BookPDF Available

Modelling Present and Climate-Shifted Distribution of Marine Fishes and Invertebrates

Authors:
ISSN 1198-6727
Fisheries Centre
Research Reports
2008 Volume 16 Number 3
Modelling Present and Climate-
Shifted Distribution of Marine
Fishes and Invertebrates
Fisheries Centre, University of British Columbia, Canada
Modelling Present and Climate-Shifted Distribution of Marine Fishes and
Invertebrates
edited by
William W. L. Cheung, Vicky W.Y. Lam and Daniel Pauly
Fisheries Centre Research Reports 16(3)
72 pages © published 2008 by
The Fisheries Centre,
University of British Columbia
2202 Main Mall
Vancouver, B.C., Canada, V6T 1Z4
ISSN 1198-6727
Fisheries Centre Research Reports 16(3)
2008
M
ODELLING
P
RESENT
A
ND
C
LIMATE
-S
HIFTED
D
ISTRIBUTION
O
F
M
ARINE
F
ISHES
A
ND
I
NVERTEBRATES
edited by
William W. L. Cheung, Vicky W.Y. Lam and Daniel Pauly
C
ONTENTS
Page
D
IRECTOR
S
F
OREWORD
................................................................................................................................... 1
A
CKNOWLEDGEMENTS
......................................................................................................................................2
L
IST OF
A
CRONYMS
............................................................................................................................................3
C
HAPTER
1
D
YNAMIC
B
IOCLIMATE
E
NVELOPE
M
ODEL TO
P
REDICT
C
LIMATE
-
INDUCED
C
HANGES IN
D
ISTRIBUTION OF
M
ARINE
F
ISHES AND
I
NVERTEBRATES
William W. L. Cheung, Vicky W. Y. Lam and Daniel Pauly
Abstract..........................................................................................................................................5
Introduction ..................................................................................................................................5
Methods.........................................................................................................................................7
Results .........................................................................................................................................24
Discussion....................................................................................................................................36
Conclusions .................................................................................................................................40
References ................................................................................................................................... 41
Appendices ..................................................................................................................................46
C
HAPTER
2
M
ODELLING
S
EASONAL
D
ISTRIBUTION OF
P
ELAGIC
M
ARINE
F
ISHES AND
S
QUIDS
Vicky W.Y. Lam, William W. L. Cheung, Chris Close, Sally Hodgson, Reg Watson and Daniel
Pauly
Abstract........................................................................................................................................ 51
Introduction ................................................................................................................................ 51
Methods.......................................................................................................................................52
Results .........................................................................................................................................57
Discussion ...................................................................................................................................61
References ...................................................................................................................................62
C
HAPTER
3
A
SYMMETRY IN
L
ATITUDINAL
,
L
ONGITUDINAL AND
B
ATHYMETRIC
D
ISTRIBUTION OF
MARINE
F
ISHES AND
I
NVERTEBRATES
Daniel Pauly, William W. L. Cheung, Chris Close, Sally Hodgson, Vicky W.Y. Lam and Reg
Watson
Abstract .......................................................................................................................................63
Introduction ................................................................................................................................64
Methods.......................................................................................................................................66
Results .........................................................................................................................................69
Discussion ....................................................................................................................................71
References ...................................................................................................................................72
A Research Report from the Sea Around Us Project, Fisheries Centre, UBC
72 pages © Fisheries Centre, University of British Columbia, 2008
F
ISHERIES
C
ENTRE
R
ESEARCH
R
EPORTS ARE ABSTRACTED IN THE
FAO
A
QUATIC
S
CIENCES AND
F
ISHERIES
A
BSTRACTS
(ASFA)
ISSN 1198-6727
1
D
IRECTOR
S
F
OREWORD
This report is the first publication of the Sea Around Us Project focusing on global warming and
its effects on marine fisheries. It is therefore a report on methodology: we had to first develop
methodological approaches before results could be presented.
The first contribution in this report describes a new approach and software for simulating the
widely documented poleward movement of marine fishes and invertebrates in response to
warming oceans. The software for simulating these movements implements a model, largely
driven by the temperature changes predicted for the next decades by coupled atmosphere-ocean
models, which considers a moderately large number of processes (reproduction, survival,
migration, etc.) and features of the organisms in question (affinities to certain habitats, depth
ranges, etc.). However, the model was designed such that it would be straightforward to
parameterize, at least for 1,500+ species and higher taxa used by the Food and Agriculture
Organization of the United Nations (FAO) to reports on global marine fisheries statistics, and for
all of which the Sea Around Us Project has basic information, including detailed distribution
range maps.
Thus, the first paper in this report will serve a starting point for several planned articles on global
warming effects on marine communities and fisheries, with the model at its core being gradually
modified and improved as applications are completed.
The second and third contributions in this report deal with the distribution range maps of marine
taxa used by the model in the first. Thus, they propose a number of simple adjustments which
help take seasonality and other modifying factors into account, both when generating present
distribution range maps, and when shifting them poleward using the model mentioned above.
The projections that we hope to generate, using these data and models, will obviously not be the
last word on the poleward migration of marine fishes and invertebrates. However, because they
cover a set of globally important species, they will enable the Sea Around Us Project to contribute
in a major way to debates on the possible impacts of global warming on marine fisheries and
biodiversity, a key environmental issue for the next decades.
Daniel Pauly
Director, UBC Fisheries Centre
4 January 2008
2
A
CKNOWLEDGEMENTS
We are grateful to the Sea Around Us Project team for their advice and help. We are thankful for
the advice from Prof. Jorge Sarmiento (the Atmospheric and Oceanic Sciences Program,
Princeton University) and his research team, and from Dr. Jessica Meeuwig (University of
Western Australia). We thankfully acknowledge partial funding from the University of Western
Australia. This is a contribution of the Sea Around Us Project, initiated and funded by the Pew
Charitable Trusts.
3
L
IST OF
A
CRONYMS
Acronym or
variable
Definition or description (dimension)
A
i
Relative abundance in cell i
a Southern boundary of the annual and winter distribution range (coordinate)
AS Actual shift in the distribution range
AvgS Latitudinal shift of centroid in summer (distance)
AvgS’ Latitudinal shift of centroid in winter (distance)
b Southern boundary of the summer distribution range (coordinate)
c' Constant for calculating the intrinsic rate of population increase
c Centroid of the annual distribution range (coordinate)
C (Annual) Latitudinal position of the centroid of species’ annual average relative
distribution (coordinate)
CA
i,t
Absolute coral abundance at grid cell (30’ x 30’) i and time (year) t (Area)
CM2.1 Coupled Model, version 2.1
C
s
Latitudinal position of the centroid of species’ summer distribution
(coordinate)
C
w
Latitudinal position of the centroid of species’ winter distribution (coordinate)
CS’ Actual shift in centroid of the distribution range (distance)
CS
s
Maximum potential shift in centroid’s latitudinal positions in summer
(distance)
CS
w
Maximum potential shift in centroid’s latitudinal positions in winter
(distance)
D Diffusion coefficient
d Northern boundary of the winter distribution range (coordinate)
DS Ratio of future to current habitat suitability
Dep Depth (length)
Dist
ij
Distance between two adjacent cells (length)
DM The Developmental type of larvae
e Northern boundary of the annual and summer distribution range (coordinate)
E Emigration of animals from a cell (relative abundance)
Er Radius of the Earth (length)
G Intrinsic population growth (time
-1
)
GFDL Geophysical Fluid Dynamics Laboratory
GR Minimum grid resolution (length)
H Habitat types
I Net adult migration (relative abundance)
Ice Sea ice coverage (Area)
I
ji
Net migrated adults at cell i from surrounding cells j (relative abundance)
K Von Bertalanffy growth parameter (time
-1
)
k Scaling factor representing the sensitivity of the calculated emigration rate to
changes in environmental suitability
KC
i
Population carrying capacity at cell i (relative abundance)
KR Carrying capacity ratio of the destination cell
Lat Latitude (coordinate)
Lav Larval abundance (relative abundance)
LC Latitude at the centre of the grid cell i where the specific area of coral occurs
(coordinate)
L Settled larvae at cell i from surrounding cells j (relative abundance)
L
U
Upper latitudinal limit (coordinate)
L
L
Lower latitudinal limit (coordinate)
Lon Longitude (coordinate)
M Natural mortality rate (time
-1
)
m
i
The instantaneous movement rate for randomly moving organisms to
emigrate across each cell boundary (time
-1
)
4
L
IST OF
A
CRONYMS
(C
ON
T
)
Acronym or
variable
Definition or description (dimension)
m
i(base)
Movement rate in the absence of any gradient (time
-1
)
MGSVA Mariano Global Surface Velocity Analysis
MI Motility Index
MP (Annual) Latitudinal position of the mid-point of the annual distribution (coordinate)
MP
S
Mid-point between the centroid and the southern bound of the annual
distribution (coordinate)
MP
W
Mid-point between the centroid and the northern bound of the annual
distribution (coordinate)
MP
S
Latitudinal position of the mid-point in summer (coordinate)
MP
W
Latitudinal position of the mid-point in winter (coordinate)
Mort
i,t
Dispersed animals that die at cell i and time step t (relative abundance)
MS’ Actual shift in the latitudinal position of the mid-point (distance)
MS
s
Maximum potential shift in the latitudinal position of the mid-point in
summer (distance)
MS
w
Maximum potential shift in the latitudinal position of the mid-point in winter
N or n Number of spatial cells where the species occurs (count)
NL’ Latitude of the northern bound in winter (coordinate)
NL Latitude of the original northern bound (coordinate)
NOAA National Oceanic and Atmospheric Administration
P Habitat suitability
PLD Pelagic larval dispersal (time)
r Intrinsic rate of population increase (time
-1
)
R Rate of larval production (time
-1
)
RE Rate of the re-entrance of the emigrated animals to the source cells (time
-1
)
S Larval settlement rate (time
-1
)
Skew Degree of skewness
SL’ Latitude of the southern bound in summer (coordinate)
SL Latitude of the original southern bound (coordinate)
SST Sea Surface Temperature (temperature)
T Average water temperature in the animal’s range (temperature)
TG Function to calculate the temperature-gradient index at the upper and lower
latitudinal limits
Ta Optimal preferred temperature of a species (temperature)
TPP Temperature Preference Profile
u East-west current velocity (length time
-1
)
v North-south current velocity (length time
-1
)
V
i
Adult dispersal rate (time
-1
)
W
inf
Asymptotic weight (weight)
x Length of a cell (length)
δ
2
Standard deviation of the normal distribution function f
λ Instantaneous rate of larval mortality and settlement (time
-1
)
5
C
HAPTER
1
D
YNAMIC BIOCLIMATE ENVELOPE MODEL TO PREDICT CLIMATE
-
INDUCED CHANGES IN DISTRIBUTION OF MARINE FISHES AND
INVERTEBRATES
1
William W. L. Cheung, Vicky W.Y. Lam and Daniel Pauly
Sea Around Us Project, Fisheries Centre, Aquatic Ecosystems Research Laboratory, 2202 Main
Mall, The University of British Columbia, Vancouver, British Columbia, Canada. V6T 1Z4.
ABSTRACT
Global climate change is recognized as an important determining factor for the future
distributions of marine organisms, notably fishes and invertebrates. Shifting of distribution range
may affect global marine fisheries and have large socio-economic implications. However, global-
scale evaluation of the impact of climate change on marine species is lacking. In this paper, we
develop a dynamic bioclimate envelope model to predict the effect of climate change on the
distributions of marine species with emphasis on commercially exploited fishes and invertebrates.
First, the model infers, for various species, bioclimate envelopes based on their current
distribution. Bioclimate envelopes are defined by sea water temperature, bathymetry, habitats
and distance from sea ice. Secondly, the model predicts the shifting of the bioclimate envelopes
induced by changes in climate variables. Simultaneously, following the shifting of the bioclimate
envelopes, the model simulates movement of relative abundance through changes in population
growth, mortality, larval dispersal and adult movement. We test the model with several
commercially exploited fish species with widely different biogeography. The model provides
reasonable and robust predictions of future distribution ranges of the four species under different
scenarios of sea water warming. Moreover, the predictions are robust to major model
assumptions and parameter uncertainty. Using realistic climate change predictions from the
NOAA/GFDL Coupled Model, this model will be used to evaluate impacts of climate change on
global marine fisheries.
