Content uploaded by Peter John Wright
Author content
All content in this area was uploaded by Peter John Wright
Content may be subject to copyright.
Fishing vs. natural recruitment variation in sandeels
as a cause of seabird breeding failure at
Shetland: a modelling approach
Elvira S. Poloczanska, Robin M. Cook, Graeme D. Ruxton, and Peter J. Wright
Poloczanska, E. S., Cook, R. M., Ruxton, G. D., and Wright, P. J. 9999. Fishing vs. natural
recruitment variation in sandeels as a cause of seabird breeding failure at Shetland:
a modelling approach. eICES Journal of Marine Science, 61: 788e797.
Sandeels represent a major component in the diet of fish, bird, and mammal predators as
well as supporting a large industrial fishery. The availability of young sandeels in coastal
waters around Shetland is generally considered a key factor influencing the breeding
success of many seabird species in the area, but the risk to the populations as a direct
consequence of the fishery is unknown. Low exploitation rates coupled with high natural
mortality rates make assessment of the Shetland sandeel stock problematic and safe
biological limits have not yet been defined. We use stochastic models to evaluate the likely
effect of varying fishing mortality on kittiwake breeding success. The models consider some
main sources of uncertainty about natural processes, such as recruitment and natural
mortality, which may affect the design of robust management strategies. The type of model
tested had a stronger influence on sandeel recruitment than the level of fishing pressure.
Even with low exploitation rates, poor years for seabird breeding were inevitable.
Ó2004 International Council for the Exploration of the Sea. Published by Elsevier Ltd. All rights reserved.
Keywords: kittiwake, sandeel, seabird breeding, Shetland.
Received 19 June 2003; accepted 23 March 2004.
E. S. Poloczanska and G. D. Ruxton: Division of Environmental and Evolutionary Biology,
Institute of Biomedical and Life Sciences, University of Glasgow, Glasgow G12 8QQ,
Scotland, UK. Present address of E. S. Poloczanska: Scottish Association for Marine
Science, Dunstaffnage Marine Laboratory, Oban, Argyll PA37 1QA, Scotland, UK. R. M.
Cook and P. J. Wright: Fisheries Research Services, Marine Laboratory, Aberdeen AB11
9DB, Scotland, UK. Correspondence to E. S. Poloczanska: e-mail: elvira.poloczanska@
sams.ac.uk.
Introduction
Sandeels (Ammodytes sp.) are small, shoaling fish that are
ubiquitous throughout the North Sea, including shallow
coastal waters over sandbanks (Macer, 1966; Reay, 1970).
The industrial fishery for sandeels in the North Sea is the
largest single-species fishery in this area. Sandeels are an
important part of the marine foodweb, forming a major
component in the diets of many fish, marine mammal, and
seabird species (Furness, 1990; Hammond et al., 1994;
Tollit and Thompson, 1996; Wright and Tasker, 1996;
Greenstreet et al., 1998; Doyle and Greenstreet, 1999).
The availability of young sandeels in the coastal waters
around Shetland appears to be a crucial factor influencing
local breeding success of many seabirds such as the kitti-
wake Rissa tridactyla (Danchin, 1992; Hamer et al., 1993),
Arctic skua Stercorarius parasiticus (Phillips et al., 1996)
and Arctic tern Sterna paradisea (Monaghan et al., 1989,
1992; Monaghan, 1992). The dependence of seabird
breeding on sandeels has resulted in conflict between the
fishing and environmental interests, most notably in the
Shetland area in the 1990s.
Around Shetland, juvenile sandeels recruit to the fishery
as 0-group fish (Wright, 1996). The fishing grounds are
close inshore and often adjacent to major breeding seabird
colonies (Monaghan, 1992). During the late 1980s, the
breeding success of the local seabird populations declined
concurrently with a decline in the sandeel population and
landings. At the time, there was considerable controversy
about the relationship between the fishery and the availabil-
ity of sandeels to breeding birds. Environmental groups
argued for a closure of the fishery as a precautionary
measure. Subsequent research indicated that the decrease in
the sandeel population was not caused by fishing but
by environmentally induced fluctuations in recruitment
(Kunzlik, 1989; Wright, 1996).
Compared with the total North Sea sandeel catch of 0.5
million to 1 million tonnes per year, the fishery around
ICES Journal of Marine Science, 61: 788e797 (2004)
doi:10.1016/j.icesjms.2004.03.030
1054-3139/$30.00 Ó2004 International Council for the Exploration of the Sea. Published by Elsevier Ltd. All rights reserved.
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
Shetland is very small, averaging approximately 20 000
tonnes annually before 1991 (ICES, 2000). Following a
peak in 1982 (52 000 tonnes), landings and recruitment
went into decline and the fishery was closed completely at
the end of the fishing season in 1990. After several years of
extremely low recruitment, the production of a large year
class in 1991 induced a recovery of the stock and the fishery
was reopened in 1995. Recruitment to the Shetland stock is
highly variable from year to year (ICES, 2000), but the
mechanisms behind recruitment variability are poorly
understood.
At present, the fishery operates with an annual Total
Allowable Catch (TAC) of 7000 tonnes and a closed season
during June and July to avoid competition with seabirds
during the chick-rearing season. However, landings since
1995 have been extremely low with only 1300 tonnes
landed in 2001. The low exploitation rate coupled with high
natural mortality makes stock assessment problematic and
safe biological limits have not been defined for this stock.
