Linking fisheries, trophic interactions and climate: Threshold dynamics drive herring Clupea harengus growth in the central Baltic Sea

Abstract

How multiple stressors influence fish stock dynamics is a crucial question in ecology in general and in fisheries science in particular. Using time-series covering a 30 yr period, we show that the body growth of the central Baltic Sea herring Clupea harengus, both in terms of condition and weight-at-age (WAA), has shifted from being mainly driven by hydro-climatic forces to an interspecific density-dependent control. The shift in the mechanisms of regulation of herring growth is triggered by the abundance of sprat, the main food competitor for herring. Abundances of sprat above the threshold of ̃18 × 10 10 ind. decouple herring growth from hydro-climatic factors (i.e. salinity), and become the main driver of herring growth variations. At high sprat densities, herring growth is considerably lower than at low sprat levels, regardless of the salinity conditions, indicative of hysteresis in the response of herring growth to salinity changes. The threshold dynamic accurately explains the changes in herring growth during the past 3 decades and in turn contributes to elucidate the parallel drastic drop in herring spawning stock biomass. Studying the interplay between different stressors can provide fundamental information for the management of exploited resources. The management of the central Baltic herring stock should be adaptive and take into consideration the dual response of herring growth to hydro-climatic forces and food-web structure for a sound ecosystem approach to fisheries.

Full text

MARINE ECOLOGY PROGRESS SERIES
Mar Ecol Prog Ser
Vol. 413: 241–252, 2010
doi: 10.3354/meps08592 Published August 26
INTRODUCTION
Disentangling the combined effects of food-web
structure and trophic interactions, climate changes and
anthropogenic pressures (e.g. fishing) on population
dynamics is of central importance for the understand-
ing of ecosystem functioning, and thus the manage-
ment of exploited resources. The quantification of the
biotic and climatic effects on ecosystem dynamics,
however, is often complicated by nonlinear relation-
ships or discontinuous responses of populations and
communities to the driving forces (Scheffer & Carpen-
ter 2003).
Herring Clupea harengus is a key species in the
Baltic Sea ecosystem, as important prey for cod Gadus
morhua (Bagge 1989) and a plankton-zoobenthos
feeder (Casini et al. 2004). Moreover, herring is one of
the most important commercial species in the area,
heavily exploited by the 9 countries surrounding the
Baltic Sea (EC 2008).
Since the late 1970s, the central Baltic herring stock
has undergone a marked decline in stock size. How-
© Inter-Research 2010 · www.int-res.com*Email: michele.casini@fiskeriverket.se
Linking fisheries, trophic interactions and climate:
threshold dynamics drive herring Clupea harengus
growth in the central Baltic Sea
Michele Casini1,*, Valerio Bartolino1, Juan Carlos Molinero2, Georgs Kornilovs3
1Swedish Board of Fisheries, Institute of Marine Research, PO Box 4, 453 21 Lysekil, Sweden
2The Leibniz Institute of Marine Sciences (IFM-GEOMAR), West Shore Campus, Duesternbrooker Weg 20,
24105 Kiel, Germany
3Latvian Fish Resources Agency, Daugavgrivas Str. 8, 1048 Riga, Latvia
ABSTRACT: How multiple stressors influence fish stock dynamics is a crucial question in ecology in
general and in fisheries science in particular. Using time-series covering a 30 yr period, we show that
the body growth of the central Baltic Sea herring Clupea harengus, both in terms of condition and
weight-at-age (WAA), has shifted from being mainly driven by hydro-climatic forces to an inter-
specific density-dependent control. The shift in the mechanisms of regulation of herring growth is
triggered by the abundance of sprat, the main food competitor for herring. Abundances of sprat
above the threshold of ~18 ×1010 ind. decouple herring growth from hydro-climatic factors (i.e. salin-
ity), and become the main driver of herring growth variations. At high sprat densities, herring growth
is considerably lower than at low sprat levels, regardless of the salinity conditions, indicative of hys-
teresis in the response of herring growth to salinity changes. The threshold dynamic accurately
explains the changes in herring growth during the past 3 decades and in turn contributes to elucidate
the parallel drastic drop in herring spawning stock biomass. Studying the interplay between differ-
ent stressors can provide fundamental information for the management of exploited resources. The
management of the central Baltic herring stock should be adaptive and take into consideration the
dual response of herring growth to hydro-climatic forces and food-web structure for a sound eco-
system approach to fisheries.
KEY WORDS: Condition · Weight-at-age · WAA · Ecological threshold · Inter-specific density-
dependence · Hydro-climate · Herring · Sprat · Ecosystem-based fisheries management · Alternative
dynamics
Resale or republication not permitted without written consent of the publisher
Contribution to the Theme Section ‘Threshold dynamics in marine coastal systems’
OPENPEN
ACCESSCCESS