INTRODUCTION
There is ample evidence from empirical observations and climate models indicating that mean
global temperatures have been increasing over the last 100 years (IPCC 2007). Global
temperature has increased by over 0.6
o
C since 1900 and it may continue to increase at a rate of
around 0.2
o
C per decade (IPCC 2007). Biological responses to this change have been observed in
both terrestrial and marine biomes (Murawski 1993; Hughes 2000; McCarty 2001; Parmesan &
Yohe 2003; Perry et al. 2005; Hobday et al. 2006). The responses include changes in physiology
(e.g. productivity), geographic range and phenology at population, species, community and
ecosystem levels (Hughes 2000; McCarty 2001). For instance, nearly two-thirds of marine fishes
in the North Sea shifted in mean latitude or depth or both over 25 years as sea temperature
increased (Perry et al. 2005). During the last century, annual growth rates for the juveniles of
eight long-lived fish species in the southwest Pacific increased in shallow waters and decreased in
deep waters where ocean warming and cooling occurred, respectively (Thresher et al 2007). This
agrees with the quantitative model of fish physiology, which predicts increasing growth
performance and fecundity in higher latitude and the converse in lower latitude as sea
1
Cited as: Cheung, W.W.L., Lam, V.W.Y., Pauly, D. 2008. Dynamic bioclimate envelope model to predict climate-induced
changes in distribution of marine fishes and invertebrates, p. 5-50. In: Cheung, W.W.L, Lam, V.W.Y., Pauly, D. (eds.)
Modelling Present and Climate-shifted Distribution of Marine Fishes and Invertebrates. Fisheries Centre Research Report
16(3). Fisheries Centre, University of British Columbia [ISSN 1198-6727].
6
temperature increases (Pörtner et al. 2001). Moreover, global warming may increase
extinction/extirpation risk of some populations and species (Pounds 2001; Thomas et al. 2004).
It may also have large implications for communities and industries that depend on these marine
species for food and income (Roessig 2004). Thus, predicting effects of climate change on marine
species is important in understanding the overall impacts of such global changes on human
society and ecosystem.
Bioclimate envelope model has been widely used to predict the effects of climate change on the
distribution range of terrestrial species (Pearson & Dawson 2003). A bioclimate envelope can be
defined as a set of physical and biological conditions that are suitable to a given species. Such a
bioclimate envelope is generally identified by studying the relationships between current species
occurrences and biogeographical attributes using statistical methods (e.g. generalized additive
model, Luoto et al. 2005) or artificial intelligence models (e.g. artificial neural network, Pearson
et al. 2002). Thus, shifts in species distributions can be predicted by evaluating changes in
bioclimate envelope under climate change scenarios. For instance, a bioclimate envelope model
based on a genetic algorithm and museum specimens was used to predict distributional shifts of
Mexican terrestrial faunas under global climate change scenarios (Peterson et al. 2002).
Bioclimate envelope models are important tools to provide guidance for policy making (Hannah
et al. 2002) although predictions from such models may be uncertain (Peason & Dawson 2003;
Araújo et al. 2005; Araújo & New 2006; Lawler et al. 2006). Critiques of bioclimatic modelling
are detailed by Pearson & Dawson (2003); they include the lack of consideration of biotic
interactions, evolutionary change and species dispersal. Understanding these processes is
important to a comprehensive evaluation of impacts of climate change on the marine ecosystem
(Harley et al. 2006). However, bioclimate envelope models are among the best tools available to
predict large scale potential ecological changes under climate change scenarios. Their applications
are particularly appropriate at large spatial scale, as will be presented below. Also, model
uncertainty can be reduced by examining results from multiple alternative models (Pearson &
Dawson 2003). Thus, predictions from such models are useful in generating hypotheses of
possible ecological impacts from climate change.
Two global databases, i.e., that compiled by the Sea Around Us Project (Watson et al. 2004), and
FishBase (Froese & Pauly 2007) provide most of the information needed for the development of a
global bioclimate envelope model for commercially exploited marine species. This is important as
predicting future responses to climate change in marine biomes lags behind those for terrestrial
species. Specifically, application of a bioclimate envelope model to large marine ecosystem is
limited. This is partly because of the general lack of large-scale biological, ecological and
biogeographical data for most marine species. However, such data are made available from the
aforementioned global databases. For instance, distributions of relative abundance of all
commercially exploited marine species are available from the Sea Around Us Project (Close et al.
2006). Combining such data with physical attributes such as global ocean temperature,
bioclimate envelopes of the marine species could be inferred. These make it possible to construct
bioclimate envelope models to predict impacts of climate change on all exploited marine species.
This contribution documents a bioclimate envelope model that aims to predict the effects of
global climate change on marine fishes and invertebrates. A major advance of the bioclimate
envelope model presented in this contribution is the incorporation of population and dispersal
dynamics for predicting impacts of climate change on distribution range. Such dynamics are
important factors in determining biogeography of marine system under climate change scenarios
(Pearson & Dawson 2003; Harley et al. 2006). Although our model does not explicitly deal with
the effects of biological interactions and evolutionary changes (Pearson & Dawson 2003), we
discuss the implications of these factors for the uncertainty of our model predictions. For the time
being, we evaluate the performance of our model by using hypothetically generated climate data,
as a first step to quantitatively evaluate the impacts of climate change on marine fishes and
invertebrates. We then discuss how this approach will be applied to the study of likely effects of
global changes to fisheries at a global scale.
METHODS
We developed a simulation model to predict changes in global distributions of commercial species
under different climate change scenarios. This model is essentially a bioclimate envelope model
combined with dynamic dispersals of animals. The Sea Around Us Project uses the distribution of
commercial species (fishes and invertebrates) to map marine fisheries (Watson et al. 2004). The
distributions have all been recently improved by Close et al. (2006), Lam et al. (this vol.) and
Pauly et al. (this vol.). The future distributions of these species were assumed to be predictable
from changes in ocean temperature, ocean advections and habitats (coral and sea ice coverage).
Details of the model are described in the following.
Current species distributions
Descriptions of current distribution of marine species are fundamental to predicting changes in
species distributions. The Sea Around Us Project produced distribution maps of over 1,200
commercially exploited fishes and invertebrates (www.seaaroundus.org). Each species’
distribution map is presented as potential relative abundance in 30’ latitude x 30’ longitude cells
of the world ocean. The map was generated by a bio-climate model that predicts the suitability of
each 30’ lat. x 30 long. cell to the studied species. Boundaries of each species’ distribution were
delineated by the following information: (1) latitudinal range; (2) depth range; (3) affinity to
certain habitats; (4) known distribution boundaries from published literature or experts’
knowledge, e.g., presence in a United Nations Food and Agriculture Organization (FAO)
statistical area. Realistic assumptions were made on distributions of relative abundance within
the above ecological limits. For instance, an ‘equatorial submergence’ filter was used to account
for the tendency of demersal species to inhabit shallower waters in higher latitude (Ekman 1957;
Close et al. 2006). It is emphasized that the Sea Around Us Project does not explicitly use
temperature and primary production for any of the procedures discussed above. A description of
the procedures to predict current species distributions are documented in Close et al. (2006).
Lam et al. (this vol.) and Pauly et al. (this vol.) present the modifications required to represent the
distribution of seasonally migrating fishes and latitudinal and longitudinal distribution
asymmetry, respectively.
Predicted current distributions of seven species: Nassau grouper (Epinephelus striatus,
Epinephelidae), Small yellow croaker (Larimichthys polyactis, Sciaenidae), Polar cod
(Boreogadus saida, Gadidae), Atlantic cod (Gadus morhua, Gadidae), Western Australian rock
lobster (Panulirus Cygnus, Palinuridae), Antarctic toothfish (Dissostichus mawsoni,
Nototheniidae), and summer and winter distributions of Australian ruff (Arripis georgianus,
Australian ruff) were shown here as examples (Figure 1). These examples represent species with
different life history and ecology, and from different geographic areas. We used these examples to
illustrate the models described in this paper.
8
a)
b)
c)
d)
Figure 1. Predicted current distributions of (a) Nassau grouper (Epinephelus striatus), (b) Small yellow croaker (Larimichthys polyactis), (c) Polar cod
(Boreogadus saida) and (d) Atlantic cod (Gadus morhua) generated from the methodology described in Close et al. (2006).
9
e)
f)
g)
h)
Figure 1. Predicted current distributions of (e) Western Australian rock lobster (Panulirus cygnus), (f) Antarctic toothfish (Dissostichus mawsoni,
Nototheniidae), and (g) summer and (h) winter distributions of Australian ruff (Arripis georgianus) generated from the methodology described in Close et al.
(2006) and Lam et al. (2007, this vol.).
10
Model algorithms
Calculating environmental preferences
Profiles of affinity to environmental and climatic attributes (i.e., seawater temperature, depth,
habitat-associations and distance from edge of sea ice) for each species were based on current
distribution maps generated by the Sea Around Us Project using the methodology documented in
Close et al. (2006). We assume that the predicted current distributions realistically depict the
bioclimate envelopes preferred by the species. In this paper, bio-climate envelopes are defined by
(a) sea water temperature; (b) bathymetry; (c) habitats and (d) distance from sea ice. For each of
these four bioclimate attributes, we expressed a species’ preference to different attribute values by
its relative abundance.
a. Sea water temperature
Distributions of marine ectothermic animals are strongly dependent on temperature, as these
animals are limited by their insufficient capacity of circulation and ventilation under low and high
temperature (Pörtner 2001). Physiological performance of marine invertebrates and fishes
changes continuously from optimum level to outside their thermal tolerance limits (Frederich &
Pörtner 2000; Pörtner 2001). Also, foraging theory predicts that animals will select areas where,
eventually, their growth rates can be maximized (Stephens & Krebs 1986). As growth is strongly
dependent on physiological performance (Pauly 1980; Elliott 1982; Regier et al. 1990), it is
reasonable to assume that ectothermic animals tend to inhabit area within their optimal
temperature range (Hughes & Grand 2000). Thus, current distributions of marine animals should
depict, at least roughly, their temperature preference.
We calculated the temperature preference profile (TPP) of each species by combining current sea
temperature with species’ predicted distribution ranges, the latter being determined by the Sea
Around Us Project algorithm (Close et al. 2006). We define TPP as the probability of occurrence
of a species at different sea water temperatures. To infer TPP from the predicted distribution
maps, firstly we converted the observed sea temperature data obtained from Met Office Hadley
Centre observations datasets (http://hadobs.metoffice.com/hadisst/) to the 30’ x 30’ resolution
of the Sea Around Us Project distribution maps. We overlaid the sea temperature over current
distribution maps and calculated species’ relative abundance in different temperatures. We
assume that relative abundances of demersal and benthopelagic species (e.g., Atlantic cod)
depend mainly on annual sea bottom temperature while pelagic species (e.g. Atlantic herring)
depend on seasonally-averaged (summer and winter) sea surface temperature (see Lam et al. this
vol.). We examined the TPP of the studied species to ensure that temperature preference by each
species is reasonable. Criteria for judging the acceptability of the temperature profiles include (1)
whether the profile is approximately unimodal; and (2) the coefficient of variation of preferred
temperature is less than 50%. Distribution maps that resulted in clearly multi-modal temperature
profiles or a wide range of preferred temperature might be predicted inaccurately and were
reviewed. Also, we assume a linear change in species preference (relative abundance) to water
temperature between consecutive temperature classes (Figure 2).
