ArticlePDF Available

Consistent global responses of marine ecosystems to future climate change across the IPCC AR5 earth system models


Abstract and Figures

We analyze for the first time all 16 Coupled Model Intercomparison Project Phase 5 models with explicit marine ecological modules to identify the common mechanisms involved in projected phytoplankton biomass, productivity, and organic carbon export changes over the twenty-first century in the RCP8.5 scenario (years 2080–2099) compared to the historical scenario (years 1980–1999). All models predict decreases in primary and export production globally of up to 30 % of the historical value. We divide the ocean into biomes using upwelling velocities, sea-ice coverage, and maximum mixed layer depths. Models generally show expansion of subtropical, oligotrophic biomes and contraction of marginal sea-ice biomes. The equatorial and subtropical biomes account for 77 % of the total modern oceanic primary production (PP), but contribute 117 % to the global drop in PP, slightly compensated by an increase in PP in high latitudes. The phytoplankton productivity response to climate is surprisingly similar across models in low latitude biomes, indicating a common set of modeled processes controlling productivity changes. Ecological responses are less consistent across models in the subpolar and sea-ice biomes. Inter-hemispheric asymmetries in physical drivers result in stronger climate-driven relative decreases in biomass, productivity, and export of organic matter in the northern compared to the southern hemisphere low latitudes. The export ratio, a measure of the efficiency of carbon export to the deep ocean, decreases across low and mid-latitude biomes and models with more than one phytoplankton type, particularly in the northern hemisphere. Inter-model variability is much higher for biogeochemical than physical variables in the historical period, but is very similar among predicted 100-year biogeochemical and physical changes. We include detailed biome-by-biome analyses, discuss the decoupling between biomass, productivity and export across biomes and models, and present statistical significance and consistency across models using a novel technique based on bootstrapping combined with a weighting scheme based on similarity across models.
Content may be subject to copyright.
1 3
DOI 10.1007/s00382-014-2374-3
Clim Dyn
Consistent global responses of marine ecosystems to future
climate change across the IPCC AR5 earth system models
Anna Cabré · Irina Marinov · Shirley Leung
Received: 20 June 2014 / Accepted: 11 October 2014
© Springer-Verlag Berlin Heidelberg 2014
climate-driven relative decreases in biomass, productivity,
and export of organic matter in the northern compared to
the southern hemisphere low latitudes. The export ratio,
a measure of the efficiency of carbon export to the deep
ocean, decreases across low and mid-latitude biomes and
models with more than one phytoplankton type, particu-
larly in the northern hemisphere. Inter-model variability is
much higher for biogeochemical than physical variables in
the historical period, but is very similar among predicted
100-year biogeochemical and physical changes. We include
detailed biome-by-biome analyses, discuss the decoupling
between biomass, productivity and export across biomes
and models, and present statistical significance and con-
sistency across models using a novel technique based on
bootstrapping combined with a weighting scheme based on
similarity across models.
Keywords CMIP5 intercomparison · Phytoplankton ·
Climate change · Earth system modeling ·
Marine ecosystems
1 Introduction
Global temperatures have risen steadily since the mid-
twentieth century (Levitus et al. 2000, 2009). This warm-
ing trend is projected to continue for the remainder of the
twenty-first century (Solomon et al. 2007; Stocker et al.
2013) and is expected to affect the growth of phytoplankton
both directly and indirectly by inducing changes in envi-
ronmental factors such as oceanic temperature, nutrient
concentrations, and light availability (e.g. Marinov et al.