Mar Ecol Prog Ser 413: 241–252, 2010
ever, the drop in biomass has been more noticeable than
the decrease in abundance (Fig. S1 in the Supplement
at
www.int-res.com/articles/suppl/m413p241_supp.pdf
),
which is partly explained by a parallel drastic decline
in mean body weight (ICES 2009a). The decrease in
individual herring weight was formerly hypothesised
as a major effect of either climate-related hydro-
graphic changes (i.e. decrease in salinity, Rönnkönen
et al. 2004) or increased competition with the enlarged
sprat Sprattus sprattus population (Cardinale & Arrhe-
nius 2000), both negatively affecting the main plank-
tonic food for herring.
In the Baltic Sea it has been shown that both top-
down and bottom-up controls (including hydro-
climatic forces) can act simultaneously on different
trophic levels (e.g. Alheit et al. 2005, Casini et al.
2008), but also that their relative strength can vary
over time as a consequence of changes in the food-web
structure (Casini et al. 2009). Specifically, zooplankton
dynamics in the offshore areas is mainly regulated by
hydrological forcing when the stock size of sprat is
below a specific abundance threshold, whereas it is
driven by sprat predation when the abundance of this
major planktivore exceeds such a threshold (Casini et
al. 2009). Similar dual ways of ecosystem functioning
and trophic control have been also found for different
species in other aquatic (Litzow & Ciannelli 2007, Stige
et al. 2009) as well as terrestrial (e.g. Wilmers et al.
2006) systems.
In this study we extend the concept of ‘threshold
dynamics’ in the Baltic Sea from Casini et al. (2009) a
step further, to investigate the body growth of the cen-
tral Baltic herring. We show that the 2 apparently con-
trasting hypotheses of herring growth regulation
(hydro-climatic factors and inter-specific density-
dependence) are not mutually exclusive, providing
evidence that the strength of the 2 forces on herring
growth varies depending on food-web structure and
interaction strength between competing species.
MATERIALS AND METHODS
Time-series of Baltic sprat stock abundance was
retrieved from stock assessment official reports (ICES
2009a). Recently compiled herring biological data
(individual total length, total weight and age) covering
the period 1978 to 2008 were collected during the
autumn international acoustic survey (ICES 2009b) by
Latvia and Sweden in offshore areas of the central
Baltic Sea (the Gotland Basin, ICES Subdivision 28,
Fig. 1).
For the estimation of condition we used the double-
logarithmic length-weight regression, in line with pre-
vious studies on clupeid growth in the Baltic Sea (e.g.
Möllmann et al. 2003, Casini et al. 2006). Condition
was estimated from the year-specific ln(L)-ln(W) re-
gression as the weight in grams at the length of
180 mm, which corresponded to the mean length of
herring in our dataset.
Fish weight-at-age (WAA) is theoretically a result of
forces acting over subsequent years from hatching to
the date of collection. However, the trends in WAA
were very similar among all age-classes, with simulta-
neous inter-annual variations (Fig. 2). This strongly
suggests that changes in external conditions affect
similarly the whole population at the same time.
Therefore, for the purpose of our analyses we aver-
aged the WAA of age 2+ fish, which represent the
part of the population that will spawn in the following
spring (i.e. at age 3+). At age 3+, central Baltic herring
are fully reproductive (ICES 2009a) and therefore
their mean body weight may have an effect on the
stock recruitment success (Cardinale et al. 2009).
Fish older than 5 yr were excluded since in some
years they were scarcely represented in our samples
(<10 ind.).
Condition and WAA (hereafter referred as growth
parameters) were used as indicators of the biological
state of the fish at each specific year, and not as prox-
ies for developmental/ontogenetic processes for which
growth rate in length would have been more indicative
(e.g. Winters & Wheeler 1994).
To account for potential differences between coun-
tries in biological parameters (length and weight
242
Fig. 1. Central Baltic Sea: ICES Subdivision 28 was the area
of fish sampling

Casini et al.: Ecological threshold and fish population dynamics
measurements as well as age estimations), and in
turn growth calculation, a Generalized Linear Model
(GLM) was used to predict the year effect on herring
condition and WAA after scaling out the country effect.
Condition data were normally distributed and, thus,
the normal distribution was used in the GLM. For
WAA, on the other hand, a Gamma distribution was
preferred to account for the moderate skewness in the
distribution of the data.
In the analyses we used growth data from ICES Sub-
division 28, because this area was used in previous
studies investigating herring growth (e.g. Möllmann et
al. 2003, Casini et al. 2006) and ecosystem dynamics
(Casini et al. 2008, 2009) and, therefore, constitutes
a valid standpoint for comparison. Moreover, Sub-
division 28 corresponds to the geographical centre of
distribution of the herring and sprat stocks used in
this study.
Salinity data for this area were provided by the
Swedish Meteorological and Hydrological Institute
(SMHI, freely available at www.smhi.se). The average
of salinity in May and August integrated between 0
and 100 m depth (samples at surface and at 10 m depth
intervals) was used in the analyses. Spring and sum-
mer are the main feeding seasons for both herring and
sprat in the central Baltic Sea (e.g. Szypu8a et al. 1997),
and spring is also the main reproduction period for the
copepod Pseudocalanus spp., one of the main prey for
both fish species (Möllmann et al. 2003, Casini et al.
2004). Therefore, spring-summer can be considered a
critical period determining fish condition and WAA
later in early autumn (Casini et al. 2006). Salinity is
related to the Baltic Sea Index (BSI), a regional climate
proxy closely coupled with the NAO (North Atlantic
Oscillation). BSI is defined as the difference of normal-
ized sea level pressure anomalies between Szczecin
(Poland) and Oslo (Norway) (Lehmann et al. 2002).
Positive values of both indices correspond to a domi-
nance of westerlies over the Baltic and increased rain-
fall and runoff, whereas a negative index corresponds
to easterlies (Hänninen et al. 2000, Lehmann et al.
2002).
The relationship between herring growth (in terms
of condition and WAA) and the predictors was first
analysed with Generalized Additive Models (GAMs).
The following additive formulation was used:
(1)
yasV sV
nn
=+ +…+ +
11
() ()ε
243
29
31
33
35
37
39
41
43
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Ye a r
Condition (g)
10
20
30
40
50
60
70
80
90
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Ye a r
Mean weight-at-age, WAA (g)
Age 5
Age 4
Age 3
Age 2
Average
0
5
10
15
20
25
30
35
40
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Year
Sprat abundance (1010 ind.)
6.5
7.0
7.5
8.0
8.5
9.0
9.5
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
Year
Salinity 0 –100 m (psu)
–0.8
–0.6
–0.4
–0.2
0.0
0.2
0.4
0.6
0.8
Baltic Sea Index, BSI
Salinity
BSI
a
dc
b
Fig. 2. Temporal trends of parameters used in the analyses: (a) herring condition, (b) herring weight-at-age (WAA), (c) sprat stock
abundance, (d) salinity between 0 and 100 m, and Baltic Sea Index, BSI (a proxy for the regional climate)