The ecological interactions between sandeels, their preda-
tors, and the fishery remain poorly understood.
Currently, the fishery is not regarded as a threat to breed-
ing seabirds, but an increase in catch may reduce sandeel
availability. It is assumed that high fishing activity would
raise the risk of breeding failure for local seabirds. By
constructing sandeel population models incorporating func-
tional responses for seabird populations, it should be pos-
sible to quantify the risk of breeding failure under different
exploitation rates and to design suitable fishery manage-
ment regimes. However, the processes involved are highly
uncertain and parameter estimates have to be based on as-
sumptions that are difficult to verify. To address this
problem, we developed a range of alternative population
models to reflect the uncertainty about processes such as
recruitment and natural mortality. The results of Monte
Carlo simulations for different models were compared
statistically. The performance of different management
regimes was evaluated in terms of consequences for the
sandeel stock, for the fishery, and for breeding seabirds.
Sandeel population biology
Recruitment to the Shetland stock is variable and is posi-
tively autocorrelated (n ¼24; r1¼0:54; p!0:05) (ICES,
2000;Figure 1). In biological time-series, positive auto-
correlations are considered indicative of environmental
forcing (Steele, 1985; Petchey et al., 1997). Trends in
spawning-stock biomass tend to follow recruitment trends
with a two-year lag (Wright, 1996).
Sandeels are relatively short-lived fish ( generally up to
eight years) compared to most exploited North Sea fish
species and become mature at two years (Macer, 1966;
Gauld and Hutcheon, 1990). Adults are thought not to
undertake extensive spawning migrations and during
spawning attach their eggs to sand grains on the seabed
(Reay, 1970; Warburton, 1982). At Shetland, larvae emerge
during the early spring, when the North Sea circulation is
largely wind-driven. Thus, larval transport by currents is
likely to vary among years (Wright and Bailey, 1996;
Proctor et al., 1998). It has been suggested that the low
recruitment around Shetland during the late 1980s was
caused by a reduction in the passive transport of pre-recruit
sandeels from spawning grounds elsewhere, such as Orkney
(Proctor et al., 1998). The size or frequency of potential
influx of pre-recruits, or of losses owing to export, remains
unquantified. Because fishing mortality is low (Fð1e3Þ!0:3)
compared to natural mortality rates (Mð1Þ¼1:2,
Mð2e7Þ¼0:6; ICES, 2000), fluctuations in stock size are
driven primarily by natural processes.
Seabird breeding biology
Seabird populations can respond to food availability in
various ways, including changes in adult survivorship, chick
growth, fledging success, and colony attendance (Cairns,
1987). Each of these parameters may be sensitive to a
0
50
100
150
200
250
300
1974 1977 1980 1983 1986 1989 1992 1995
Year
SSB (000 tonnes)
0
50
100
150
200
250
300
350
Recruits (billions)
(a)
0
50
100
150
200
250
300
350
0 50 100 150 200 250 300
SSB
Recruitment
(b)
Figure 1. (a) Time-series of spawning-stock biomass (SSB) and
recruitment and (b) stock-recruitment plot (drawn line: Ricker
stock-recruitment model estimated from 10 000 bootstrap simu-
lations; broken lines: 95% confidence interval) for sandeels at
Shetland (data from ICES, 2000).
789A cause of seabird breeding failure at Shetland
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
particular range of prey availability values and integrates
food availability over a specific time scale. Seabirds, being
relatively long-lived, should be buffered against short
sporadic periods of adverse environmental conditions such
as low prey availability or bad weather by varying reproduc-
tive effort (Monaghan et al., 1989, 1992; Phillips et al., 1996).
At Shetland, the entire breeding season (arrival to
departure) for small seabirds lasts from March to September
(Furness, 1990) and overlaps with the sandeel fishery
operating from April to September (except for the closure
during June and July since 1995). Small surface feeding
seabirds, such as kittiwakes (Harris and Wanless, 1990;
Harris and Wanless, 1997; Rindorf et al., 2000) and Arctic
terns (Monaghan et al., 1989, 1992), are considered
particularly sensitive to fluctuations in prey availability
(Furness and Tasker, 2000), and breeding success (number
of chicks fledged per apparently occupied nest) of these may
be regarded as a useful indicator of sandeel biomass over
the breeding season. A relationship with sandeel abundance
has been proposed before for kittiwakes in North Sea areas
(Furness and Tasker, 1997, 2000).
Breeding success of kittiwakes for colonies around
Shetland over 1986e1993 was taken from Dunnet and
Heubeck (1995). Data for additional sites and years were
supplied by Martin Heubeck, University of Aberdeen.
Breeding success was plotted against four biomass measures
of sandeel availability: 0-group, 0+1-group, 1-group, and
0 to 7-group (Figure 2). Biomass was calculated as a multiple
of the number by age group and year estimated by stock
assessment and mean weight-at-age (ICES, 2000). To take
account of the influence of colony size on mean breeding
success, a weighted mean breeding success was determined
for each year and a logistic model was fitted through the data.
Young kittiwakes at two colonies, Kettleness and Eshaness,
have been subject to particularly high predation by great
skuas (Catharacta skua), leading to very few chicks being
fledged (Dunnet and Heubeck, 1995). All data for Kettleness
and 1993e1996 data for Eshaness were excluded from
analysis. Positive relationships between weighted mean
breeding success and sandeel abundance were found for all
four measures of biomass, but the relationship for 0+1-group
was strongest.