Understanding how global phytoplankton popula-
tions will respond to climate change is critical, since
Abstract We analyze for the first time all 16 Coupled
Model Intercomparison Project Phase 5 models with
explicit marine ecological modules to identify the com-
mon mechanisms involved in projected phytoplankton
biomass, productivity, and organic carbon export changes
over the twenty-first century in the RCP8.5 scenario (years
2080–2099) compared to the historical scenario (years
1980–1999). All models predict decreases in primary and
export production globally of up to 30 % of the historical
value. We divide the ocean into biomes using upwelling
velocities, sea-ice coverage, and maximum mixed layer
depths. Models generally show expansion of subtropical,
oligotrophic biomes and contraction of marginal sea-ice
biomes. The equatorial and subtropical biomes account for
77 % of the total modern oceanic primary production (PP),
but contribute 117 % to the global drop in PP, slightly com-
pensated by an increase in PP in high latitudes. The phy-
toplankton productivity response to climate is surprisingly
similar across models in low latitude biomes, indicating a
common set of modeled processes controlling productiv-
ity changes. Ecological responses are less consistent across
models in the subpolar and sea-ice biomes. Inter-hemi-
spheric asymmetries in physical drivers result in stronger
Electronic supplementary material The online version of this
article (doi:10.1007/s00382-014-2374-3) contains supplementary
material, which is available to authorized users.
A. Cabré (*) · I. Marinov · S. Leung
Department of Earth and Environmental Science, University
of Pennsylvania, 251 Hayden Hall, 240 South 33rd St.,
Philadelphia, PA 19104, USA
S. Leung
School of Oceanography, University of Washington,
Seattle, WA 98105, USA
A. Cabré et al.
1 3
phytoplankton provide the ultimate food source for all
marine organisms and draw down atmospheric CO2 by fix-
ing inorganic carbon into organic matter via photosynthesis.
Sinking and subsequent remineralization of organic plank-
ton results in net transport of nutrients and carbon from the
euphotic zone to the deep ocean in a process known as the
biological pump. By affecting net deep ocean carbon stor-
age, changes in biological pump efficiency can provide a
feedback mechanism on climate and atmospheric pCO2
levels (e.g. Marinov et al. 2008). Thus, an important recent
development in Earth-System modeling has been the inclu-
sion of marine phytoplankton and ecology. The study of the
projected biological response to climate change globally
and on a centennial time scale has only recently become
possible with the inclusion of explicit phytoplankton func-
tional types (PFTs) in Global Circulation Models (GCMs).
Our goal here is to inter-compare the newest coupled
Earth System GCM simulations, run as part of the Coupled
Model Intercomparison Project Phase 5 (CMIP5) (Taylor
et al. 2012). We analyze for the first time all 16 CMIP5
models with explicit marine ecological modules, in order to
identify the common mechanisms and sensitivities involved
in projected phytoplankton biomass and productivity
changes over the twenty-first century, while simultane-
ously improving the statistical significance of multi-model
100-year trends. While the performance of various aspects
of ocean–atmosphere dynamics and the global carbon
cycle across multiple GCMs has been widely studied (e.g.
Friedlingstein et al. 2006; Arora et al. 2013), little research
has looked into what multiple GCMs can collectively say
about present and future ocean ecology across the different
To date, global modeling studies seem to agree on
a predicted decrease in global export production in the
twenty-first century (e.g. Bopp et al. 2001; Fung et al.
2005; Schmittner et al. 2008; Frölicher et al. 2009), but
offer contradictory projections for global primary pro-
duction (PP). Steinacher et al. (2010) found a decline in
both biological production and export production under
twenty-first century global warming in 4 different global
models under the SRES A2 scenario (Special Report on
Emission Scenarios, Nakićenović et al. 2000). Bopp et al.
(2013) presented an ecological comparison of 10 CMIP5
models, in which they found declines of both primary
and export production from the 1990s to the 2090s for all
Representative Concentration Pathways (RCP) scenarios
studied. In contrast, Schmittner et al. (2008) and Sarm-
iento et al. (2004a) saw declines in export production
but increases in projected global twenty-first century PP.
Whether models predict global PP increases or decreases
may depend on the parameterized sensitivity of metabolic
processes (growth and remineralization rates) to increases
in temperature.
