Modelling the effects of climate change on the distribution and production of marine fishes: Accounting for trophic interactions in a dynamic bioclimate envelope model

Abstract

Climate change has already altered the distribution of marine fishes. Future predictions of fish distributions and catches based on bioclimate envelope models are available, but to date they have not considered inter-specific interactions. We address this by combining the species-based Dynamic Bioclimate Envelope Model (DBEM) with a size-based trophic model. The new approach provides spatially and temporally resolved predictions of changes in species' size, abundance and catch potential that account for the effects of ecological interactions. Predicted latitudinal shifts are, on average, reduced by 20% when species interactions are incorporated, compared to DBEM predictions, with pelagic species showing the greatest reductions. Goodness-of-fit to biomass data from fish stock assessments in the North Atlantic between 1991 and 2003 is improved slightly by including species interactions. The differences between predictions from the two models may be relatively modest because, at the North Atlantic basin scale, (1) predators and competitors may respond to climate change together; (2) existing parameterization of the DBEM might implicitly incorporate trophic interactions; and/or (3) trophic interactions might not be the main driver of responses to climate. Future analyses using ecologically-explicit models and data will improve understanding of the effects of inter-specific interactions on responses to climate change, and better inform managers about plausible ecological and fishery consequences of a changing environment. This article is protected by copyright. All rights reserved.

Full text

TECHNICAL ADVANCE
Modelling the effects of climate change on the
distribution and production of marine fishes: accounting
for trophic interactions in a dynamic bioclimate envelope
model
JOSE A. FERNANDES*†, WILLIAM W. L. CHEUNG‡, SIMON JENNINGS*§,
MOMME BUTENSCH €
ON †, LEE DE MORA†,THOMASL.FR€
OL I C H E R ¶,
MANUEL BARANGE†and ALASTAIR GRANT*
*School of Environmental Sciences, The University of East Anglia, Norwich, NR4 7TJ, UK, †Plymouth Marine Laboratory,
Prospect Place, The Hoe, Plymouth, PL13 DH, UK, ‡Changing Ocean Research Unit, The University of British Columbia,
Fisheries Centre, 2202 Main Mall, Vancouver, BC V6T 1Z4, Canada, §Centre for Environment, Fisheries and Aquaculture
Science, Lowestoft, NR33 0HT, UK, ¶Atmospheric and Oceanic Sciences Program, Princeton University, Princeton,
NJ 08544, USA
Abstract
Climate change has already altered the distribution of marine fishes. Future predictions of fish distributions and
catches based on bioclimate envelope models are available, but to date they have not considered interspecific interac-
tions. We address this by combining the species-based Dynamic Bioclimate Envelope Model (DBEM) with a size-
based trophic model. The new approach provides spatially and temporally resolved predictions of changes in species’
size, abundance and catch potential that account for the effects of ecological interactions. Predicted latitudinal shifts
are, on average, reduced by 20% when species interactions are incorporated, compared to DBEM predictions, with
pelagic species showing the greatest reductions. Goodness-of-fit of biomass data from fish stock assessments in the
North Atlantic between 1991 and 2003 is improved slightly by including species interactions. The differences between
predictions from the two models may be relatively modest because, at the North Atlantic basin scale, (i) predators
and competitors may respond to climate change together; (ii) existing parameterization of the DBEM might implicitly
incorporate trophic interactions; and/or (iii) trophic interactions might not be the main driver of responses to climate.
Future analyses using ecologically explicit models and data will improve understanding of the effects of inter-specific
interactions on responses to climate change, and better inform managers about plausible ecological and fishery conse-
quences of a changing environment.
Keywords: biological feedback, climate change, competition, ecosystem approach, fisheries management, model validation,
modelling, size spectrum, species interactions
Received 23 January 2013; revised version received 5 April 2013 and accepted 14 April 2013
Introduction
Climate change affects ocean conditions, including tem-
perature, salinity, ice coverage, currents, oxygen level,
acidity and consequently growth, body size, distribution,
productivity and abundance of marine species, includ-
ing those that are exploited by fisheries (Perry et al.,
2005; Behrenfeld et al., 2006; Brander, 2007; P€
ortner,
2010; Simpson et al., 2011; Cheung et al., 2013). Over a
range of greenhouse gas emission scenarios (IPCC,
2007), changes in the marine environment are predicted
to be more rapid in the 21st century with implications
for marine ecosystems and dependent industries (Roes-
sig et al., 2004; Lam et al., 2012; Merino et al., 2012).
A range of modelling approaches have been devel-
oped to predict the potential effects of future climate
change on species distributions and abundance (Stock
et al., 2011). One class of models, species-based biocli-
mate envelope models, have been used to predict redis-
tribution of both terrestrial and aquatic species (Pearson
& Dawson, 2003; Jones et al., 2012). The Dynamic Biocli-
mate Envelope Model (DBEM) developed by Cheung
et al. (2008a,b, 2009, 2011) projects changes in marine
species distribution, abundance and body size with
Correspondence: Jose A. Fernandes, tel. +44 (0)1603591375,
fax +44 (0)1752633101, e-mail: j.fernandes@uea.ac.uk
©2013 John Wiley & Sons Ltd 1
Global Change Biology (2013), doi: 10.1111/gcb.12231