Methods
We used an age-structured population model for sandeels
linked to breeding success of kittiwakes being dependent on
juvenile biomass. Effects of the fishery were simulated by
changing fishing mortality on the sandeel stock. Uncertainty
was addressed by examining three different recruitment
models and two patterns of natural mortality.
Population structure
The model time-step is one year. Eight age classes of
sandeel were distinguished (0-group recruits through to 7+
fish). Individuals were transferred from one age class to
the next in January (the last class being a cumulative one).
The initial population (age 1 and older) conformed to the
estimated stock numbers on 1 January 1981 (ICES, 2000;
Table 1). Numbers of 0-group were calculated within
the model using a stock-recruitment relationship. All
sandeels Rage 2 are assumed to be mature and all 0- and
1-group to be immature. Spawning-stock biomass (S)
was calculated as the numbers of the mature part of
the population multiplied by weight-at-age in the stock
(Table 1).
Mortality
Age-specific instantaneous fishing mortality rates (F) are
available from stock assessment (ICES, 2000). F was
decomposed into an age-specific selectivity effect (s
a
) and
a year-specific effect ( f
y
;Cook and Reeves, 1993):
Fay ¼safyð1Þ
Estimates of s
a
and f
y
and the associated covariance matrix
were obtained by fitting a fully separable model (Cook
and Reeves, 1993) over the years, 1981e1990 inclusive
(Table 1). In the simulations, stochastic values for s
a
were
selected from the covariance matrix using a multivariate
correlation random number generator.
Models were tested with three levels of fishing mortality:
F
low
,F
med
, and F
high
based on spawning-stock biomass per
recruit values (Sissenwine and Shepherd, 1987; Jakobsen,
1992; Caddy and Mahon, 1995). F
med
corresponds to
a mortality rate that on average balances recruitment. At
F
high
, recruitment is sufficient to replace losses in only 10%
of the years, which may be regarded as a high-risk exploita-
tion strategy. At F
low
, recruitment exceeds the losses in
90% of the years, which is probably a safe exploitation
strategy.
As for F, age-specific natural mortality rates (M) can also
be decomposed into an age effect (m
a
) and a year effect (k
y
;
Cook, 1993):
May ¼makyð2Þ
The m
a
are effectively the assumed age-specific natural
mortality rates if k ¼1(ICES, 2002a;Table 1). In some
simulations, interannual stochasticity was introduced into
M assuming positive autocorrelation as a worst-case
scenario. A first-order autoregressive model was used to
generate autocorrelated k
y
values:
kt¼f0þf1ðkt1f0Þþ3t;ð3Þ
where f
0
is a constant equal to mean of k
t
(0.10), f
1
is the
first-order autoregressive coefficient (+0.5), k
t
is the
variable k
y
at time t, and 3
a
is a random variable drawn
790 E. S. Poloczanska et al.
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
from the normally distributed residual distribution with
a mean of zero and standard deviation s
3
such that:
s3¼skffiffiffiffiffiffiffiffiffiffiffiffiffi
1f2
1
qð4Þ
where s
k
is standard deviation of the m
a
values (0.276).
Recruitment
Because of uncertainty about the underlying mechanisms,
three stock-recruitment models were applied. The base case
was a Ricker (1954, 1975) model with gamma-distributed
error terms. Secondly, a Ricker model with autocorrelated
error terms was used to simulate environmental forcing on
recruitment variability. The last case considered was a Ricker
model with weak curvature near the origin as a ‘‘worst-case’’
scenario. The stock-recruit data contain few observations in
the range of low stock sizes and therefore the shape of the
relationship near the origin cannot be reliably determined.
However, if the curvature near the origin had been over-
estimated in the base case, the model population would
appear more resilient to high exploitation rates than the
population in the real world would be. Recruitment variabil-
ity caused by emigration and immigration was assumed to be
incorporated in the residuals of the stock-recruit models.
Case 1
The Ricker stock-recruit model is widely used in fishery
science and is simple and robust:
R¼aS ebS
;ð5Þ
where R is recruitment, S is spawning-stock biomass, and
a and b are constants. The model is dome-shaped, implying
that density-dependent processes tend to dominate at high
values of spawning-stock biomass.
We assumed that the variability around the functional
relationship followed a gamma-distribution function instead
0.0
0.5
1.0
1.5
2.0
0.0
0.5
1.0
1.5
2.0
Kittiwake breeding success
0.0
0.5
1.0
1.5
2.0
0.0
0.5
1.0
1.5
2.0
Kittiwake breeding success
Biomass Biomass
0 50 100 150
0 50 100 150
0 50 100 150 200
0 100 200 300
(d)
(a) (b)
(c)
Figure 2. Kittiwake breeding success ( young fledged per apparently occupied nest) at Shetland (1986e1993 from Dunnet and Heubeck,
1995; 1994e1996 from Heubeck, pers. comm.) against measures of sandeel biomass from ICES (2000): (a) 0+1-group with fitted logistic
model; (b) 0-group; (c) 1-group; and (d) 0 to 7-group (+: mean breeding success estimated for individual colonies; ,: weighted mean
breeding success across all colonies per year).
Table 1. Input parameters for sandeel population model (based on
ICES, 2000; P
in
: initial population size in millions; ma: proportion
mature; w
a
: weight-at-age in grams; s
a
,m
a
: ln age-specific selectiv-
ity effect of fishing and natural mortality, respectively).
