Effects of sea temperature and stratification changes on seabird breeding success

Abstract

As apex predators in marine ecosystems, seabirds may primarily experience climate change impacts indirectly, via changes to their food webs. Observed seabird population declines have been linked to climate-driven oceanographic and food web changes. However, relationships have often been derived from relatively few colonies and consider only sea surface temperature (SST), so important drivers, and spatial variation in drivers, could remain undetected. Further, explicit climate change projections have rarely been made, so longer-term risks remain unclear. Here, we use tracking data to estimate foraging areas for 11 black-legged kittiwake Rissa tridactyla colonies in the UK and Ireland, thus reducing reliance on single colonies and allowing calculation of colony-specific oceanographic conditions. We use mixed models to consider how SST, the potential energy anomaly (indicating density stratification strength) and the timing of seasonal stratification influence kittiwake productivity. Across all colonies, higher breeding success was associated with weaker stratification before breeding and lower SSTs during the breeding season. Eight colonies with sufficient data were modelled individually: higher productivity was associated with later stratification at 3 colonies, weaker stratification at 2, and lower SSTs at one, whilst 2 colonies showed no significant relationships. Hence, key drivers of productivity varied among colonies. Climate change projections, made using fitted models, indicated that breeding success could decline by 21 to 43% between 1961-90 and 2070-99. Climate change therefore poses a longer-term threat to kittiwakes, but as this will be mediated via availability of key prey species, other marine apex predators could also face similar threats.

Full text

CLIMATE RESEARCH
Clim Res
Vol. 66: 75– 89, 2015
doi: 10.3354/cr01332 Published October 6
1. INTRODUCTION
Ecological impacts of climate change are increas-
ingly well-understood, with changes in species’
ranges and phenology predicted and observed in
both terrestrial and marine environments (Parmesan
2006, Doney et al. 2012). Some species may be pri-
marily affected via changed biotic interactions (e.g.
© Inter-Research 2015 · www.int-res.com*Corresponding author: matthew.carroll@rspb.org.uk
Effects of sea temperature and stratification changes
on seabird breeding success
M. J. Carroll1,*, A. Butler2, E. Owen3, S. R. Ewing4, T. Cole4, J. A. Green5,
L. M. Soanes5, J. P. Y. Arnould6, S. F. Newton7, J. Baer7,11, F. Daunt8, S. Wanless8,
M. A. Newell8, G. S. Robertson9,12, R. A. Mavor10, M. Bolton1
1RSPB Centre for Conservation Science, The Lodge, Sandy, Bedfordshire SG19 2DL, UK
2Biomathematics and Statistics Scotland, The King’s Buildings, Edinburgh EH9 3JZ, UK
3RSPB Centre for Conservation Science, Etive House, Beechwood Park, Inverness IV2 3BW, UK
4RSPB Centre for Conservation Science, Scotland Headquarters, 2 Lochside View, Edinburgh Park, Edinburgh EH12 9DH, UK
5School of Environmental Sciences, University of Liverpool, Nicholson Building, Brownlow Street, Liverpool L69 3GP, UK
6School of Life and Environmental Sciences, Deakin University, Melbourne Burwood Campus, 221 Burwood Highway,
Burwood, VIC 3125, Australia
7BirdWatch Ireland, 20D Bullford Business Campus, Kilcoole, Co. Wicklow, Republic of Ireland
8Centre for Ecology & Hydrology, Bush Estate, Penicuik, Midlothian EH26 0QB, UK
9Institute of Biodiversity, Animal Health and Comparative Medicine, Graham Kerr Building, University of Glasgow,
Glasgow G12 8QQ, UK
10Joint Nature Conservation Committee, Inverdee House, Baxter Street, Aber deen AB11 9QA, UK
11Present address: BioConsult SH, Schobüller Str. 36, 25813 Husum, Germany
12Present address: Game and Wildlife Conservation Trust, Forest-in-Teesdale, Barnard Castle DL12 0HA, UK
ABSTRACT: As apex predators in marine ecosystems, seabirds may primarily experience climate
change impacts indirectly, via changes to their food webs. Observed seabird population declines
have been linked to climate-driven oceanographic and food web changes. However, relationships
have often been derived from relatively few colonies and consider only sea surface temperature
(SST), so important drivers, and spatial variation in drivers, could remain undetected. Further, ex -
plicit climate change projections have rarely been made, so longer-term risks remain unclear.
Here, we use tracking data to estimate foraging areas for 11 black-legged kittiwake Rissa tridac -
ty la colonies in the UK and Ireland, thus reducing reliance on single colonies and allowing calcu-
lation of colony-specific oceanographic conditions. We use mixed models to consider how SST, the
potential energy anomaly (indicating density stratification strength) and the timing of seasonal
stratification influence kittiwake productivity. Across all colonies, higher breeding success was
associated with weaker stratification before breeding and lower SSTs during the breeding season.
Eight colonies with sufficient data were modelled individually: higher productivity was associated
with later stratification at 3 colonies, weaker stratification at 2, and lower SSTs at one, whilst 2
colonies showed no significant relationships. Hence, key drivers of productivity varied among
colonies. Climate change projections, made using fitted models, indicated that breeding success
could decline by 21 to 43% between 1961−90 and 2070−99. Climate change therefore poses a
longer-term threat to kittiwakes, but as this will be mediated via availability of key prey species,
other marine apex predators could also face similar threats.
KEY WORDS: Black-legged kittiwake · Oceanography · Potential energy anomaly · Productivity ·
Rissa tridactyla · SST · Tracking data
Resale or republication not permitted without written consent of the publisher
This authors' personal copy may not be publicly or systematically copied or distributed, or posted on the Open Web,
except with written permission of the copyright holder(s). It may be distributed to interested individuals on request.

Clim Res 66: 75– 89, 2015
Pearce-Higgins et al. 2010), but such impacts can be
harder to predict and observe (Tylianakis et al. 2008,
Gilman et al. 2010). These ‘indirect’ impacts are
likely to be widespread and bring with them substan-
tial extinction risks (Cahill et al. 2013, Ockendon et
al. 2014), but they also pose considerable conserva-
tion challenges. Species at higher trophic levels
attract most attention (Sergio et al. 2008), but their
populations may depend more on species at lower
trophic levels and their abiotic drivers.
Seabirds are the world’s most threatened group of
birds (Croxall et al. 2012). As apex predators, they
are likely to experience indirect climate change
impacts through their supporting food webs (Syde-
man et al. 2012). Their populations are responsive to
changes in breeding success (Sandvik et al. 2012),
which is influenced by prey availability during the
breeding period (Hamer et al. 1993, Regehr & Mon-
tevecchi 1997, Wanless et al. 2004). Under poorer
feeding conditions, body condition is lower, nest
attendance falls, and chicks can starve (Wanless &
Harris 1992, Frederiksen et al. 2004b, Vincenzi &
Mangel 2013). Hence, climatic and oceanographic
changes affecting food webs could impact seabird
productivity. Whilst identifying underlying mecha-
nisms is challenging, it is informative to examine
relationships between physical ocean conditions and
demographic parameters (e.g. Frederiksen et al.
2004b, Wanless et al. 2007), as these can indicate the
ultimate drivers of population declines.
In the UK and Ireland, abundances of several sea-
bird species have fallen substantially since the mid-
1980s (JNCC 2014). Some declines have been linked
to rising sea surface temperatures (SSTs) (e.g. Fred-
eriksen et al. 2004b, 2007). A possible mechanism
behind this is reduced prey availability and nutri-
tional value due to changing zooplankton communi-
ties (Arnott & Ruxton 2002, Wanless et al. 2004, van
Deurs et al. 2009). Although strong relationships with
SST have been derived for individual colonies (Fred-
eriksen et al. 2004b), its importance varies spatially,
with colonies in some regions showing only weak
SST relationships (Frederiksen et al. 2007, Lauria et
al. 2012). Further, other oceanographic drivers, not -
ably density stratification, may also be important
(Scott et al. 2006). Stratification occurs when temper-
ature or salinity differences cause pronounced den-
sity differences between deep and shallow waters.
Associated changes in nutrient availability and light
regimes influence plankton growth, and in turn fish
activity and growth (Scott et al. 2006, Sharples et al.
2006). Under earlier stratification, key fish species
may be available too early or be less nutritious
(Wright & Bailey 1996, Wanless et al. 2004, Scott et al.
2006), whilst abundance of key zooplankton and fish
species may fall under stronger stratification (Beare
et al. 2002, Jensen et al. 2003). To improve under-
standing of the physical drivers of seabird productiv-
ity and identify underlying biological mechanisms, it
is therefore necessary to consider multiple colonies
across multiple regions (Lauria et al. 2012, Sydeman
et al. 2012), and multiple oceanographic variables.
With improved understanding of physical drivers of
productivity, longer-term climate change impacts can
be considered. Longer-term impacts have been im -
plied from observed changes, but few studies have
made explicit projections (but see Frederiksen et al.
2013, Sandvik et al. 2014).A clearer understanding of
future impacts is essential when considering possible
conservation strategies in a changing climate, espe-
cially in light of legislative frameworks that consider
seabird productivity under prevailing climatic condi-
tions (HM Government 2012). Therefore, both ob ser -
ved relationships and explicit climate change pro -
jections are necessary to provide a more complete
under standing of the impacts of oceanographic
change and stochasticity on seabird populations.
Here, we examine drivers of productivity for multi-
ple seabird colonies, considering SST and stratifica-
tion. We consider the black-legged kittiwake Rissa
tridactyla (hereafter ‘kittiwake’), as it is a sensitive
indicator of environmental conditions (Wanless et al.
2007, Cook et al. 2014). We focus on the UK and Ire-
land, which support around 14% of the biogeo-
graphic kittiwake population and for which popula-
tion data are routinely collected (JNCC 2014).
Specifically, we consider the following hypotheses:
(1) Higher SSTs are associated with reduced kitti-
wake breeding success
(2) Strong, early stratification is associated with
reduced kittiwake breeding success
(3) Modelledkittiwake productivity will bere duced
in future scenarios due to the impacts of climate
change
2. MATERIALS AND METHODS
2.1. Study species
Despite being one of the most abundant seabirds in
the UK and Ireland, kittiwakes have declined sub-
stantially since 1986 (JNCC 2014). They nest on cliffs
in colonies of up to tens of thousands of pairs (Coul-
son 2011). Egg-laying occurs from April to June, and
incubation and fledging each take approximately
76
Author copy

