihood estimation methods as a function of sur-
vival probabilities between 2-day sampling periods. Stream residence was also estimated on the basis of individual mark–recapture histo-
ries (±SD). Pre-spawning mortality was assumed in ﬁsh that failed to demonstrate in-stream survival over a minimum of two sampling
periods (3 days). Sensitivity to this threshold stream residence is shown as a range of estimated pre-spawning mortality given threshold
values of 1–9 days.
Fig. 2. Fish with Petersen disc tag (photograph: Michael Webster).
Unaccounted mortality in salmon ﬁsheries 755
2009 The Authors. Journal compilation 2009 British Ecological Society, Journal of Applied Ecology,46,752–761
INCIDENCE AND SEVERITY OF GILLNET INJURY
Fishery-related injuries due to gillnet entanglement were evi-
dent in 11% of ﬁsh sampled at Pick Creek in 2005. Subsequent
sampling in 2006 and 2007 recorded gillnet injury rates of 29%
and 18% respectively. Fungal infection was strongly associ-
ated with the severity of gillnet injury. No infection was
observed in uninjured ﬁsh in 2005. Nearly half (43%) of gill-
net-injured ﬁsh were infected, with rates of 6%, 76% and
100% for ﬁsh with minor, moderate and severe injuries respec-
tively. Similar patterns were noted in 2006 and 2007.
In multi-year sampling at 10 streams, the incidence of gillnet
injury ranged between 4–37% (mean = 18 ± 13Æ1%) in 2006
and 7–29% (mean = 14 ± 6Æ5%) in 2007 (Fig. 3). Among
injured ﬁsh, both sexes exhibited 68% minor injury, 23% mod-
erate injury and 9% severe injury in 2006 and 80% minor
injury, 18% moderate injury and 2% severe injury in 2007.
Fungal infection was associated with severity of gillnet injury
(2 ·3 contingency tables: 2006: v
=91Æ90, P<0Æ001). Infection rates for ﬁsh with
minor, moderate and severe injuries were 9%, 41% and 77%
(2006) and 5%, 33% and 62% (2007) respectively. Although
excluded from our mark–recapture analyses, 2% of sockeye
salmon sampled across 10 streams in both 2006 and 2007 also
exhibited damage from boat propellers.
STREAM ENTRY AND IN-STREAM OBSERVATIONS
Fish must enter and maintain residence in the stream to suc-
cessfully spawn. We tested the independence of severity of
injury and whether or not ﬁsh entered the stream and found
signiﬁcant diﬀerences between groups (v
P<0Æ001). Virtually all (98%) uninjured ﬁsh and most (92%)
ﬁsh with minor injuries entered the stream in contrast to 33%
of ﬁsh with moderate injuries and 10% of ﬁsh with severe inju-
ries. The presence of fungal infection was also a strong indica-
tor of whether ﬁsh entered the stream (v
P<0Æ001). Nearly all (96%) ﬁsh without fungal infection
were observed in-stream in contrast to a minority (14%) with
infection. Whether or not a ﬁsh was observed in-stream was
independent of sex in the control group (v
Diﬀerences were also noted in the date of stream entry. Most
control ﬁsh entered the stream 4 days after tagging. Fish with
minor injuries held oﬀ the mouth more than twice as long.
Both the mean (t
=4Æ21, P<0Æ001) and variance
=0Æ327, P<0Æ001) in stream entry date were distin-
guishable from control ﬁsh. There was no detectable diﬀerence
=0Æ60, P=0Æ549) in mean stream entry date between
control males (mean days to stream entry = 4Æ4±4Æ8) and
control females (3Æ8±3Æ9). Similarly, no detectable diﬀerence
was found (t
=0Æ22, P=0Æ829) between males with
minor injury (mean days to stream entry = 9Æ4±7Æ8) and
females with minor injury (9Æ0±7Æ6). Few ﬁsh with moder-
ate-to-severe injury entered the stream, which prevented accu-
SURVIVAL AND STREAM RESIDENCE TIME
Survival and detection probabilities
Using the maximum likelihood estimates of survival between
sampling occasions, we calculated cumulative in-stream sur-
vival across sampling intervals as a function of entry date
(Fig. 4). In-stream survival declined precipitously for ﬁsh with
moderate to severe gillnet injury. Trends were even more pro-
nounced for comparisons of ﬁsh with and without fungal infec-
tion. On any given sampling occasion, the probability of
detecting a control ﬁsh known to have entered the stream was
estimated at 0Æ718, taken as an average of MLE estimates over
20 sampling events. No diﬀerences were noted between males
(0Æ700) and females (0Æ698). To enable estimation of detection
probabilities independent of survival we also employed the
Manly & Parr (1968) approach and recorded a detection prob-
ability of 0Æ702.