Age P
in
ma W
a
s
a
s(s
a
)m
a
0e0 0.746 2.4599 2.7371 0.8
1 23 279 0 3.095 2.1144 2.3163 1.2
2 8 328 1 5.409 2.0847 2.3905 0.6
3 3 637 1 8.585 1.8717 2.4813 0.6
4 1 543 1 11.143 2.0775 2.2935 0.6
5 747 1 13.705 2.1885 2.6955 0.6
6 380 1 15.605 2.4615 3.2034 0.6
7+ 517 1 21.254 2.1885 2.6955 0.6
791A cause of seabird breeding failure at Shetland
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
of the more commonly used lognormal distribution, because
the latter may result in bias if the assumption is not exactly
met or sample size is small (MacCall and Ralston, 2002).
The stock-recruit model assuming gamma-distributed
error terms (residuals distributed randomly and indepen-
dently ewhite noise) was fitted by minimizing the sum of
the deviances:
Dev ¼X
n
i¼1
ln Ri
R
ˆi
þRiR
ˆ
R
ˆi
;ð6Þ
where R
i
is observed recruitment (billions) for a specific S
i
,
R
ˆiis predicted recruitment (billions) for the same S
i
, and n
is the number of observations i. The best-fit Ricker model
with 95% confidence intervals estimated from 10 000
bootstrap simulations is shown in Figure 1b.
In the stochastic version, a bootstrap stock-recruit data set
was generated for each simulation of n years. Parameters and
residuals of the fitted Ricker model were recorded and used to
calculate recruitment in each model year. Recruitment
variability was generated by randomly selecting a residual
from the bootstrapped data set with replacement to preserve
the parameter covariance matrix.
Case 2
As before, the parameters of the Ricker model and associated
residuals were determined by 10 000 bootstrap simulations.
Autocorrelation was introduced into the residual sets by a
two-stage process. Firstly, a first-order autoregressive model
was used to generate a random autocorrelated series with
a mean of 0 and standard deviation of 1 for each population
simulation of n years:
yt¼b1yt1;ð7Þ
where y
t
is the observation at time t and b
1
is the auto-
correlation coefficient.
The log-transformed historical recruit time-series for
the Shetland sandeel stock was positively and strongly
(b1¼0:6) autocorrelated (red noise) at a lag of one year.
This value was selected here as representing strong forcing
on the time-series. Secondly, the bootstrap data sets were
ranked in ascending order. Values in each generated auto-
correlated series were assigned a rank position (ascending
order) without re-arranging the series. These were then used
to re-order the bootstrap sets so following the pattern in the
autocorrelated series.
Case 3
Bootstrap output for the model parameters and residuals
were generated as described for Case 1. A subset was then
selected from this output in which parameter a fell in the
lower quartile of the range of estimates. Recruitment
variability was generated by randomly selecting a residual
from the bootstrap data set.
Seabird breeding success
The fitted logistic model (Figure 2a) was used to identify
a range of sandeel biomasses at which kittiwake breeding
success may be poor, not to produce actual predictions of
breeding success. Management of the sandeel stock in the
North Sea applies a decision rule stating that all sandeel
fishing within 50 km of the UK coast is halted if the
breeding success of kittiwakes drops below 0.5 chicks per
apparently occupied nest for three consecutive years (ICES,
2002b). In the logistic model fitted to kittiwake breeding
success against 0+1-group sandeel biomass (Figure 1d),
a breeding success R0.5 was obtained when biomass was
R60 000 tonnes and this value was selected as the
‘‘threshold’’ biomass to indicate failure of kittiwake breed-
ing. In the historical data set, the biomass of 0+1-group
dropped below this value in 7 out of 24 years (1986e1990
inclusive, 1993 and 1995; ICES, 2000).
Output statistics and simulations
The selected summary statistics to compare model output
were mean annual recruitment and SSB to describe the
sandeel population status, and mean catch to describe
fishery performance. In addition, two output statistics were
calculated for potential effects on seabird populations: the
number of years in a simulation with 0+1-group biomass
below the threshold and the percentage of simulations
producing series of three or more consecutive years with
0+1-group biomass below the threshold.
Simulations were run with a burn in a period of 100
years, without stochasticity, to reach equilibrium. Thereaf-
ter, stochasticity was switched on and a further 20 years
were simulated. Output statistics were calculated from the
stochastic period. To determine necessary sample size to
infer stability in results, the running total mean (mean of the
simulation means) for each output statistic was calculated
during 5000 consecutive simulations. Because these means
stabilized rapidly after about 1000 runs, we inferred that
5000 simulations was an adequate number. Simulations
with and without stochasticity in natural mortality were
carried out for each of the three levels of fishing mortality
and for each of the three recruitment models (18 trials).
Results
The mean output statistics from the 18 sets of simulations
are shown in Table 2. The sandeel population was
considered to have ‘‘crashed’’ if recruitment, and thus
SSB, dropped to (near) zero for a number of consecutive
years and showed no indication of increasing again. The
incidence of crashing varied among F levels, being highest
at F
high
(17e20%), and was slightly higher among
recruitment models with low curvature near the origin
(Case 3). Even for F
low
, simulations still crashed in 1e3%
792 E. S. Poloczanska et al.
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
of trials. In calculating output statistics, simulations with
population crashes were excluded.