explicit consideration of population dynamics, dispersal
(larval and adult) and ecophysiology (Cheung et al.,
2008a,b, 2009, 2011, 2013). Projections suggest that there
will be a high rate of species invasions in high-latitude
regions and a potential high rate of local extinction in
the tropics and semi enclosed seas in the 21st century
(Cheung et al., 2009). Moreover, as a result of predicted
changes in range and primary productivity, Cheung
et al. (2010) project that maximum catch potential of
exploited species is expected to decrease in the tropics
and to increase in high latitudes. However, these
projections do not account for the effects of species
interactions on redistribution and abundance, thus
introducing a source of structural uncertainty (Cheung
et al., 2010).
Rates of primary production and transfer efficiency
influence production and biomass of consumers. ‘Size-
spectrum’ models have been developed to describe
energy transfer from primary producers to consumers
of progressively larger body size (e.g. Dickie et al.,
1987) and variants of these models have been devel-
oped and applied to predict potential biomass, produc-
tion and size structure of fish in the world’s oceans
from estimates of primary production and temperature
(Jennings et al., 2008), and to predict the responses of
fish communities to fishing and climate change (Blan-
chard et al., 2011, 2012). These size-based models are
not taxonomically resolved, and this limits the range of
applications, given that species identity is usually a key
consideration for management, monitoring and regula-
tory purposes.
Here, we combine the strengths of the DBEM
(i.e. focus on identified species) with those of the size
spectrum model (i.e. focus on trophic interactions) to
predict spatial and temporal changes in species’ abun-
dance and distribution in response to predicted future
changes in temperature and primary production. Forty-
eight of the most abundant and commercially impor-
tant marine fishes in the North Atlantic, here defined as
Food and Agriculture Organization (FAO) statistical
area 27, are included. The size spectrum is used to
determine resource limits in a given geographical area
and these limits, along with habitat suitability for a
given species, determine the biomass of that species
that can be supported in this area.
Materials and methods
A modelling approach that integrates the species-based DBEM
model with the size spectrum approach, hereafter called size-
spectrum DBEM (SS-DBEM) was developed. The SS-DBEM:
(i) estimates potential biomass supported by the system; (ii)
predicts habitat suitability; and (iii) models species interac-
tions. Predictions from the SS-DBEM are then compared with
those from a DBEM model that does not incorporate species
interactions (NSI-DBEM, where NSI denotes no species inter-
actions).
Potential biomass supported at each body size class
The size-spectrum is described as a log-log relationship
between abundance and body size. The slope of the spectrum
is determined by trophic transfer efficiency and the relation-
ships between the body sizes of predators and their prey
(Borgmann, 1987; Jennings & Mackinson, 2003). The height of
the spectrum is determined by primary production and
describes the total abundance of individuals from all species
that can be supported in any defined body size class (e.g.
Boudreau & Dickie, 1992).
As predator-prey mass ratios and transfer efficiencies in
marine food chains do not depend systematically on the mean
rate of primary production or mean temperature (Barnes et al.,
2010), less energy is transferred to consumers of a given body
size when food webs are supported by smaller primary pro-
ducers (Barnes et al., 2010). Much of the variation in the body
size distribution of primary producers depends on the abso-
lute rate of primary production, with picoplankton, the small-
est phytoplankton, dominating when primary production is
low (Agawin et al., 2000). Thus, the median and mean body
sizes of phytoplankton decrease with decreasing rates of pri-
mary production (Barnes et al., 2011). To account for this, the
position of the median body mass class for phytoplankton (m)
was calculated as:
m¼ ½ð6:1PsÞ8:25=log10ð2Þð1Þ
where P
s
is the predicted contribution of picophytoplankton
net production to total net Primary Production (PP) as calcu-
lated using the empirical equation
Ps¼½ð12:19 log10 PP þ37:248=100 ð2Þ
derived by Jennings et al. (2008) using the data from Agawin
et al. (2000).
Once the median body mass class of phytoplankton was
defined, we calculated the consumer biomass at body size
following the approach of Jennings et al. (2008). Assumptions
about trophic transfer efficiency and the predator-prey mass
ratio (e=0.125 and l=3 respectively) followed Jennings et al.
(2008), but the spectrum was discretized using a log
2
series of
body mass from 2
1
to 2
19
g. Subsequent evidence suggests
that the predator-prey mass ratio may increase with body
mass and that transfer efficiency may decrease, but the
changes are not expected to affect the time-averaged slope of
the size-spectrum (Barnes et al., 2010).
Habitat suitability
The prediction of habitat suitability in SS-DBEM was based on
the algorithm implemented in NSI-DBEM (Cheung et al.,
2008a,b, 2009, 2011; Cheung et al., 2013). The NSI-DBEM
defines the relative preferences of the modelled species for
temperature and other environmental variables based on the
relationship between current distributions and gridded envi-
ronmental data. The initial distribution of relative abundance
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
2J. A. FERNANDES et al.

(representing 1970–2000) of the modelled marine species on a
30′930′latitude-longitude grid map of the world ocean is
predicted using the Sea Around Us project algorithm (Close
et al., 2006; Jones et al., 2012) based on parameters describing
range limits, association with major habitat types and known
occurrence boundaries. Parameter values for each species
were derived from data in online databases, mainly FishBase
(www.fishbase.org) and SeaLifeBase (www.sealifebase.org).
Environmental variables incorporated into the NSI-DBEM
include sea surface temperature, sea bottom temperature,
coastal upwelling, salinity, sea-ice extent, depth and habitat
types (Cheung et al., 2011). NSI-DBEM first calculates changes
in growth and other life history traits in response to changes
in temperature and oxygen concentration based on algorithms
derived from growth and metabolic functions and empirical
equations (Cheung et al., 2011, 2013). Second, NSI-DBEM pre-
dicts size-frequency distributions for each species in each spa-
tial cell using a size-structured ‘per recruit’ model. Finally, the
model simulates spatial and temporal changes in relative
abundance within a cell based on carrying capacity of a cell,
density-dependent population growth, larval dispersal and
adult migration (Cheung et al., 2008b, 2011).
Species interactions
A new algorithm was developed to describe resource competi-
tion between different species co-occurring in a cell by com-
paring the energy (in biomass) that can be supported in the
cell (estimated with the SS model) with the energy demanded
by the species predicted to inhabit the given cell (estimated
with the NSI model). The algorithm comprises two stages: (i)
an initialization stage where competition parameters are esti-
mated; and, (ii) a recurrent stage where the competition
parameters are used to resolve conflicts between energy (bio-
mass) demands and biomass that can be supported. One
advantage of this approach is that it focuses on competition
for the energy available within a cell, thus negating the need
for a diet matrix that describes species-specific feeding interac-
tions. Data to develop such matrices are scarce at the scale of
FAO Area 27 and the persistence and emergence of feeding
interactions through time, and in response to future climate
change, is highly uncertain.
First stage. The model uses the NSI-DBEM approach to
establish an initial distribution for each species. The approach
assumes that predicted habitat suitability is a proxy for the
distribution of relative abundance of a given species. Thus,
multiplying the initial relative biomass by the estimated abso-
lute biomass from empirical data, initial species distribution is
expressed in terms of absolute biomass in each cell. Because
biomass estimates from assessment data are not available for
some of the species considered (Table 1), the initial biomass
estimates were approximated by the predicted unexploited
biomass (B
∞
) from maximum reported fisheries catch (MC)
since 1950 and an estimate of the intrinsic growth rate (r)of
the population (Schaefer, 1954):
B1¼MSY 4=rð3Þ
Maximum sustainable yield (MSY) was calculated using the
algorithm documented in Cheung et al. (2008a) that used the
average maximum values of the catch time series of a species
as an approximated MSY. Values for r, estimated based on an
empirical equation that was dependent on asymptotic length
of the species, were obtained from FishBase (www.fishbase.
org). Although this is an approximation and not as reliable as
estimates of biomass using survey-based methods (Pauly
et al., 2013), we show that, consistent with similar findings by
Froese et al. (2012), biomass estimates from maximum catch
data were significantly correlated with those from aggregated
stock assessments (Table 1; Fig. 1). These biomass estimates
were used for model initialization only.
The initial absolute biomass estimates, based on habitat
suitability in the cells where they are distributed (Fig. 2), are
used to generate a matrix of species’ energy demand (expressed
as biomass). Matrix elements define the proportion of total
energy obtained by a species at each habitat suitability bin
and size class. The amount of energy is determined by the
average proportion of energy that a species gets in cells with
the same habitat suitability.
Energy demanded (E_D) by a species in a cell is compared
with the total biomass or energy (E_S) that can be supported
in the cell (see Table 2 for a summary of abbreviations). E_D is
determined based on the predicted habitat suitability from the
DBEM algorithm, whereas E_S is determined by the SS model.
Thus, the average proportion of energy that a species demands
in cells with the same habitat suitability can be calculated as
follows:
resoucesSpp;Suit;Size ¼E DSuit
Spp;w;i
E S ð4Þ
To convert from biomass (B) distribution to numbers (N) and
vice versa, the mean body mass (W) at each size class (i)is
used as shown below:
B¼X
n
i¼1
NiWið5Þ
where nis the number of size classes considered in the model.
The initial habitat suitability value is converted using a square
root data transformation, to ensure a balanced distribution of
the cells across the habitat suitability classes, and then normal-
ized to a range from 0 to 1 relative to minimum and maximum
value of habitat suitability for each species. The model then
groups habitat suitability into six classes (bins) of values:
0–0.3, >0.3–0.4, >0.4–0.5, >0.5–0.6, >0.6–0.7 and >0.7–1. The use
of discretized bins of habitat suitability, a nonparametric
methodology, does not require the specification of explicit
distribution functions and is more computationally efficient
(Fayyad & Irani, 1993; Dougherty et al., 1995). The effects of
such discretization are minimized here by square root trans-
formation of the predicted habitat suitability, the low number
of bins and the choice of bin boundaries (Uusitalo, 2007;
Fernandes et al., 2010).
Available energy in a size class which is not demanded by
the modelled species was assigned to a group called ‘Other
groups’, because species that were not modelled explicitly
would also have an energy demand. This group has its own
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
CLIMATE CHANGE EFFECTS ON MARINE FISHES 3