Stream residence by entry date
Maximum likelihood estimates of stream residence time ð^
were calculated as a function of survival probabilities between
2-day sampling periods. Gillnet injury had a direct bearing on
stream residence time. We assumed ﬁsh that were never
observed in the stream, never entered the stream. Among ﬁsh
that entered the stream, uninjured ﬁsh had a mean stream resi-
dence of 11Æ01 (95% CI = 9Æ44–12Æ58) days in contrast to 8Æ85
(95% CI = 5Æ85–11Æ84) days for ﬁsh with minor injury and
4Æ11 (95% CI = 2Æ28–5Æ95) days for ﬁsh with moderate injury.
Too few ﬁsh with severe injury entered the stream to estimate
stream residence. Stream residence was also estimated as a
2005 2006 2007
Percentage gillnet injury at natal streams
Fig. 3. Incidence of gillnet injury averaged across 10 streams in the
Wood River system (2005–2007). Only one site was sampled in 2005
756 M. R. Baker & D. E. Schindler
function of all ﬁsh in each category (regardless of whether ﬁsh
entered the stream) by integrating maximum likelihood esti-
mates of stream residence for ﬁsh observed in the stream with
estimates of zero for those never observed.
Longevity and stream residence by calendar date
Due to standardization by stream entry date, our maximum
likelihood estimates do not provide estimates of survival for
individual ﬁsh in real time nor allow us to characterize the
number of ﬁsh in-stream at any given time. To analyse diﬀer-
ences by calendar date, we estimated longevity for individual
ﬁsh on the basis of the last observation for that individual. We
estimated stream residence as the diﬀerence between the ﬁrst
and last in-stream observations. These methods conﬁrmed the
results achieved through maximum likelihood methods
Longevity (survival in days post-tagging) was greatly
=15Æ03, P<0Æ001) among moderately and
severely injured ﬁsh relative to controlﬁsh.Interestingly,ﬁsh
with minor injuries lived somewhat longer than the uninjured
=1Æ36, P=0Æ179; Fig. 5a), but exhibited reduced
stream residence (t
=2Æ02, P=0Æ046), due to later stream
entry (Fig. 5c). Pair-wise comparisons of stream residence
between categories of gillnet injury conﬁrmed signiﬁcant
diﬀerences between all groups (P<0Æ050) except between
those with moderate and severe injuries (anova,post hoc
Tukey HSD: P=0Æ912). Distinct patterns in longevity
were also noted as a function of fungal infection. Fish
without fungal infections lived more than 15 times longer
=16Æ95, P<0Æ001; Fig. 5b) and, among ﬁsh observed
in-stream, spent more than twice as long in-stream (t
P=0Æ005; Fig. 5d). The longevity of control females
(mean = 19Æ6±7Æ7, n= 50) was signiﬁcantly longer
=2Æ50, P=0Æ014) than control males (mean =
15Æ3±9Æ2, n= 49) and among those that entered the stream,
females demonstrated longer stream residence (t
P=0Æ009). Overall, however, males and females displayed
similar patterns of decline in stream residence as a function of
severity of gillnet injury (Fig. 6).