Taking the output of the model with stochastic recruit-
ment (Case 1), F
med
, and constant M as the baseline, the
percentage change in sandeel summary statistics for the
other models was calculated relative to the baseline values
(Figure 3). Introducing autocorrelation in the recruitment
residuals (Case 2), using stock-recruit models with low
curvature near the origin (Case 3) or autocorrelating M
values all decreased mean recruitment. The largest declines
in mean recruitment, at all levels of F tested, occurred in
Case 3 and with autocorrelated M. Spawning-stock biomass
(SSB) decreased at F
high
and increased at F
low
.AtF
high
,
SSB was more affected by autocorrelated M than by
constant M while the reverse was true for F
low
.
Mean catch increased with increased F by 33e41% in all
models except in Case 3, where the increase was only
8e13%. In Case 3 and with constant M, F
med
actually
resulted in a larger average catch than F
high
. Reducing
fishing pressure decreased mean catch by 38e47% in all
models (Figure 3).
All models showed a high incidence of years with
juvenile sandeel biomass below the threshold for successful
seabird breeding. This happened on average during 3.3e5.9
years out of the 20 years simulated (Table 2). The
recruitment model with autocorrelated residuals (Case 2)
tended to produce runs with a higher incidence of years
below the threshold and this is also the case for
incorporating autocorrelation in M (Figure 4).
For all models, at least 22% of simulations (excluding
crashed simulations) contained at least one period of three
or more consecutive years below the threshold. Values were
highest for models with autocorrelated recruitment resid-
uals at all three levels of F. Introducing autocorrelation in
M also greatly increased the incidence of such events for all
recruitment models at all F.
Discussion
One of the main concerns in managing the Shetland sandeel
fishery is the potential impact on breeding birds. The
simulations suggest that even with the most benign sandeel
population dynamics model and low F, the probability of
poor seabird breeding success would be once every six
years (3.4 years in a 20-year period). Using conventional
criteria used in management elsewhere in the North Sea, the
fishery would have to be closed during a 20-year period in
22% of runs for the most optimistic model. What is of
particular relevance, therefore, is the marginal increase in
the need to close the fishery as F increases. The simulations
indicate that for any of the stock-recruitment and natural
mortality models considered, events requiring fishery
closures increase marginally as F increases from F
low
to
F
high
, but that the probability of the sandeel stock collapsing
rises sharply (Table 2). This result would suggest that the
need to protect the sandeel stock from collapse would occur
before the threat to seabird breeding was itself elevated.
Our results illustrate the importance of understanding the
mechanisms driving recruitment for managing the sandeel
fishery. Models of recruitment processes are based on our
perceptions of underlying mechanisms, and the type of
model chosen strongly influences predictions of population
fluctuations and persistence. Generally, stock-recruit mod-
els incorporate random variation such that each randomized
value is assumed to be independent of other values.
However, this is rarely the case in biological systems as
the effects of environmental stochasticity may persist longer
than one season or one year (Moran, 1953; Grenfell et al.,
1998; Thompson and Ollason, 2001).
Positive autocorrelation noise in a single-species pop-
ulation model has been shown to decrease extinction risk
because a random value at time t is likely to be similar to
the value drawn at t 1(Ripa and Lundberg, 1996).
However, autocorrelating the residuals around the stock-
recruit curve (Case 2) seemed to have little influence on the
Table 2. Summary statistics from stochastic simulations of the
Shetland sandeel stock for 18 different models: Cases 1, 2, and 3:
recruitment (R) models with random gamma-distributed error
terms, autocorrelated error terms, and weak curvature at the origin,
respectively; Ct, A: constant and autocorrelated natural mortality
(M); Med, High, and Low: instantaneous fishing mortality rates
(F); %cr: percentage simulations where the population ‘‘crashed’’
(see text); R, SSB, C: mean values for recruitment (billions),
spawning-stock biomass (!10
3
tonnes), catch (!10
3
tonnes),
respectively (means for 20-year simulations, excluding crashed
populations); y, %3y: mean number of years per 20-year
simulation, and percentage of simulations with R3 consecutive
years, having juvenile sandeel biomass below threshold value,
respectively (bold: baseline).
Model specification Output statistics
R M F %cr R SSB C y %3y
Case 1 C Med 6 73 G18 134 G45 18 G16 3.4 22
High 17 74 G22 115 G50 25 G17 3.9 27
Low 1 71 G14 150 G37 11 G12 3.3 22
A Med 7 68 G19 146 G59 19 G17 4.5 42
High 17 69 G22 124 G61 26 G19 4.6 40
Low 2 66 G17 162 G55 11 G14 4.5 44
Case 2 C Med 6 70 G16 130 G44 17 G15 4.7 54
High 17 71 G19 113 G47 24 G17 4.9 55
Low 1 69 G13 146 G37 11 G13 4.4 53
A Med 7 66 G17 142 G59 18 G16 5.6 67
High 17 66 G20 121 G59 25 G18 5.9 66
Low 3 65 G16 160 G55 11 G13 5.5 67
Case 3 C Med 8 65 G18 125 G47 20 G12 4.0 28
High 20 63 G20 106 G63 20 G13 4.6 34
Low 2 66 G16 141 G49 10 G10 3.5 23
A Med 8 60 G18 136 G63 15 G13 5.0 45
High 20 58 G19 116 G41 21 G14 5.5 46
Low 2 60 G16 154 G59 10 G11 4.6 42
793A cause of seabird breeding failure at Shetland
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
number of simulations considered to have crashed (become
extinct) or on the simulated population parameters compared
to the base case. This does have strong implications for
breeding seabirds. The frequency of years of juvenile sandeel
biomass below the threshold set for ‘‘sufficient’’ seabird
breeding success increased when the residuals around the
stock-recruit curve were positively autocorrelated. As resid-
uals below the curve are likely to occur in sequence, this
produces a series of years of low recruitment to the sandeel
population. This explains why more trials met closure condi-
tions in stochastic simulations of 20 years. Similarly, in-
troducing positive autocorrelation in natural mortality rates
increased the frequency of years with conditions for poor
seabird breeding success.