Table 1 List of modelled fish species. Stocks that have been aggregated to provide species abundance estimates are identified by
their stock ID codes (STOCKID) in the RAM Legacy database (upper case codes). For some ICES assessed stocks not listed in the
RAM Legacy database, stock ID codes that were based on ICES Stock Summary Database were used (lower case codes)
Common name Scientific name Type Stock ID code
Albacore Thunnus alalunga Pelagic ALBANATL
American plaice/
long rough dab
Hippoglossoides platessoides Demersal
Angler Lophius piscatorius Demersal
Atlantic cod Gadus morhua Demersal CODNEAR, CODBA2224, CODBA2532, CODVIa, CODIS,
CODICE, CODNS and CODKAT
Atlantic herring Clupea harengus Pelagic HERRIsum, HERRNS, HERR2224IIIa, HERR2532, HERR30,
HERRRIGA, HERRNIRS, HERRNWATLC, HERR4VWX,
HERR4RFA, HERR4RSP, HERR4TFA, HERR4TSP, HERR31,
her-noss, hervian and her-vasu
Atlantic horse mackerel Trachurus trachurus Pelagic hom-west
Atlantic mackerel Scomber scombrus Pelagic MACKNEICES
Baltic sprat Sprattus sprattus Pelagic SPRAT22-32
Blue whiting Micromesistius poutassou Pelagic whb-comb
Boarfish Capros aper Demersal
Capelin Mallotus villosus Pelagic CAPEICE and CAPENOR
Common sole Solea solea Demersal SOLENS, SOLEVIId, SOLEIS, SOLEIIIa, SOLEVIIe, SOLECS,
and SOLEVIII
Cuckoo ray Leucoraja naevus Demersal
Dab Limanda limanda Demersal
European anchovy Engraulis encrasicolus Pelagic ANCHOBAYB
European hake Merluccius merluccius Demersal HAKESOTH and HAKENRTN
European pilchard Sardina pilchardus Pelagic sar-soth
European plaice Pleuronectes platessus Demersal PLAIC7d, PLAICIIIa, PLAICNS, PLAICIS, PLAICECHW
and PLAICCELT
European sprat Sprattus sprattus Pelagic SPRATNS
Flounder Platichthys flesus Demersal
Fourbeard rockling Enchelyopus cimbrius Demersal
Fourspotted megrim Lepidorhombus boscii Demersal mgb-8c9a
Greenland halibut Reinhardtius hippoglossoides Demersal GHALNEAR, GHALBSAI and GHAL23KLMNO
Haddock Melanogrammus aeglefinus Demersal HAD4X5Y, HAD5Y, HAD5Zejm, HADICE, HADNEAR,
HADFAPL, HADNS-IIIa, HADVIa, HADVIIb-k, HADROCK
and HADGB
John dory Zeus faber Demersal
Lemon sole Microstomus kitt Demersal
Ling Molva molva Demersal
Megrim Lepidorhombus whiffiagonis Demersal mgw-8c9a
Northern bluefin tuna Thunnus thynnus Pelagic ATBTUNAEATL and ATBTUNAWATL
Norway pout Trisopterus esmarkii Demersal nop-34
Golden Redfish Sebastes norvegicus Demersal GOLDREDNEAR
Pearlsides Maurolicus muelleri Pelagic
Piked dogfish/Spurdog Squalus acanthias Demersal
Pollack Pollachius pollachius Demersal
Poor cod Trisopterus minutus Demersal
Pouting/Bib Trisopterus luscus Demersal
Red bandfish Cepola macrophthalma Demersal
Saithe/Pollock Pollachius virens Demersal POLL5YZ, POLLNEAR, POLLFAPL, POLL4X5YZ and
POLLNS-VI-IIIa
Smallspottedcatshark Scyliorhinus canicula Demersal
Splendid alfonsino Beryx splendens Demersal
Spotted ray Raja montagui Demersal
Striped red mullet Mullus surmuletus Demersal
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
4J. A. FERNANDES et al.