Carroll et al.: Seabird breeding success and oceanographic change 77
1 mo (Coulson 2011). During breeding, kittiwakes
feed primarily on fish, with sandeels (Ammodytidae;
particularly the lesser sandeel Ammodytes marinus)
a key prey resource (Furness & Tasker 2000, Wanless
et al. 2007). However, clupeids (e.g. herring, sprat),
gadids (e.g. cod, pollock) and planktonic crustacea
can also be important (e.g. Lewis et al. 2001, Chivers
et al. 2012). Colonies with diverse diets may be
buffered from fluctuating prey availability (Coulson
2011), while those dependent upon a single species
are more likely to be sensitive to climatic variability.
2.2. Kittiwake foraging areas
Previous analyses have extracted oceanographic
predictor variable values from arbitrary areas near
colonies (e.g. Frederiksen et al. 2004b, Burthe et al.
2012, Sandvik et al. 2014). However, seabird tracking
has indicated variability among colonies in the size
and shape of areas used (e.g. Wakefield et al. 2013),
so the area of sea influencing breeding success is also
likely to vary. Hence, here, tracking data were used
to define colony-specific areas.
Data were acquired for 11 colonies where kitti-
wakes were tracked during the 2010−12 breeding
seasons, and for which productivity data were avail-
able (Table 1, Fig. 1). Tracked birds had high-
resolution GPS tags (modified IgotU GT 120, Mobile
Action) attached with adhesive tape to back feathers
whilst at the colony. Tags recorded a location fix
Site Map Region Coordinates Years of Years of Total
site breeding success tracking birds
number data overlapping data tracked
oceanography
Fair Isle 1 Shetland 1.65°W, 59.52°N 19 3 11
Boddam to Collieston 2 East Scotland 1.85°W, 57.42°N 15 1 25
Fowlsheugh 3 East Scotland 2.20°W, 56.92°N 17 1 15
Isle of May NNR 4 East Scotland 2.57° W, 56.18° N 18 1 17
St. Abb’s Head NNR 5 East Scotland 2.13° W, 55.91°N 18 1 15
Coquet Island 6 East England 1.52° W, 55.34°N 12 2 36
Flamborough Head and Bempton Cliffs 7 East England 0.08° W, 54.12° N 18 3 51
Bardsey Island NNR 8 Irish Sea 4.83° W, 52.76° N 17 1 8
Puffin Island 9 Irish Sea 4.03° W, 53.32° N 1 3 70
Lambay 10 Irish Sea 6.03° W, 53.50° N 1 2 14
Isle of Colonsay 11 West Scotland 6.21° W, 56.08° N 6 3 59
Table 1. Sites included in analyses of effects of sea temperature and stratification changes on kittiwake breeding success. ‘Site’
refers to the name in the SMP database, and ‘map site number’ to the location shown in Fig. 1. Regions listed are based on
Frederiksen et al. (2005). Oceanographic data were available up to 2004, whilst breeding success data were available from
1986, meaning that the maximum possible overlap was 19 yr. NNR: National Nature Reserve
Fig. 1. Locations of kittiwake colonies included in the
analyses. Numbers refer to colony descriptions in Table 1
Author copy

Clim Res 66: 75– 89, 2015
accurate to 20 m approximately every 100 s, and
remained attached for 2 to 5 d. Tracking occurred
from May to July, but mostly in June, covering late
incubation and chick rearing periods.
It was assumed that oceanographic conditions pri-
marily affect kittiwake productivity via food webs, so
the most relevant areas from which to extract oceano-
graphic data were those associated with foraging.
Therefore, GPS records were filtered to identify rele-
vant locations. Records within 1 km of the colony cen-
tre were removed to exclude fixes associated with be-
haviours around the nest, which are rarely associated
with foraging (Suryan et al. 2002). Travel speeds be-
tween points were calculated; these formed a bimodal
distribution, with lower speeds likely to be associated
with foraging (e.g. Kotzerka et al. 2010). Based on pre-
liminary analysis of a subset of data, records with
speeds over 14 km h−1 were removed (see Supple-
ment 1 at
www. int-res. com/ articles/ suppl/ c066 p075 _
supp. pdf
). Filtering left 192 638 records. Although fil-
tering did not exclude behaviours such as resting on
the sea, the range of kittiwake foraging behaviours
(Coulson 2011) makes a more inclusive approach
preferable. A sensitivity analysis indicated that
threshold selection made little difference to extracted
oceanographic variable values (Supplement 1), so
analyses presented here should be robust to threshold
specification within the ranges considered.
Kernel density estimates (KDEs) were calculated to
convert GPS records into estimated foraging areas.
For each colony, data were pooled across all birds
and years to estimate the ‘core’ foraging area; whilst
interannual variation was found, most colonies used
similar areas each year (Supplement 2), so pooling
was considered appropriate. Although kernel density
estimation is sensitive to the number of birds in -
cluded, all colonies had at least the number re quired
to describe >50% of the ‘true’ foraging area (Soanes
et al. 2013). Kernel densities were evaluated on a
regular 30 arc-second by 30 arc-second rectangular
grid with limits 1.25° away from the most extreme
observations.
KDEs were based on a bivariate Gaussian kernel,
and were evaluated using the ‘ks’ R package (Duong
2013). A bivariate plug-in estimator (Duong & Hazel-
ton 2003) and a rule-of-thumb approach (Silverman
1986) were considered for choosing the degree of
smoothing. The rule-of-thumb approach took band-
width to be 1.06·σx·n(–0.2) and 1.06 · σy·n(−0.2),where n
denotes sample size and σxand σydenote standard
deviations of longitudes and latitudes; this is derived
in a univariate setting under an assumption of nor-
mality, so should be interpreted cautiously here.
However, the plug-in was highly computationally
intensive for datasets of this size, so the approaches
were compared using a subset of sites: extracted
oceanographic data were highly correlated (r ≥0.99),
so the rule-of-thumb approach was used for all sites.
Foraging areas were defined by the 90% density
contour, which has been recommended for home
range estimates (Börger et al. 2006). Kernels are pre-
sented in Supplement 2.
2.3. Kittiwake breeding success data
Breeding success data were acquired from the sea-
bird monitoring programme (SMP; http://jncc. defra.
gov.uk/smp; Walsh et al. 1995). The SMP is an an -
nual sample survey of seabird breeding population
size and productivity, which started in 1986 and is
coordinated by the Joint Nature Conservation Com-
mittee (JNCC). Data from the Isle of May National
Nature Reserve were acquired from the Centre for
Ecology&Hydrology (https://eip.ceh.ac.uk/, ac ces sed
12 Apr 2013). Productivity data were not available
for all years for all colonies, leaving 142 site-by-year
combinations (Table 1).
SMP breeding success is often analysed as mean
fledged chicks per nest (e.g. Frederiksen et al. 2007).
However, it was preferable to avoid this here, as
Gaussian responses could become negative in pro-
jections, and varying numbers of nests contributed to
observations (range 21 to 1446). Therefore, numbers
of fledged and failed chicks were modelled as a bino-
mial response, with fledged chicks taken from the
data, and failed chicks estimated as [(2 × nests) −
fledged], based on the mean and modal UK kittiwake
clutch size of 2 (range 1 to 3; Coulson & Porter 1985,
Coulson 2011), thus preventing negative predictions
and allowing prior weights to account for varying
nest numbers. Hence, breeding success was mod-
elled as chicks fledged per egg (Cook et al. 2014). To
ensure results were robust to these assumptions,
fledged chicks were also modelled as a Poisson re -
sponse with an offset of log(nests); results were very
similar to the binomial analysis (Supplement 3).
2.4. Oceanographic data
Two oceanographic datasets were acquired: one
covered recent years (hereafter, ‘hindcast’), whilst
the other covered 30 yr periods for the mid 20th and
late 21st centuries (hereafter, ‘projections’). Both
were produced from the Proudman Oceanographic
78
Author copy