Evidence suggests that recruitment to the Shetland sandeel
stock may be largely driven by hydroclimatic factors. Wright
(1996) suggested that the poor recruitment during the late
1980s may have been caused by unfavourable oceanic
currents reducing larval advection to Shetland. Analysis of
North Sea data from the Continuous Plankton Recorder
indicated a large anomalous period during these years,
signifying unusual ocean-climate conditions (Edwards et al.,
2002). Furthermore, Arnott and Ruxton (2002) showed
a relationship between recruitment to North Sea sandeels and
the winter index of the North Atlantic Oscillation (NAO).
Fluctuations in large-scale climatic phenomena, such as
NAO, tend to persist over long time scales (decades or more).
Fluctuations in the NAO have been linked with timing of
biological events and/or population dynamics in marine
(Planque and Taylor, 1998; Sims et al., 2001; Attrill and
Power, 2002), freshwater (Straile, 2000; Gerten and Adrian,
2001), and terrestrial (Post and Forchhammer, 2002; Huppop
R SSB C
1
2
3
4
5
6
1
2
3
4
5
6
-30 -20 -10 0 10 20 30
1
2
3
4
5
6
F Low
F High
F Med
1
2
3
4
5
6
1
2
3
4
5
6
-50 -30 -10 10 30 50
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
-30 -20 -10 0 10 20 30
1
2
3
4
5
6
% change from baseline
Figure 3. Percentage change in sandeel summary statistics (R: recruitment; SSB: spawning-stock biomass; and C: catch) from baseline
model for three recruitment models (Case 1: 1/2; Case 2: 3/4; Case 3: 5/6) and two natural mortality models (constant M: 1/3/5;
autocorrelated M: 2/4/6) at three levels of fishing mortality (F
med
,F
high
, and F
low
).
794 E. S. Poloczanska et al.
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
and Huppop, 2003; Post, 2003) systems. The positive
autocorrelation in the Shetland sandeel recruit time-series
and apparent synchrony in recruitment across Shetland
fishing grounds suggests that environmental forcing may be
strong (Poloczanska, 2001).
The set of recruitment models with low curvature near
the origin (Case 3) had the highest rates of population
crashes, as the incidence of high recruitment was reduced
for any given SSB. Determining a realistic stock-recruit
relationship is impossible with a limited data set over
1 3 5 7 9 1113151719
M Stochastic
(b)
(d)
(f)
0
5
10
15
20
FMed
FLow
FHigh
0
5
10
15
20
0
5
10
15
20
135791113151719
M Constant
(a)
(c)
(e)
Frequency %
Years below threshold Years below threshold
Figure 4. Frequency distribution of simulations with 1e20 years of juvenile sandeel biomass below the threshold value for three
recruitment models (Case 1: 1/2; Case 2: 3/4; Case 3: 5/6) and two natural mortality models (constant M: 1/3/5; autocorrelated M: 2/4/6) at
three levels of fishing mortality (F
med
,F
high
, and F
low
), excluding simulations where sandeel populations crashed.
795A cause of seabird breeding failure at Shetland
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
a restricted SSB range. Choosing a wrong model may over-
predict the range of potential recruitment for any given SSB
and might make the simulated population appear more
resilient to exploitation. However, our results show that
allowing for autocorrelation in the recruitment residuals has
much more impact on most summary statistics than chang-
ing the curvature parameter (Case 2;Table 2).
If fluctuations in the sandeel stock at Shetland are driven
largely by extraneous factors, this has important implica-
tions for fishery management. Even if exploitation rates are
kept low, the risk of population crash exists. The risk
increases with exploitation pressure. Of the simulations that
did not crash, the type of model selected had stronger
influence on sandeel recruitment and, as a derivative, seabird
breeding success than fishing pressure. The frequency of
years with sandeel biomass below the threshold increased
with exploitation rate, but the occurrence of a series of three
consecutive low years did not necessarily increase. Thus, the
management system evaluated here would not seem to
require additional measures on controlling fishing pressure
to limit future closures to the fisheries. Poor breeding years
for seabirds are inevitable, even with a greatly reduced
fishery. It is up to seabird ecologists to understand the
implications of breeding failure years on the long-term
dynamics of seabird populations.
Acknowledgements
The study was carried out as part of a PhD funded by the
Fisheries Research Services. Thanks to Martin Heubeck
for supplying data and advice on kittiwake breeding at
Shetland.
References
Arnott, S. A., and Ruxton, G. D. 2002. Sandeel recruitment in the
North Sea: demographic, climatic and trophic effects. Marine
Ecology Progress Series, 238: 199e210.
Attrill, M. J., and Power, M. 2002. Climatic influence on a marine
fish assemblage. Nature, 417: 275e278.