The average stream residence for Pick Creek ﬁsh not killed
through predation is 10–25 days (Hendry et al. 1999). We
adopted a conservative estimate of pre-spawning mortality,
a minimum of 3 days failed to spawn. Using maximum likeli-
hood estimates, pre-spawning mortality was calculated as a
function of ﬁsh known alive at the second sampling occasion
). According to our model, stream entry is considered the
release date for each individual. Subsequent in-stream observa-
tions are in reference to this standardized release. Thus the per-
centage known alive at the second sampling occasion (v
includes all ﬁsh that survive a minimum of two sampling inter-
vals (3 days) from stream entry. Given this criteria, the major-
ity (51%) of ﬁsh with gillnet injuries were predicted to fail to
spawn in contrast to 6% of control ﬁsh. Nearly all ﬁsh (93%)
with fungal infection at the time of tagging failed to spawn
To account for predation eﬀects, we surveyed all carcasses
to determine the cause ofdeath.BrownbearsUrsus arctos are
a major source of in-stream predation and pre-spawning mor-
tality on sockeye salmon in south-west Alaska (Shuman 1950;
Gard 1971) and are known to preferentially select ﬁsh in better
condition in environments that facilitate foraging (Gende,
Quinn & Willson 2001). We noted higher predation on control
ﬁsh. Among ﬁsh with known fates (n= 76), bear predation
was observed for 31% of uninjured males (n= 17) in contrast
to 17% of gillnet-injured males (n= 7) and in 11% of unin-
jured females (n= 39) in contrast to 8% of gillnet-injured
females (n= 13). While a signiﬁcant portion of pre-spawning
Fig. 4. Plots of in-stream survival according to severity of injury and presence of fungal infection. These estimates standardize by stream entry
date, such that the plots illustrate total in-stream survival regardless of the timing of stream entry. The ﬁrst interval reﬂects the number of ﬁsh
tagged and released. The second interval is the percentage estimated to have entered the stream according to in-stream observations. Subsequent
intervals (S1–S18) are calculated as the product of the number alive at the previous period and our MLE estimate for survival between the previ-
ous and the current period (95% conﬁdence intervals are contained within error bars). Fish not observed in the stream were presumed dead.
Unaccounted mortality in salmon ﬁsheries 757
mortality in our control group may be attributable to bear pre-
dation, it is unlikely that predation alone accounts for the high
rates of pre-spawning mortality in ﬁsh injured through non-
retention in gillnets.
MODEL PERFORMANCE AND ASSUMPTIONS
To analyse model performance, we utilized the release–recap-
ture software SURPH 2Æ1 (Survival Under Proportional Haz-
ards. 2002). To determine whether survival and detection are
the same across treatment groups, we applied TEST 1 devel-
oped by Burnham et al. (1987) and conﬁrmed that survival
parameters diﬀer between ﬁsh with and without evident gillnet
=117Æ73, P=0Æ000). To determine whether sex
impacts survival or detection, we compared males and females
within the control group and found no signiﬁcant diﬀerences
=34Æ23, P=0Æ687). Because our analysis standardized
survival estimates according to stream entry date, we tested
whether detection probabilities hold constant across sampling
occasions to ensure diﬀerent conditions at diﬀerent sampling
occasions would not bias this approach. Speciﬁcally we analy-
sed mark–recapture data by calendar date and compared the
relative performance of: (i) a model assuming unique detection
parameters for each sampling period and (ii) a model assuming
a common detection parameter across sampling periods. On
the basis of the Akaike Information Criterion (Akaike 1973),
vided the best ﬁt to the data (Table 2).
IMPLICATIONS FOR NON-RETENTION AND DELAYED
MORTALITY IN EXPLOITED STOCKS
Our results suggest that disentanglement from gillnets is a reg-
ular occurrence in commercial ﬁsheries in Bristol Bay, Alaska.
As a consequence, ﬁshery-related injuries are common in
spawning stocks of sockeye salmon. Mark–recapture results
Fig. 5. (a–d) Longevity (post-tagging sur-
vival) and stream residence time according to
severity of injury and presence of fungal infe-
ction. These estimates illustrate survival and
stream residence by calendar date. Longevity
estimates include ﬁsh known alive at any
given sampling occasion. Stream residence
estimates include ﬁsh known alive and
known to have entered the stream.
Fig. 6. Box plots of stream residence time according to sex and sever-
ity of gillnet injury.
758 M. R. Baker & D. E. Schindler
demonstrate that survival on the spawning grounds is mark-
edly reduced among gillnet-injured ﬁsh and inversely corre-
lated with the severity of injury. Conservative estimates
suggest that more than half of the ﬁsh that reach natal spawn-
ing grounds after contracting injuries in the ﬁshery fail to
reproduce. The incidence and severity of gillnet injury also
appear to vary between years, probably as a function of ﬁshing
intensity and run size. Due to constant escapement targets, lar-
ger runs will experience higher rates of exploitation. During
smaller runs managers implement more closures, which inad-
vertently improves the relative condition of the escaped stocks.