Carroll et al.: Seabird breeding success and oceanographic change 79
Laboratory Coastal Ocean Modelling System (POL-
COMS), which simulates ocean hydrodynamics as
driven by atmospheric inputs (Holt & James 2001).
Data acquired were monthly mean temperature and
salinity on a 1/6° longitude × 1/9° latitude grid (~12 ×
12 km) over multiple vertical layers.
Hindcastdatawereacquired from the MyOcean pro-
ject (http://marine.copernicus.eu/, product NORTH
WESTSHELF_REANALYSIS_PHYS_004_005,access -
ed 23 Apr 2013), and represented an estimate of con-
ditions experienced between 1967 and 2004, so could
be used to establish relationships with kittiwake pro-
ductivity. Further information on this dataset is pro-
vided by Holt et al. (2012). Projection data were
acquired from the British Atmospheric Data Centre
(http://badc.nerc.ac.uk/data/link, accessed 01 Mar
2013; access provided by the UK Met Office), and
represented baseline (1961−90) and future (2070−99,
SRES Scenario A1B) periods. Projections did not cor-
respond to conditions in specific years, so could only
be used to predict breeding success under average
conditions in each period. Further information on this
dataset is provided by Lowe et al. (2009).
2.5. Explanatory variables
Three oceanographic variables that could influ-
ence kittiwake productivity were calculated: SST (e.g.
Frederiksen et al. 2004b), stratification strength, and
the timing of seasonal stratification onset (e.g. Scott
et al. 2006, 2010). SST was calculated by ex tracting
the top layer of temperature data.
Stratification strength was expressed using the
potential energy anomaly (PEA), as defined by Holt
et al. (2010). PEA indicates the energy per unit depth
required to mix the water column. Hence, higher val-
ues indicate stronger stratification. PEA was calcu-
lated as
(1)
Here, g= gravitational acceleration, h= water
depth (or 400 m if hexceeds this), z= the vertical
coordinate (0 indicating the surface, negative values
indicating deeper water), ρ= density (calculated
using a polynomial function [Jackett et al. 2006]), T=
temperature, S= salinity; the overbar indicates that
the quantity is averaged from hto the surface. As
data were available for discrete depths, the integral
was evaluated numerically using Simpson’s rule.
Seasonal stratification onset was calculated simi-
larly to previous analyses of POLCOMS data (Lowe
et al. 2009, Holt et al. 2010), but as daily outputs were
unavailable, additional assumptions were made. Stra -
tification onset was defined as the first day of the year
with mixed layer depth (MLD) < 50 m (Holt et al. 2010).
MLD was defined as the depth at which density dif-
fered from surface density by an amount equivalent
to a 0.5°C temperature reduction. Only monthly out-
puts were available, so daily MLD values were inter-
polated by fitting a cubic spline through monthly val-
ues; whilst this retains the seasonal pattern of MLD,
it may underestimate true variability. Hence, whilst
the stratification onset metric is relatively coarse,
variability among years and sites should be ade-
quately described.
For SST and PEA, winter and spring means were
calculated. Winter (December, January, February)
corresponded to the period important for sandeel
spawning and egg hatching (Arnott & Ruxton 2002).
Spring (March, April, May, June) corresponded to
the period when kittiwakes commence breeding,
sandeel larvae grow and sandeel abundance peaks
(Wright & Bailey 1996, Coulson 2011). For stratifica-
tion onset, only annual means could be defined.
As well as oceanographic influences, breeding suc-
cess could be influenced by density-dependence,
with reduced productivity at higher population sizes
(Furness & Birkhead 1984). Therefore, for the subset
of sites and years with SMP data on kittiwake breed-
ing population size available (9 colonies; 78 site-by-
year combinations), log(population) was considered
as a further predictor variable (Supplement 4). Across
all sites and at 3 of 4 individual colonies, there was no
significant relationship between population size and
breeding success; at the remaining colony, a positive
relationship was found.Relationships be tween breed-
ing success and oceanographic variables were not
influenced by inclusion of population size. Conse-
quently, in the present study there is little evidence
of density-dependent effects on breeding success
(Supplement 4); due to the much-restricted dataset
involved in this analysis, further discussion relates to
models excluding population size.
2.6. Statistical analysis
Analyses were conducted in R version 3.1.0 (R Core
Team 2014). Mean oceanographic variable values
within foraging areas were calculated using the
‘raster’ R package (Hijmans 2013). Variables were ex-
plored for collinearity and temporal trends (Supple-
ment 5). PEA values displayed skewed distributions,
so logged and untransformed values were compared
,,d
0
g
hzTzSz TS z
zh
∫
[]
{}
[]
() ()
−ρ −ρ
=−
Author copy

Clim Res 66: 75– 89, 2015
80
in preliminary productivity models (Sup ple ment 5):
logged PEA performed better, so further models used
log(PEA). Previous analyses have shown that vari-
ables with and without a 1 yr lag may influence pro-
ductivity (Frederiksen et al. 2004b), so both were tri-
alled: relationships were similar, but lagged variables
produced higher AICs (Supplement 5), so further
analyses considered unlagged variables.
Breeding success was modelled using generalised
linear mixed models (GLMMs) with binomial error
and logit link. Models were fitted using the ‘lme4’ R
package (Bates et al. 2014), with time as a predictor
to identify temporal trends, and then with oceano-
graphic predictors to explore drivers of productivity.
Models were first fitted for individual sites, consider-
ing single predictors only. Then, equivalent single-
predictor models were fitted using data from all sites.
Finally, multiple-site models were fitted with multi-
ple predictors, to allow a more complete examination
of oceanographic drivers.
For single-site models, only colonies with ≥10 yr of
productivity and oceanography data were used. Data
were deemed insufficient to include multiple ex plan -
atory variables (minimum 12 data points, maximum
19), so only single predictors were considered. An
observation-level factor was included as a random
effect to model overdispersion in the response (e.g.
Browne et al. 2005). Variable influence was as sessed
by comparing sample-size-corrected Akaike infor-
mation criterion (AICc) to that from a null model with
intercept and random effects only: ΔAICc ≤0 was
considered to indicate some support, and ΔAICc ≤−2
to indicate substantial support.
To account for spatial and temporal structuring of
data, models including data from all sites were fitted
with ‘site’, ‘region’, ‘year’, ‘site × year’ and ‘region ×
year’ random effects. ‘Site × year’ was an observa-
tion-level factor to model overdispersion. ‘Region’
was in cluded to account for spatial clustering of
colonies, and was based on regions previously iden-
tified from kittiwake productivity trends (Frederik-
sen et al. 2005); if a region was not stated for a spe-
cific site, the nearest region was used. These models
were as ses sed by comparing uncorrected AIC (due
to the larger sample size) to that from a null model.
Next, models were fitted with multiple predictors.
Interaction terms were not considered, as this would
lead to over fitting and reduce interpretability.
Model comparison was conducted using the
‘MuMIn’ R package (Barton 2014); performance was
assessed by comparing AIC values to that from the
model with lowest AIC, with ΔAIC ≤2 considered to
indicate similar support.
2.7. Climate change projections
Climate change impacts were estimated using the
multiple-predictor models. To account for model and
parameter uncertainty, a randomisation procedure
with 1 000000 runs was used: on each run, one model
was picked with probability equal to its Akaike
weight, and new parameter estimates were simu-
lated. Fixed effect estimates were simulated from a
multivariate normal distribution, with mean and
covariance matrix taken from the chosen model, using
the ‘mvtnorm’ R package (Genz et al. 2014). ‘Site’ and
‘region’ effects were extracted from the model, whilst
‘year’, ‘site × year’ and ‘region × year’ were simulated
from normal distributions with mean = 0 and standard
deviations taken from the model.
Simulated parameters were applied to oceano-
graphic projections to produce breeding success esti-
mates for ‘baseline’ and ‘future’ periods. As these
periods represented average conditions, the mean
across all years in each period was calculated. Pro-
portional change in breeding success was calculated
as [(future − baseline)/baseline]; probability of de -
cline was examined by calculating the proportion of
randomisation runs that did not show a decline
between baseline and future periods. Differences
between periods were tested using Wilcoxon rank
sum tests.
3. RESULTS
3.1. Temporal trends and cross correlations
Across all sites, breeding success showed no signif-
icant temporal trend (p = 0.141; Supplement 5).
Spring SST increased significantly (p = 0.026), and
winter SST showed a non-significant increase (p =
0.054). Winter PEA showed a weakly significant
increase (p = 0.046), but spring PEA (p = 0.173) and
stratification onset (p = 0.096) showed no significant
change.
Breeding success decreased significantly at Flam-
borough Head, Fowlsheugh and St. Abb’s Head, but
increased at Bardsey Island (0.003 ≤p ≤0.047). Win-
ter SST increased significantly at Bardsey Island,
Coquet Island, Flamborough Head and Lambay
(0.029 ≤p ≤0.043), whilst spring SST increased sig-
nificantly at Bardsey Island, Flamborough Head and
Puffin Island (p < 0.01). Winter PEA increased signif-
icantly at Isle of May (p = 0.016) and St. Abb’s Head
(p = 0.048), but spring PEA showed no trends. Strati-
fication onset became significantly earlier at Boddam
Author copy