In some cases, sea-water temperature preference by a species is not uni-modal, i.e., there is more
than one distinct peak of relative abundance in different temperatures (e.g., Atlantic cod and
Western Australian rock lobster, Figure 2d and 2e). Physiological performance of marine
ectotherms generally peaks at certain optimum temperature from where it declines to their
thermal tolerance limits (Frederich & Pörtner 2000; Pörtner 2001). As we assume that each
species distribution represents a single uniform distribution, we consider the multi-modal
temperature preference distributions as artifacts which resulted from uncertainties of the original
species distribution. As this may lead to unrealistic predictions of species’ responses to changes in
sea temperature, we smoothed the TPP with running-mean to ensure that the distributions were
generally uni-modal. To minimize distortion to the original temperature preference distribution,
the number of temperature class averaged in the running mean calculation was increased from 3
until a uni-modal distribution was obtained. In the case of Western Australian rock lobster, a 3-
temperature-classes running-mean is required to change the original bi-modal distribution to
uni-modal (Figure 3).
11
a)
0
5
10
15
20
25
30
0 3 6 9 12 15 18 21 24 27 30
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Nassau grouper
b)
0
2
4
6
8
10
12
14
16
18
20
4 6 7 9 10 12 13 15 16 18 19 21 22 24 25
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Small yellow croaker
c)
0
5
10
15
20
25
30
35
40
-2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Polar cod
d)
0
2
4
6
8
10
12
14
16
-3 -1 1 3 5 7 9 11 13 15
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Atlantic cod
e)
0
5
10
15
20
25
30
4 6 7 9 10 12 13 15 16 18 19 21 22 24 25 27
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Western Australian
rock lobster
f)
0
20
40
60
80
100
-2.7 -1.2 0.3 1.8 3.3
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Antarctic toothfish
g)
0
10
20
30
40
50
60
70
12 13 14 15 16 17 18 19 20 21 22 23 24
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Australia ruff (summer)
h)
0
5
10
15
20
25
30
35
40
17 18 19 20 21 22 23 24 25 26 27 23 24
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Australia ruff (winter)
Figure 2. Temperature Preference Profile (TPP) expressed as relative abundance in areas with
different sea temperature of: (a) Nassau grouper, (b) Small yellow croaker, (c) Polar cod, (d)
Atlantic cod, (e) Western Australian rock lobster, (f) Antarctic toothfish, (g) and (h) summer and
winter distributions of Australian ruff. We assume linear changes in a species’ preference to sea
water temperature between consecutive temperature classes (lines).
a)
0
5
10
15
20
25
30
4 6 7 9 10 12 13 15 16 18 19 21 22 24 25 27
Sea w ater tempe rature (C
o
)
Relative abundance (%)
b)
0
5
10
15
20
25
30
4 6 7 9 10 12 13 15 16 18 19 21 22 24 25 27
Sea w ater tempe rature (C
o
)
Relative abundance (%)
Figure 3. Temperature Preference Profile of Western Australian rock lobster: (a) original
distribution, (b) smoothing by 3 temperature classes.
b. Depth limits
We assume that a species’ distribution is also limited indirectly by depth. Thus, there are lower
and upper limits of water depth outside of which a species does not occur. Different levels of
temperature, oxygen concentration, food availability and predation pressure exist at different
water depths. Vertical distributions of marine fishes and invertebrates were suggested to be
limited partly by these factors in both freshwater and marine environments (Matthews et al. 1985;
Pihl et al. 1991; Orlowski 1999). Also, animals inhabiting extreme depth generally develop special
morphological and physiological adaptations (Helfman et al. 1997). Thus, species that are
adapted to surface waters cannot occur in deep water and conversely. These create limits to the
range of water depth where different species occur.
12
We estimate lower and upper depth limits from species’ current distribution. The Sea Around Us
Project algorithm uses depth limits as criteria to predict a species’ distribution. However, because
of other physiological or ecological limitations (e.g., temperature) that restrict the distribution of
the species, the current species distribution predicted by the Sea Around Us Project algorithm
(Close et al. 2006) may not depict the extreme bathymetric limits of the species. Also, depth
limits used by Close et al. (2006) were imprecise in most cases. Therefore, we accounted for the
high uncertainty of the depth limits by widening them in the current species distribution by 100%.
For instance, the original depth limits of Small yellow croaker used by Close et al. (2006) is 1 –
105 m, but wider limits of 0.5 210 m were used in our model. We assume that a species would
maintain its depth limits when its distribution changes as a result of global climate change.
c. Habitats
Each species is assigned an index of association to one or more of the four habitat types: coral
reefs, estuaries, seamounts and other habitats. The index represents relative density of a species
in the particular habitat. It was assigned based on qualitative descriptions of the ecology of the
species from FishBase or other publications and literature (Close et al. 2006; Cheung et al. 2007).
Distribution of relative abundance obtained from the Sea Around Us Project algorithm had been
adjusted based on the habitat-association index and global distributions of the four habitat types
(Close et al. 2006, Table 1).
Table 1. Habitat categories for which global maps are available in the Sea Around Us Project.
Categories Origin of global map Terms often used to describe
these categories
Estuary Alder (2003) Estuaries, mangroves, river mouth.
Coral UNEP World Conservation
Monitoring Centre (2005)
Coral reef, coral, atoll, reef slope.
Seamounts Kitchingman and Lai (2004) Seamounts.
Other habitats --- Muddy/sandy/rocky bottom.
d. Distance from sea ice
Polar ecosystems, and distributions of their associated species, are largely shaped by the
dynamics of sea ice (Longhurst 1981). In both the Arctic and Antarctic, primary productivity
under and around sea ice is generally high. For instance, in the Antarctic, phytoplankton growth
is enhanced by dynamics of mixed-layer of water affected by influx of lower salinity water from
ice-melting. Sea ice cover also provides a habitat that allows for maximum utilization of sunlight
and thus enhances primary production (Eicken 1992). Zooplankton such as krill (Euphasia
superba) forage and take refuge under sea ice (Daly & Macaulay 1988; Brierley et al. 2002). The
zooplanktons, in turn, form the basis of the foodweb which support fishes and mammals in polar
ecosystems (Eicken 1992; Legendre et al. 1992; Longhurst 1981). Thus, polar fishes such as Arctic
cod and Antarctic toothfish are physiologically adapted to polar environments (Farrell &
Steffensen 2005) and generally range close to sea ice (Legendre et al. 1992; Fuiman et al. 2002).
It is therefore reasonable to assume that polar species are partly dependent on the presence of
sea ice at least at a certain distance. To be consistent with the current species distributions, which
represent annual average (except for pelagic fishes), annual average sea ice distribution was used.
In this study, we used average monthly ice extent (19179-1999) with its border defined by a
minimum of 50% sea ice coverage. The sea ice data were obtained from the US National Snow &
Ice Data Centre web site (http://nsidc.org/data/smmr_ssmi_ancillary/trends.html#gis) and
processed by Kaschner (2004). We calculated the distance between the nearest sea ice and the
centre of each 30’ lat. x 30’ long. cell. Overlaying species’ distribution from the Sea Around Us
Project algorithm on maps of nearest distance from sea ice, we calculated polar species’ relative
abundance at different distances from sea ice.
In this study, we used a biogeographical definition of polar species (Møller et al. 2005), i.e.,
species that range mostly (≥ 75% of their relative abundance) within the Arctic or Antarctic were
categorized as polar. Maps of the Arctic and Antarctic from ller et al. (2005) were employed
(Figure 4). A list of exploited polar fishes and invertebrates is shown in Appendix 1.
13
Figure 4. Map of the Arctic and Antarctic (in grey), based on Møller et al. (2005).
Dynamics of climate change-induced range shift
We developed a model to predict climate change-induced shift in species distributions. Spatial
and temporal dynamics of populations are assumed to be determined by larval and adult
dispersals, immigration, intrinsic population growth and extirpation. The rates of these processes
are dependent on the ‘carrying capacity’ in each area (30’ lat. x 30 long. cell). Here, carrying
capacity is defined as the maximum relative abundance of a species in a cell. It is largely
dependent on the environmental suitability of the cell to the species. All these processes are
incorporated into the model. Since the model aims to predict changes in relative species
distributions while accurate predictions of absolute changes in abundance are not necessary, we
simplified the population dynamic models to reduce the number of required parameters that are
otherwise unavailable for most of the studied species. Details of the population dynamic model,
along with all the assumptions and simplifications, are reported in the following.
Changes in relative abundance of a species in each 30’ lat. x 30’ long. cell (i) at each time step t
can be expressed as:
=
++=
N
j
jijii
i
ILG
dt
dA
1
…1)
where A
i
is the relative abundance of cell i, G is the intrinsic population growth, and L
ji
and I
ji
are
settled larvae and net migrated adults from surrounding cells j, respectively.
Growth (G)
We modelled intrinsic population growth through a logistic growth function (Hilborn & Walters
1992):
)1(
i
i
ii
KC
A
ArG =
…2)
where r is intrinsic rate of population increase, A
i
and KC
i
are the relative abundance and
population carrying capacity at cell i, respectively. A major assumption of our model is that
current species distributions calculated from the algorithm of Close et al. (2006) as modified by
14
Pauly et al. (this vol.) and Lam et al. (this vol.) are in equilibrium, and every spatial cell at time-
step (t) 0 is at its carrying capacity, i.e., KC
i,t=0
= A
i,t=0
and, therefore, G
i, t=0
=0.
We used an indirect method to approximately estimate intrinsic rate of population increase (r).
Empirically estimated intrinsic rates of increase of most exploited marine species were not
available because of a lack of time-series population data (e.g., abundance, catch rate). Therefore,
we calculated r based on the estimated natural mortality rate (M):
Mcr
=
'
…3)
where c’ is a constant that commonly ranges between 1 and 3. Natural mortality rate was
estimated from an empirical equation (Pauly 1980):
)log(4687.0)log(6757.0)log(0824.04851.0
inf
TKWM ++=
…4)
where W
inf
is asymptotic weight, K is the von Bertalanffy growth parameter and T is the average
water temperature in the animal’s range. Acknowledging the high uncertainty of r estimated from
this method, we compared the simulation results obtained from a range of r to evaluate the
sensitivity of our model to the uncertainty of the population growth rate.
The carrying capacity of a cell varies positively with the habitat suitability to the studied species.
As habitat in a cell (defined here by temperature, bathymetry, habitat types and ice-coverage)
becomes more suitable for the animal, carrying capacity should also increase. Thus, in our model,
carrying capacity KC in cell i is modified according to the change in habitat suitability (P) between
time (t), i.e.,
t
t
tt
P
P
KCKC
1
1
+
+
=
…5)
and
)()()()(
IcePHPDepPTPP
=
…6)
where T, Dep, H and Ice refer to temperature, bathymetry, habitat types and sea ice coverage,
respectively.
If an initially unoccupied spatial cell (A
0
= KC
0
= 0) or a cell without suitable habitat becomes
suitable for the survival of a species as global climate changes, its new carrying capacity is
assumed to be the average carrying capacity of other ‘occupied’ cells with similar habitat
suitability.
Spawning and larval dispersal (L)
Immigration consists of two components: larval dispersal and migration of adults. We assume
larval production (Lav) to be directly proportional to the relative abundance in a cell (i):
ii
ARLav =
…7)
where R is an assumed rate of larval production. Our model does not have an explicit stock-
recruitment relationship. The main focus of the model is to simulate changes in cells’ carrying
capacity and extent of dispersal by the species. The absolute amount of larval production affects
only the rate at which carrying capacity level in a cell is being approached, and should not affect
the general prediction by the model. On the other hand, we tested the sensitivity of the model
outputs to this simplification by comparing model results obtained from a range of larval
production rate (R).
Our model calculates dispersal of larvae through ocean current and diffusion. We assume that
pelagic larvae disperse passively from surrounding source’ areas through ocean surface current
15
and diffusion. Thus, the magnitude of larval recruitment is dependent on pelagic larval duration
(PLD), strength and direction of ocean currents and diffusivity. PLD, expressed in days, can be
calculated from an empirical equation established from a meta-analysis of PLD from 72 species of
fish and invertebrates (O’Connor et al. 2007):
2
20
))/(ln(283.0))/(ln(1368)ln(
cc
TTTTPLD =
β
…8a)
)(714.0)ln((739.0739.0
0
DMT ++=
β
…8b)
N
T
T
N
i
i
=
=
1
)ln(
)ln(
…8c)
where T is sea surface temperature, T
c
=15
o
C, N is the number of spatial cells (i) where the
species occurs. DM is the developmental type of larvae. DM is 0 or 1 for lecithotrophic (non-
feeding development) or planktotrophic (feeding development) larvae, respectively (O’Connor et
al. 2007). Thus, PLD is shorter in areas where the sea water temperature is higher. On the other
hand, PLD is longer when the average temperature over the entire occurrence range of a species is
higher, reflecting its evolutionary adaptation to higher environmental temperature (O’Connor et
al. 2007).