Mar Ecol Prog Ser 413: 241–252, 2010
where ais the intercept, sthe thin plate smoothing
spline function (Wood 2003), V1…Vnthe predictors
and εthe random error. As predictors, we used sprat
stock abundance and salinity. These parameters
were chosen because of their acknowledged impor-
tance for herring growth in the Baltic Sea, acting on
the main planktonic prey for herring (Möllmann et
al. 2003, Rönkkönen et al. 2004, Casini et al. 2006).
We did not test for intra-specific density-dependence
(i.e. effect of herring abundance on herring growth)
as no evidence has been found for such relationship
(Rönkkönen et al. 2004, Casini et al. 2006). The lack
of intra-specific density-dependence is also sup-
ported by the absence of a negative relation between
herring abundance and zooplankton in the offshore
areas of the central Baltic Sea (Casini et al. 2008),
that could be due to the omnivorous nature of her-
ring which also feed on nektobenthos (Casini et al.
2004), reducing the interaction strength with zoo-
plankton. Unfortunately, nektobenthos time-series
covering our study period are not available for this
area and could not be used in the growth models.
The numerical dominance of sprat compared to her-
ring in a large part of the study period (ICES 2009a)
could also partly explain the lack of detectable intra-
specific density-dependence within the herring pop-
ulation.
Furthermore, to test for the existence of a potential
‘phase transition’ (Litzow & Ciannelli 2007) in the
response of growth to the predictors, we applied a
Threshold Generalized Additive Model (TGAM, Cian-
nelli et al. 2004) to condition and WAA. This kind of
non-additive model can be formulated as:
(2)
where 2 specific additive formulations are adopted for
different values of the threshold variable Vx. We tested
the occurrence of a threshold on sprat abundance,
under the assumption that the zooplankton community
in the offshore Baltic Sea is dominated by environmen-
tal variability for low abundances of sprat, while it is
driven by the dynamic of this pelagic fish when its
abundance is high (Casini et al. 2009).
TGAMs are an extension of non-parametric regres-
sion techniques (Hastie &Tibshirani 1990) and were
chosen here for their ability to represent an abrupt
change in the relationships between dependent and
independent variables (i.e. a phase transition) at a spe-
cific threshold value. The threshold value was selected
minimizing the Generalized Cross Validation score
(GCV) of the whole model (Ciannelli et al. 2004). The
searching algorithm runs the model for 100 possible
threshold values between the 0.1 lower and the 0.9
upper quantiles. To evaluate the robustness of the esti-
mated threshold, we performed a sensitivity analysis
(Saltelli et al. 2000). Our objective was to test the
dependency of the identified threshold value to the
specific set of data and years available. We generated
1000 random replicates of the dataset, for different
proportions of the original dataset. Then, a TGAM was
fitted on each replicate and the corresponding thresh-
old estimated. Residuals of the TGAMs were analysed
to inspect for potential deviation from the normality
assumption and other anomalies in the data or in the
model fit (i.e. autocorrelation) using graphical methods
(Cleveland 1993). Autocorrelation of the model residu-
als was examined using the autocorrelation function
(ACF).
The normal distribution was used in the condition
models, whereas a Gamma distribution was preferred
for the WAA models due to the moderate skewness in
the distribution of the data. Additive and threshold for-
mulations were compared using the Akaike Informa-
tion Criterion (AIC, Akaike 1973). To calculate the AIC
for TGAMs, an extra penalty was introduced to ac-
count for the threshold parameter.
Additionally, an estimate of the significance of the
threshold was obtained through a wild bootstrap
approach (Mammen 1993). In this case, a null sce-
nario of no-threshold effect was simulated reshuffling
the threshold variable. Then, the same threshold
model formulation was fitted and the penalized max-
imum log-likelihood (PML) of the model was calcu-
lated. This procedure was repeated 1000 times
reshuffling the threshold variable in each bootstrap
sample. In this way we obtained a reference distribu-
tion of the PML for the null scenario of no-threshold
effect. Finally, the bootstrap estimate of the p-value
for the threshold was calculated as the percentage of
the PML values of the reference distribution for the
null scenario that was larger than the original model
PML.
The strength of the link between herring growth and
sprat abundance (as well as salinity) in the 2 configura-
tions identified by the TGAM was also assessed, by
quantifying the probability density distribution of the
correlation coefficients obtained by bootstrap resam-
pling (Casini et al. 2009). This analysis involved a ran-
dom pairwise sampling with replacement where each
time-series was resampled 5000 times. The number of
elements in each bootstrap sample equals the number
of elements in the original dataset. The probability
density distribution of the corresponding correlation
coefficients was then computed using nonparametric
kernel smoothing (Casini et al. 2009). In the simple lin-
ear correlation analyses, the potential occurrence of
temporal autocorrelation in the growth time-series was
tested in the 2 configurations using the autocorrelation
function (ACF).
ya sV s V V t
sV s
nn x
=+ +…+ + >
+…+
11 1 1
21 1 2
,,
,,
() ()
()
εfor
nnn x
VVt()+≤
⎧
⎨
⎩εfor
244