resource allocation matrix based on the average habitat
suitability of the modelled species, allowing the inclusion of
resource demand from species that are not explicitly
modelled. As the species assemblage in the boundary of the
geographical domain of the model is likely to be underrepre-
sented by the modelled species, the matrix for ‘Others group’
is only computed for cells where the number of species pres-
ent is more than the square root of the total number of species
modelled.
Second stage. Abundance of each species in each cell was
predicted using the algorithm in the NSI-DBEM. The model
runs uses an annual time-step for bottom-dwelling (demersal)
species and two seasonal time-steps (summer and winter) for
species in the water-column (pelagic). The energy demand of
each species is compared with energy demands of other spe-
cies co-occurring in the same cell (Fig. 2). If the energy
demanded by all organisms in the cell exceeds the energy
available, then the available energy is allocated to each species
in proportion to its energy demands. If the energy demanded
by all the species is lower than the energy available, the sur-
plus energy is allocated according to the proportional energy
demand of the species present, including the ‘Others group’. To
represent population growth that is limited by factors other
than available energy, the rate at which energy can be assimi-
lated by a species is limited as shown below:
res opSpp;Suit;W¼2std.devðE DSuitÞ
meanðEDSuitÞð6Þ
where, E_D
Suit
denotes the energy demanded in all the cells in
each bin of habitat suitability. Therefore, the amount of addi-
tional energy that can be taken by the species is limited by
two times the standard deviation (std.dev) of energy that each
species gets in the initial distribution to each habitat suitability
bin. Any energy that is left after these allocations is assumed
to be used by the ‘Others group’.
Model testing
The results from the model that includes competition were
compared with results from the NSI-DBEM and “empirical”
time series of abundance data from fish stock assessments for
the Northeast Atlantic (FAO area 27), as extracted from the
RAM Legacy Stock Assessment Database (Ricard et al., 2012;
http://ramlegacy.marinebiodiversity.ca/) and ICES Stock
Summary Database (http://www.ices.dk). To compare pro-
jected changes with observations, abundance data for each
species were normalized by dividing them by their mean
value. While the models were applied to a set of 48 fish spe-
cies, comparison with empirical data was conducted for 24
species for which data were available from the RAM Legacy
and ICES databases (Table 1). The output of the DBEM mod-
els were compared with the “empirical” time-series values for
each species and distribution of absolute error (AE) was calcu-
lated as follows:
AE ¼jpjxjjð7Þ
where p is the total biomass predicted in a DBEM model in
a particular year for a species j, and x is the total biomass
from the assessments. The comparison was done for the
years with available assessment data for all the 24 species
considered (1991–2003). To compare the performance of the
SS-DBEM and NSI-DBEM, the Percent Reduction in Error
(PRE) was calculated (Hagle & Glen, 1992; Fernandes et al.,
2009), but weighted by the maximum catch of each species
(WPRE):
Table 1 (continued)
Common name Scientific name Type Stock ID code
Thickback sole Microchirus variegatus Demersal
Thornback ray Raja clavata Demersal
Tub gurnard Chelidonichthys lucerna Demersal
Tusk/Torsk/Cusk Brosme brosme Demersal CUSK4X
Whiting Merlangius merlangus Demersal WHITNS-VIId-IIIa, WHITVIa and WHITVIIek
Witch Glyptocephalus cynoglossus Demersal
Fig. 1 Relationship between the maximum assessed biomass
(log) and the estimated carrying capacity of fish population (B
∞
,
log) for 22 species in the 27 FAO area (after removing extreme
values, the lowest and highest B
∞
).
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
CLIMATE CHANGE EFFECTS ON MARINE FISHES 5

WPRE ¼1
Pl
k¼1MaxCatchkXl
k¼1
100ðAENSIkAESSkÞ
AENSIk

MaxCatchk
ð8Þ
where AENSI is the absolute error in the NSI-DBEM model,
AESS is the absolute error in the SS-DBEM model, k the
number of species and MaxCatch the maximum catch of the
species.
These models were also compared with empirical data
describing latitudinal and depth centroid shifts of species in
response to climate change (Dulvy et al., 2008; Cheung et al.,
2011). Distribution centroid (DC
t
) for each year (t) was calcu-
lated as:
Dynamic bioclimate envelop model
Cheung et al., 2008; 2009; 2011
Size distribution for each species
Size spectrum model
Jennings et al., 2008
Earth system models
ERSEM, GFDL, ...
Biomass supported
at each size bin
Ram Legacy Stock Assessment database
ICES Stock Summary database
Algorithm for
species interactions
Size distribution for each species
Primary production
Temperature
LINK BETWEEN MODELS
Species interactions table
Taxa
parameters
Species
distribution
Algorithm pseudocode for species interactions
1: Calculate =
>
3: TotalResW, i =
4: If TotalResW, i< 1 then = ·
5: If TotalResW, i> 1 then Normalize: TotalResW, i
6: If > then =
7: Adjust biomass, abundance and size distributions based on
2: then = / E_SW,i
If
Initial
biomass
Relative
distribution
across cells
Size
distribution
Total biomass
supported at
each size bin
in each cell
% resources demanded
at each size bin
Habitat
suitability
in each cell
Species interactions table for each species:
i.e. % resources demanded at each
size bin by habitat suitability
Temperature
Salinity
Advection
Other
FishBase
SeaAroundUsBase Species
initial
biomass
Fig. 2 Framework to calculate the matrix of energy demand at each size class for each species and to calculate the effects of species
interactions.
Table 2 Summary of abbreviations
Abbreviation Description Details
DBEM Dynamic Bioclimate Envelope Model
ECSuit
Spp;w;iBiomass by competition resSuit
Spp;w;iE S
EDSuit
Spp;w;iBiomass demanded Calculated at each yearly shift
ERSEM European Regional Seas Ecosystem Model
E_S
size,i
Total biomass supported in a cell Calculated from Primary production
GFDL Geophysical Fluid Dynamic Laboratory Earth System model
I Index of cell From 0 to 250 200
NSI No species interactions
resSuit
Spp;w;iActual proportion of resources by competition See Fig. 2
res
Spp,Suit,w
Proportion resources at matrix of energy demand See Eqn (4)
Res-op
Spp,Suit,w
Proportion of resources by opportunity See Eqn (6)
Spp Index of species From 0 to 48 species
SS Size-spectrum (based interactions)
Suit Index of the habitat suitability bin Between 0 and 4 bins
TotalRes
W, i
Total proportion of resources demanded PSpp resSuit
Spp;w;i
W Index of the size spectrum 21 log
2
classes from 2
1
to 2
19
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
6J. A. FERNANDES et al.