Based on the calculated PLD and ocean current velocity data, the model calculates dispersal of
pelagic larvae over time through diffusion and advection. Diffusion and advection of ocean
current are important factors determining dispersal of pelagic larvae of marine organisms
(Possingham & Roughgarden 1990; Gaylord & Gaines 2000; Bradbury & Snelgrove 2001; Gaines
et al. 2003). The temporal and spatial patterns of pelagic larval dispersal were modelled by a
two-dimensional advection-diffusion equation (e.g., Sibert et al. 1999; Gaylord & Gaines 2000;
Hundsdorfer & Verwer 2003):
( ) ( )
LavNv
y
Lavu
xy
Lav
D
yx
Lav
D
xt
Lav
+
=
λ
…9)
where change in relative larvae abundance over time (∂Lav/∂t) is determined by diffusion (i.e.,
the first two terms on the right-hand side of eq. 9) and current-driven movements (i.e., the third
and fourth terms of eq. 9). Diffusion is characterized by a diffusion parameter D, while advection
is characterized by the two current velocity parameters (u, v) which describe the east-west and
north-south current movement. Diffusion coefficient, expressed in m
2
s
-1
, is assumed to be a
function of length scale of the spatial grid: D = (1.1 x 10
-4
GR
1.33
where GR is the minimum grid
resolution (Nahas et al. 2003).
Annual average current fields were obtained from the NOAA/GFDL Coupled Model. Thus, we
implicitly assume that larvae remain within a single horizontal layer of the water column or are
well mixed vertically in water of nearly constant depth. The instantaneous rate of larval mortality
and settlement is represented by λ = M + S where M and S are the natural mortality and
settlement rates of larvae, respectively. The default larval survival and larvae retention rate are
0.15 day
-1
and 0.2 day
-1
, respectively. Alternative values were used to test for the sensitivity of
simulation results to these parameters.
We employed a numerical solution of the partial differential equation (eq. 9) provided by Sibert &
Fournier (1994). Basically, eq. 9 is solved using implicit alternating direction method (Press et al.
1988). This method solves the partial differential equation for each direction (u and v) after half a
time step sequentially (Figure 5). The implicit solution can be expressed as systems of linear
equations with tridiagonal matrices of coefficients. Therefore, the 30’ x 30’ grid world map was
firstly segregated into horizontal and vertical segments with consecutive sea cells. Movement of
larvae from diffusion and advection was then calculated for each east-west and north-south
16
segment. The tridiagonal system of equations can be robustly solved by the recursive algorithm
(Press et al. 1988) (see Appendix 3 and Sibert & Fournier 1993 for details of equations).
Figure 5. Schematic diagram of a computational grid. The horizontal (u) and vertical (v)
direction movements are solved separately at half a time step sequentially.
A daily time step was used in simulating larval dispersal. The simulation time frame is
determined by the calculated PLD from eq. 8. Displacement at each time step (∆x and ∆y) was
assumed to be the distance between two adjacent cells (Dist
ij
), which is calculated by:
ErLonLonLatLatLatLataDist
jiijijij
+=
))cos()cos()cos()sin()cos(sin(
/2
…10)
where Lat and Lon are the latitude and longitude of cells i and j, respectively; Er is the radius of
the Earth (6378.2 km).
In addition to average larval dispersal as described above, we also modelled extreme dispersal
events. Here, extreme dispersal events refer to sporadic dispersal of larvae as a result of ocean
advection anomalies. Such events may influence dispersal pattern and meta-population structure
(Lockwood et al. 2002). In the model, extreme dispersal events are assumed to be random events
and are represented by a doubling of the average dispersal distance. We assume that extreme
dispersal events occur once every 5 years, but alternative values were used to test the sensitivity of
model outputs to this parameter.
Net adult migration (I)
In the model, animals disperse by actively swimming to surrounding areas. Animals were
considered to have reached a cell by active dispersal if the dispersal distance in a simulation time-
step was greater than the nearest distance between source and destination cells. Distance
travelled by active dispersal from a source cell was calculated from a dispersal rate (km·year
-1
).
Generally, species that are pelagic, large-bodied, fusiform-shaped and metabolically more active
have higher dispersal rate. The aspect ratio of a fish’s caudal fin (i.e., the ratio between the square
of the height of fish’s caudal fin to the caudal fin area) is a proxy of its ‘motility’ (Palomares &
Pauly 1998). For invertebrates, ‘motility’ values were estimated from their general shape, and
scaled after fishes with similar shapes and habits.
a. Adult dispersal rate
We developed a fuzzy logic expert system that used ‘rules-of-thumb’ to predict dispersal rate from
species’ life history, habitats and ‘motility’. The ‘rules-of-thumb’ represented our general
understanding of the relationship between some easily-obtainable parameters with dispersal
ability of marine animals. These rules are expressed as IF-THEN clauses that link premises
(maximum body length, aspect ratio and habitat) to conclusions (dispersal rate) (Appendix 2).
Ordinal categories of input attributes were categorized based on pre-defined fuzzy membership
functions (Figure 6). For instance, if a fish has a maximum body length of 75 cm, it is classified as
small- and medium-sized fish with degrees of membership equal to 0.5 (Figure 6a). Memberships
to habitat types (pelagic, demersal and coral reef) are either 0 or 1.
17
a)
0.0
0.2
0.4
0.6
0.8
1.0
0 100 200 300 400 500
Maxi mum body length (cm)
Degree of membership
S M Large
b)
0.0
0.2
0.4
0.6
0.8
1.0
0 2 4 6 8 10 12
Aspect ratio
Degree of membership
L M H
c)
0.0
0.2
0.4
0.6
0.8
1.0
0 100 200 300 400 500 600
Dispersal rate (km ye ar
-1
)
Degree of membership
L M H
Figure 6. Fuzzy membership functions of the inputs (a.
maximum body length and b. aspect ratio) and output
(dispersal rate). S – Small, M – Medium, H – High. A fish with
maximum body length of 75 cm is calculated to have
membership to ‘Small’ and ‘Medium’ body size with equal
degree of membership of 0.5 (dotted line in 6a).
b. Modelling adult dispersal
Emigration (adult animals moving out of a cell) was calculated from the dispersal or movement
rate using an algorithm employed in an Eulerian spatial ecosystem simulation model – Ecospace
(Walters et al. 1999). Basically, if animals are distributed randomly within a cell (with length of a
cell side x and distance from the boundary of a cell y) at the start of a time interval dt, a
proportion xdy/x
2
will be a candidate for emigration across each cell boundary (Figure 7). Also,
the proportion of average organism within this cell to move a length dy over a short time interval
dt, in a completely random direction, is 1/π (Walters et al. 1999). Thus, the instantaneous
emigration rate m
i
for randomly moving organisms to emigrate across each cell boundary can be
calculated by:
x
V
m
i
i
=
π
…11)
where V
i
and x are the adult dispersal rate and length of a cell, respectively (see Walters et al.
1999 for details).
18
Figure 7. Schematic diagram of emigration and immigration movements by adult on a spatial
grid. A is the animal abundance in a cell. Emigration rate is inversely proportional to the length
of a cell (x). Absolute emigration of animals is directly proportional to emigration rate (m) and
animal abundance.
Animals generally have higher emigration rate towards cells with more preferable environment
(e.g., temperature, depth, habitat types). We modelled such behaviour by incorporating a
hyperbolic function in calculating the emigration rate m
i
:
)(
)(
Dk
km
m
base
i
i
+
=
…12)
where m
i(base)
is the movement rate in the absence of any gradient and k is a scaling factor
representing the sensitivity of the calculated emigration rate to changes in environmental
suitability (as measured by D) (Walters et al. 1999). Small values of k (e.g. 0.1) result in high
sensitivity to change in D while large values (e.g. 10) render adult dispersal rate insensitive to D.
We used an intermediate value of k (= 2), but we also compared model outputs from alternative k
values.
D is assumed to be the ratio of habitat suitability (P), as defined in eq. 6, between the source (i)
and destination (j) cells. That is,
j
i
ij
P
P
D=
…13)
Thus, animals would have higher emigration rate to cells with higher habitat suitability relative to
the source cell. Finally, emigration of animals (E) from a cell is calculated by multiplying
emigration rate (m
i
) by abundance of the cell (A
i
) (Figure 7).
Emigrated animals can move back to the source cells. The rate of such re-entry to the source cells
(RE) is dependent on the abundance relative to its carrying capacity in the destination cells:
)]1([
j
ii
ji
KRk
kAm
RE +
=
…14a)
j
j
j
KC
A
KR =
…14b)
where k is the scaling factor in eq. 12, KR is the carrying capacity ratio of the destination cell; A
and KC are animal abundance and carrying capacity of a cell, respectively. Thus, we assume that
19
the re-entry rate of animals is lower from cells with lower abundance relative to carrying capacity.
This can also reflect lower density-dependency (e.g., intra-specific competition) in destination
cells which favour successful establishment of new migrants. On the other hand, an animal that
enter a cell with population abundance at its carrying capacity would return to its original
(‘source’) cell (which is equivalent to stating that cells at carrying capacity cannot accept new
animals).
Thus, the net dispersal of adult animals (I
ij
) to a cell j from cell i is given by:
jiijij
REEI =
…15)
However, some of the dispersed animals (larvae and adults) may not establish themselves in the
destination cell and die because of density-dependent/independent factors (e.g., increased
predation, competition). Dispersed animals that die at cell i and time step t (Mort
i,t
) are
calculated in our model.
( )
0,,0,,
1
0
0, jiji
N
j
I
i
LIMMort +=
=
…16)
where N is the total number of cells from where the dispersed adult animal (I) and settled larvae
(L) come. A major assumption of our model is that current species distributions are in
equilibrium, i.e., dA/dt = 0. Also, for computation purposes, we have to assume that population in
every spatial cell at time-step 0 is at its carrying capacity. These assumptions mean that there is
no net emigration or dispersal at time-step 0 and Mort
i,0
.
The ‘mortality’ of dispersed animals at cell i and time step t (Mort
i,t
) changes as cells’ habitat
suitability changes with climate. Change in mortality of dispersal animals was modelled by
ti
i
i
I
ti
I
P
P
MM
,
0,
0,,
=
…17)
where P is the calculated habitat suitability to the species at time step 0 and t. Thus, as habitat
suitability changes, spatial dynamics of the modelled species move away from equilibrium and net
movement of animals between cells occurs
Predicting changes in species distributions
Using equations 2 to 17, we calculated changes in abundance per time step for each spatial cell for
each species. Since we assume that current species distribution is in equilibrium, abundance in
each cell remains unchanged if physical conditions (e.g., sea water temperature, habitat types) are
constant. However, as global climate changes, habitat suitability in each cell, and thus its carrying
capacity, and the growth, net migration and mortality of the organisms therein, change
accordingly. For instance, for any cell, if temperature becomes more favourable to a species, its
habitat suitability may increase according to eq. 6. Carrying capacity (KC) and population growth
(G) at the cell increase according to eqs. 2 and 5. Simultaneously, increase in habitat suitability
reduces the mortality rates of larval and adult immigrants. This also allows the successful
establishment of larvae and adult migrants in previously un-occupied cells. Thus, abundance in
this particular cell would increase. On the other hand, as temperature becomes less favourable to
the species, extirpation/emigration increases, and population growth becomes negative (because
of reduced carrying capacity), leading to a decrease in relative abundance.