Casini et al.: Ecological threshold and fish population dynamics
RESULTS
Time-series of herring condition and WAA are pre-
sented in Fig. 2a,b. Condition showed a continuous
decline from the late 1970s, reaching a minimum
in 1996. Thereafter, herring condition increased, al-
though oscillating at relatively low levels. WAA
showed a similar pattern, except the decline started
some years later than for condition, and the increase
after 1996 was not as marked as for condition. WAA
trends showed a tight covariation among all age-
classes, with inter-annual changes coincident for the
whole population. Sprat stock was mostly at low levels
between the late 1970s and the early 1990s, followed
by a drastic increase that peaked in the years 1995 to
1996. Subsequently, the sprat stock has oscillated at
relatively high levels (Fig. 2c). Salinity between 0 to
100 m showed a constant decrease from the 1970s to
the early 1990s, followed by a steady increase, and
eventually reached values almost as high as at the
beginning of the time-series (Fig. 2d). Salinity was sig-
nificantly correlated with the winter Baltic Sea Index
(r = –0.43, p = 0.012; after correcting for temporal auto-
correlation by first-order differencing, r = –0.38, p =
0.028), evidencing the strong influence of atmospheric
oscillations on the hydrological processes in the Baltic
Sea (Hänninen et al. 2000, Lehmann et al. 2002).
In general, both for the GAMs and TGAMs, the
effect of sprat abundance on herring growth was
stronger than the effect of salinity (Table 1). The 2
predictors were significant in all the models, except
salinity in the GAM for WAA. Threshold models per-
formed generally better than additive formulations
both in terms of deviance explained and AIC (Table
1). However, the improvements including a threshold
were more evident for the WAA model than for the
condition model, which showed a weaker non-addi-
tivity in the dynamics (Table 1), as also shown by the
shallower minimum in the GCV profile (see Figs. 3b
& 4b). The PML calculated on the original data was
smaller than the lower 5% of the PML distribution for
the bootstrap under the no-threshold null hypothesis
in both the condition and WAA models (Fig. S2 in the
Supplement). Consequently, the thresholds identified
were considered statistically significant for both
growth parameters. In addition, the same significant
threshold value of ~18 ×1010 sprat individuals was
obtained for both the response variables (Table 1).
For sprat abundances lower than the identified
threshold (low-sprat configuration), salinity showed a
significant positive relationship with both condition
and WAA. In contrast, for sprat abundances higher
than the identified threshold (high-sprat configura-
tion), a significant negative effect of sprat on herring
growth was found (Table 1, Figs. 3c,d & 4c,d). The
TGAM formulation correctly captured the general
pattern of herring condition along the whole time
period investigated (Fig. 3a). The model performed
better during the second half of the time series,
tightly modelling the wide inter-annual fluctuations
that characterized herring condition. The TGAM
formulation well described the rapid drop in WAA
observed in the mid 1990s and the following fluctua-
tions around values almost half of those observed at
the beginning of the time-series (Fig. 4a). The model
underestimated 1982 to 1983 values and overesti-
mated the peak in WAA observed in 2002. The resid-
uals of the TGAMs did not violate the normality
assumption and did not present temporal autocorrela-
tion (Fig. S3 in the Supplement).
The sensitivity analysis showed that a decrease in
the dataset size did not affect the median of the esti-
mated threshold in the TGAMs (Fig. 5). A reduction of
the dataset size down to 72% of the original size had
only minor effect on the precision of the threshold esti-
mate in the WAA model. A sensible loss in the preci-
sion of the estimate was observed only for levels <70%
of the dataset size, with an asymmetric dispersion
245
Model Herring response Threshold (t) Dev. Expl. AIC n Factors Fedf p n (t)
GAM Condition – 77.3 102.07 29 Salinity 3.55 2.22 0.0295 –
Sprat 15.13 1.72 < 0.0001 –
WAA – 79.0 189.90 29 Salinity 1.95 1.71 0.149 –
Sprat 18.79 2.05 < 0.0001 –
TGAM Condition 17.85 81.4 101.67 29 Salinity for sprat < t 5.75 2.76 0.0045 16
Sprat for sprat ≥t 22.14 2.82 < 0.0001 15
WAA 17.85 88.8 173.96 29 Salinity for sprat < t 9.86 1.00 0.0044 16
Sprat for sprat ≥t 47.33 2.95 < 0.0001 15
Table 1. Generalized Additive Model (GAM) and Threshold Generalized Additive Model (TGAM) analyses. For each model, de-
viance explained (Dev. Expl.), Akaike Information Criterion (AIC) and no. observations (n) are given. For each predictor, effective
degrees of freedom (edf) and significance value (p) are provided. For TGAMs, threshold value (t) and no. of observations above
and under the threshold are also given (n(t)). WAA: weight-at-age

Mar Ecol Prog Ser 413: 241–252, 2010
towards low threshold estimates. The sensitivity analy-
sis on the condition model showed a more symmetric
dispersion but poorer stability in the estimates of the
threshold. Threshold estimates showed a visible dis-
persion already omitting 7% of the original condition
dataset (Fig. 5).
The linear relationships between variations in her-
ring growth and the predictors (salinity and sprat
abundance) in the 2 configurations identified by the
TGAM are shown in Fig. 6 (condition) and Fig. 7
(WAA). Considering the whole period, the analysis evi-
denced a strong negative Pearson’s correlation coeffi-
cient between sprat abundance and growth parame-
ters. However, the analysis of the relationship in the
low-sprat configuration did not provide statistical sup-
port that variations in herring growth are related to
sprat abundance, as illustrated by the simple linear
correlations and the probability density distribution of
correlations coefficients (Figs. 6a,c & 7a,c). On the
other hand, in the high-sprat configuration, the link
was significantly enhanced and herring growth was
closely coupled to sprat variations (Figs. 6a,c & 7a,c).
Reversed results were found in the assessment of the
link between herring growth and salinity. In fact, con-
sidering the whole period, the analysis showed a posi-
tive Pearson’s correlation coefficient, which increased
in the low-sprat configuration (Figs. 6b,d & 7b,d). In
contrast, in the high-sprat configuration, no relation-
ship between salinity and herring growth was notice-
able (Figs. 6b,d & 7b,d). Herring growth showed lower
values in the high-sprat configuration than in the low-
sprat configuration, regardless of the salinity level. In
246
Fig. 3. Threshold Generalized Additive Model (TGAM) for herring condition. (a) Time-series of observed values and values pre-
dicted by TGAM, (b) Generalized Cross Validation (GCV) score profile as function of threshold variable (sprat abundance) with
vertical dotted line = estimated threshold, (c) effect (unitless) of salinity on condition, (d) effect of sprat abundance on condition.
Grey-shaded regions represent 95 % confidence intervals