DCt¼Pn
iBt;iAiLati
Pn
iBt;iAi
;ð9Þ
where, B
i
is the predicted relative abundance in cell i,Ais the
area of the cell, Lat is the latitude at the centre of the cell and n
is the total number of cells where the species was predicted to
occur. We calculated the rate of range shift as the slope of a
fitted linear regression between the distribution centroid of
the species and time. We expressed latitudinal range shift (LS)
as poleward shift in distance from the equation:
LS ¼DS p=180 6378:2;ð10Þ
where DS is the distribution shift in degree latitude per year.
The models were run for 35 years, from 1970 to 2004, with
environmental forcing predicted from two modelling systems:
(i) the National Oceanographic and Atmospheric Administra-
tion Geophysical Fluid Dynamic Laboratory Earth System
Model (ESM) 2.1 (GFDL); and (ii) the European Regional Seas
Ecosystem Model (ERSEM). GFDL ESM2.1 is a global atmo-
sphere-ocean general circulation model (Delworth et al., 2006)
coupled to a marine biogeochemistry model (TOPAZ; Dunne
et al., 2010) which includes major nutrients and three phyto-
plankton functional groups with variable stoichiometry. For
the GFDL hindcast simulations (Henson et al., 2010), air tem-
perature and incoming fluxes of wind stress, freshwater,
shortwave and longwave radiation are prescribed as bound-
ary conditions from the CORE- version 2 reanalysis effort
(Large & Yeager, 2009). ERSEM is a biogeochemical model
that uses a functional-groups approach incorporating four
phytoplankton and three zooplankton functional groups and
decouples carbon and nutrient dynamics (Blackford et al.,
2004). Data from two different configurations of ERSEM were
applied here: on the global scale a hindcast of the NEMO-ER-
SEM model forced with DFS 4.1 reanalysis for the atmosphere
(Dunne et al., 2010) and on the regional scale a hindcast of the
POLCOMS-ERSEM model for the NW-European shelf forced
with ERA 40 reanalysis (extended with operational ECMWF
reanalysis until 2004) for the atmosphere and global ocean
reanalysis for the open ocean boundaries (more details on the
configuration can be found in Holt et al., 2012; Artioli et al.,
2012). The domain of this global model overlapped the
domain of a regional model of the North Sea area.
Results and discussion
Performance of SS-DBEM and NSI-DBEM
Predicted biomasses from SS-DBEM were generally
lower than those projected from NSI-DBEM (Fig. 3).
The reason is that the energy available from primary
producers limits species’ biomass in SS-DBEM but not
in NSI-DBEM, where species’ carrying capacity depends
mainly on the habitat suitability of the cell. The algo-
rithm in SS-DBEM explicitly modelled interspecific
competition for energy, based on size considerations,
without specifying these interactions (e.g. no diet matrix).
Our approach allows for the development of scenarios
of large-scale shift in species distribution and catches,
complementing other models that have been designed
to achieve this (Cheung et al., 2010; Metcalfe et al.,
2012).
Outputs from SS-DBEM explain slightly more of the
variation in biomass estimated from stock assessments
(FAO area 27) than those from the NSI-DBEM. The
error weighted by maximum catch predicted across
species from SS-DBEM against empirical data is 3.7%
lower than those predicted from NSI-DBEM using
GFDL environmental forcing data and 0.6% lower
using ERSEM data. GFDL might be more accurate
(Fig. 4) for the time period considered since the model
run was forced by re-analysis data such as surface
Fig. 3 Species biomass by body mass class supported in a single coastal cell (30′930′), used as an example. Open circles represent the
biomass that can be supported in this cell using only the size-spectrum component of the model.
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
CLIMATE CHANGE EFFECTS ON MARINE FISHES 7

temperature and wind fields, which is not the case for
ERSEM. However, the differences in mean absolute
error are not significant and might not hold when the
models are used for forecasting. Future work will
explore the causes of this difference, which may not
depend on the modelling itself but on input data such
as environmental forcing, or even on the adequacy of
the assessment data used for the comparison. Finally, a
lower variance in the absolute error in SS-DBEM with
respect to NSI-DBEM model (Fig. 4) is indicative of a
higher precision of simulated biomass from SS-DBEM
(Taylor, 1999). This also supports the view that the pro-
posed modelling approach is a potential advance over
models that do not account for species interactions.
Distribution shifts
Both NSI-DBEM and SS-DBEM projected poleward lati-
tudinal shift of species distributions (Fig. 5), and the
projected shifts are generally consistent between simu-
lations forced by the two sets of Earth System Model
outputs (Table 3). In addition, the projected shift of
pelagic species by the model with interactions is consis-
tently lower than if no interactions are considered
(Table 3). With ERSEM forcing, the median projected
rates of poleward shift are 63.5 km and 54.9 km over
35 years, or 18.1 and 15.7 km decade
1
, from NSI-
DBEM and SS-DBEM respectively. Similar to previous
analysis using NSI-DBEM, all sets of simulations show
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.2 0.4 0.6 0.8
Absolute error distribution Absolute error distribution
0 1 2 3 4 5
0 1 2 3 4 5
Error frequency
Error frequency
ERSEM NSI-DBEM
(0.35 ± 0.09)
ERSEM SS-DBEM
(0.36 ± 0.08)
GFDL NSI-DBEM
(0.347 ± 0.082)
GFDL SS-DBEM
(0.368 ± 0.095)
Fig. 4 Distribution of absolute error of predicted biomass for SS-DBEM and NSI-DBEM and the biomass estimated from stock assess-
ments for the 1991–2003 period in the Northeast Atlantic (FAO Area 27). The time series have been normalized between 0 and 1 before
calculating the absolute error, to ensure that species’ absolute abundances do not affect the results. The comparison is presented for
European Regional Seas Ecosystem Model (ERSEM) (left) and Geophysical Fluid Dynamic Laboratory (GFDL) (right) showing in the
legend mean and standard deviation of the absolute error. A narrower distribution of error (lower standard deviation) in SS-DBEM is
indicative of a higher precision.
1971 1977 1983 1989 1995 2001 1971 1977 1983 1989 1995 2001
NSI-DBEM SS-DBEM
Latitude distance (km)
Latitude distance (km)
Fig. 5 Predicted latitudinal shift of distribution centroids of 49 species of fishes from 1971 to 2004 using European Regional Seas Eco-
system Model (ERSEM) climatic dataset for the NSI-DBEM and SS-DBEM. The thick dark bar represents the median shift of all the spe-
cies in a year, the lower and upper boundaries of the box represent the 25% and 75% quartiles respectively. Positive value indicates
poleward shift relative to species distribution in 1971.
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
8J. A. FERNANDES et al.