Diﬀerences in the size of returning ﬁsh may also inﬂuence
retention, given a relatively constant range of mesh sizes used
in the ﬁshery. For these reasons, distinguishing between total
escapement (all ﬁsh that migrate past escapement towers) and
eﬀective escapement (ﬁsh that survive the migration and
spawn) should be considered.
There are also broader ecological implications to decreased
spawning activity in coastal watersheds. Recent attention has
focused on the consequences to habitat and community struc-
ture related to the overexploitation of ecosystem engineers by
commercial ﬁsheries (Coleman & Williams 2002). Habitat
modiﬁcation by spawning salmon alters community organiza-
tion in stream ecosystems and strongly inﬂuences the down-
stream ﬂux of nutrients and resource subsidies (Moore,
Schindler & Scheuerell 2004). Non-retention mortality in
spawning stocks will reduce these eﬀects relative to expecta-
tions based on escapement counts.
POTENTIAL FOR UNDERESTIMATING INCIDENCE OF
Our estimates of the incidence of gillnet injury are almost cer-
tainly lower than actual rates of non-retention in escaped
stocks of spawning salmon. To assess ﬁsh from discrete popu-
lations and minimize the inclusion of strays or migrants, our
sampling was conducted at natal streams immediately prior to
stream entry, roughly 2 weeks after stocks had migrated
through the ﬁshery and were enumerated at escapement count-
ing towers. During this period, many injured ﬁsh probably do
not survive the challenges associated with migration, osmoreg-
ulation, sexual maturation and maintenance metabolism.
Experimental studies of maturing sockeye salmon disentangled
from gillnets found that 80% died within 1 week (Thompson
et al. 1971; Thompson & Hunter 1973). Our estimates of the
incidence of non-retention fail to account for ﬁsh that survive
long enough to migrate past escapement towers but perish
before our sampling occurs at natal streams. It is therefore rea-
sonable to assume our estimate of 11–29% gillnet injury is con-
servative. Actual rates of injury in escaped stocks may be
considerably higher (for further research, see Appendix S3).
PRE-SPAWNING MORTALITY AND PROXIMATE
It is clear that virtually all ﬁsh with moderate to severe gillnet
injury fail to spawn. In the case of ﬁsh with minor injuries, the
delay in stream entry, abbreviated stream residence and the
inhibition of morphological traits associated with sexual matu-
ration (Fig. 1) suggest that even minor injuries retard matura-
tion and reduce reproductive ﬁtness. This delay in maturity
may explain why ﬁsh with minor injuries live longer than unin-
jured ﬁsh despite reduced stream residence. Pre-spawning mor-
tality was highly correlated with and was likely facilitated by
fungal infection, caused by Saprolegnia spp., a facultative para-
site common in freshwater ecosystems. Saprolegnia spp.causes
tissue damage, loss of epithelial integrity and osmoregulatory
failure (Bruno & Wood 1999). It is associated with damaged
epidermal tissue (Hatai & Hoshiai 1994; Pickering 1994), sug-
gesting ﬁsh with gillnet injuries are particularly susceptible to
such infections. Fish with severe infections generally fail to
recover (Pickering & Willoughby 1982). Of 43 ﬁsh with fungal
infection at the time of our tagging, only one successfully
spawned. Many injured ﬁsh without Saprolegnia spp. at tag-
ging presumably developed infections subsequently. Due to
the close correlation between fungal infection and pre-spawn-
ing mortality, Saprolegnia spp. is likely to be the proximate
cause of pre-spawning mortality in gillnet injured ﬁsh.
BROADER APPLICATION OF NON-RETENTION
MORTALITY AND SUSTAINABLE FISHERIES
Commercial gillnet ﬁsheries harvest Paciﬁc salmon on their
return migration and are managed to ensure suﬃcient numbers
of adults spawn and perpetuate discrete stocks. Complicating
management, many salmon enumerated in escapement counts
suﬀer injuries in the ﬁshery and fail to spawn. Estimates of
spawning potential based on such escapement counts fail to
consider this loss. Our study indicates that gillnet injury aﬀects
a minimum of 11–29% of escaped ﬁsh. Roughly half of the
injured ﬁsh fail to spawn. Even minor injuries may lead to
adverse consequences, such as delayed maturation. The
number of viable spawners in escapement counts may be over-
estimated by 5–15%, with repercussions for stock-recruitment
analyses (Fig. 7). Currently, non-retention and delayed mortal-
ity are neither measured nor explicitly incorporated into
stock assessment. The magnitude and inter-annual variation of
non-retention in spawning stocks suggest that this source of
Table 2. SURPH model comparison for unique vs. common
detection parameters appliedacross sampling occasions
Unique detection parameters
for each sampling occasion
39 )852Æ618 1783Æ24
Common detection parameters
for every sampling occasion
21 )786Æ366 1614Æ73
This analysis conﬁrms our assumption that in-stream detection
remained constant throughout the sampling period. It suggests
that standardizing individual capture histories by stream entry
date (rather than calendar date) did not bias survival estimates.