Casini et al.: Ecological threshold and fish population dynamics
general, the improvement of the relationship between
growth and sprat abundance in the high-sprat configu-
ration (and between growth and salinity in the low-
sprat configuration) was more evident for herring
WAA than condition, as shown by the Pearson’s corre-
lation coefficients and their probability density distrib-
utions. The growth time-series did not present tempo-
ral autocorrelation in either of the 2 configurations
(Fig. S4 in the Supplement).
DISCUSSION
In this study, we have shown that the main factors
driving herring growth in the central Baltic Sea may
switch following changes in other key food-web com-
ponents. Specifically, we found that the dominant fac-
tors affecting variations in the growth of herring
switched from salinity to inter-specific competition
depending on the size of the sprat population. There-
fore, during the past 3 decades, a combination of
changes in hydro-climatic factors and trophic interac-
tions after the sprat outburst, has led to a drastic reduc-
tion in the individual growth of herring.
The factors affecting the body growth of other her-
ring populations worldwide have been intensively
investigated. These factors range from internal popu-
lation control, e.g. intra-specific density-dependence
(Winters & Wheeler 1994, Tanasichuk 1997, Shin &
Rochet 1998, Melvin & Stephenson 2007), physical
forcing as wind-induced turbulence (Shin & Rochet
1998), to temperature (Watanabe et al. 2008) and/or
247
Fig. 4. TGAM for herring WAA. (a) Time-series of observed values and values predicted by TGAM, (b) GCV profile as function of
threshold variable (sprat abundance) with vertical dotted line = estimated threshold, (c) effect of salinity on WAA, (d) effect of
sprat abundance on WAA. Grey-shaded regions represent 95 % confidence intervals

Mar Ecol Prog Ser 413: 241–252, 2010
anthropogenic stress as fishery-related size selection
(Wheeler et al. 2009). Density-dependence and hydro-
climatic forces have also been suggested to act at dif-
ferent temporal (Heath et al. 1997) and spatial (Husebø
et al. 2007) scales.
Compared to other herring stocks, the central Baltic
Sea herring population inhabits an enclosed brackish
environment, which makes organisms particularly sen-
sitive to large salinity variations determined by cli-
mate-mediated irregular water inflows from the North
Sea and precipitation (Voipio 1981, Lehmann et al.
2002). Moreover, the large diet overlap between her-
ring and sprat in this region (Casini et al. 2004) pro-
duces a strong inter-specific feeding competition
among clupeids. Accordingly, the growth of the central
Baltic Sea herring has been previously linked to salin-
ity variations and sprat stock size (Möllmann et al.
248
Fig. 5. Sensitivity analysis for (a) herring condition and (b)
WAA. Boxplot of threshold values estimated sampling differ-
ent proportions of original dataset (randomly sampling from
97 to 66% of the time-series). Bold lines identify the median,
and box-whiskers = approximate 2-side 95% confidence in-
terval and first and third quartiles
29
31
33
35
37
39
41
43
0 5 10 15 20 25 30 35 40 45
Sprat abundance (1010 ind.)
Herring condition (g)
02
92
90
89
88
87
86
85
84
83
82
80
79
78
08
07
06 05
04
03
01
00
99
98
97
96
95
94
93
29
31
33
35
37
39
41
43
7.2 7.4 7.6 7.8 8.0 8.2 8.4 8.6 8.8
Salinity 0–100 m (psu)
Low sprat
High sprat
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
–1.00 –0.75 –0.50 –0.25 0.00 0.25 0.50 0.75 1.00 –1.00 –0.75 –0.50 –0.25 0.00 0.25 0.50 0.75 1.00
r (relation condition-sprat abundance)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
r (relation condition-salinity)
Low sprat Low sprat
High sprat High sprat
Whole period Whole period
a
dc
b
Low sprat High sprat
Density distribution
Herring condition (g)Density distribution
Fig. 6. Alternative dynamics in regulation mechanism of
herring condition. (a) Relationships between sprat abundance
and herring condition in the 2 configurations. Low-sprat: r =
–0.56, p = 0.27; high-sprat: r = – 0.82, p = 0.00022; whole
period: r = –0.82, p < 0.0001; (b) relationships between salinity
and herring condition in the 2 configurations. Low-sprat: r =
0.58, p = 0.030; high-sprat: r = 0.22, p = 0.43; whole period: r =
0.54, p = 0.0024. Nos. (each point) = observation year. (c,d)
Density distribution of correlation coefficients between her-
ring condition and sprat abundance, as well as salinity, in the
2 configurations and whole study period