To predict changes in species distribution that are solely a result of global climate change, we
subtract from the predicted changes in distribution with climate change scenario a ‘baseline’ run
in which temperature was assumed to change only in the first year, then remain constant
throughout the duration of the rest of the simulation. The reason for this is that, in the model, we
assume that current (predicted) species distributions are in equilibrium and habitat suitability is
20
solely dependent on the specified environmental factors (see above). However, the original
species distribution was constructed from environmental boundaries such as maximum and
minimum depth limits, northern and southern latitudinal limits and habitat associations. Thus
‘favourable’ habitats, as defined by the specified environmental factors (e.g., temperature, depth)
may be available outside these boundaries. In reality, however, there may be other factors
unaccounted for in our model that restrict the species from occupying these areas. Therefore, in
parallel to simulating changes in relative abundance of a species with changes in sea water
temperature, we simulated abundance changes by allowing the system to follow the change in sea
water temperature in the first simulation time step (year 1), after which temperature was kept
constant. Such procedure ‘perturbs’ the distribution of the species so it moves out of its
equilibrium distribution range (at time step = 0). (Other environmental factors such as ocean
advection patterns are treated the same way as temperature changes.) The difference in predicted
relative abundance between the two simulations was subtracted from the result of simulation with
temperature changes. For example, if relative abundance increased by 3 units under a given
climate change scenario and increased by 1 unit without climate change, the predicted change in
relative abundance became 3 1 units = 2 units. This procedure enables us to minimize the
contribution of factors that are not explicitly accounted for in our model to the change in species’
relative abundance distribution.
Modelling the dynamics of pelagic species
The model is adapted to represent the seasonal (summer and winter) patterns of distributions for
pelagic species. Current summer and winter distributions of pelagic species are predicted based
on the algorithm described in Lam et al. (this vol.). Also, current and future seasonally-averaged
sea water temperatures can be obtained from the NOAA/GFDL Coupled Model. Thus, for pelagic
species, the model is initiated with the current summer and winter distributions separately.
Subsequently, changes in species distributions in the two seasons are simulated independently
using the algorithm described in the above sections. However, instead of sea bottom temperature,
predicted sea surface temperature is used to identify the species’ thermally-preferred habitats.
Moreover, the model does not consider bathymetry in predicting the potential distribution range
of pelagic species. Furthermore, seasonality of spawning is considered. Generally, fishes in higher
latitudes have stronger seasonality in reproduction than those in lower latitudes. In addition,
marine fishes often spawn in spring and fall, with spring being the dominant spawning season
(Helfman et al. 1997). Thus, we assume that larvae production in the tropics (0
o
10
o
N/S) is
similar between seasons, but the proportion of total annual larval production in spring increases
linearly with latitude until 50
o
N/S after which the proportion remains constant at 90% in spring
(or 10% in fall) (Figure 8).
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60 70 80 90
Latitude
Prop. of reproductive output
Spring spawning
Fall spawning
Figure 8. Proportion of total annual reproductive output (larvae production) in spring (solid
line) and fall (dotted line) assumed in the dynamic bioclimatic envelope model.
21
Model implementation
The numerical model to simulate climate change-induced distribution shift is implemented in
Visual Basic.net environment. The overall structure of the model is summarized in Figure 9.
Figure 9. Schematic representation of the structure of the dynamic bioclimate envelope
model developed in this study, implemented in Visual Basic.Net environment.
System evaluation
We evaluated the functioning of our model by undertaking model simulation with global sea
water temperature generated from simple assumptions on rate of temperature increase in the
next 30 years. We considered two scenarios of global increase in sea water (bottom and surface)
temperature (Table 2). In each scenario, sea water is warming up slowly near the equator and
quickly toward the poles. Ocean advection fields were based on the annual average current
velocity data from the Mariano Global Surface Velocity Analysis (Mariano et al. 1995). In the test
simulations, ocean advection was assumed to be constant throughout the simulation time-frame.
However, in the future, ocean advection current data predicted by the NOAA/GFDL Coupled
Model will be used in predicting realistic effects of global climate change on species distributions.
Table 2. Parameters used to generate hypothetical scenarios of global sea water temperature increase for
model testing.
Rate of temperature increase (year
-1
)
Scenario At 0
o
lat At 90
o
N/S lat
Interpolation of temperature in other
latitude
1 0.025 0.075
2 0.050 0.150
Linear change in rate of temperature
increase from the equator to the poles.
To test the performance of our model, we simulated changes in distributions of four commercially
exploited species in 30 years under the above two scenarios of global sea temperature change
(Table 2). The evaluated species are: Nassau grouper, Small yellow croaker, Polar cod, Atlantic
cod, Western Australia rock lobster, Antarctic toothfish and Australian ruff. Their life history
parameters (e.g., L
inf
, W
inf
, K) were obtained from FishBase and SeaLifeBase (Froese & Pauly
22
2007; www.sealifebase.org) and from Phillips et al. (1992) for the Western Australian rock lobster
(Table 3).
Table 3. Input parameters used to simulate changes in distributions of four commercially exploited
marine species.
State variables (units) Nassau
grouper
Small yellow
croaker
Polar
cod
Atlantic
cod
L
inf
(cm) 90 29.2 31.3
129
W
inf
(g) 13279
403
258 22,300
K (year
-1
) 0.09
0.44
0.22 0.20
R (year
-1
) 0.51 1.58 0.94
0.39
Diffusion coefficient (m
2
s
-1
) 100 100 100 100
Movement rate (km year
-1
) 50 100 100 200
Larval mortality rate (day
-1
) 0.15 0.15 0.15 0.15
Larval settlement rate (day
-1
) 0.20 0.20 0.20 0.20
Table 3. Con’t
State variables (units) Western Australian
rock lobster
Antarctic
toothfish
Australian
ruff
L
inf
(cm) 10.4
*
185 41
W
inf
(g) - 75,600 -
K (year
-1
) 0.15 0.06
0.24
R (year
-1
) 0.3
0.02 1.50
Diffusion coefficient (m
2
s
-1
) 100 100 100
Movement rate (km year
-1
) 50 50 100
Larval mortality rate (day
-1
) 0.15 0.15 0.15
Larval settlement rate (day
-1
) 0.20 0.20 0.20
* Carapace length
We also evaluated the possible effects of climate change-induced shifting of coral reefs on the
distribution of reef-associated species. Coral reefs occur in areas with sea water temperature
between 18
o
C to 30
o
C (Veron 2000). Instances of increased temperature over the physiological
tolerance limits resulting from climate anomalies had led to large-scale coral bleaching events
(Glynn 1991, 1993; Hoegh-Guldberg 1999; Bellwood et al. 2004). Intensity and frequency of such
bleaching events appeared to be increasing since the 1970s (Hoegh-Guldberg 1999; Walther et al.
2002). The impacts of large-scale bleaching would be particularly prominent to coral reefs in area
at or near the coral’s upper temperature limits (Hoegh-Guldberg 1999). Some fishes and
invertebrates are obligatorily or strongly dependent on coral reefs (e.g., some species of butterfly
fishes, fam. Chaetodontidae). Thus, their distributions may be strongly affected by changes in
coral reef distribution (Bellwood et al. 2004). On the other hand, the high-latitude limits of coral
distribution may not shift much because coral growth at high-latitude is generally limited by
factors other than temperature, e.g., light (Hoegh-Guldberg 1999). Also, consistent increase in
water temperature from global warming may result in changes in species composition to more
heat-resistant species, and acclimation or evolution to higher heat tolerance. Thus, distribution of
coral reefs may not change as global sea temperature changes (Polsenberg 2003). These may
dampen the effects of global warming on coral reef distribution (Hughes et al. 2003).
To test the potential effects of climate-induced changes on coral reef distribution, we attempted to
mimic the effects of global warming on coral reefs. Based on a global map of coral reefs (UNEP-
World Conservation Monitoring Centre 2005), we plotted the distribution of relative coral reef
abundance over latitudinal zones (Figure 10). Relative coral reef abundance was calculated from
the ratio of the estimated coral reef area to the area of sea with average depth below 50 m at each
latitudinal zone. The latter was calculated from a spatial grid of the world ocean at 30’ latitude x
30’ longitude resolution, and was used to indicate the availability of waters for potential coral
growth. Thus, the calculated relative coral reef abundance is an approximate measure of coral reef
density. We assume that relative coral abundance is bimodal and could be approximated by two
normal distributions representing relative coral abundance in the northern and southern
hemisphere. We fitted the two normal distributions to the observed relative coral reef abundance
distribution using least-square methods so that the sum-of-square difference between the
observed distribution and the sum of the two predicted distributions was minimized (Figure 10).
We then assumed that, under global warming, the mean latitude of each abundance distribution
23
in the northern and southern hemisphere would shift north and south, respectively, while the
higher latitudinal limits and the standard deviations of the distribution remained constant.
Future absolute coral abundance at specific area was then calculated from:
),(
),,(
2
1
2
1,,
δ
δ
=
t
i
t
i
titi
LCLCf
LCLCf
CACA
…18)
where CA
i,t
is absolute coral abundance (km
2
) at grid cell (30’ x 30’) i and time (year) t. LC
i
is the
latitude at the centre of grid cell i where the specific area of coral occurs, and δ
2
is the standard
deviation of the normal distribution function f (Figure 11).
a)
0
0.1
0.2
0.3
0.4
-40-35-30 -25 -20-15-10 -5 0 5 10 15 20 2 5 30 35 40 4 5
Latitude
Relative coral reef abundance
b)
0
0.1
0.2
0.3
0.4
-45-40-35-30-25-20-15-10 -5 0 5 10 15 20 25 30 35 40 4 5
Latitude
Relative coral reef abundance
Figure 10. Relative coral reef abundance at different latitudinal zones based on: (a) observed coral reef
abundance (UNEP-World Conservation Monitoring Centre 2005); (b) predicted abundance by fitting
two normally-distributed relative abundance – latitude relationship with means = 10.4
o
and -16.4
o
and
standard deviations = 12.36
o
and 6.88
o
. The observed northern and southern limits of coral
distribution were maintained in the predicted distribution.
0 10 20 30 40 50
Latitude
Relative coral abundance
L C (t) L C (t+1)
Figure 11. Schematic diagram showing the calculation of the hypothetical effects of global
warming on coral abundance in the northern hemisphere. The solid line is a normal distribution
obtained from fitting with observed relative coral abundance with a mean of LC(t). Assuming
that mean relative coral abundance shift at a rate of (LC(t+1)-LC(t)) per year, the dotted line with
open circles represents a predicted distribution of relative coral reef abundance at year t + 1.
Predicted absolute coral reef area in a grid cell at latitude of 25
o
N would increase by a ratio as
indicated by the arrows.
We simulated hypothetical changes in distribution of coral reef abundances under three
scenarios: (a) no change; (b) mean relative coral reef abundance shifted at a rate of 20 km year
-1
(northward and southward in the northern and southern hemisphere, respectively) and (c) mean
relative coral reef abundance shifted at 50 km year
-1
(Figure 12). We used the Sohal surgeonfish
(Acanthurus sohal) as a case study to evaluate potential influence of changes in coral abundance
under global warming on distributions of coral reef-associated species.
24
a)
b)
c)
Figure 12. Simulated hypothetical changes in distribution of coral reef abundances in the Indo-Pacific
region after 30 years under three scenarios: (a) no change, (b) mean relative coral reef abundance
shifted at a rate of 20 km year
-1
(northward and southward in the northern and southern hemisphere,
respectively) and (c) mean relative coral reef abundance shifted at 50 km year
-1
.
We also tested the effect of change in sea ice coverage on polar species. Both empirical and
climate models suggest that sea ice coverage will continue to decrease as global temperature
increases as predicted (Johannessen et al. 1999, 2004; Vinnikov et al. 1999; Flato & Boer 2001;
Comiso 2002). Change in sea ice coverage can greatly affect polar ecosystems and the distribution
of the associated species (Eicken 1991). For example, the reproductive grounds of krill
(Euphausia superba), a key food source for higher predators such as penguins and whales, can be
affected by reducing the area of sea ice formed near the Antarctic Peninsula (Loeb et al. 1997;
Walther et al. 2002). This may also affect the distributions of the predators that depend on krill.
Thus, we evaluated the sensitivity of predicted polar species distributions to changes in sea ice
coverage. We assumed that the polar sea ice edges retreat at a rate of 5 km year
-1
. This is not an
attempt to mimic realistic sea ice changes. Instead, the hypothetical scenario allows us to explore
the potential effects of sea ice change on distribution of polar species. Realistic changes in sea ice
coverage predicted from climate model will be used in the future.