Carroll et al.: Seabird breeding success and oceanographic change
to Collieston, Fowlsheugh and Isle of May (0.014 ≤
p≤0.020).
Correlations between variables were moderate or
weak (Supplement 5), with the highest between win-
ter and spring PEA (ρ= 0.669), winter and spring SST
(ρ= 0.672), and stratification onset and PEA (spring
ρ= −0.559; winter ρ= −0.485), so it was considered
acceptable to include multiple predictors in the same
model. Strong or moderate correlations were found
between lagged and unlagged forms of all variables
(0.647 ≤ρ≤0.950).
3.2. Single predictor variable models
The strongest predictor of breeding success dif-
fered among sites (Table 2; Supplement 6). Stratifica-
tion onset provided the best model at Isle of May and
St. Abb’s Head, with higher productivity associated
with later stratification. Spring PEA provided the best
model at Flamborough Head, whilst winter PEA pro-
vided the best model at Bardsey Island, with both
showing higher productivity to be associated with
lower PEA; winter PEA attained significance at
Coquet Island but was not supported over the null
model. Spring SST provided the best model at Fair
Isle, showing higher breeding success was associ-
ated with lower SSTs. Winter SST did not perform
better than the null model at any site. At Boddam to
Collieston and Fowlsheugh, no variable performed
better than the null model.
The best all-sites single-predictor model showed
higher breeding success with lower winter PEA
(Table 2; Fig. 2). A similar relationship was found
with spring PEA, but the model received less sup-
port. There was also evidence of a negative relation-
ship with spring SST and a positive relationship with
stratification onset (Table 2). Therefore, breeding
success was higher under lower SSTs, later stratifica-
tion and when the water column was better mixed
early in the year.
3.3. Multiple predictor variable models
The best multiple-predictor model (Table 3; Sup-
plement 6) contained significant, negative coeffi-
cients for winter PEA and spring SST, showing that
higher breeding success was associated with weaker
stratification before breeding and lower SSTs during
breeding. Three other models showed similar empir-
ical support: all contained significant, negative coef-
81
Null model Spring PEA Spring SST Strat. onset Winter PEA Winter SST
Bardsey Island AICc = 187.621 −1.719 ± 2.266 1.311 ± 0.877 0.041 ± 0.019 −1.645 ± 0.693 0.928 ± 0.579
ΔAICc = 2.420 ΔAICc = 0.807 ΔAICc = −1.356 ΔAICc = −2.090 ΔAICc = 0.516
Boddam to Collieston AICc = 178.476 −0.123 ± 1.414 0.057 ± 0.593 0.024 ± 0.018 −0.141 ± 0.488 0.175 ± 0.498
ΔAICc = 3.174 ΔAICc = 3.172 ΔAICc = 1.561 ΔAICc = 3.099 ΔAICc = 3.059
Coquet Island AICc = 103.824 1.228 ± 0.968 −0.061 ± 0.351 0.018 ± 0.014 −0.697 ± 0.346 −0.075 ± 0.315
ΔAICc = 2.140 ΔAICc = 3.636 ΔAICc = 1.992 ΔAICc = 0.109 ΔAICc = 3.610
Fair Isle AICc = 278.788 −13.414 ± 5.33 −4.280 ± 1.189 0.042 ± 0.058 −0.942 ± 1.295 −3.661 ± 1.474
ΔAICc = −3.316 ΔAICc = −8.679 ΔAICc = 2.348 ΔAICc = 2.336 ΔAICc = −2.561
Flamborough Head AICc = 225.489 −2.502 ± 0.909 −0.663 ± 0.300 −0.023± 0.029 0.253 ± 0.509− 0.434 ± 0.393
and Bempton Cliffs ΔAICc = −3.417 ΔAICc = −1.416 ΔAICc = 2.321 ΔAICc = 2.668 ΔAICc = 1.733
Fowlsheugh AICc = 214.311 −1.176 ± 1.244 −0.239 ± 0.407 0.013 ± 0.020 −0.388 ± 0.451 −0.270 ± 0.366
ΔAICc = 2.117 ΔAICc = 2.264 ΔAICc = 2.561 ΔAICc = 2.263 ΔAICc = 2.453
Isle of May AICc = 254.784 0.689 ± 2.371 −0.488 ± 0.601 0.092 ± 0.030 −1.478 ± 1.192 −0.283 ± 0.535
ΔAICc = 2.830 ΔAICc = 2.264 ΔAICc = −4.855 ΔAICc = 2.738 ΔAICc = 2.636
St. Abb’s Head AICc = 230.539 −1.177 ± 1.241 −0.024 ± 0.361 0.034 ± 0.013 −1.085 ± 0.613 −0.122 ± 0.328
ΔAICc = 2.034 ΔAICc = 2.910 ΔAICc = −2.665 ΔAICc = 0.029 ΔAICc = 2.777
All sites AIC = 1803.730 −0.602 ± 0.285 −0.700 ± 0.264 0.014 ± 0.007 −0.641 ± 0.201 −0.240 ± 0.231
ΔAIC = −2.669 ΔAIC = −5.242 ΔAIC = −3.383 ΔAIC = −11.502 ΔAIC = 0.994
Table 2. Results from models relating breeding success of kittiwakes to single oceanographic predictor variables. See text for
model fitting details. Parameter estimates (±SE) are given, followed by ΔAIC (for all-sites models) or ΔAICc (for individual site
models) relative to a null model fitted with intercept and random effects only. Bold: significantly different from 0 at p < 0.05, as in-
dicated by Wald Ztests; italics : close to significance with 0.05 ≤p < 0.1. Full model details are given in Supplement 6. PEA:
potential energy anomaly; SST: sea surface temperature
Author copy

Clim Res 66: 75– 89, 2015
ficients for winter PEA and spring SST, and one non-
significant variable. The second-ranked model
(ΔAIC = 1.649) contained a non-significant positive
ef fect of winter SST, contrasting with single predictor
models; this possibly reflects collinearity between
winter and spring SST. The third-ranked model
(ΔAIC = 1.861) contained a non-significant positive
coefficient for stratification onset, whilst the fourth-
ranked model (ΔAIC = 1.926) showed a non-signifi-
cant negative effect of spring PEA. Therefore, results
highlighted the importance of lower winter PEAs and
spring SSTs for kittiwake productivity.
3.4. Climate change
projections
Projections indicated that cli-
mate change could drive sub-
stantial productivity declines
(Table 4; Fig. 3). For the baseline
period, mean projected breeding
success across all sites was 0.560
(~1.12 chicks per pair); by 2070−
99, this had de clined by 32.6% to
0.377 (~0.754 chicks per pair).
Only 3.0% of simulations did not
predict a decline.
All sites showed projected de-
clines (Table 4). The largest pro-
portional decline was for Fair Isle
(43.2%), whilst the smallest was
at Coquet Island (21.4%). The
largest absolute decline was at
Flamborough Head (−0.214), and
the smallest was at Boddam to
Collieston (−0.161). At Bardsey
Island and Fair Isle, only 1.8 and
1.1% of simulations respectively
did not predict a decline, whilst
for Boddam to Collieston, Coquet
Island, Fowlsheugh, Isle of May
and St. Abb’s Head, between 7.9
and 16.9% of simulations did not
predict declines. Therefore, the
magnitude and probability of de-
clines varied among sites.
Neither stratification onset nor
winter PEA changed signifi-
cantly between periods (Fig. 3).
Spring PEA increased signifi-
cantly (Fig. 3), but the absolute
changewassmall (1961−90 mean
10.02 J m−3 [log scale 2.034];
2070− 99 mean 12.13 J m−3 [log
scale 2.215]) and spring PEA coefficients in high-
ranking models were small. Hence, these 3 variables
changed too little or had too little an effect on pro-
ductivity to drive the projected productivity declines.
SST increased significantly in spring (1961− 90 mean
7.95°C; 2070−99 mean 10.46°C; Fig. 3) and winter
(1961− 90 mean 7.08°C; 2070−99 mean 9.58°C;
Fig. 3); spring SST model coefficients were large and
negative, whilst winter SST coefficients were small
and positive or large and negative. Hence, rising
SSTs appeared to be the major driver of projected
declines.
82
Fig. 2. Plots of breeding success against
oceanographic predictor variables with
no lag, along with fitted lines from bino-
mial generalised linear mixed models
(GLMMs) including the ‘site’ and ‘re-
gion’ random effects. Each point repre-
sents one site-by-year observation; point
sizes are scaled by log(nests surveyed) to
reflect weightings of observations in
models. Dashed lines: fitted relationship
from GLMMs, taking into account 'site'
and 'region' random effects; each line
represents a single site
Author copy