Caddy, J. F., and Mahon, R. 1995. Reference points for fisheries
management. FAO Fisheries Technical Paper 347, FAO, Rome.
Cairns, D. K. 1987. Seabirds as indicators of marine food supplies.
Biological Oceanography, 5: 261e271.
Cook, R. M. 1993. The use of sensitivity analysis to quantify
uncertainties in stock projections. ICES CM 1993/D: 66.
Cook, R. M., and Reeves, S. 1993. Assessment of North Sea
industrial fish stocks with incomplete catch-at-age data. ICES
Journal of Marine Science, 50: 425e434.
Danchin, E. 1992. Food shortage as a factor in the 1988 Kittiwake
Rissa tridactyla breeding failure in Shetland. Ardea, 80: 92e98.
Doyle, C., and Greenstreet, S. 1999. Importance of sandeels to
marine predators. In Sandeel Biology and its implications to
Management, p. 13. Ed. by P. J. Wright and F. M. Kennedy.
Fisheries Research Services Report No 12/99, Marine Labora-
tory, Aberdeen. 37 pp.
Dunnet, G. M., and Heubeck, M. 1995. The monitoring of breeding
seabirds and eiders. Proceedings of the Royal Society of
Edinburgh, 103B: 137e164.
Edwards, M., Beaugrand, G., Reid, P. C., Rowden, A. A., and
Jones, M. B. 2002. Ocean climate anomalies and the ecology of
the North Sea. Marine Ecology Progress Series, 239: 1e10.
Furness, R. W. 1990. A preliminary assessment of the quantities of
Shetland sandeels taken by seabirds, seals, predatory fish and the
industrial fishery in 1981e83. Ibis, 132: 205e217.
Furness, R. W., and Tasker, M. L. 1997. Seabird consumption in
sand lance MSVPA models for the North Sea, and the impact of
industrial fishing on seabird population dynamics. In Forage
Fishes in Marine Ecosystems, pp. 147e170. Alaska Sea Grant
College Program Report No. AK-SG-97-01, University of
Alaska, Fairbanks 1997. 807 pp.
Furness, R. W., and Tasker, M. L. 2000. Seabirdefishery
interactions: quantifying the sensitivity of seabirds to reductions
in sandeel abundance, and identification of key areas for
sensitive seabirds in the North Sea. Marine Ecology Progress
Series, 202: 253e264.
Gauld, J. A., and Hutcheon, J. R. 1990. Spawning and fecundity in
the lesser sandeel, Ammodytes marinus Raitt, in the north-
western North Sea. Journal of Fish Biology, 36: 611e613.
Gerten, D., and Adrian, R. 2001. Differences in the persistency of
the North Atlantic Oscillation signal among lakes. Limnology
and Oceanography, 46: 448e455.
Greenstreet, S. P. R., McMillan, J. A., and Armstrong, E. 1998.
Seasonal variation in the importance of pelagic fish in the diet of
piscivorous fish in the Moray Firth, NE Scotland: a response to
variation in prey abundance? ICES Journal of Marine Science,
55: 121e133.
Grenfell, B. T., Wilson, K., Finkensta
¨dt, B. F., Coulson, T. N.,
Murray, S., Albon, S. D., Pemberton, J. M., Clutton-Brock, T. H.,
and Crawley, M. J. 1998. Noise and determinism in synchronized
sheep dynamics. Nature, 394: 674e677.
Hamer, K. C., Monaghan, P., Uttley, J. D., Walton, P., and
Burns, M. D. 1993. The influence of food supply on the
breeding ecology of kittiwakes Rissa tridactyla in Shetland.
Ibis, 135: 255e263.
Hammond, P. S., Hall, A. J., and Prime, J. H. 1994. The diet of
grey seals in the inner and outer Hebrides. Journal of Applied
Ecology, 31: 737e746.
Harris, M. P., and Wanless, S. 1990. Breeding success of British
kittiwakes Rissa tridactyla in 1986e88: evidence for changing
conditions in the northern North Sea. Journal of Applied
Ecology, 27: 172e187.
Harris, M. P., and Wanless, S. 1997. Breeding success, diet, and
brood neglect in the kittiwake (Rissa tridactyla) over an 11-year
period. ICES Journal of Marine Science, 54: 615e623.
Huppop, O., and Huppop, K. 2003. North Atlantic Oscillation and
timing of spring migration in birds. Proceedings of the Royal
Society of London Series B, 270: 233e240.
ICES. 2000. Report of the working group on the assessment of
demersal stocks in the North Sea and Skagerrak. CM
2000/ACFM: 7. pp. 597.
ICES. 2002a. Report of the working group on the assessment of
demersal stocks in the North Sea and Skagerrak. ICES CM
2002/ACFM: 01.
ICES. 2002b. Report of the ICES Advisory Committee on
Ecosystems 2002. ICES Cooperative Research Report 254.
ICES Copenhagen. pp 138.
Jakobsen, T. 1992. Biological reference points for North-East
Arctic cod and haddock. ICES Journal of Marine Science, 49:
155e166.
Kunzlik, P. 1989. Small fish around Shetland. In Seabirds and
Sandeels: Proceedings of a seminar held in Lerwick, Shetland,
15e16th October 1988, pp. 38e47. Ed. by M. Heubeck.
Shetland Bird Club, Lerwick, Shetland. 81 pp.