RESULTS
Simulated shift in distribution
Small yellow croaker (Larimichthys polyactis)
Simulations using the two hypothetical scenarios of increase in global sea water temperature
predicted that distributions of the Small yellow croaker would shift northward in 30 years (Figure
13, 14). Both the centroid and latitudinal range limits of the distribution shifted in all positive
warming scenarios. The original distribution of Small yellow croaker was restricted to the East
China Sea. However, as sea water temperature increased, the northern range limit reached into
the Bohai Sea and the coast of Japan (main islands), which were not previously occupied by this
species. Simultaneously, the southern range limits shifted north from the Taiwan Strait. The
degree of range shift also differs between coast and offshore regions. Under a stronger warming
scenario (scenario 2), the northern range limit shifted by 5
o
latitude while the centroid of the
distribution shifted by around 3
o
northward (Figure 14).
25
Figure 13. Simulated changes in distribution of Small yellow croaker after 1 year (upper left), 10 years (upper
right), 20 years (lower left) and 30 years (lower right) under a hypothetical mild level of ocean warming
(scenario 1).
Figure 14. Simulated changes in distribution of Small yellow croaker after 1 year (upper left), 10 years (upper
right), 20 years (lower left) and 30 years (lower right) under a hypothetical strong level of ocean warming
(scenario 2).
Year 1 Year 10
Year 20 Year 30
Year 20 Year 30
Year 1 Year 10
26
Nassau grouper (Epinephelus striatus)
Similar patterns of range shift were observed for Nassau grouper, which ranges across the equator
(Figure 15 & 16). Under both milder and stronger warming scenarios, Nassau grouper generally
moved away from the equator after 30 years, while their relative abundance increased in higher
latitudes. For example, abundance increased almost five-fold in southern coast of Brazil. Also, the
southern range limit extended further into Uruguay. Particularly, in the stronger warming
scenario (Figure 16), relative abundance of inshore populations from Venezuela to northern Brazil
was much reduced. On the other hand, relative abundance in the southeast and east coast of USA
increased.
Figure 15. Simulated changes in distribution of Nassau grouper after 1 year (upper left), 10 years (upper right),
20 years (lower left) and 30 years (lower right) under a hypothetical mild level of ocean warming (scenario 1).
Figure 16. Simulated changes in distribution of Nassau grouper after 1 year (upper left), 10 years (upper right),
20 years (lower left) and 30 years (lower right) under a hypothetical strong level of ocean warming (scenario 2).
Year 1 Year 10
Year 20
Year 1 Year 10
Year 20
Year 30
Year 30
27
Polar cod (Boreogadus saida)
Polar cod was found to be sensitive to the warming scenarios and the model predicted that it
would be extirpated in most of its range even under the milder warming scenario (Figure 17). This
is due to its occurrence in the Arctic Ocean, which largely precludes it from moving northwards.
Polar cod was predicted to be extirpated around Greenland and its abundance was largely
reduced in other parts of the Arctic Ocean after 30 years of hypothetical warming.
Figure 17. Simulated changes in distribution of Polar cod after 1 year (upper left), 10 years (upper right), 20 years
(lower left) and 30 years (lower right) under hypothetical scenarios of ocean warming (scenario 1) and retreating
sea ice edge at a rate of 5 km per year. Polar cod is extirpated from most of its range in 30 years.
Year 1 Year 10
Year 20 Year 30
Year 30
28
Antarctic toothfish (Dissostichus mawsoni)
Under a mild ocean warming and with the ice edge retreating at a rate of 5 km per year, the
distribution range of Antarctic toothfish was predicted to contract (Figure 18). As this species only
occurs around Antarctica, it cannot expand its southern limits when sea water temperature
increases. Also, we assume that Antarctic toothfish has an affinity to sea ice edge. The retreating
sea ice also contributed to the range contraction. Under the stronger warming scenario, the
distribution range of Antarctic toothfish becomes so restricted that it would induce extinction in
30 years.
Figure 18. Simulated changes in distribution of Antarctic toothfish after 1 year (upper left), 10 years (upper right),
20 years (lower left) and 30 years (lower right) under hypothetical scenarios of ocean warming (scenario 2) and
retreating sea ice edge at a rate of 2 km per year. Antarctic toothfish is predicted to become extinct in 30 years
under the specified scenario.
Year 1 Year 10
Year 20 Year 30
29
Atlantic cod (Gadus morhua)
A strong global warming scenario (scenario 2) resulted in a general northward shift of
distribution of Atlantic cod (Figure 19). In the northwest Atlantic, our model predicted that the
abundance of the southern cod stocks (Georges Bank, Gulf of Maine, and Scotian Shelf) would
decline. In the northeast Atlantic, relative abundance of cod declines in the North Sea, Irish Sea,
Celtic Sea and Norwegian Sea. On the other hand, the relative abundance of the Icelandic, Faroe
Island and Barents Sea cod stock increased. Also, the distribution of cod extends further into the
Arctic as the ice sheet retreats.
Figure 19. Simulated changes in distribution of Atlantic cod after 1 year (upper left), 10 years (upper right),
20 years (lower left) and 30 years (lower right) under hypothetical scenarios of ocean warming (scenario 2).
Year 1
Year 10
Year 20
Year 30
30
Western Australian Rock lobster (Panulirus cygnus)
Distribution range of Western Australian rock lobster was predicted to shift southward under the
sea temperature warming scenarios (Figure 20). The centroid of its latitude distribution shifted
south by approximately 1
o
and 3
o
in 30 years under the milder and stronger warming scenarios,
respectively. Moreover, the southern range limit extended further into the southwest of the
Australia continent.
Figure 20. Simulated changes in distribution of Western Australian rock lobster after 1 year (upper left), 10 years
(upper right), 20 years (lower left) and 30 years (lower right) under a hypothetical strong ocean warming (scenario
2).
Year 1 Year 10
Year 20 Year 30
31
Australian ruff (Arripis georgianus)
Similar to Western Australian rock lobster, when global sea water temperature increases, the
distribution range of Australian ruff moves to cooler waters in the south (Figure 21). Australian
ruff is a pelagic species and its seasonal movements are accounted for our simulation model. Our
results suggest that the summer distribution of Australian ruff is more sensitive to temperature
changes in summer (as indicated from the faster rate of southward shift) than in winter.
Summer
Winter
Figure 21. Simulated changes in distribution of Australian ruff in summer (left panel) and winter (right panel)
after 1 year, 10 years, 20 years and 30 years under a hypothetical strong ocean warming (scenario 2).
Year 1 Year 1
Year 30 Year 30
Year 10
Year 20
Year 10
Year 20
32
Sensitivity analysis
Larval dispersal
Dispersal and recruitment of fish larvae are sensitive to the larval mortality and settlement rates
specified in the larval dispersal model (Figure 22). Again, recruitment is defined here as the
settlement of pelagic larvae. Absolute level of total recruitment is controlled by the natural
mortality rate of the larvae. Thus, species with extended pelagic larval duration are generally
more sensitive to the specified mortality rate. Larval settlement rate strongly affects the
distribution of settled larvae. In the case of Small yellow croaker, a high settlement rate (15% day
-
1
) results in strong local recruitment (i.e., recruitment to the adult distribution range), while a low
settlement rate (1% day
-1
) results in wider range of larval dispersal (Figure 22).
a)
b)
c)
Figure 22. Simulated dispersal of larvae of Small yellow croaker from the current species distribution
predicted by Close et al. (2006). Alternative larval settlement rates (a) 15% day
-1
, (b) 7.5% day
-1
and
(c) 1% day
-1
representing scenarios with strong and weak local larval recruitment, respectively.
33
Low larval settlement generally accelerates the movement rate and increases the extent of
distribution shift (Figure 23). Under a lower larval settlement rate of 0.01 year
-1
(i.e., larvae were
allowed to travel further before settlement), distribution of Small yellow croaker extends further
northeast to the coast of Japan after 30 years under a strong warming (scenario 2). The higher
dispersal ability enables the fish to occupy such habitat. On the other hand, the general pattern of
distribution shift is similar to the scenario with higher larval settlement rate (0.15 year
-1
).
Figure 23. Predicted distribution of Small yellow croaker with low larval dispersal rate
(0.01 year
-1
) after 30 years of strong warming (scenario 2).
A high larval production rate or alternative settings of migration sensitivity parameter (k) results
in slight changes in the prediction distribution of Small yellow croaker after 30 years under a
strong warming scenario (Figure 24). Relative abundance is slightly higher in the newly-occupied
areas in the north. The effect is similar to those obtained from a low larval settlement rate or low
larval mortality rate. Alternative settings of migration sensitivity parameter (k) (default = 2, low =
0.5, high = 5) only slightly affect the prediction distributions of Small yellow croaker (Figure 25).
Figure 24. Predicted distribution of Small yellow croaker with high larval production
rate (0.5 times parent biomass per year) after 30 years of strong warming (scenario 2).
34
Figure 25. Predicted distribution of Small yellow croaker after 30 years of strong warming (scenario 2) with
(a) low migration sensitivity (k = 0.5) and high migration sensitivity (k = 5) (default value of k = 2).
Intrinsic rate of increase
The intrinsic rate of population increase (r) specified in the model does not appear to have strong
effects on the predicted change in distribution range of Small yellow croaker (Figure 26). Under
both low and high values of r (0.8 and 3 year
-1
, respectively) and hypothetical strong warming
(scenario 2), the extents of range shift after 30 years of simulation are generally the same.
However, with lower r, relative abundance in the newly-occupied northern region is lower than
predictions with higher r. Predicted relative abundance around the southern limits is similar
between simulations with low and high r (Figure 26).
Figure 26. Predicted distribution of Small yellow croaker after 30 years of strong warming (scenario 2) with
intrinsic rate of population increase of (a) 0.08 year
-1
and (b) 3 year
-1
.
a) b)
a) b)
35
Coral reef shift
A hypothetical shift in coral reef distribution has little effect on the predicted distribution of Sohal
surgeonfish (Acanthurus sohal) (Figure 27). Sohal surgeonfish is a coral reef fish, although it also
occurs in habitats such as rocky bottoms. Under 30 years of hypothetical warming, the relative
abundance of inshore populations of Sohal surgeonfish at low latitude is predicted to be much
reduced. Conversely, its relative abundance at higher latitude increased. However, coral reefs,
overall, had small effects on the simulated distributions.
a)
b)
c)
Figure 27. Predicted distribution of Sohal surgeonfish (Acanthurus sohal) after 30 years under three
hypothetical scenarios (a) no change in coral abundance, (b) coral reef shifts at a rate of 20 km year
-1
,
and (c) coral reef shifts at a rate of 50 km year
-1
. A mild warming (scenario 2) was used in the
simulations.
36
Discussion
Test simulations
The simulation model presented in this report allows quantitative predictions of the effects of
climate change on distributions of all commercially exploited marine species at a global scale.
Unlike conventional bioclimate envelope models that predict changes in species distribution only
from identified bioclimate envelope (Pearson & Dawson 2003), this model incorporated
population and ocean current dynamics in simulating distribution changes. These components
are important in shaping distributions of marine organisms (Gaylord & Gaines 2000; Bradbury &
Snelgrove 2001). For instance, range limits could be caused largely by oceanographic
discontinuities (Gaylord & Gaines 2000). Also, organisms’ life history traits may affect their
responses (in terms of rate and magnitude) to environmental changes (Perry et al. 2005). Thus,
their inclusions allowed more realistic simulations of responses to climate change scenarios.
Our model explicitly represents both non-interactive (additive) and interactive (multiplicative)
effects of climate influence on population dynamics (Stenseth et al. 2002). Example of non-
interactive or additive effects included increased net influx of new migrants (through adult
movement or larval dispersal) as environmental conditions in an area became more favourable to
the species of interest. In the case of Small yellow croaker, increased temperature reduced the
mortality of migrants to the previously unoccupied Bohai Sea. Thus, relative abundance of the
species increased in Bohai Sea through the addition of new migrants. Simultaneously, carrying
capacity of Small yellow croaker increased in the northern range limit and decreased in the
southern range limit as temperature increased. These reduced the strength of density dependence
in the north, and increased it in the south. Thus, relative abundance of Small yellow croaker
shifted gradually towards the north under both (mild and strong) warming scenarios. Responses
to the warming scenarios were similar in the case of Nassau grouper.