a higher rate of range shift for pelagic species than
bottom-dwelling species (Cheung et al., 2009; Jones
et al., 2013). A reduction in the expected geographical
shift of particular populations as a result of ecological
interactions is consistent with the perception of
compensatory ecological processes (Frank et al., 2011).
Shifts in depth are also observed and are strongly dri-
ven by the forcing model considered. The shift in depth
is also dependent on the spatial domain considered.
For example, for demersal species in FAO Area 27, out-
puts from SS-DBEM driven by ERSEM data project a
shift to deeper waters of 1.3 m decade
1
. However,
when we consider North Sea only, the projected shift to
deeper waters is higher at 5.7 m decade
1
.
The slower rates of projected shifts from the
SS-DBEM relative to NSI-DBEM are consistent with
previous literature based on recent observations.
Specifically, Perry et al. (2005) projected a mean rate of
latitudinal shift of 22 km decade
1
from 1980 to 2004 in
the North Sea for six fish species. Comparable rates
of shift (between 18.5 and 18.8 km decade
1
) are
projected here for our modelled subset of species,
which includes four of these species (bib, blue whiting,
Norway pout and witch). Also, Dulvy et al. (2008) esti-
mated that bottom-dwelling species were moving into
deeper waters at an average rate of 3.1 m decade
1
from 1980 to 2004 (19 species of 28 species are common
between this study and Dulvy et al., 2008), which is
slower than our prediction of 5.7 m decade
1
. These
direct comparisons between predicted and observed
shifts need to be interpreted with caution because of
differences in the species included, the spatial domains,
and the time period considered. In addition, our simu-
lations represent average species-level changes without
consideration for stock structure, owing to incomplete
biological data to address the latter. The trend in abun-
dance or range shift of a given species may not neces-
sarily be equivalent to that of every stock of that species
(Petitgas et al., 2012).
Maximum catch
The maximum catch predicted by both DBEM models
(SS and NSI) broadly follows multi-decadal variability
in empirical estimates of total catches for the 1970–2004
time period in the ICES areas (Fig. 6). This is demon-
strated by maximum and minimum points in similar
years, with the highest discrepancy in years around
1985. All the time series show higher maximum values
in the first half of the time period and consistently
lower maximum values in the second half. However,
this negative trend in catches in all the time series is not
statistically significant. The empirical catch data are
aggregated catches by all species reported in ICES areas
as collected in the Eurostat/ICES database on catch sta-
tistics (http://www.ices.dk). The predicted maximum
catch is based on the aggregation of the potential catch
of the 48 modelled species in ICES areas. Despite some
Table 3 Average latitudinal shift in different simulations.
NSI corresponds to simulations where the model does not
incorporate species interactions through the size-spectrum,
whereas SS denotes the use of the species interactions algo-
rithm. Geophysical Fluid Dynamic Laboratory (GFDL) and
European Regional Seas Ecosystem Model (ERSEM) corre-
spond to two different Earth System Models
Latitudinal Shift (km decade
1
)
Projection All species Demersal Pelagic
NSI-DBEM GFDL 16.7 14.1 26.0
SS-DBEM GFDL 13.7 12.6 18.4
NSI-DBEM ERSEM 18.1 15.2 28.2
SS-DBEM ERSEM 15.7 15.3 16.9
Fig. 6 Predicted changes in maximum catch compared with empirical catch data. Time series has been normalized between 0 and 1 to
compare interannual variability.
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
CLIMATE CHANGE EFFECTS ON MARINE FISHES 9

discrepancies, the models are able to reproduce general
trends in observed fisheries productivity in the North
East Atlantic, providing some confidence in their utility.
Catches predicted from SS and NSI approaches show
similar patterns when the most abundant and commer-
cially important species are aggregated. Further work
will focus on examining the effects of different model-
ling approaches on catch predicted for specific species,
areas (e.g. ICES areas) or size classes.
Model uncertainty
Projections from NSI- and SS-DBEM are sensitive to the
environmental variables projected by the Earth System
Models and used to force the ecological models. Earth
System Models have a number of limitations when
applied to fisheries problems (Stock et al., 2011). Their
resolution is relatively coarse to capture ecological pro-
cesses (generally ca. 1 degree in the ocean) and they
also do not capture well the coastal and continental
shelf ocean dynamics. As a result, Earth System Models
are known to systematically project lower primary
production in coastal areas (Steinacher et al., 2010).
Inter-model spread arises from diverse sources, such as
the parameters chosen for sub-grid-scale parametriza-
tions. In addition, there is overall limited availability of
reliable data to calibrate the models. Efforts to improve
the understanding and projections for primary produc-
tion are ongoing (e.g. Holt et al., 2012; Krause-Jensen
et al., 2012), which will likely contribute to improved
performance of DBEM models.
An assumption of the size-spectrum component of
the model is the linear relationship between log-abun-
dance and log-body size classes in the cell. Such an
assumption is made mainly for computational perfor-
mance. In reality, it may be violated by species’ migra-
tions that lead to energy losses and subsidies from
given cells, and by seasonal fluctuations in primary
production (Blanchard et al., 2011).
The relative abundance of individuals at size can be
modified by the overall constraints on energy availabil-
ity. In general, these have limited effect on the projec-
tions because the changes account for a small
percentage of the total abundance of species in the cell
(an average of 0.03% of abundance decrease). However,
the effects are larger and occur in more cells for whit-
ing, blue whiting, Atlantic cod, Norway pout, European
plaice, saithe and Atlantic horse mackerel.
The DBEM modelling approaches have a number of
inherent assumptions and uncertainties that may affect
the performance of the models (Cheung et al., 2009).
First, the models are based on the assumption that
the predicted current species distributions depict the
environmental preferences of the species and are in
equilibrium. Second, the underlying biological hypoth-
esis, represented by the model structure and input
parameters, may be uncertain. Moreover, the models
did not consider the potential for phenotypic and
evolutionary adaptations of the species. As these
assumptions apply to both NSI- and SS- DBEM, they do
not affect the comparison of projections between the
two models. We used theory and empirical data to
model trophic interactions. The modelling approach
does not incorporate the full range or complexity of
interactions among species. This simplification avoids
the difficulties of formalizing transient and complex
species-specific predatory interactions at large-scales. It
also requires no assumptions about the extent to which
species-specific trophic interactions that are seen today
will persist in the future. Furthermore, at the system
level, size-based processes account for much of the
variation in prey choice and trophic structure.
Survey data can provide an alternative way of vali-
dating model outputs (Simpson et al., 2011). However,
there are scale reasons why we did not pursue this type
of validation in this study. Fisheries surveys tend to
focus on particular species assemblages (e.g. pelagic or
bottom-dwelling species), and are designed to provide
a geographical and temporal snapshot that fits with the
life history of target species. As such survey data are
not directly comparable to model outputs for a large
geographical area (FAO area 27).
There are small but quantifiable improvements in
goodness-of-fit with stock assessment abundance esti-
mates, predictions of latitudinal shifts and comparisons
with predicted maximum catch and observed catches.
However, we need to be cautious about our interpreta-
tions of model performance at this stage owing to struc-
tural and parameter uncertainties, and uncertainties in
the models used to generate the environmental forcing.
The similarity of predictions might reflect incorrect
assumptions. For example, we assume that single spe-
cies models do not account for species interaction
because there is no explicit mechanism, even though
species interactions might already be implicitly incor-
porated in its parameterization (e.g. habitat suitability
calculation from observed distribution data). The simi-
larity of predictions might also be attributed to the simi-
lar effects of changing climate on many predators and
competitors and the implicit assumption of the NSI-
DBEM approach that the importance of interspecific
interactions remains the same. In addition, trophic
interactions might not be the main driver of responses
to climate at the basin scale. However, our results at the
scale of the North Atlantic basin, or aggregated ICES
areas, does not mean that trophic interactions may not
have more influence on regional and local responses.
Unfortunately, the earth system and ecological models
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
10 J. A. FERNANDES et al.