MacCall, A. D., and Ralston, S. 2002. Is logarithmic transformation
really the best procedure for estimating stock-recruitment
796 E. S. Poloczanska et al.
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
relationships? North American Journal of Fisheries Manage-
ment, 22: 339e350.
Macer, C. T. 1966. Sand eels (Ammodytidae) in the south-western
North Sea; their biology and fishery. Fishery Investigations
Series II Volume XXIV Number 6, Ministry of Agriculture,
Fisheries and Food.
Monaghan, P. 1992. Seabirds and sandeels: the conflict between
exploitation and conservation in the northern North Sea.
Biodiversity and Conservation, 1: 98e111.
Monaghan, P., Uttley, J. D., and Burns, M. D. 1992. Effect of
changes in food availability on reproductive effort in Arctic terns
Sterna paradisaea. Ardea, 80: 71e80.
Monaghan, P., Uttley, J. D., Burns, M. D., Thaine, C., and
Blackwood, J. 1989. The relationship between food supply,
reproductive effort and breeding success in Arctic terns Sterna
paradisaea. Journal of Animal Ecology, 58: 261e274.
Moran, P. A. P. 1953. The statistical analysis of the Canadian lynx
cycle ii. synchronization and meteorology. Australian Journal of
Zoology, 1: 291e298.
Petchey, O. L., Gonzalez, A., and Wilson, A. 1997. Effects on
population persistence: the interaction between environmental
noise colour, intraspecific competition and space. Proceedings of
the Royal Society, London, B264: 1841e1847.
Phillips, R. A., Caldow, R. W. G., and Furness, R. W. 1996. The
influence of food availability on the breeding effort and
reproductive success of Arctic skuas Stercorarius parasiticus.
Ibis, 138: 410e419.
Planque, B., and Taylor, A. H. 1998. Long-term changes in
zooplankton and the climate of the North Atlantic. ICES Journal
of Marine Science, 55: 644e654.
Poloczanska, E. 2001. Modelling the Shetland sandeel stock. PhD
thesis, University of Glasgow.
Post, E. 2003. Large-scale climate synchronizes the timing of
flowering by multiple species. Ecology, 84: 277e281.
Post, E., and Forchhammer, M. C. 2002. Synchronization of animal
population dynamics by large-scale climate. Nature, 420:
168e171.
Proctor, R., Wright, P. J., and Everitt, A. 1998. Modelling the
transport of larval sandeels on the north-west European shelf.
Fisheries Oceanography, 7: 347e354.
Reay, P. J. 1970. Synopsis of biological data on North Atlantic
sand eels of the genus Ammodytes. FAO Fisheries Synopsis
No. 82.
Ricker, W. E. 1954. Stock and recruitment. Journal of the Fisheries
Resource Board of Canada, 11: 559e623.
Ricker, W. E. 1975. Computation and interpretation of biological
statistics of fish populations. Bulletin of the Fisheries Resource
Board of Canada, 191:382 pp.
Rindorf, A., Wanless, S., and Harris, M. P. 2000. Effects of changes
in sandeel availability on the reproductive output of seabirds.
Marine Ecology Progress Series, 202: 241e252.
Ripa, J., and Lundberg, P. 1996. Noise colour and the risk of
population extinctions. Proceedings of the Royal Society London
Series B, 263: 1751e1753.
Sims, D. W., Genner, M. J., Southward, A. J., and Hawkins, S. J.
2001. Timing of squid migration reflects North Atlantic
variability. Proceedings of the Royal Society of London Series
B, 268: 2607e2611.
Sissenwine, M. P., and Shepherd, J. G. 1987. An alternate
perspective on recruitment overfishing and biological reference
points. Canadian Journal of Fisheries and Aquatic Science, 44:
913e918.
Steele, J. H. 1985. A comparison of terrestrial and marine
ecological systems. Nature, 313: 355e358.
Straile, D. 2000. Meteorological forcing of plankton dynamics in
a large and deep continental European lake. Oecologia, 122:
44e50.
Thompson, P. M., and Ollason, J. C. 2001. Lagged effects of ocean
climate on fulmar population dynamics. Nature, 413: 417e420.
Tollit, D. J., and Thompson, P. M. 1996. Seasonal and between-
year variations in the diet of harbour seals in the Moray Firth,
Scotland. Canadian Journal of Zoology, 74: 1110e1121.
Warburton, K. 1982. Sandeels ethe elusive species. Scottish
Fisheries Bulletin, 47: 22e27.
Wright, P. J. 1996. Is there a conflict between sandeel fisheries and
seabirds? A case study at Shetland. In Aquatic Predators
and their Prey, pp. 154e165. Ed. by S. P. R. Greenstreet, and
M. L. Tasker. Fishing News Books, Oxford. 191 pp.
Wright, P. J., and Bailey, M. C. 1996. Timing of hatching in
Ammodytes marinus from Shetland waters and its significance to
early growth and survivorship. Marine Biology, 126: 143e152.
Wright, P. J., and Tasker, M. L. 1996. Analysis of fish consumption
by seabirds by age class of prey fish. In Hunt, G.L., and Furness,
R.W. (Eds.), Seabird/Fish Interactions, with Particular Reference
to Seabirds in the North Sea. ICES Cooperative Research
Report, 216: 2e5.
797A cause of seabird breeding failure at Shetland
at Fisheries Research Services on June 9, 2010 http://icesjms.oxfordjournals.orgDownloaded from