Casini et al.: Ecological threshold and fish population dynamics
2003, Casini et al. 2006), both factors acting on the
abundance of one of the main preys for herring, the
copepod Pseudocalanus spp. In particular, while the
growth and reproductive performance of this zoo-
plankter are enhanced by high salinity, sprat predation
can operate a top-down control on this food resource
(Möllmann & Köster 2002, Renz & Hirche 2006). Our
results add to this general ecological understanding a
crucial aspect, showing that the relative strength of the
2 main drivers (i.e. salinity and inter-specific competi-
tion) on herring growth may switch depending on the
stock size of the sprat, the main planktivorous fish in
the offshore areas of the central Baltic Sea. We specifi-
cally identified an ecological threshold of ~18 ×1010
sprat individuals which separated one low-sprat con-
figuration characterized by a close link between her-
ring growth and salinity variations, and one high-sprat
configuration in which herring growth appears decou-
pled from salinity and becomes strongly controlled by
inter-specific density-dependence.
The ecological explanation of this switch in the main
regulatory mechanisms of herring growth is provided
by Casini et al. (2009) who showed that the external
drivers of zooplankton dynamics switch from hydro-
climatic forcing to predation pressure depending on
the population size of sprat. Also for zooplankton, the
shift from one regulation mechanism to the other is
triggered when the sprat population exceeds the
threshold of ~17 ×1010 sprat individuals (Casini et al.
2009), very close to the threshold found in our study. In
particular, at low levels of sprat population, Pseudo-
calanus spp. appears to be driven by salinity variations,
a link that is disrupted when the abundance of sprat
exceeds the ecological threshold (Casini et al. 2009).
The evidence provided by our investigation in com-
bination with Casini et al. (2009) underlines that in low
diverse systems, as the Baltic Sea, variations in key
species such as sprat can have implications for ecosys-
tem functioning detectable across trophic levels.
Specifically, although salinity is an important factor for
herring condition and WAA (e.g. Flinkman et al. 1998,
Möllmann et al. 2003, Rönnkönen et al. 2004), our
study provides evidence that food-web interactions,
i.e. competition with sprat, can dampen the positive
effect of high salinities, and become the main regulator
of herring growth. This is exemplified by the increase
249
20
30
40
50
60
70
02
92 90
89
88
87
86 84
83
82
80
79 78
08 07
06
05
04
0301
00
99
98
97 96
95
94
93
20
30
40
50
60
70 Low sprat
High sprat
Low sprat High sprat
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Low sprat
High sprat
Whole period
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
Low sprat
High sprat
Whole period
a
dc
b
0 5 10 15 20 25 30 35 40 45
Sprat abundance (1010 ind.)
Herring WAA (g)
7.2 7.4 7.6 7.8 8.0 8.2 8.4 8.6 8.8
Salinity 0–100 m (psu)
–1.00 –0.75 –0.50 –0.25 0.00 0.25 0.50 0.75 1.00 –1.00 –0.75 –0.50 –0.25 0.00 0.25 0.50 0.75 1.00
r (relation WAA-sprat abundance) r (relation WAA-salinity)
Density distribution
Herring WAA (g)Density distribution
Fig. 7. Alternative dynamics in regulation mechanism of herring WAA. (a) Relationships between sprat abundance and herring
WAA in the 2 configurations. Low-sprat: r = –0.20, p = 0.49; high-sprat: r = –0.76, p = 0.001; whole period: r = –0.81, p < 0.0001;
(b) relationships between salinity and herring WAA in the 2 configurations. Low-sprat: r = 0.60, p = 0.032; high-sprat: r = 0.11,
p = 0.69; whole period: r = 0.53, p = 0.0031. Nos. (each point) = observation year. (c,d) Density distribution of the correlation co-
efficients between herring WAA and sprat abundance, as well as salinity, in the 2 configurations and whole study period

Mar Ecol Prog Ser 413: 241–252, 2010
in salinity from the beginning of the 1990s, which was
not translated into the expected increase in herring
growth (especially WAA), despite the high biomass of
the primary producers (Casini et al. 2008). This demon-
strates the occurrence of a discontinuous pattern (i.e.
hysteresis) in the response of herring growth to salinity
variations (Fig. 8) that may be indicative of alternative
stable states in the system (Scheffer & Carpenter 2003).
The shift between the 2 configurations during the past
3 decades has been triggered by the dramatic increase
in the sprat stock, which has been linked before to pre-
dation release from cod (Casini et al. 2008) whose
stock decreased due to high fishing pressure (ICES
2009a) and adverse hydro-climatic factors (i.e. low
salinity and anoxic conditions in the deep waters)
(Köster et al. 2005). However, the rise in temperature,
which enhances sprat egg and larval survival, has also
contributed to the sprat stock increase (Nissling 2004,
Alheit et al. 2005).
The occurrence of shifts in the functioning of marine
ecosystems has been generally seldom reported, al-
though some examples exist (e.g. Hunt et al. 2002,
Ciannelli & Litzow 2007, Stige et al. 2009). In the Gulf
of Alaska it was found that the Pacific cod–prey inter-
action is either top-down or bottom-up regulated
depending on temperature conditions and community
state (Litzow & Ciannelli 2007). Hunt et al. (2002) also
suggested that temperature and ice dynamics modu-
late the strength of top-down and bottom-up forces on
the recruitment of walleye pollock population Thera-
gra chalcogramma in the Barents Sea. Stige et al.
(2009), on the other hand, showed for the Barents Sea
that at low densities of planktivorous fish (capelin
Mallotus villosus) the climate forcing on zooplankton is
stronger, similar to what we found in our study for
herring growth. In our study, however, we also indicate
a putative ecological threshold (~18 ×1010 ind. sprat)
responsible for the shift in the control of herring
growth.
The threshold dynamic evidenced in our study by
the TGAMs was more evident for herring WAA than
condition, as shown by all the statistics used in the
analyses. Although at the moment we do not have
robust elements to provide a comprehensive explana-
tion for this difference, the common threshold identi-
fied indicates that common mechanisms are behind the
threshold dynamic of both growth parameters. It is
worth noting that using additive models (GAMs) sprat
abundance was the main explanatory factor, whereas
salinity appeared of minor importance. Also using
TGAMs, the strength of the relationship between sprat
abundance and herring growth in the high-sprat con-
figuration appeared higher than the relationship
between salinity and herring growth in the low-sprat
configuration. These considerations point to that, while
a significant threshold was identified by our analyses,
the density-dependent mechanism seems overall to
have been the major regulator of herring growth dur-
ing the past 3 decades.
The direct effect of fishing on the body size of target
organisms has been shown by several studies (Shin et
al. 2005). Also for the central Baltic Sea, it may be sug-
gested that the high fishing mortalities during the
1980s and 1990s (ICES 2009a) have contributed to the
reduction in the mean size of herring by selecting the
larger individuals (e.g. Vainikka et al. 2009). However,
the increase in fishing mortality could be partly an
effect, and not only a cause, of the decrease in individ-
ual herring size (ICES 2009a). In fact, for a given catch
in biomass, a decrease in fish individual weight would
imply a higher number of fish caught every year. We
opted, therefore, not to introduce fishing mortality as a
potential predictor in our analyses. The question
whether changes in fishing mortality are mainly a
cause, or a result, of the variations in herring mean size
is worth further investigation.
The central Baltic herring stock decreased steadily
from the end of the 1970s (ICES 2009a). The drop in
WAA and condition of the spawners could have indi-
rectly contributed to the general decrease in the her-
ring stock, by hampering the population recruitment
success (Cardinale et al. 2009), likely through reduced
fecundity and hatching success of the eggs (Laine &
Rajasilta 1999). The decrease in herring body growth
also directly contributes to explain the sharper de-
crease in herring spawning stock biomass, than in
stock abundance, since the early 1980s (Fig. S1 in the
Supplement). This is because a decrease in the mean
weight is automatically translated into a reduction in
biomass, for comparable levels of abundance. How-
ever, the change in age structure of the spawning pop-
ulation towards a relatively higher abundance of
250
Low sprat
High sprat
Stressor (salinity)
Biological feature
(herring growth)
High cod
Low cod
Herring
growth Sprat
Salinity
Herring
growth Sprat
Salinity
Fig. 8. Discontinuity (hysteresis) in the response of herring
growth to variations in salinity. In low-sprat configuration,
salinity explains changes in herring growth, whereas in
high-sprat configuration the link between herring growth
and salinity becomes weaker and inter-specific density-
dependence becomes the main factor affecting herring
growth