Carroll et al.: Seabird breeding success and oceanographic change
4. DISCUSSION
Weaker, later stratification and lower SSTs were
associated with higher kittiwake productivity. Indi-
vidual colonies also showed such relationships, but
the most important driver varied among colonies.
Projections indicated that climate change could drive
longer-term productivity declines. The analytical
ap proach reduced reliance on intensively-studied
colonies, accounted for colony-specific habitat use,
allowed examination of spatial heterogeneity, and
considered short- and longer-term effects; thus pro-
viding a more complete examination of drivers of kit-
tiwake productivity. The study therefore provides an
example of how changing physical conditions, pre-
sumably acting via supporting food webs, can influ-
ence apex predators, leading to indirect climate
change impacts.
83
Intercept Spring PEA Spring SST Stratification Winter PEA Winter SST AIC ΔAIC Weight
onset date
4.429 ± 2.181 − −0.539 ± 0.244 − −0.602 ± 0.190 − 1789.734 0 0.263
p = 0.042 p = 0.027 p = 0.002
4.308 ± 2.185 − −0.674 ± 0.336 − −0.609 ± 0.192 0.173 ± 0.295 1791.383 1.649 0.115
p = 0.049 p = 0.045 p = 0.001 p = 0.556
4.206 ± 2.269 − −0.544 ± 0.245 0.003 ± 0.008 −0.566 ± 0.214 − 1791.595 1.861 0.104
p = 0.064 p = 0.027 p = 0.712 p = 0.008
4.706 ± 2.408 −0.090 ± 0.333 −0.541 ± 0.244 − −0.574 ± 0.217 − 1791.659 1.926 0.100
p = 0.051 p = 0.786 p = 0.027 p = 0.008
−0.677 ± 0.268 − − − − − 1803.730 15.336 0.000
p = 0.012
Table 3. Top-ranked models from the all-sites analysis relating breeding success of kittiwakes to oceanographic variables.
Those shown have ΔAIC ≤2 relative to the best model; the null model, fitted with intercept and random effects only, is shown
at base of table for comparison. See text for details of model fitting. Parameter estimates (±SE) are given, followed by p values
from Wald Ztests. Bold: p < 0.05; italics: 0.05 ≤p < 0.1. AIC values are shown, and corresponding ΔAIC values relative to the
best model. Weight: Akaike weight of each model in the full set, with higher values indicating higher relative support for that
model. Full details are in Supplement 6
Site Breeding success (±SD) Absolute Percentage Proportion
1961−1990 2070−2099 change change not showing
decline
Bardsey Island 0.426 ± 0.090 0.246 ± 0.121 −0.181 −42.4 0.018
Boddam to Collieston 0.578 ± 0.109 0.418 ± 0.107 −0.161 −27.8 0.169
Coquet Island 0.776 ± 0.077 0.610 ± 0.123 −0.166 −21.4 0.125
Fair Isle 0.431 ± 0.091 0.245 ± 0.068 −0.186 −43.2 0.011
Flamborough Head and Bempton Cliffs 0.591 ± 0.108 0.378 ± 0.112 −0.214 −36.1 0.028
Fowlsheugh 0.606 ± 0.106 0.442 ± 0.109 −0.164 −27.0 0.168
Isle of Colonsay 0.535 ± 0.101 0.350 ± 0.104 −0.185 −34.6 0.035
Isle of May 0.492 ± 0.097 0.308 ± 0.084 −0.183 −37.3 0.098
Lambay 0.500 ± 0.077 0.318 ± 0.139 −0.182 −36.4 0.087
Puffin Island 0.633 ± 0.106 0.437 ± 0.158 −0.197 −31.0 0.026
St. Abb’s Head 0.592 ± 0.088 0.401 ± 0.097 −0.191 −32.2 0.079
Across all sites 0.560 ± 0.074 0.377 ± 0.095 −0.183 −32.6 0.030
Table 4. Projected breeding success for the UKCP09 climatic baseline period of 1961−90 and for 2070−99 under the SRES A1B
scenario. Reported breeding success values are the mean of 100 000 randomisation runs, where each run produces a mean
breeding success across all years in the time period; breeding success is here defined as the proportion of successfully fledged
chicks. The standard deviation of 1 000 000 projections is also given. Percentage change is calculated as [(future –
baseline)/baseline]×100, based on the mean for each period. To indicate the probability of decline, the difference between the
baseline and future projections was calculated for each run, and the proportion of these differences > 0 (i.e. those not showing
a decline) was calculated. See ‘Materials and methods’ for randomisation procedure details
Author copy

Clim Res 66: 75– 89, 2015
4.1. Use of colony-specific areas
Previous analyses have extracted oceanographic
data from arbitrary areas or broad regions (e.g. Fred-
eriksen et al. 2004a, Lauria et al. 2012), but here,
colony-specific areas were produced. This allowed
the analysis to reflect observed habitat use, but sev-
eral caveats apply when interpreting results. It was
assumed that colonies use foraging areas consis-
tently, but foraging locations may
vary (e.g. Ainley et al. 2003, Robert-
son et al. 2014). However, kitti-
wakes can display high foraging
site fidelity (Irons 1998) and kernels
were often similar among years
(Supplement 2), indicating that
‘core’ foraging areas may re tain
importance. Further, kernel density
estimation is sensitive to the num-
ber of birds, trips and years in -
cluded (Soanes et al. 2013, Bogda -
nova et al. 2014), so areas estimated
here may not adequately represent
‘whole colony’ foraging areas.
However, all colo nies passed the
threshold required to estimate
>50% of the core foraging area,
and many passed that required for
estimating 95% (Soanes et al.
2013). Collection of further tracking
data could resolve such issues, pro-
viding increased understanding of
spatiotemporal variability in forag-
ing areas and more robust kernel
estimates. Finally, if prey species
are transported or migrate into for-
aging areas, physical conditions
elsewhere could be more important
in determining prey availability.
However, after settlement, adult
sand eels do not move to other
areas, and larval sandeel transport
towards the UK is limited (Chris-
tensen et al. 2008), so local condi-
tions are likely to remain important
in areas where sandeels dominate
seabird diets. Improved under-
standing of seabird diet, and the
population dynamics of key prey
species, could help to clarify such
uncertainties.
4.2. Drivers of kittiwake productivity
As in previous analyses (e.g. Frederiksen et al.
2004b) a negative relationship between breeding
success and SST was found. However, the strongest
relationship showed a negative relationship with
winter PEA. This suggests that examining multiple
variables is necessary to improve our understand-
ing of physical drivers of kittiwake productivity,
84
Fig. 3. Boxplots comparing oceanographic variables and projected breeding
success between 1961−90 and 2070−99. For plots of oceanographic variables,
input values were 30 yr of projection data for each foraging area used in all-
sites analyses; for breeding success, input values were 1 000000 annual breed-
ing success projections (see Section 2.7 for details). Boxes indicate interquartile
range (IQR) and median; whiskers indicate 1.5 × IQR; outliers indicate points
outside 1.5 × IQR. Results of Wilcoxon rank sum tests are shown at top of each
diagram, indicating whether there is a significant difference between periods
Author copy

Carroll et al.: Seabird breeding success and oceanographic change
and the biological mechanisms through which they
act.
Stratification timing and strength are likely to
inter act to influence feeding conditions. Seasonal
stratification influences plankton growth, which can
in turn cause fish to move towards the surface to feed
(e.g. Greenstreet et al. 2006, Buren et al. 2014).
Hence, early stratification can cause a mismatch be -
tween peak fish availability or size and the seabird
breeding period (Scott et al. 2006, Burthe et al. 2012).
Although seabirds can adjust the timing of breeding,
such changes may not be sufficient to track prey
availability, leading to phenological mismatch (Bur-
the et al. 2012). Relationships with winter PEA may
themselves reflect timing effects, with high PEA val-
ues simply indicating areas likely to stratify early.
However, kittiwakes avoid foraging in very strongly
stratified areas (Scott et al. 2010), suggesting that
stratification strength could directly affect breeding
success. Strong stratification could reduce sandeel
availability, as larvae are more abundant in weakly-
stratified surface waters (Jensen et al. 2003), and
oxygen deficits under stronger stratification reduce
habitat suitability (Behrens et al. 2009). Stronger
stratification is also associated with lower abundance
of Calanus finmarchicus (Beare et al. 2002), a key
prey species for North Sea forage fish (e.g. van
Deurs et al. 2009). As stratification is likely to become
stronger and earlier under climate change (Lowe et
al. 2009), investigating mechanisms linking stratifi-
cation, fish and seabirds is a priority.
It has been suggested that SST relationships could
reflect stratification conditions (Scott et al. 2006), but
the best models here included both PEA and SST, in -
dicating that temperature has an independent effect.
For sandeels, increased metabolic costs at higher
temperatures may inhibit growth or cause them to
remain buried in the sediment (Greenstreet et al.
2006), and can reduce recruitment (Arnott & Ruxton
2002). Higher temperatures also influence plankton
communities, with smaller, less nutritious species
replacing larger, cold-adapted species (Beau grand et
al. 2002, Morán et al. 2010); such changes could
reduce fish survival or growth. It should also be noted
that if climate change affects the distribution of tem-
perature through the water column, stratification
could itself be affected by temperature increases
(Lowe et al. 2009). It therefore appears beneficial to
consider both temperature and stratification effects
on food webs when considering drivers of seabird
productivity.
Single-site models highlighted spatial variation in
oceanographic drivers of productivity, but where for-
aging areas overlapped, similar patterns were ob -
served. At Isle of May and St. Abb’s Head, which
overlapped somewhat (Supplement 2), stratification
onset provided the best model, whilst at Boddam to
Collieston and Fowlsheugh, which overlapped sub-
stantially, no relationships were significant. This sup-
ports the idea that clustering of kittiwake population
trends is driven by local foraging conditions (Fred-
eriksen et al. 2005). Further, only Isle of May and St.
Abb’s Head showed a lagged variable to perform
better than the unlagged equivalent (Supplement 5).
Similar results have previously been taken to show
that 1-group sandeels influence productivity more
than 0-group (Frederiksen et al. 2004b); weak lagged
effects elsewhere imply that other colonies may rely
more on 0-group sandeels or other species. More sea-
bird diet data are required to improve understanding
of such spatial patterns.
4.3. Climate change impacts
Projections indicated that kittiwake productivity
could decline by 21 to 43% between the mid 20th
and late 21st centuries. The largest absolute decline
was projected for Flamborough Head, likely reflect-
ing the strong warming forecast there (Lowe et al.
2009). Smaller declines, with lower probabilities of
occurrence, were projected for colonies further up
the east coast, but the largest proportional decline
occurred at Fair Isle, indicating that larger impacts
may not be limited to southerly colonies. Indeed, as
dramatic declines have already occurred in northern
Scotland (JNCC 2014), these colonies are likely to
face the greatest climate change threats.
Between 1986 and 2008, UK kittiwake productiv-
ity declined by 31% (Cook & Robinson 2010), com-
parable to declines projected here over longer
timescales. This does not, however, indicate that
declines have reached their maximum: realised
magnitudes of longer-term declines will be deter-
mined by factors including anthropogenic influences
(e.g. Furness & Tasker 2000) and adult condition
(Frederiksen et al. 2004a). Notably, although no
density-dependence was found in the present study
or several previous studies of kittiwakes (Frederik-
sen et al. 2005, Sandvik et al. 2014), density-depen-
dence could exacerbate or ameliorate productivity
de clines, through processes such as reduced local
competition for food in smaller populations (Furness
& Birkhead 1984), or reduced threats from predators
in larger populations (Massaro et al. 2001). There-
fore, further information about how kittiwake be -
85
Author copy