Some species may be unable to adapt to global warming by range shifting and may be extirpated.
Antarctic toothfish occurs around the Antarctic. In our model, as sea temperature increased, the
species was unable to shift its range further south, into cooler waters. As sea temperature
increased outside the tolerance limits of Antarctic toothfish, suitable habitat that was reachable
by the species gradually declined. Eventually, the species was extirpated as suitable habitats
disappeared. Thus, if the magnitude of warming is large enough, it is likely that some species will
be rendered extinct.
The predicted distribution of Atlantic cod under hypothetical warming scenarios agreed with
predictions independently conducted by others (e.g., Drinkwater 2005). Based on observed
relationship between recruitment strength and sea-bottom temperature, Drinkwater (2005)
predicts that, if sea water temperature increased, the Celtic and Irish Seas stocks of Atlantic cod
would disappear, while the southern North Sea and Georges Bank stocks would decline. Cod
distribution may also shift northwards along coastal Greenland and Labrador and to the Barents
Sea and the Arctic Ocean (Drinkwater 2005). Such predictions generally agree with our
simulation results, which provided some support to the validity of our model.
Changes in distribution ranges predicted from climate model generated data will obviously be
more complex than the test simulations presented in this study. Here, we assumed a monotonic
increase in sea temperature (bottom and surface). However, predictions from climate models are
more complex than our hypothetical scenarios. For instance, warming may show hemispheric
asymmetry, with more warming in the northern high latitude than in the south, although the
evidence is not yet conclusive (Flato & Boer 2001). Also, rate of warming in different ocean basins
may vary. Recorded sea water temperature over the last half-century showed different rates of
increase in different oceans (Levitus et al. 2000). Moreover, we assumed that ocean currents
were in a steady-state. However, ocean currents will change with global climate (Rahmstorf &
Ganopolski 1999; Vellinga & Wood 2002), which will affect distributions and population
dynamics of marine species (Gaylord & Gaines 2000; Walther et al. 2002).
In the future, predicted changes in physical attributes (e.g., sea bottom and surface temperature,
ocean advection fields) from year 2000 to 2100 will be provided through the kindness of Dr Jorge
37
Sarmiento and his collaborators at the Atmospheric and Oceanic Sciences Program, Princeton
University. The data are simulated from the NOAA’s Geophysical Fluid Dynamics Laboratory
Coupled Model, version 2.1 (GFDL’s CM2.1 model) under three scenarios of future CO
2
emission:
(1) drastic reduction of CO
2
from the present, (2) moderate reduction of CO
2
from the present, (3)
maintenance at year 2000 level. When we use such data in our simulations, predicted patterns of
range-shifting of the studied species should be more realistic.
Model uncertainty
Because of its broad geographic and taxonomic scope, it was unavoidable that we had to make
various assumptions in our model, to reduce the number of required parameters and simplify the
dynamics of the system to a practical level. Many assumptions are in fact inherent in most
bioclimate envelope models developed to study the effects of climate change (Pearson & Dawson
2003). The major assumptions, and the potential implications for predictions from our model, are
detailed here.
Biotic interactions
The model did not explicitly account for inter-specific interactions. Species within a community
may respond differently to climate change (Walther et al. 2002). For instance, distributions of
predators and their preys may shift at different rates as climate-linked oceanographic conditions
change. This can result in reduced range-overlap and may disrupt existing biotic interactions
(e.g., predation) in a community (Murawski 1993; Hughes 2000; McCarty 2001; Walther et al.
2002). Thus, a predator with high prey specificity may not find enough food if its prey’s range
does not shift along with its own. Moreover, food web interactions may affect the rate of climate-
induced distribution shift. For example, the rapid expansion of distribution range of the
Humboldt squid (Dosidicus gigas) in the eastern North Pacific may be linked to changes in
climate-related oceanographic conditions and the depletion of their competitors and predators
(Zeidberg & Robison 2007). Rapid invasion of predators such as the Humboldt squid may
destabilize the ecosystem as potential preys may be exposed to increased predation risk. These
trophic dynamics may modify the patterns of range shifts predicted by our model.
Incorporation of biotic interactions in predicting effects of climate change at multi-species or
ecosystem levels would be the next step of this modelling exercise. The model developed in this
study targets a wide range of exploited marine species at global scale and is aimed to predict
general patterns of potential responses of individual species to climate change scenarios.
Predictions on individual species can later be incorporated into dynamic multi-species/ecosystem
models such as Ecopath with Ecosim and Ecospace (Walters et al. 1997, 1999; Pauly et al. 2000).
The tropho-dynamic models can then evaluate the potential effects of climate changes on biotic
interactions and on structure of the ecosystem.
Evolutionary changes
Species with high rates of evolutionary changes may adapt to changing climates and thus affect
their patterns of range-shifting. Our model implicitly assumes that rates of adaptation are slower
than extinction rates (niche conservatism) (Pearson & Dawson 2003). On the other hand,
evolutionary responses to climate changes are shown in some insects (Thomas et al. 2001). For
instance, two butterfly species have increased the variety of habitat types that they can colonize.
Also, fractions of longer-winged (dispersive) individuals of two bush cricket species increased
(Thomas et al. 2001). In these examples, the dispersal ability of the species was enhanced by
evolutionary changes. Although evidence on genetic changes in marine species that are directly
related to climate change is scarce, changes in life history traits in response to extrinsic factors
(e.g. fishing) that may be phenotypic or genetic are more common (Law 2000). Phenotypic
changes in response to fishing included increased somatic growth, reduced age at first maturity
and decreased body size (Rijnsdorp 1993; Law 2000; Kraak 2007). However, these changes most
likely affect the rate of range-shifting, but not the magnitude of the shift. On the other hand,
evidence from Pleistocene glaciation indicates that species more often responded to climate
change ecologically, by shifting their ranges, rather than evolutionarily, through local adaptation
(Parmesan et al. 2000). Thus, although our model may underestimate the rate of range shift by
38
ignoring potential evolutionary changes that may increase species’ dispersal ability, model results
can be considered conservative predictions of responses to climate change.
Population dynamics
The complexity of population dynamics may affect species’ responses to climate change. The
incorporation of population dynamics in our bioclimate envelope model is an advance from
previous attempts to predict climate change-induced distribution shift (Pearson & Dawson 2003).
However, considering the resolution of the model, the scale of the study and the limited
availability of model parameters, simplifications were made to the population dynamic models.
Firstly, we modelled populations as homogenous biomass pools. Thus the model did not explicitly
represent population and age structures. Populations of a species may have different life history
and show different adaptations to environmental changes. For instance, in some species,
individuals with bigger body size tend to have a more unstable range limit and to be more likely to
shift in range than those with a smaller body size (Roy et al. 2001). Secondly, we represented
larval production and recruitment by a linear relationship with biomass-pool and a constant
larval survival and settlement rate. However, dynamics of larval production, survival and
recruitment are generally non-linear and controlled by complex factors that include climate,
oceanography and ecology. Thirdly, except for pelagic species, our model simulated annual
averaged dynamics of populations. However, seasonal dynamics may be important in
understanding responses to climate change for non-pelagic species as well (Barbraud &
Weimerskirch 2003).
Given the uncertainties of the parameters, the predicted range-shifts were generally robust. The
magnitude of predicted distribution range-shift was similar under alternative values for the
intrinsic rate of population increase, larval dispersal and settlement. The extent of the
distribution range in the model is determined primarily by the available bioclimate envelope and
is thus less dependent on population dynamics. However, the rate of range-shifting may be more
sensitive to the specified population dynamics. Species with higher dispersal ability are able to
colonize suitable habitats at a faster rate. Also, higher larval production (or reduced larval
mortality) also increases the rate of colonization. On the other hand, climate change (e.g.,
temperature changes) should generally operate at a much longer time scale (decade) than species’
population dynamics (year). Thus, predictions from our model should be largely insensitive to
population parameters. This is backed by the results from the sensitivity analysis conducted in
this study, which generally showed that simulated changes in distribution ranges in 30 years were
insensitive to the specified population parameter values.
Our model did not represent density-dependent changes in environment preferences. Species’
preferences for habitats may change with population density. For instance, Atlantic cod in the
southern Gulf of St. Lawrence tended to occupy colder water when its abundance was high. Such
density-dependent shifts in spatial patterns may be a response to decreased food availability by
lowering density-independent energy costs (Swain & Kramer 1995). However, if the current
species distribution range has captured the full temperature tolerance range of the species, such
density-dependent responses should have small effect on predicted distribution range from our
model under various climate change scenarios. On the other hand, should the species be able to
move to colder habitats from those depicted by the current range because of density-dependent
effects, the prediction from our models will be conservative. Density-dependent movement to
colder habitats may increase the distribution range if climate change increases the strength of
density-dependent competition for food.
Extrinsic factors
Our model assumes that the current distribution of a species reflects its temperature preference
and tolerance limits. However, factors other than those explicitly considered here may determine
species distribution (Samways et al. 1999). Thus, a species’ distribution range may be well within
its physiological temperature limits. Then, predictions from our model would over-estimate the
responses of the species to temperature changes. On the other hand, sea temperature is shown to
be a principal factor determining distributions, abundance and physiology of fish species (Pörtner
2001; Roessig et al. 2004). Other climate factors that may potentially affect marine fishes and
39
invertebrates but which are not explicitly incorporated in our model include productivity, salinity,
pH, seawater chemistry, UV radiation and sea level. Productivity is partly dealt with in our model
as it is correlated with temperature. Salinity, pH, seawater chemistry and sea level may have
stronger influence on intertidal or strongly coastal species. Effects on more oceanic species, which
represent the majority of commercially-exploited species, are generally less important than
temperature. UV radiation may affect survival of pelagic larvae. However, predicted range-shifts
from our model appeared to be largely insensitive to larval survival rate. Thus, although fine-scale
predictions from our model may be uncertain, predictions at large (global) scale should be robust
to parameter uncertainty.
Potential shifting of habitats
Predictions from our model were generally robust to the potential shift of physical habitats.
Physical habitats that marine species may be associated to can shift in space under climate
change, e.g. coral reefs (Hoegh-Guldberg 1999; Hughes et al. 1999). However, the overwhelming
majority of the commercially-exploited species that are included in the UN Food and Agriculture
Organization (FAO) fisheries statistics, and hence in the Sea Around Us database, do not occur
exclusively on coral reefs. They were mostly reef-associated species that could inhabit other
complex physical structures such as rocky reefs. Distributions of these alternative physical
structures were generally not affected by climate change. Thus, at the scale of our model (30’
latitude x 30’ longitude grid), effects of shift in physical habitat such as coral reefs on the
associated (commercially-exploited) species should not be apparent. On the other hand, such
habitat effects on species distribution may be important at local scale.
Future improvements
In the future, the model will account for the effect of changes in salinity on species’ distribution
ranges. Salinity can determine the distribution of marine fishes and invertebrates (Helfman et al.
1997; Edgar and Last 1999; Blaber 2000) and global warming may result in changes in sea water
salinity (Munk 2003). For example, regional climate models predict reduction in the salinity of
the Baltic Sea following increase in river run-off (Omstedt et al. 2000; Meier 2002; Meier &
Kauker 2003). Species that are adapted to waters with high salinity may disappear from such
regions (Roessig et al., 2005). To incorporate the salinity factor into the model, we will categorize
marine fishes and invertebrates into different salinity tolerance levels. We will also generate
habitat maps of species’ preferred sea water salinity levels based on current and predicted sea
water salinity and species’ salinity tolerance limits. Then, we can incorporate salinity as a habitat
factor in the algorithm that account for species habitat affinity (e.g. coral reef) in simulating
future distribution ranges.