Casini et al.: Ecological threshold and fish population dynamics
younger spawners (ICES 2009a) has likely also con-
tributed to the drastic decrease in spawning stock bio-
mass.
The results presented here have important implica-
tions for an ecosystem-approach to fisheries manage-
ment. The threshold in the mechanisms of regulation
of herring growth implies that effective management
should take in consideration both hydro-climatic cir-
cumstances and food-web structure, and be adaptive
to their variations with prompt management actions.
We have specifically detected here a clear-cut and
easy-to-understand threshold value which could be
used to reach the goal of a healthy central Baltic her-
ring stock. The restraint of the sprat population below
the abundance threshold would release herring from
inter-specific competition and enhance herring growth,
increasing spawning stock biomass and stock repro-
ductive potential. The best way to achieve this goal is
to allow the Baltic cod stock to recover to a level capa-
ble of controlling the sprat population.
Acknowledgements. This paper is dedicated to the memory
of our colleague and dear friend Johan Modin, who inspired
Fig. 8.
LITERATURE CITED
Akaike H (1973) Information theory and an extension of
maximum likelihood principle. In: Petrov BN, Csáki F
(eds) Proc Sec Int Symp Info Theory. Akademia Kiado,
Budapest
Alheit J, Möllmann C, Dutz J, Kornilovs G, Loewe P,
Mohrholz V, Wasmund N (2005) Synchronous ecological
regime shifts in the central Baltic and the North Sea in the
late 1960s. ICES J Mar Sci 62:1205–1215
Bagge O (1989) A review of investigations of the predation of
cod in the Baltic. Rapp P-V Reun Cons Int Explor Mer 190:
51–56
Cardinale M, Arrhenius F (2000) Decreasing weight-at-age of
Baltic herring (Clupea harengus) between 1986 and 1996:
a statistical analysis. ICES J Mar Sci 57:882– 893
Cardinale M, Möllmann C, Bartolino V, Casini M and others
(2009) Effect of environmental variability and spawner
characteristics on the recruitment of Baltic herring Clupea
harengus populations. Mar Ecol Prog Ser 388:221–234
Casini M, Cardinale M, Arrhenius F (2004) Feeding prefer-
ences of herring (Clupea harengus) and sprat (Sprattus
sprattus) in the southern Baltic Sea. ICES J Mar Sci 61:
1267–1277
Casini M, Cardinale M, Hjelm J (2006) Inter-annual variation
in herring (Clupea harengus) and sprat (Sprattus sprattus)
condition in the central Baltic Sea: What gives the tune?
Oikos 112:638– 650
Casini M, Lövgren J, Hjelm J, Cardinale M, Molinero JC,
Kornilovs G (2008) Multi-level trophic cascades in a heav-
ily exploited open marine ecosystem. Proc R Soc B Biol Sci
275:1793–1801
Casini M, Hjelm J, Molinero JC, Lövgren J and others (2009)
Trophic cascades promote threshold-like shifts in pelagic
marine ecosystems. Proc Natl Acad Sci USA 106:197–202
Ciannelli L, Chan KS, Bailey K, Stenseth NC (2004) Nonaddi-
tive effect of the environment on the survival of a large
marine fish population. Ecology 85:3418– 3427
Cleveland WS (1993) Visualizing data. Hobart Press, Summit,
NJ
EC (Council of the European Union) (2008) Council Regula-
tion (EC) No 1322/2008 of 28 November 2008 fixing the
fishing opportunities and associated conditions for certain
fish stocks and groups of fish stocks applicable in the
Baltic Sea for 2009. Off J Eur Union L 345:1– 9
Flinkman J, Aro E, Vuorinen I, Viitasalo M (1998) Changes in
northern Baltic zooplankton and herring nutrition from
1980s to 1990s: top-down and bottom-up processes at
work. Mar Ecol Prog Ser 165:127–136
Hänninen J, Vuorinen I, Hjelt P (2000) Climatic factors in the
Atlantic control the oceanographic and ecological changes
in the Baltic Sea. Limnol Oceanogr 45:703– 710
Hastie TJ, Tibshirani RJ (1990) Generalized additive models.
Chapman & Hall, New York
Heath M, Scott B, Bryant AD (1997) Modelling the growth of
herring from four different stocks in the North Sea. J Sea
Res 38:413– 436
Hunt GL Jr, Stabeno P, Walters G, Sinclair E, Brodeur RD,
Napp JM, Bond NA (2002) Climate change and control of
the southeastern Bering Sea pelagic ecosystem. Deep-Sea
Res II 49:5821–5853
Husebø Å, Slotte A, Stenevik EK (2007) Growth of juvenile
spring-spawning herring in relation to latitudinal and
interannual differences in temperature and fish density in
their coastal and fjord nursery areas. ICES J Mar Sci 64:
1161–1172
ICES (2009a) Report of the Baltic Fisheries Assessment Work-
ing Group. ICES CM 2009/ACOM:07
ICES (2009b) Report of the Baltic International Fish Survey
Working Group. International Council for the Exploration
of the Sea. ICES CM 2009/LRC:05
Köster FW, Möllmann C, Hinrichsen HH, Wieland K and oth-
ers (2005) Baltic cod recruitment: the impact of climate
variability on key processes. ICES J Mar Sci 62:1408–1425
Laine P, Rajasilta M (1999) The hatching success of Baltic her-
ring eggs and its relation to female condition. J Exp Mar
Biol Ecol 237:61–73
Lehmann A, Krauss W, Hinrichsen HH (2002) Effects of
remote and local atmospheric forcing on circulation and
upwelling in the Baltic Sea. Tellus 54A:299–316
Litzow MA, Ciannelli L (2007) Oscillating trophic control
induces community reorganization in a marine ecosystem.
Ecol Lett 10:1124–1134
Mammen E (1993) Bootstrap and wild bootstrap for high
dimensional linear models in resampling. Ann Stat 21:
255–285
Melvin GD, Stephenson RL (2007) The dynamics of a recover-
ing fish stock: Georges Bank herring. ICES J Mar Sci 64:
69–82
Möllmann C, Köster FW (2002) Population dynamics of
calanoid copepods and the implications of their predation
by clupeid fish in the Central Baltic Sea. J Plankton Res 24:
959– 977
Möllmann C, Kornilovs G, Fetter M, Köster FW, Hinrichsen
HH (2003) The marine copepod Pseudocalanus elongatus,
as a mediator between climate variability and fisheries in
the Central Baltic Sea. Fish Oceanogr 12:360– 368
Nissling A (2004) Effects of temperature on egg and larval
survival of cod (Gadus morhua) and sprat (Sprattus sprat-
tus) in the Baltic Sea: implications for stock development.
Hydrobiologia 514:115–123
Renz J, Hirche HJ (2006) Life cycle of Pseudocalanus acuspes
251
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤➤
➤
➤
➤
➤
➤