Clim Res 66: 75– 89, 2015
86
haviour and breeding success interact with popula-
tion size could be important in understanding popu-
lation-scale impacts of climate change. Further, var-
ious methodological processes and assumptions
influence the magnitude of projected declines. Pro-
jections describe 30 yr means for 11 colonies,
whereas observed decline estimates are based on
individual years of data for over 50 colonies (Cook &
Robinson 2010). Data were extracted from recent
foraging areas, but birds might shift their foraging
areas under climate change to track prey. However,
if kittiwakes remain reliant upon sandeels, it is
unlikely that important new areas will emerge due
to patchy distribution of sandeels, the sparse distri-
bution of sandeel habitat and limited transport
among sandbanks (Christensen et al. 2008); shifts to
new dominant prey species cannot be predicted
using currently-available data. Finally, climate pro-
jections represented only one possible future sce-
nario, so cannot account for the full range of condi-
tions that may be experienced, and whilst the
projections present a plausible future scenario, they
are subject to uncertainty (Holt et al. 2012) so
realised future conditions may differ from projec-
tions. Overall, however, results indicate that climate
change is expected to reduce kittiwake productivity
in the longer term.
Although projections suggest that climate change
will drive declines in breeding success, the conser-
vation status of kittiwake populations will be influ-
enced by more than just productivity. Adult and
juvenile survival declines under higher SSTs (Fre -
derik sen et al. 2004b, Sandvik et al. 2014), and pop-
ulation size is sensitive to declining survival (Sand -
vik et al. 2012). Hence, if rising temperatures drive
declines in both productivity and survival, abun-
dances could fall very rapidly. If, by contrast,
warmer temperatures cause higher adult survival,
as has been found in some cases (Sandvik et al.
2014), population trends may be somewhat buffered
from declining productivity. There may also be
impacts on individual-level responses such as stress
hormone levels (Brewer et al. 2008) and chick
development rates (Vincenzi & Mangel 2013); such
responses could combine to produce substantial
population-level effects. Collection of data on these
other demographic parameters, and examination of
how they interact with SST and stratification, could
prove highly informative in understanding popula-
tion-level climate change impacts.
Global SSTs are projected to increase by 1 to 3°C
by the end of the 21st century (Collins et al. 2013),
so further impacts on seabirds may be unavoidable.
However, appropriate marine management could
ameliorate some negative effects. Sandeel fisheries
can reduce seabird productivity (Frederiksen et al.
2004b, Daunt et al. 2008), so any action that reduces
prey abundance in key foraging areas is also likely
to affect seabirds. With improved knowledge of for-
aging locations, it may be possible to grant impor-
tant areas enhanced environmental protection, min-
imising negative anthropogenic influences on fish
populations, and thus providing a more resilient
food web; this is in line with previous recommenda-
tions for marine climate change adaptation (Mawds-
ley et al. 2009). Establishing marine management
strategies to promote healthy forage fish populations
may provide the best approach for conserving kitti-
wakes and other apex predators under uncertain
future conditions.
5. CONCLUSIONS
This study suggests that weaker, later stratification
and lower SSTs are beneficial for kittiwake produc-
tivity, and that climate change is a longer-term
threat. Kittiwakes are surface-feeding apex preda-
tors, so some findings may be primarily relevant to
similar species: if oceanographic changes reduce
prey availability near the surface, this may explain
why surface feeders such as kittiwakes and Arctic
terns Sterna paradisaea appear most sensitive to
changing conditions (Enstipp et al. 2006). If, how-
ever, overall prey abundance or quality is reduced,
more species could be affected. Indeed, declines
have been observed in North Sea harbour seal Phoca
vitulina populations (Lonergan et al. 2007), increased
harbour porpoise Phocoena phocoena starvation
might be linked to reduced sandeel availability
(MacLeod et al. 2007b; but see MacLeod et al. 2007a,
Thompson et al. 2007), and productivity of guillemots
Uria aalge and razorbills Alca torda has declined
(JNCC 2014). These findings suggest that impacts of
changing oceanographic conditions on marine food
webs affect more than just surface-feeding birds. Cli-
mate change could therefore have substantial eco-
system-wide impacts.
This study provides an example of possible indirect
climate change impacts, with effects mediated via
supporting food webs. Such impacts are possible
whenever predators depend upon prey species that
are sensitive to climate change, and may be more
important than previously understood (Cahill et al.
2013, Ockendon et al. 2014). Given the complexity
as sociated with identifying and understanding these
Author copy