Moreover, we will incorporate coastal upwelling as a factor in determining the present and future
distributions of marine species. Coastal upwelling is an important factor that determines the
productivity and distribution of many marine species (Horn and Allen 1978; Barber and Smith
1981). Some species are solely associated with coastal upwelling systems (e.g., the Peruvian
anchoveta), while some may avoid such systems. It is predicted that climate change may lead to
acceleration of coastal upwelling (Bakun 1990; McGregor et al. 2007). Such changes will affect
the physical and biological structure of coastal upwelling systems (Barth et al. 2007) and,
therefore, may contribute to shifts in distribution ranges of their associated species. To
incorporate this factor into the dynamic bioclimate envelope model, we will construct a
distribution map of major coastal upwelling regions of the world ocean and calculate their
intensity using sea surface temperature anomalies. We will then assign an affinity index of coastal
upwelling to marine fishes and invertebrates. Using the affinity index and the present and
projected upwelling intensity (calculated from predictions from global circulation models), we
will then be able to incorporate coastal upwelling as a factor in predicting distribution shifts of
marine fishes and invertebrates given different climate change scenarios.
We will also attempt to predict global distribution maps of kelp forests and simulate how climate
change may affect the distribution of kelp forests and their associated fauna. Kelp forests are
among the world’s most productive ecosystems and are important habitats for a wide variety of
species (Mann 1973). Their distribution is sensitive to oceanographic conditions such as sea
40
temperature (Graham et al. 2007). Thus, climate change may have considerable impact on the
distributions of kelp forest and their associated fauna. A previous study successfully uses
ecophysiological and oceanographic model to predict kelp populations (Graham et al. 2007). Our
model can adopt a similar approach in predicting the effects of climate change on the distribution
of these kelp populations. Moreover, similar to coral reef species, we can assign a kelp-affinity
index to species that are associated with kelp forests. Combining this index with a global habitat
map of kelp, we can improve our predicted distribution ranges of kelp-associated species. By
simulating the changes in distribution of kelp forest under climate change scenarios, we can
predict how distribution ranges of kelp-associated species may change.
Our model can also incorporate hypotheses relating to climate-induced changes in physiology and
population dynamics. The metabolism of fish is generally limited by the intake and availability of
oxygen to body cells. These processes are strongly temperature-dependent. The metabolism
affects somatic growth, body size and other life history traits that are closely correlated with each
other (Pauly 1980, 1981, 1998). These may then affect population dynamics (e.g., reproduction,
mortality, density-dependence etc.). Thus, incorporation of such dynamics in the simulation
model may allow the model to make more accurate predictions on the impacts of climate change.
We do not explicitly account for the effects of changes in ocean chemistry in our model; this can
be an area for future improvement. Physiology and population dynamics of marine organisms
may also be affected by changes in ocean chemistry. Specifically, the potential impacts of ocean
acidification by anthropogenic CO
2
on the balance of calcium carbonate system in the Oceans
have been raised (Feely et al. 2004). Ocean acidification may lead to the dissolution of calcium
structure such as shells or coral skeleton and render the maintenance of such structures more
difficult (Feely et al. 2004; Orr et al. 2005). High acidity may also increase the mortality of fish
eggs and larvae (Kikkawa et al. 2003).Under various projected scenarios of future CO
2
emission,
conditions of ocean acidification may have stronger impacts on high-latitude ecosystems and the
impacts could develop within decades. In the future, we may potentially include ocean chemistry
as a factor that affects the organisms’ growth and evaluate its effect on their distribution.
To assess the accuracy of predictions from our model, past climate and species distribution data
can be used. Based on these past data, we can use the model to simulate changes in distribution
ranges of the species. The accuracy of the model can be assessed by comparing the model-
reconstructed changes with empirically-observed data. For instance, detailed climate, abundance
and distribution data of many commercially-important species in the North Sea are well-
documented (Perry et al. 2005). Such data can be used for model validation.
CONCLUSIONS
The dynamic bioclimate envelope model developed in this study provides a tool to quantitatively
evaluate the effects of climate scenarios on a wide range of marine species. Using the data from
global databases such as FishBase and the Sea Around Us Project database, the model can be
applied to all commercially-exploited marine species, and indeed to any species for which
distribution range maps exist. Given the targeted scale of the model and the trade-off between
data availability and model details, the model appears to provide reasonable predictions that are
robust to major model assumptions. Used in conjunction with models that translate biogeography
to fishery productivity, it can be used to inform policy makers about potential socio-economic
implications of climate change for global fisheries.
41
REFERENCES
Alder, J. 2003. Putting the Coast in the Sea Around Us Project. The Sea Around Us Newsletter. No. 15: 1-2.
Araújo, M. B. and New, M. 2006. Ensemble forecasting of species distributions. Trends in Ecology and Evolution, 22: 42-
47.
Araújo, M. B., Pearson, R. G., Thuiller, W. and Erhard, M. 2005. Validation of species-climate impact models under
climate change. Global Change Biology, 11: 1504-1513.
Bakun, A. 1990. Coastal ocean upwelling. Science, 247: 198-201.
Barber, R.T. and Smith, R.L. 1981. Coastal Upwelling Ecosystems. In: Longhurst, A.R. (ed). Analysis of Marine Ecosystems,
p. 31-68.
Barbraud, C. and Weimerskirch, H. 2003. Climate and density shape population dynamics of a marine top predator.
Proceedings of the Royal Society: Series B, 270: 2111-2116.
Barth, J.A., Menge, B.A., Lubchenco, J., Chan, F., Bane, J.M., Kirincich, A.R., McManus, M.A., Nielsen, K.J., Pierce, S.D.
and Washburn, L. 2007. Delayed upwelling alters nearshore coastal ocean ecosystems in the northern California
current. PNAS, 104: 3719-3724.
Bellwood, D. R., Hughes, T. P., Folke, C. and Nyström, M. 2004. Confronting the Coral Reef Crisis. Nature, 429: 827-833
Blaber, S.J. 2000. Tropical Estuarine Fishes: Ecology, Exploitation and Conservation. Blackwell Science, Oxford, 372 p.
Bradbury, I. R. and Snelgrove, P. V. R. 2001. Contrasting larval transport in demersal fish and benthic invertebrates: the
roles of behaviour and advective processes in determining spatial pattern. Canadian Journal of Fisheries and Aquatic
Science, 58: 811-823.
Brierley, A. S., Fernandes, P. G., Brandon, M. A., Armstrong, F., Millard, N. W., McPhail, S. D., Stevenson, P., Pebody, M.,
Perrett, J., Squires, M., Bone, D. G., Griffiths, G. 2002. Antarctic krill under sea ice: elevated abundance in a narrow
band just south of ice edge. Science, 295: 1890-1892.
Cheung, W. W. L., Watson, R., Morato, T., Pitcher, T. J. and Pauly, D. 2007. Intrinsic vulnerability in the global fish catch.
Marine Ecology Progress Series, 333: 1-12.
Close, C., Cheung, W. W. L., Hodgson, S., Lam, V., Watson, R., Pauly, D. 2006. Distribution ranges of commercial fishes
and invertebrates. In: Palomares, M. L. D., Stergiou, K. I., Pauly, D. (eds). Fishes in Databases and Ecosystems.
Fisheries Centre Research Reports 14 (4). Fisheries Centre, University of British Columbia, Vancouver, p 27-37.
Comiso, J. C. 2002. A rapidly declining perennial sea ice cover in the Arctic. Geophysical Research Letters, 29(20), 1956,
doi:10.1029/2002GL015650.
Cox E (1999) The fuzzy systems handbook: a practitioner's guide to building, using, and maintaining fuzzy systems. AP
Professional, San Diego.
Daly, K. L., and Macaulay, M. C. 1988. Abundance and distribution of krill in the ice edge zone of the Weddell Sea, austral
spring 1983. Deep-Sea Research, 35: 21-41.
Drinkwater, K. F. 2005. The response of Atlantic cod (Gadus morhua) to future climate change. ICES Journal of Marine
Science, 62: 1327-1337.
Edgar, G.J., Barrett, N.S., and Last, P.R. 1999. The distribution of macroinvertebrates and fishes in Tasmanian estuaries.
Journal of Biogeography, 26: 1169-1189.
Feely, R.A., Sabine, C.L., Lee, K., Berelson, W., Kleypas, J., Fabry, V., Millero, F.J. 2004. Impact of anthropogenic CO
2
on
the CaCo
3
system in the oceans. Science 305, 362-366.
Eicken, H. 1992. The role of sea ice in structuring Antarctic ecosystems. Polar Biology, 12: 3-13.
Ekman, S. 1967. Zoogeography of the Sea. Sidgwick and Jackson. London.
Farrell, A. P. and Steffensen, J. F. (eds.) 2005. The Physiology of Polar Fishes. Elsevier Academic Press, London. 396 p.
Flato, G. M. and Boer, G. J. 2001. Warming asymmetry in climate change simulations. Geophysical Research Letters, 28:
195-198.
Frank, K. T., Perry, R. I. and Drinkwater, K. F. 1990. Predicted response of Northwest Atlantic invertebrate and fish stocks
to CO
2
-induced climate change. Transactions of the American Fisheries Society, 119: 353-365.
42
Frederich, M. and Pörtner, H. O. 2000. Oxygen limitation of thermal tolerance defined by cardiac and ventilatory
performance in the spider crab, Maja squinado. American Journal of Physiology, 279: R1521-R1528.
Fuiman, L. A., Davis, R. W., and Williams, T. M. 2002. Behavior of midwater fishes under the Antarctic ice: observations
by a predator. Marine Biology, 140: 815-822.
Gaines, S. D., Gaylord, B. and Largier, J. L. 2003. Avoiding current oversights in marine reserve design. Ecological
Applications, 13(1) Supplement: S32-S46.
Gaylord, B. and Gaines, S. D. 2000. Temperature or transport? Range limits in marine species mediated solely by flow.
The American Naturalist, 155(6): 769-789.
Glynn, P. W. 1991. Coral reef bleaching in the 1980s and possible connections with global warming. Trends in Ecology and
Evolution, 6: 175-179.
Glynn, P. W. 1993. Coral reef bleaching: ecological perspectives. Coral Reefs, 12: 1-17.
Graham, M.H., Kinlan, B.P., Druehl, L.D., Garske, L.E., and Banks, S. 2007. Deep-water kelp refugia as potential hotspots
of tropical marine diversity and productivity. PNAS, 104: 16576-16580
Hannah, L., Midgley, G. F. and Millar, D. 2002. Climate change-integrated conservation strategies. Global Ecology and
Biogeography, 11: 485-495.
Harley, C. D. G., Hughes, A. R., Hultgren, K. M., Miner, B. G., Sorte, C. J. B., Thornber, C. S., Rodriguez, L. F., Tomanek, L.
and Williams, S. L. 2006. The impacts of climate change in coastal marine systems. Ecology Letters, 9: 228-241.
Haugen, T. O. and Vøllestad, L. A. 2001. A century of life-history evolution in grayling. Genetica, 475-491.
Helfman, F. S., Collette, B. B. and Facey, D. E. 1997. The Diversity of Fishes. Blackwell Science, London. 528 p.
Hellbery, M. E., Balch, D. P. and Roy, K. 2001. Climate-driven range expansion and morphological evolution in a marine
gastropod. Science, 292: 1707-1710.
Hilborn, R., Walters, C.J. 1992. Quantitative fisheries stock assessment: choice, dynamics and uncertainty. Chapman &
Hall, New York
Hobday, A. J., Okey, T. A., Poloczanska, E. S., Kunz, T. J. and Richardson, A. J. (eds) 2006. Impacts of climate change on
Australian marine life: Part C. Literature Review. Report to the Australian Greenhouse Office, Canberra, Australia.
September 2006.
Horn, M.H. and Allen, L.G. 1978. A distributional analysis of California coastal marine fishes. Journal of Biogeography, 5:
23-42.
Hughes, L. 2000. Biological consequences of global warming: is the signal already apparent? Trends in Ecology and
Evolution, 15: 56-61.
Hughes, N. F. and Grand, T. C. 2000. Physiological ecology meets the ideal-free distribution: predicting the distribution of
size-structured fish populations across temperature gradients. Environmental Biology of Fishes, 59: 285-298.