Mar Ecol Prog Ser 413: 241–252, 2010
Giesbrecht (Copepoda, Calanoida) in the Central Baltic
Sea: seasonal and spatial distribution. Mar Biol 148:
567–580
Rönkkönen S, Ojaveer E, Raid T, Viitasalo M (2004) Long-
term changes in Baltic herring (Clupea harengus mem-
bras) growth in the Gulf of Finland. Can J Fish Aquat Sci
61:219–229
Saltelli A, Chan K, Scott EM (2000) Sensitivity analysis. Prob-
ability and statistics series, John Wiley & Sons, Somerset,
NJ
Scheffer M, Carpenter SR (2003) Catastrophic regime shifts in
ecosystems: linking theory to observation. Trends Ecol
Evol 18:648– 656
Shin YJ, Rochet MJ (1998) A model for the phenotypic plastic-
ity of North sea herring growth in relation to trophic con-
ditions. Aquat Living Resour 11:315– 324
Shin YJ, Rochet MJ, Jennings S, Field JG, Gislason H (2005)
Using size-based indicators to evaluate the ecosystem
effects of fishing. ICES J Mar Sci 62:384–396
Stige LC, Lajus DL, Chan KS, Dalpadado P, Basedow SL,
Berchenko I, Stenseth NC (2009) Climatic forcing of zoo-
plankton dynamics is stronger during low densities of
planktivorous fish. Limnol Oceanogr 54:1025–1036
Szypu8a J, Ostrowski J, Margo?ski P, Krajewska-So8tys A
(1997) Food of Baltic herring and sprat in the years 1995-
1996 in light of the availability of components. Bull Sea
Fish Inst (Gdynia) 2:19– 31
Tanasichuk RW (1997) Influence of biomass and ocean cli-
mate on the growth of Pacific herring (Clupea pallasi)
from the south-west coast of Vancouver Island. Can J Fish
Aquat Sci 54:2782–2788
Vainikka A, Mollet F, Casini M, Gårdmark A (2009) Spatial
variation in growth, condition and maturation reaction
norms of the Baltic herring Clupea harengus membras.
Mar Ecol Prog Ser 383:285–294
Voipio A (ed) (1981) The Baltic Sea. Elsevier, Amsterdam
Watanabe Y, Dingsør GE, Tian Y, Tanaka I, Stenseth NC
(2008) Determinants of mean length at age of spring
spawning herring off the coast of Hokkaido, Japan. Mar
Ecol Prog Ser 366:209–217
Wheeler JP, Purchase CF, Macdonald PDM, Fill R, Jacks L,
Wang H, Ye C (2009) Temporal changes in maturation,
mean length-at-age, and condition of spring-spawning
Atlantic herring (Clupea harengus) in Newfoundland
waters. ICES J Mar Sci 66:1800–1807
Wilmers CC, Post E, Peterson RO, Vucetich JA (2006) Preda-
tor disease out-break modulates top-down, bottom-up and
climatic effects on herbivore population dynamics. Ecol
Lett 9:383– 389
Winters GH, Wheeler JP (1994) Length-specific weight as a
measure of growth success of adult Atlantic herring (Clu-
pea harengus). Can J Fish Aquat Sci 51:1169–1179
Wood SN (2003) Thin-plate regression splines. J R Stat Soc B
65:95–114
252
Submitted: December 11, 2009; Accepted: March 17, 2010 Proofs received from author(s): August 5, 2010
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