AIC, Akaike Information Criterion.
Unaccounted mortality in salmon ﬁsheries 759
mortality is not adequately considered under current manage-
ment assumptions. This additional unaccounted source of
ﬁshing mortality has not prevented sustainability in the Bristol
Bay ﬁshery due to a precautionary approach to management.
It does, however, suggest that the productivity of these stocks
has been systematically underestimated and indicates a means
to improve eﬃciency if retention can be increased or mortality
due to non-retention reduced. Management agencies across a
wide range of commercial ﬁsheries should carefully consider
the potential for non-retention mortality in target stocks
and instances where such mortality can be estimated and ⁄or
We gratefully acknowledge J. Skalski, J. Lady and R. Hilborn for technical
advice and guidance, G. Holtgrieve, T. Reed and T. Branch for review of the
manuscript, managers and biologists at the Alaska Department of Fish and
Game for data, permitting and consultation, and friends and colleagues in the
University of Washington, Alaska Salmon Program for assistance in the ﬁeld.
Support and funding for this research was provided by the Gordon and Betty
Moore Foundation, the National Science Foundation, the Environmental
Protection Agency STAR fellowship programme, the H. Mason Keeler
Endowment for Excellence, the John G. Peterson Scholarship, the Galen and
Helen MaxﬁeldFisheries Scholarship, and Alaska salmon processors.
Akaike, H. (1973) Information theory and an extension of the maximum likeli-
hood principle. Second International Symposium on Information Theory (eds
B.N. Petrov & F. Csaki), pp. 267–281. Hungarian Academy of Sciences,
Alverson, D. (1997)Global assessment of ﬁsheries bycatchand discards: a sum-
mary overview. Global Trends: Fisheries Management (eds E.K. Pikitch,
D.D. Huppert & M.P. Sissenwine), pp. 115–125. American Fisheries Society
Symposium 20, Seattle, WA, USA.
ASFEC (Ad Hoc Selective Fishery Evaluation Committee). (1995) Selective
ﬁshery evaluation. Paciﬁc SalmonCommission, Vancouver, BC.
Booth, R.K.,Kieﬀer, J.D., Davidson,K., Bielak, A. & Tufts, B.L. (1995)Eﬀects
of late season catch and release angling on anaerobic metabolism, acid-base
status, survival, and gamete viability in wild Atlantic salmon (Salmo salar).
Canadian Journal of Fisheriesand Aquatic Sciences,52, 283–290.
Bruno, D.W. & Wood, B.P. (1999) Saprolegnia and other Oomycetes.Fish Dis-
eases and Disorders:Viral, Bacterial and Fungal Infections, Vol. 3 (eds P.T.K.
Woo & D.W. Bruno), pp 599–659. CABI Publishing,Wallingford.
Buchanan, S., Farrell, A.P., Freser, J., Gallaugher, P.E., Joy, R. & Routledge,
R. (2002) Reducing gillnet-mortality of incidentally caught coho salmon.
North AmericanJournal of Fisheries Management,22(4), 1270–1275.
Burgner, R.L. (1991) Life history of the sockeye salmon (Oncorhynchus nerka).
Paciﬁc Salmon Life Histories (eds C. Groot & L. Margolis), pp. 3–117. Uni-
versity of British Columbia Press, Vancouver, BC.
Burnham, K.P., Anderson, D.R., White, G., Brownie, C. & Pollock, K.H.
(1987) Design and Analysis Methods for Fish Survival Experiments Based on
Release–Recapture. AmericanFisheries Society, Bethesda, MD.
Chopin, F.S. & Arimoto, T. (1995) The condition of ﬁsh escaping from ﬁshing
gears – a review. FisheriesResearch,21, 325–327.