In the subtropics, downwelling and strong water column
stratification hinder the supply of nutrients from nutrient-
rich deep waters to the surface. Consequently, phytoplank-
ton production is primarily nutrient-limited. Some studies,
though not all, have suggested that recent warming has
enhanced stratification, reduced surface nutrients and ulti-
mately productivity (e.g., satellite-based studies of Gregg
et al. 2005; Behrenfeld et al. 2006; Boyce et al. 2014),
and has resulted in an expansion of the low-chlorophyll
subtropical regions observed by satellites from 1998 to
2007 (Polovina et al. 2008; Irwin and Oliver 2009). How-
ever, these trends may be due to natural climate variabil-
ity related to the Pacific Decadal Oscillation or other cli-
mate indices rather than a long-term warming signal (Irwin
and Oliver 2009; Yoder et al. 2010). Further supporting
this notion, Martinez et al. (2009) find an inverse relation-
ship between changes in SST and chlorophyll (Chl) over
20 years of satellite data (1980–2000), synchronous with
decadal oscillations and indistinguishable from climate
warming. Henson et al. (2010) point out that we might need
40 more years of continuous satellite color data to detect an
anthropogenic warming signal in the time series of Chl (see
also Yoder et al. 2010; Beaulieu et al. 2013).
In the nutrient-replete high-latitude regions, phyto-
plankton tend to be co-limited by light and iron (Moore
et al. 2013). Therefore, an expected increase in stratifica-
tion (shoaling of the mixed layer) and decrease in sea-ice
coverage with climate change (Curran et al. 2003; Serreze
et al. 2007) might result in higher production rates and a
longer growing season by increasing light availability to
phytoplankton (Sarmiento et al. 2004a; Bopp et al. 2005;
Doney 2006; Steinacher et al. 2010). However, Martinez
et al. (2009) found decreasing chlorophyll concentrations
with surface warming and increased stratification at high
latitudes. Thus, phytoplankton productivity projections
at high latitudes are rather uncertain due to the complex-
ity of light and nutrient co-limitation, as well as the highly
variable timing of various processes that initiate blooms
in strongly seasonal regions (Martinez et al. 2011). For
example, high-nutrient low-chlorophyll (HNLC) areas
such as the Southern Ocean and subpolar Pacific Ocean
are known to be iron limited, and the iron supply from both
atmospheric and oceanic sources can change with climate
change. Recent research by Behrenfeld et al. (2010, 2013)
introduces the dilution-recoupling hypothesis, which states
that the deepening of the mixing layer depth during winter
creates favorable conditions for phytoplankton to grow due
to physical dilution of grazers, adding complexity to the
standard theory of light limitation at high latitudes.
As an added complication, interaction with the envi-
ronment varies among different phytoplankton groups. In
nutrient-poor waters (e.g., subtropical gyres), small phyto-
plankton outcompete larger ones such as diatoms because
Consistent global responses of marine ecosystems
1 3
they can acquire nutrients more efficiently due to their
higher surface area-to-volume ratio. Conversely, in nutri-
ent-rich, turbulent waters (e.g., subpolar domain), diatoms
dominate because of their higher intrinsic growth rates.
Therefore, with future climate warming-induced changes
in nutrient distributions, we expect a shift in phytoplankton
sizes to smaller or bigger species depending on the region.
Stronger stratification might lead to a decrease in diatom
abundance in low latitude regions, as suggested by both
modeling (e.g. Bopp et al. 2005; Marinov et al. 2013) and
observations on geological timescales (e.g., Finkel et al.
2005). This would affect the sinking rate of organic mat-
ter and thus export production, which is largely driven by
the relatively heavier diatoms, and the export or e-ratio,
which is the ratio between export production at the bottom
of the euphotic layer and the euphotic layer primary pro-
duction. A decrease in the relative abundance of large phy-
toplankton over the twenty-first century could thus result in
a shift to lower e-ratios and a consequent decrease in the
efficiency of the biological carbon pump. See Finkel et al.
(2010) for a review of how phytoplankton cell size affects
PP and export.
With climate change, the temperature contrast between
the Northern Hemisphere (NH) and Southern Hemisphere
(SH) is predicted to increase, with important implications
for the Hadley circulation and tropical rainfall patterns.
Globally, surface temperatures are increasing faster in the
NH than in the SH, an asymmetry largely attributed to less
ocean area and hence less heat capacity and thermal iner-
tia in the NH compared to the SH (e.g., Bryan et al. 1988;
Manabe et al. 1991; Flato and Boer 2001; Xu and Ramana-
than 2012). Higher evaporative latent heat losses over the
oceans and a stronger positive ice-albedo feedback in the
Arctic compared to the Antarctic further enhance this inter-
hemispheric temperature asymmetry (Wang and Overland
2012). Part of this asymmetry can also be explained by the
thermal isolation of the Southern Ocean imposed by the
Antarctic Circumpolar Current (Hutchinson et al. 2013).