Carroll et al.: Seabird breeding success and oceanographic change
impacts, there is an urgent need to investigate biotic
mechanisms linking physical drivers to higher con-
sumers. By identifying the specific physical condi-
tions, prey species and community changes that
drive population-level responses in apex predators,
we may be better-able to target conservation actions.
If appropriate management allows apex predators to
maintain high productivity in some years, it may still
be possible to ameliorate population-level impacts of
climate change.
Acknowledgements. This study was jointly funded by the
RSPB and Natural England (NE) through the Action for
Birds in England partnership. Tracking data collected under
FAME and STAR projects were funded by the EU regional
development fund through its Atlantic area program and by
Marine Scotland, Scottish Natural Heritage (SNH) and
JNCC. Bardsey Island and Puffin Island tracking was
funded by a Natural Environment Research Council (NERC)
CASE studentship, Environment Wales and Natural
Resources Wales (NRW). Coquet Island tracking was funded
by a NERC CASE studentship. Flamborough Head tracking
was funded by the LEADER programme and NE. Isle of May
tracking was jointly funded by NERC and the RSPB. Lambay
tracking was funded by the EU regional development fund
through its Atlantic area program and BirdWatch Ireland
(Seabird Appeal).
Licences to fit GPS devices were issued by the British
Trust for Ornithology. We thank NRW for access to Bardsey
Island and Puffin Island, NE for permission to work on
Coquet Island, East Riding of Yorkshire Council for access to
Flamborough Head, SNH for access to Isle of May NNR, the
Trustees of the Lambay Estate for permission to work on
Lambay Island, Sir Richard Williams-Bulkeley for permis-
sion to work on Puffin Island, and the National Trust for
Scotland for access to St. Abb’s Head. We thank Bardsey
Island Trust, Bardsey Island Bird and Field Observatory, Fair
Isle Bird Observatory and Margaret and Patrick Kelly for
facilitating fieldwork.
We thank Chris Bell, Antony Bellamy, Maria Bogdanova,
Helen Boland, Andy Brown, Sarah Burthe, Kendrew Col-
houn, Stephen Dodd, Carrie Gunn, Maggie Hall, Mike Har-
ris, Robert Hughes, Becky Langton, Liz Mackley, Mara
Nydegger, Kat Snell, Jenny Sturgeon, Jennifer Taylor and
Ashley Tweedale for collecting tracking data. We thank
Wesley Davies, David Jardine, Paul Morrison and the East
Yorkshire Ringing Group for fieldwork help.
We are grateful to JNCC for providing access to SMP
data. Data used were extracted from the Seabird Monitoring
Programme Database at http://jncc.defra.gov.uk/smp/. Data
are provided to the SMP by the generous contributions of
nature conservation and research organisations, and of
many volunteers throughout the British Isles.
This study has been conducted using MyOcean products,
and we thanks all organisations involved. We thank the UK
Met Office and BADC for access to climate projection data.
Isle of May NNR data: © Database Right/Copyright NERC -
Centre for Ecology & Hydrology, all rights reserved.
We thank Dr S. Wakelin for providing advice on oceano-
graphic variables. We thank 3 anonymous reviewers for
helpful comments.
LITERATURE CITED
Ainley DG, Ford RG, Brown ED, Suryan RM, Irons DB (2003)
Prey resources, competition, and geographic structure of
kittiwake colonies in Prince William Sound. Ecology 84:
709−723
Arnott SA, Ruxton GD (2002) Sandeel recruitment in the
North Sea: demographic, climatic and trophic effects.
Mar Ecol Prog Ser 238: 199−210
Barton K (2014) MuMIn: multi-model inference. R pack -
age version 1.10.0. http: //R-Forge.R-project.org/pro-
jects/ mumin/
Bates D, Maechler M, Bolker B, Walker S (2014) lme4: linear
mixed-effects models using Eigen and S4. R package
version 1.1-6. http: //lme4.r-forge.r-project.org/
Beare DJ, Batten S, Edwards M, Reid DG (2002) Prevalence
of boreal Atlantic, temperate Atlantic and neritic zoo-
plankton in the North Sea between 1958 and 1998 in
relation to temperature, salinity, stratification intensity
and Atlantic inflow. J Sea Res 48: 29−49
Beaugrand G, Reid PC, Ibañez F, Lindley JA, Edwards M
(2002) Reorganization of North Atlantic marine copepod
biodiversity and climate. Science 296: 1692−1694
Behrens JW, Ærtebjerg G, Petersen JK, Carstensen J (2009)
Oxygen deficiency impacts on burying habitats for lesser
sandeel, Ammodytes tobianus, in the inner Danish
waters. Can J Fish Aquat Sci 66: 883−895
Bogdanova MI, Wanless S, Harris MP, Lindström J and oth-
ers (2014) Among-year and within-population variation
in foraging distribution of European shags Phalacrocorax
aristotelis over two decades: implications for marine spa-
tial planning. Biol Conserv 170: 292−299
Börger L, Franconi N, De Michele G, Gantz A and others
(2006) Effects of sampling regime on the mean and vari-
ance of home range size estimates. J Anim Ecol 75:
1393−1405
Brewer JH, O’Reilly KM, Dean Kildaw S, Loren Buck C
(2008) Interannual variation in the adrenal responsive-
ness of black-legged kittiwake chicks (Rissa tridactyla).
Gen Comp Endocrinol 156: 361−368
Browne WJ, Subramanian SV, Jones K, Goldstein H (2005)
Variance partitioning in multilevel logistic models that
exhibit overdispersion. J R Stat Soc Ser A Stat Soc 168:
599−613
Buren AD, Koen-Alonso M, Pepin P, Mowbray F and others
(2014) Bottom-up regulation of capelin, a keystone for-
age species. PLoS ONE 9: e87589
Burthe S, Daunt F, Butler A, Elston D and others (2012) Phe-
nological trends and trophic mismatch across multiple
levels of a North Sea pelagic food web. Mar Ecol Prog
Ser 454: 119−133
Cahill AE, Aiello-Lammens ME, Fisher-Reid MC, Hua X and
others (2013) How does climate change cause extinction?
Proc R Soc B 280: 20121890
Chivers LS, Lundy MG, Colhoun K, Newton SF, Reid N
(2012) Diet of Black-legged Kittiwakes (Rissa tridactyla)
feeding chicks at two Irish colonies highlights the impor-
tance of clupeids. Bird Study 59: 363−367
Christensen A, Jensen H, Mosegaard H, St. John M, Schrum
C (2008) Sandeel (Ammodytes marinus) larval transport
patterns in the North Sea from an individual-based
hydrodynamic egg and larval model. Can J Fish Aquat
Sci 65: 1498−1511
Collins M, Knutti R, Arblaster J, Dufresne JL and others
(2013) Long-term climate change: projections, commit-
87
Author copy
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤

Clim Res 66: 75– 89, 2015
ments and irreversibility. In: Intergovernmental Panel on
Climate Change (IPCC) Climate Change 2013: the phys-
ical science basis. Contribution of Working Group I to the
Fifth Assessment Report of theIPCC. Cambridge Univer-
sity Press, Cambridge, p 1029−1136
Cook A, Robinson R (2010) How representative is the cur-
rent monitoring of breeding seabirds in the UK? BTO
Research Report 573. British Trust for Ornithology, Thet-
ford
Cook ASCP, Dadam D, Mitchell I, Ross-Smith VH, Robinson
RA (2014) Indicators of seabird reproductive perform-
ance demonstrate the impact of commercial fisheries on
seabird populations in the North Sea. Ecol Indic 38: 1−11
Coulson JC (2011) The kittiwake. Poyser, London
Coulson JC, Porter JM (1985) Reproductive success of the
Kittiwake Rissa tridactyla: the roles of clutch size, chick
growth rates and parental quality. Ibis 127: 450−466
Croxall JP, Butchart SHM, Lascelles B, Stattersfield AJ, Sul-
livan B, Symes A, Taylor P (2012) Seabird conservation
status, threats and priority actions: a global assessment.
Bird Conserv Int 22: 1−34
Daunt F, Wanless S, Greenstreet SPR, Jensen H, Hamer KC,
Harris MP (2008) The impact of the sandeel fishery clo-
sure on seabird food consumption, distribution, and pro-
ductivity in the northwestern North Sea. Can J Fish
Aquat Sci 65: 362−381
Doney SC, Ruckelshaus M, Emmett Duffy J, Barry JP and
others (2012) Climate change impacts on marine ecosys-
tems. Annu Rev Mar Sci 4: 11−37
Duong T (2013) ks: kernel smoothing. R package version
1.8.13. https: //CRAN.R-project.org/package=ks
Duong T, Hazelton M (2003) Plug-in bandwidth matrices for
bivariate kernel density estimation. J Nonparametr Stat
15: 17−30
Enstipp M, Daunt F, Wanless S, Humphreys E, Hamer K,
Benvenuti S, Grémillet D (2006) Foraging energetics of
North Sea birds confronted with fluctuating prey avail-
ability. In: Boyd IL, Wanless S, Camphuysen CJ (eds) Top
predators in marine ecosystems. Cambridge University
Press, Cambridge, p 191−210
Frederiksen M, Harris MP, Daunt F, Rothery P, Wanless S
(2004a) Scale-dependent climate signals drive breeding
phenology of three seabird species. Glob Change Biol
10: 1214−1221
Frederiksen M, Wanless S, Harris MP, Rothery P, Wilson LJ
(2004b) The role of industrial fisheries and oceano-
graphic change in the decline of North Sea black-legged
kittiwakes. J Appl Ecol 41: 1129−1139
Frederiksen M, Wright PJ, Harris MP, Mavor RA, Heubeck
M, Wanless S (2005) Regional patterns of kittiwake Rissa
tridactyla breeding success are related to variability in
sandeel recruitment. Mar Ecol Prog Ser 300: 201−211
Frederiksen M, Edwards M, Mavor RA, Wanless S (2007)
Regional and annual variation in black-legged kittiwake
breeding productivity is related to sea surface tempera-
ture. Mar Ecol Prog Ser 350: 137−143
Frederiksen M, Anker-Nilssen T, Beaugrand G, Wanless S
(2013) Climate, copepods and seabirds in the boreal
Northeast Atlantic — current state and future outlook.
Glob Change Biol 19: 364−372
Furness RW, Birkhead TR (1984) Seabird colony distribu-
tions suggest competition for food supplies during the
breeding season. Nature 311: 655−656
Furness R, Tasker M (2000) Seabird-fishery interactions:
quantifying the sensitivity of seabirds to reductions in
sandeel abundance, and identification of key areas for
sensitive seabirds in the North Sea. Mar Ecol Prog Ser
202: 253−264
Genz A, Bretz F, Miwa T, Mi X, Leisch F, Scheipl F, Hothorn T
(2014) mvtnorm: multivariate normal and t distributions. R
package version 0.9-99992. http: //CRAN.R-project. org/
package=mvtnorm
Gilman SE, Urban MC, Tewksbury J, Gilchrist GW, Holt RD
(2010) A framework for community interactions under
climate change. Trends Ecol Evol 25: 325−331
Greenstreet SPR, Armstrong E, Mosegaard H, Jensen H and
others (2006) Variation in the abundance of sandeels
Ammodytes marinus off southeast Scotland: an evalua-
tion of area-closure fisheries management and stock
abundance assessment methods. ICES J Mar Sci 63:
1530−1550
Hamer KC, Monaghan P, Uttley JD, Walton P, Burns MD
(1993) The influence of food supply on the breeding ecol-
ogy of Kittiwakes Rissa tridactyla in Shetland. Ibis 135:
255−263
Hijmans R (2013) raster: geographic data analysis and mod-
eling. R package version 2.1-49. http: //CRAN.R-project.
org/package=raster
HM Government (2012) Marine strategy part one: UK initial
assessment and good environmental status. https://www.
gov. uk/ government/policies/marine-fisheries
Holt JT, James ID (2001) An scoordinate density evolving
model of the northwest European continental shelf. 1.
Model description and density structure. J Geophys Res-
Oceans 106: 14015−14034
Holt J, Wakelin S, Lowe J, Tinker J (2010) The potential
impacts of climate change on the hydrography of the
northwest European continental shelf. Prog Oceanogr
86: 361−379
Holt J, Butenschön M, Wakelin SL, Artioli Y, Allen JI (2012)
Oceanic controls on the primary production of the north-
west European continental shelf: model experiments
under recent past conditions and a potential future sce-
nario. Biogeosciences 9: 97−117
Irons DB (1998) Foraging area fidelity of individual seabirds
in relation to tidal cycles and flock feeding. Ecology 79:
647−655
Jackett DR, McDougall TJ, Feistel R, Wright DG, Griffies SM
(2006) Algorithms for density, potential temperature,
conservative temperature, and the freezing temperature
of seawater. J Atmos Ocean Technol 23: 1709−1728
Jensen H, Wright PJ, Munk P (2003) Vertical distribution of
pre-settled sandeel (Ammodytes marinus) in the North
Sea in relation to size and environmental variables. ICES
J Mar Sci 60: 1342−1351
JNCC (Joint Nature Conservation Committee) (2014) Sea-
bird population trends and causes of change: 1986−2013
report. Updated August 2014. Joint Nature Conservation
Committee. http: //.jncc.defra.gov.uk/page-3201 (access -
ed 16 Feb 2015)
Kotzerka J, Garthe S, Hatch S (2010) GPS tracking devices
reveal foraging strategies of black-legged kittiwakes.
J Ornithol 151: 459−467
Lauria V, Attrill MJ, Pinnegar JK, Brown A, Edwards M,
Votier SC (2012) Influence of climate change and trophic
coupling across four trophic levels in the Celtic Sea. PLoS
ONE 7: e47408
Lewis S, Wanless S, Wright P, Harris M, Bull J, Elston D
(2001) Diet and breeding performance of black-legged
kittiwakes Rissa tridactyla at a North Sea colony. Mar
88
Author copy
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤

Carroll et al.: Seabird breeding success and oceanographic change 89
Ecol Prog Ser 221: 277−284
Lonergan M, Duck CD, Thompson D, Mackey BL, Cunning-
ham L, Boyd IL (2007) Using sparse survey data to inves-
tigate the declining abundance of British harbour seals. J
Zool 271: 261−269
Lowe J, Howard T, Pardaens A, Tinker J and others (2009)
UK climate projections science report: marine and
coastal projections. Met Office Hadley Centre, Exeter
MacLeod CD, Pierce GJ, Begoña Santos M (2007a) Starva-
tion and sandeel consumption in harbour porpoises in
the Scottish North Sea. Biol Lett 3: 535−536
MacLeod CD, Santos MB, Reid RJ, Scott BE, Pierce GJ
(2007b) Linking sandeel consumption and the likelihood
of starvation in harbour porpoises in the Scottish North
Sea: could climate change mean more starving por-
poises? Biol Lett 3: 185−188
Massaro M, Chardine JW, Jones IL (2001) Relationships be -
tween black-legged kittiwake nest-site characteristics
and susceptibility to predation by large gulls. Condor
103: 793−801
Mawdsley JR, O’Malley R, Ojima DS (2009) A review of cli-
mate-change adaptation strategies for wildlife manage-
ment and biodiversity conservation. Conserv Biol 23:
1080−1089
Morán XAG, López-Urrutia Á, Calvo-Díaz A, Li WKW
(2010) Increasing importance of small phytoplankton in a
warmer ocean. Glob Change Biol 16: 1137−1144
Ockendon N, Baker DJ, Carr JA, White EC and others
(2014) Mechanisms underpinning climatic impacts on
natural populations: altered species interactions are
more important than direct effects. Glob Change Biol 20:
2221−2229
Parmesan C (2006) Ecological and evolutionary responses to
recent climate change. Annu Rev Ecol Evol Syst 37:
637−669
Pearce-Higgins JW, Dennis P, Whittingham MJ, Yalden DW
(2010) Impacts of climate on prey abundance account for
fluctuations in a population of a northern wader at the
southern edge of its range. Glob Change Biol 16: 12−23
R Core Team (2014) R: a language and environment for sta-
tistical computing. <www.R-project.org/>
Regehr HM, Montevecchi WA (1997) Interactive effects of
food shortage and predation on breeding failure of
black-legged kittiwakes: indirect effects of fisheries
activities and implications for indicator species. Mar Ecol
Prog Ser 155: 249−260
Robertson GS, Bolton M, Grecian WJ, Monaghan P (2014)
Inter- and intra-year variation in foraging areas of breed-
ing kittiwakes (Rissa tridactyla). Mar Biol 161: 1973−1986
Sandvik H, Erikstad K, Sæther B (2012) Climate affects
seabird population dynamics both via reproduction and
adult survival. Mar Ecol Prog Ser 454: 273−284
Sandvik H, Reiertsen T, Erikstad K, Anker-Nilssen T and
others (2014) The decline of Norwegian kittiwake popu-
lations: modelling the role of ocean warming. Clim Res
60: 91−102
Scott BE, Sharples J, Wanless S, Ross ON, Frederiksen M,
Daunt F (2006) The use of biologically meaningful
oceanographic indices to separate the effects of climate
and fisheries on seabird breeding success. In: Boyd IL,
Wanless S, Camphuysen CJ (eds) Top predators in mar-
ine ecosystems: their role in monitoring and manage-
ment. Cambridge University Press, Cambridge, p 46–62
Scott BE, Sharples J, Ross ON, Wang J, Pierce GJ, Cam-
phuysen CJ (2010) Sub-surface hotspots in shallow seas:
fine-scale limited locations of top predator foraging habi-
tat indicated by tidal mixing and sub-surface chloro-
phyll. Mar Ecol Prog Ser 408: 207−226
Sergio F, Caro T, Brown D, Clucas B and others (2008) Top
predators as conservation tools: ecological rationale,
assumptions, and efficacy. Annu Rev Ecol Evol Syst 39:
1−19
Sharples J, Ross ON, Scott BE, Greenstreet SPR, Fraser H
(2006) Inter-annual variability in the timing of stratifica-
tion and the spring bloom in the North-western North
Sea. Cont Shelf Res 26: 733−751
Silverman BW (1986) Density estimation for statistics and
data analysis. CRC Press, London
Soanes LM, Arnould JPY, Dodd SG, Sumner MD, Green JA
(2013) How many seabirds do we need to track to define
home-range area? J Appl Ecol 50: 671−679
Suryan RM, Irons DB, Kaufman M, Benson J, Jodice PGR,
Roby DD, Brown ED (2002) Short-term fluctuations in
forage fish availability and the effect on prey selection
and brood-rearing in the black-legged kittiwake Rissa
tridactyla. Mar Ecol Prog Ser 236: 273−287
Sydeman W, Thompson S, Kitaysky A (2012) Seabirds and
climate change: roadmap for the future. Mar Ecol Prog
Ser 454: 107−117
Thompson P, Ingram S, Lonergan M, Northridge S, Hall A,
Wilson B (2007) Climate change causing starvation in
harbour porpoises? Biol Lett 3: 533−535
Tylianakis JM, Didham RK, Bascompte J, Wardle DA (2008)
Global change and species interactions in terrestrial eco-
systems. Ecol Lett 11: 1351−1363
van Deurs M, van Hal R, Tomczak M, Jónasdóttir S, Dolmer
P (2009) Recruitment of lesser sandeel Ammodytes mari-
nus in relation to density dependence and zooplankton
composition. Mar Ecol Prog Ser 381: 249−258
Vincenzi S, Mangel M (2013) Linking food availability, body
growth and survival in the black-legged kittiwake Rissa
tridactyla. Deep-Sea Res II 94: 192−200
Wakefield ED, Bodey TW, Bearhop S, Blackburn J and oth-
ers (2013) Space partitioning without territoriality in gan-
nets. Science 341: 68−70
Walsh P, Halley D, Harris M, Del Nevo A, Sim I, Tasker M
(1995) Seabird monitoring handbook for Britain and Ire-
land: a compilation of methods for survey and monitoring
of breeding seabirds. JNCC/RSPB/ITE/Seabird Group,
Peterborough
Wanless S, Harris MP (1992) Activity budgets, diet and
breeding success of Kittiwakes Rissa tridactyla on the
Isle of May. Bird Study 39: 145−154
Wanless S, Wright PJ, Harris MP, Elston DA (2004) Evidence
for decrease in size of lesser sandeels Ammodytes mari-
nus in a North Sea aggregation over a 30-yr period. Mar
Ecol Prog Ser 279: 237−246
Wanless S, Frederiksen M, Daunt F, Scott BE, Harris MP
(2007) Black-legged kittiwakes as indicators of environ-
mental change in the North Sea: evidence from long-
term studies. Prog Oceanogr 72: 30−38
Wright PJ, Bailey MC (1996) Timing of hatching in Ammo -
dytes marinus from Shetland waters and its significance
to early growth and survivorship. Mar Biol 126: 143−152
Editorial responsibility: Mauricio Lima,
Santiago, Chile
Submitted: March 23, 2015; Accepted: July 16, 2015
Proofs received from author(s): September 11, 2015
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
Author copy
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