described in this article are too complicated to allow
comprehensive explorations of the effects of changing
model structures and parameterisation. Such explora-
tions could be achieved in the long-term by comparing
projections from the DBEMs with alternative parameter
settings for larger datasets of time series of changes in
distribution and abundance from different ocean
regions.
The main benefit of our model comes from unifying
two modelling approaches providing spatially and tem-
porally resolved species and size predictions, with full
consideration for the effects of ecological interactions.
Future development of the DBEM will also attempt to
incorporate other key biological processes that are
likely to be important in determining the responses of
marine fishes and invertebrates to climate change. Our
model has provided new insight into the effects of
ecological interactions on responses to climate and
provides a new tool for further exploring the effects of
future climate change. Predictions, in conjunction with
those from other models, will inform managers about
the range of possible ecological and fishery responses
to a changing environment, thus supporting the devel-
opment of management systems that take account of
the effects of climate change (Perry et al., 2011) and the
ongoing implementation of an ecosystem approach to
fisheries (Garcia & Cochrane, 2005; Rice, 2011). Predic-
tions of the long-term effects of climate currently need
to be considered alongside those used for operational
management, to prepare policy makers and fisheries
governance systems for changes in target fisheries and
dependent communities and economies (Perry et al.,
2011).
Acknowledgements
The research was funded by EURO-BASIN of the European
Union’s 7th Framework Program (Grant Agreement No.
264933). W. Cheung acknowledges funding support from Natu-
ral Sciences and Engineering Research Council of Canada and
National Geographic Society. The funders had no role in study
design, data collection and analysis, decision to publish, or
preparation of the manuscript.
References
Agawin NSR, Duarte CM, Agust S (2000) Nutrient and temperature control of the
contribution of picoplankton to phytoplankton biomass and production. Limnology
Oceanography,45, 591–600.
Artioli Y, Blackford JC, Butensch€
on M et al. (2012) The carbonate system in the North
Sea: sensitivity and model validation. Journal of Marine Systems,102–104,1–13.
Barnes C, Maxwell D, Reuman DC, Jennings S (2010) Global patterns in predator-prey
size relationships reveal size dependency of trophic transfer efficiency. Ecology,91,
222–232.
Barnes C, Irigoien X, De Oliveira JA, Maxwell D, Jennings S (2011) Predicting marine
phytoplankton community size structure from empirical relationships with remo-
tely sensed variables. Journal of Plankton Research,33,13–24.
Behrenfeld MJ, O’Malley RT, Siegel DA et al. (2006) Climate-driven trends in contem-
porary ocean productivity. Nature,444, 752–755.
Blackford JC, Allen JI, Gilbert FJ (2004) Ecosystem dynamics at six contrasting sites: a
generic modelling study. Journal of Marine Systems,52, 191–215.
Blanchard JL, Law R, Castle MD, Jennings S (2011) Coupled energy pathways and the
resilience of size-structured food webs. Theoretical Ecology,4,1–12.
Blanchard JL, Jennings S, Holmes R et al. (2012) Potential consequences of climate
change for primary production and fish production in large marine ecosystems.
Philosophical Transactions of the Royal Society B,367, 2979–2989.
Borgmann U (1987) Models of the slope of, and biomass flow up, the biomass size
spectrum. Canadian Journal of Fisheries and Aquatic Sciences,44 (Suppl. 2), 136–140.
Boudreau PR, Dickie LM (1992) Biomass spectra of aquatic ecosystems in relation to
fisheries yield. Canadian Journal of Fisheries and Aquatic Science,49, 1528–1538.
Brander KM (2007) Global fish production and climate change. Proceedings of the
National Academy of Sciences,104, 19709–19714.
Cheung WWL, Close C, Lam V, Watson R, Pauly D (2008a) Application of macroeco-
logical theory to predict effects on climate change on global fisheries potential.
Marine Ecology Progress Series,365, 187–197.
Cheung WWL, Lam VWY, Pauly D (2008b) Modelling Present and Climate-Shifted Dis-
tribution of Marine Fishes and Invertebrates. Fisheries Centre Research Report 16(3),
University of British Columbia, Vancouver, Canada.
Cheung WWL, Close C, Kearney K, Lam V, Sarmiento J, Watson R, Pauly D (2009)
Projections of global marine biodiversity impacts under climate change scenarios.
Fish and Fisheries,10, 235–251.
Cheung WWL, Lam VW, Sarmiento JL, Kearney K, Watson R, Zeller RK, Pauly D
(2010) Large-scale redistribution of maximum fisheries catch potential in the
global ocean under climate change. Global Change Biology,16, 24–35.
Cheung WWL, Dunne J, Sarmiento JL, Pauly D (2011) Integrating ecophysiology and
plankton dynamics into projected maximum fisheries catch potential under
climate change in the Northeast Atlantic. ICES Journal of Marine Science,68,
1008–1018.
Cheung WWL, Sarmiento JL, Dunne J et al. (2013) Shrinking of fishes exacerbates
impacts of global ocean changes on marine ecosystems. Nature Climate Change,3,
254–258.
Close C, Cheung WWL, Hodgson S, Lam V, Watson R, Pauly D (2006) Distribution
ranges of commercial fishes and invertebrates. In: Fishes in Databases and Ecosys-
tems. Fisheries Centre Research Reports 14 (4). (eds Palomares MLD, Stergiou KI,
Pauly D), pp. 27–37. Fisheries Centre, University of British Columbia, Vancouver.
Delworth TL, Rosati A, Stouffer RJ et al. (2006) GFDL’s CM2 global coupled climate
models. part I: formulation and simulation characteristics. Journal of Climate,19,
643–674.
Dickie LM, Kerr SR, Boudreau PR (1987) Size-dependent processes underlying
regularities in ecosystem structure. Ecological Monographs,57, 233–250.
Dougherty J, Kohavi R, Sahami M (1995) Supervised and unsupervised discretization
of continuous features. In: Proceedings of International Conference on Machine Learn-
ing (eds Prieditis A, Russell S), pp. 194–202. Morgan Kaufmann, San Francisco,
CA, USA.
Dulvy NK, Rogers SI, Jennings S et al. (2008) Climate change and deepening of the
North Sea fish assemblage: a biotic indicator of warming seas. Journal of Applied
Ecology,45, 1029–1039.
Dunne JP, Gnanadesikan A, Sarmiento JL, Slater RD (2010) Technical description of
the prototype version (v0) of tracers of phytoplankton with allometric zooplankton
(TOPAZ) ocean biogeochemical model as used in the Princeton IFMIP model.
Biogeosciences,7(Suppl.), 3593.
Fayyad U, Irani K (1993) Multi-interval discretization of continuous-valued attributes
for classification learning. In: Proceedings of the Thirteenth International Joint Confer-
ence on Artificial Intelligence, pp. 1022–1027. Morgan Kaufmann, San Francisco, CA.
Fernandes JA, Irigoien X, Boyra G, Lozano JA, Inza I (2009) Optimizing the number of
classesin automated zooplankton classification. Journal of Plankton Research,31,19–29.
Fernandes JA, Irigoien X, Goikoetxea N, Lozano JA, Inza I, P
erez A, Bode A (2010)
Fish recruitment prediction, using robust supervised classification methods.
Ecological Modelling,221, 338–352.
Frank KT, Petrie B, Fisher JA, Leggett WC (2011) Transient dynamics of an altered
large marine ecosystem. Nature,477,86–89.
Froese R, Zeller D, Kleisner K, Pauly D (2012) What catch data can tell us about the
status of global fisheries. Marine biology,159, 1283–1292.
Garcia SM, Cochrane KL (2005) Ecosystem approach to fisheries: a review of
implementation guidelines. ICES Journal of Marine Science,62, 311–318.
Hagle TM, Glen EM (1992) Goodness-of-fit measures for Probit and Logit. American
Journal of Political Science,36, 762–784.
Henson SA, Sarmiento JL, Dunne JP et al. (2010) Detection of anthropogenic climate
change in satellite records of ocean chlorophyll and productivity. Biogeosciences,7,
621–640.
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
CLIMATE CHANGE EFFECTS ON MARINE FISHES 11