Coleman, F.C. & Williams, S.L. (2002) Overexploiting marine ecosystem engi-
neers: potential consequences for biodiversity. Trends in Ecology & Evolu-
Cormack, R.M. (1964) Estimates of survival from sighting of marked animals.
Davis, M.W. (2002) Key principlesfor understanding ﬁsh bycatchdiscard mor-
tality. CanadianJournal of Fisheries and Aquatic Sciences,59, 1834–1843.
Foote, C.J. (1990) An experimental comparison of male and female spawning
territoriality in a Paciﬁc salmon.Behaviour,115, 283–313.
French, R.R., Craddock, D., Bakkala, R., Dunn, J. & Sutherland, D. (1970)
Ocean distribution and migration of salmon. International North Paciﬁc
Fisheries Commission Annual Report 1968 (also Annual Reports: 1965–1967),
Gard, R. (1971) Brown bear predation on sockeye salmon at Karluk Lake.
Alaska Journalof Wildlife Management,35, 193–204.
Gende, S.M., Quinn, T.P. & Willson, M.F. (2001) Consumption choice by
bears feeding on salmon.Oecologia,127, 372–382.
Hall, M.A., Alverson, D.L. & Metuzals, K.I. (2000) By-catch: problems and
solutions.Marine Pollution Bulletin,41, 204–219.
Harrington, J.M., Myers, R.A. & Rosenberg, A.A. (2005) Wasted ﬁshery
resources: discarded by-catch in the USA. Fish andFisheries,6, 350–361.
Hartt, A.C. (1963)Problems in tagging salmon at sea. International Commission
for the Northwest Atlantic Fisheries, Special Publication,4, 144–155.
Hatai, K. & Hoshiai, G.I. (1994) Pathogenicity of Saprolegnia parasitica coker.
Salmon Saprolegniasis (ed. G.J. Mueller), pp. 87–89. U.S. Department of
Energy, Bonneville Power Administration, Portland.
Hendry, A.P. (1998) Reproductive energetics of Paciﬁc salmon: strategies, tac-
tics, and trade-oﬀs. PhD thesis,University of Washington,Seattle, WA.
Hendry, A.P. & Berg, O.K. (1999) Secondary sexual characters, energy use,
senescence, and the cost of reproduction in sockeye salmon. Canadian Jour-
nal of Zoology,77, 1663–1675.
Hendry, A.P., Berg, O.K. & Quinn, T.P. (1999) Condition dependence and
adaption-by-time: breeding date, life history, and energy allocation within a
population of salmon.Oikos,85, 499–514.
Hilborn, R., Quinn,T.P., Schindler, D.E. & Rogers,D.E. (2003) Biocomplexity
and ﬁsheries sustainability. Proceedings of the National Academy of Sciences
of the United Statesof America,100(11), 6564–6568 .
Lady, J.M. & Skalski,J.R. (1998) Estimators of streamresidence time of Paciﬁc
salmon (Oncorhynchus spp.) basedon release-recapture data.Canadian Jour-
nal of Fisheries andAquatic Sciences,55, 2580–2587.
Manly, B.F.J. & Parr, M.J.(1968) A new method of estimating populationsize,
survivorship and birth rate from capture-recapture data. Transactions of the
Society for British Entomology,18,8 1–89.
Moore, J.W., Schindler, D.E. & Scheuerell, M.D. (2004) Disturbance of fresh-
water habitatsby anadromous salmon in Alaska.Oecologia,139, 298–308.
Parker, R.R., Black, E.C. & Larkin, P.A. (1959) Fatigue and mortality in troll-
caught Paciﬁc salmon (Oncorhynchus spp.). Journal of the Fisheries Research
Board of Canada,16, 429–488.
Pickering, A.D. (1994) Factors which predispose salmonid ﬁsh to Saprolegnia-
sis. Salmon Saprolegniasis (ed. G.J. Mueller), pp. 67–84. U.S. Department of
Energy, Bonneville Power Administration, Portland,OR.
Fig. 7. Plots and Ricker (1954) model ﬁt to spawner-recruit data in
Wood River stocks (1956–2001). Failure to account for non-retention
mortality in escaped stocks of salmonids will inﬂate estimates of
viable spawners and underestimate recruits-per-spawner. We plot the
stock recruitment relationship with spawning stock as enumerated at
escapement towers (d) and discounted ()10%) for non-retention
mortality ( ) Mean recruits per spawner are 2Æ81 (escapement
estimates) in contrast to 3Æ21 (discounted estimates). While a constant
discount rate illustrates a signiﬁcant diﬀerence in estimated produc-
tivity, accounting for interannual variance in non-retention (as a
function of ﬁshing intensity and size of returning ﬁsh) would be more
informative to management and may improve our understanding of
the relationship between spawning stock size and recruitment.
760 M. R. Baker & D. E. Schindler
Pickering, A.D. & Willoughby, L.G. (1982)Saprolegnia infections of salmonid
ﬁsh. Microbial Diseases of Fish (ed. R.J. Roberts), pp. 271–297. Academic
Quinn, T.P. & Foote, C.J. (1994) The eﬀects of body size and sexual dimor-
phism on the reproductive behaviour of sockeye salmon (Oncorhynchus ner-
ka). Animal Behaviour,48, 751–761.
Quinn, T.P., Adkison, M.D. & Ward, M.B. (1996) Behavioral tactics of male
sockeye salmon (Oncorhynchus nerka) under varying operational sex ratios.
Rich, H.B., Carlson, S.M., Chasco, B.C., Briggs, K.C. & Quinn, T.P. (2006)
Movements of male sockeye salmon on spawning grounds: eﬀects of in-
stream residencydensity and body size. Animal Behaviour,71, 971–981.
Ricker, W.E. (1954)Stock and recruitment. Journal of Fisheries ResearchBoard
Ricker, W.E. (1976) Review of the rate of growth and mortality of Paciﬁc sal-
mon in salt water, and noncatch mortality caused by ﬁshing. Journal of the
Fisheries Research Board Canada,33, 1483–1524.
Rogers, L.A. & Schindler, D.E. (2008) Asynchrony in population dynamics of
sockeye salmonin southwest Alaska. Oikos,117(10),1578–1586.
Shuman, R.F. (1950) Bear depredations on red salmon spawning populations
in the Karluk River system, 1947. Journal of Wildlife Management,14,1–9.
SURPH 2.1. Survival Under Proportional Hazards. (2002) Developed by J.
Lady, P. Westhagen & J.R. Skalski. Prepared for the U.S. Department of
Energy. Bonneville Power Administration Division of Fish and Wildlife.
Contract No. DE-B179-90BP02341.
Thomas & AssociatesLtd (1997) North Coast Seine Mortality Study, p. 25.Pre-
pared for Fisheries and Oceans, Prince Rupert, BC.
Thompson, W.F. & Burgner, R.L. (1952) On the eﬀect of net marks. Bristol
Bay MemorandumNo.3, Circular No. 25.
Thompson, R.B. & Hunter, C.J. (1973) Viability of adult sockeye salmon that
disentangle from gillnets. International North Paciﬁc Fisheries Commission,
Annual Report,1971, 107–109.
Thompson, R.B., Hunter, C.J. & Patten, B.G. (1971) Studies of live and dead
salmon that unmeshfrom gillnets. International North Paciﬁc FisheriesCom-
mission, AnnualReport,1969, 108–112.
Vander Haegen,G.E., Ashbrook, C.E., Yi, K.W. & Dixon, J.F. (2004) Survival
of spring Chinook salmon captured and released in a selective commercial
ﬁshery using gillnetand tangle nets. Fisheries Research,68,1 23–133.
Vincent-Lang, D., Alexandersdottir, M. & McBride, D. (1993) Mortality of
coho salmon caught and released using sport tackle in Little Susitna River,
Alaska. FisheriesResearch,15, 339–356.
Received 26 November 2008; accepted 11 May 2009
Handling Editor:Nick Dulvy
Additional supporting information may be found in the online ver-
sion of this article.
Appendix S1. Map of the Nushagak ﬁshing district and Wood River
system, Bristol Bay, Alaska
Appendix S2. Inﬂuence of spawning date on the senescence schedule
Appendix S3. Our results in the context of past research
Please note: Wiley-Blackwell are not responsible for the content or
functionality of any supporting materials supplied by the authors.
Any queries (other than missing material) should be directed to the
corresponding author for the article.
Unaccounted mortality in salmon ﬁsheries 761