Holt J, Butensch€
on M, Wakelin SL, Artioli Y, Allen JI (2012) Oceanic controls on the
primary production of the northwest European continental shelf: model
experiments under recent past conditions and a potential future scenario.
Biogeosciences,9,97–117.
IPCC (2007) Summary for policymakers. In: Climate Change 2007: The Physical Science
Basis. Working Group I Contribution to the Fourth Assessment Report of the IPCC, (eds
Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller
HL), pp. 1–18. Cambridge University Press, Cambridge.
Jennings S, Mackinson S (2003) Abundance- body mass relationships in size struc-
tured food webs. Ecology Letters,6, 971–974.
Jennings S, M
elin F, Blanchard JL, Forster RM, Dulvy NK, Wilson RW (2008) Glo-
bal-scale predictions of community and ecosystem properties from simple
ecological theory. Proceedings of the Royal Society B: Biological Sciences,275, 1375–
1383.
Jones MC, Dye SR, Pinnegar JK et al. (2012) Modelling commercial fish distributions:
prediction and assessment using different approaches. Ecological Modelling,225,
133–145.
Jones MC, Dye SR, Fernandes JA, Fr€
olicher TL, Pinnegar JK, Warren R, Cheung
WWL (2013) Predicting the impact of climate change on threatened species in UK
waters. PLoS ONE,8, e54216.
Krause-Jensen D, Markager S, Dalsgaard T (2012) Benthic and pelagic primary
production in different nutrient regimes. Estuaries and Coasts,35, 527–545.
Lam VW, Cheung WWL, Swartz W, Sumaila UR (2012) Climate change impacts on
fisheries in West Africa: implications for economic, food and nutritional security.
African Journal of Marine Science,34, 103–117.
Large W, Yeager S (2009) The global climatology of an interannually varying air-sea
flux data set. Climate Dynamics,33, 341–364.
Merino G, Barange M, Blanchard JL et al. (2012) Can marine fisheries and aquaculture
meet fish demand from a growing human population in a changing climate?
Global Environmental Change,22, 795–806.
Metcalfe JD, Le Quesne WJF, Cheung WWL, Righton DA (2012) Conservation
physiology for applied management of marine fish: perspectives on the role
and value of telemetry. Philosophical Transactions of the Royal Society B,367,
1746–1756.
Pauly D, Hilborn R, Branch TA (2013) Fisheries: does catch reflect abundance? Nature,
494, 303–306.
Pearson RG, Dawson TP (2003) Predicting the impacts of climate change on the
distribution of species: are bioclimate envelope models useful? Global Ecology and
Biogeography,12, 361–371.
Perry AL, Low PJ, Ellis JR, Reynolds JD (2005) Climate change and distribution shifts
in marine fishes. Science,308, 1912–1915.
Perry RI, Ommer RE, Barange M, Jentoft S, Neis B, Sumaila UR (2011) Marine social–
ecological responses to environmental change and the impacts of globalization.
Fish and Fisheries,12, 427–450.
Petitgas P, Alheit J, Peck MA et al. (2012) Anchovy population expansion in the North
Sea. Marine Ecology-progress Series,444,1–13.
P€
ortner HO (2010) Oxygen- and capacity-limitation of thermal tolerance: a matrix for
integrating climate-related stressor effects in marine ecosystems. Journal of Experi-
mental Biology,213, 881–893.
Ricard D, Minto C, Jensen OP, Baum JK (2012) Evaluating the knowledge base and
status of commercially exploited marine species with the RAM Legacy Stock
Assessment Database. Fish and Fisheries,13, 380–398.
Rice J (2011) Managing fisheries well: delivering the promises of an ecosystem
approach. Fish and Fisheries,12, 209–231.
Roessig JM, Woodley CM, Cech JJ, Hansen LJ (2004) Effects of global climate change
on marine and estuarine fishes and fisheries. Reviews in Fish Biology and Fisheries,
14, 251–275.
Schaefer MB (1954) Some aspects of the dynamics of populations important to the
management of the commercial marine fisheries. Bulletin of the Inter-American
Tropical Tuna Commission,1,25–56.
Simpson SD, Jennings S, Johnson MP, Blanchard JL, Sch€
on PJ, Sims DW, Genner MJ
(2011) Continental shelf-wide response of a fish assemblage to rapid warming of
the sea. Current Biology,21, 1565–1570.
Steinacher M, Joos F, Fr€
olicher TL et al. (2010) Projected 21st century decrease in
marine productivity: a multi-model analysis. Biogeosciences,7, 979–1005.
Stock CA, Alexander MA, Bond NA et al. (2011) On the use of IPCC-class models to
assess the impact of climate on living marine resources. Progress in Oceanography,
88,1–27.
Taylor JR (1999) An Introduction to Error Analysis: The Study of Uncertainties in Physical
Measurements. pp. 128–129. University Science Books, Sausalito, CA.
Uusitalo L (2007) Advantages and challenges of Bayesian networks in environmental
modelling. Ecological Modelling,203, 312–318.
©2013 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12231
12 J. A. FERNANDES et al.