On the other hand, sulfate aerosols are expected to cool
the NH more than the SH (Dufresne et al. 2005), reduc-
ing the asymmetry in warming set up by the aforemen-
tioned factors. Overall, Friedman et al. (2013) found that
from 1980 onward the NH-SH asymmetry in temperature
has been growing as the contribution of greenhouse gases
to the asymmetry began to overtake the contribution from
aerosols. The increasing temperature contrast between the
NH and SH combined with continued overall warming is
predicted to strengthen the southern Hadley cell, weaken
the northern Hadley cell, and lead to a pronounced north-
ward shift in tropical precipitation (Yoshimori and Broccoli
2009; Kang et al. 2009; Frierson and Hwang 2012; Fried-
man et al. 2013).
In both hemispheres, mid-latitude westerly winds
have intensified and shifted poleward in recent dec-
ades as a result of changes in atmospheric pressure pat-
terns, caused by a combination of elevated greenhouse
gas concentrations and ozone depletion (Thompson and
Solomon 2002; Strong and Davis 2007). These westerly
winds are intensifying more strongly in the SH, partially
offsetting buoyancy-driven increases in stratification
and circulation changes in the Southern Ocean (Russell
et al. 2006; Toggweiler 2009), with potentially signifi-
cant implications for the ocean carbon uptake there (Le
Quéré et al. 2007; Lovenduski and Ito 2009; Behrenfeld
et al. 2013; Bernardello et al. 2014, submitted). This dif-
ference in the magnitude of westerly wind intensifica-
tion between the NH and SH will likely contribute to a
north–south asymmetric response of ocean stratification,
nutrient supply to the surface, light limitation, and ulti-
mately phytoplankton biomass and productivity to cli-
mate change.
We expect that with twenty-first century climate
change across all models, decreased vertical mixing and
shallower mixed layer depths throughout the ocean will
result in (a) reduced phytoplankton primary production
in the tropics and nutrient-limited subtropics as well as
a shift to lower e-ratios, and (b) increased primary pro-
duction in the nutrient and light co-limited subpolar and
marginal sea-ice regions. The first objective of this paper
is to quantify the agreement among CMIP5 models as
to the effects of climate-driven physical changes on pri-
mary and export production. We thus compare the bio-
logical sensitivity to physical changes across models and
ecological biomes within models. We hypothesize that
across all models, global primary production and export
production will decrease with climate change, a signal
dominated by reductions at low latitudes. The second
objective is to study whether the asymmetry in physical
properties between the northern and southern hemispheres
propagates up to primary and export production, both in
terms of historical values and climate-driven centennial
In Sect. 2, we introduce the models and scenarios used
in our analysis, define the method used to separate the
ocean into biomes, and describe the statistical techniques
used to assess multi-model significance. In Sect. 3, we pre-
sent the main results. In Sect. 3.1, we analyze the physi-
cal mechanisms responsible for changes in phytoplankton
production one biome at a time. In Sect. 3.2, we explore
the relationships between PP, phytoplankton biomass, and
export across models. In Sect. 3.3 we focus on the asym-
metric response to climate change between the equivalent
Northern and Southern Hemispheres biomes. Conclusions
and discussion are presented in Sect. 4.
A. Cabré et al.
1 3
2 Methodology
2.1 Description of models and scenarios
We examine a group of Earth System simulations from
the recent Coupled Model Intercomparison Project
CMIP5 (Taylor et al. 2012) that can be downloaded at In order to estimate
100-year trends, our analysis compares 20 years of “pre-
sent” historical output (from 1980 to 1999) to 20 years of
“future” projected output (from 2080 to 2099). The “pre-
sent” output is based on the historical scenario (years 1850
to 2005) forced by observed atmospheric changes (both
anthropogenic and natural), and the “future” output is based
on the RCP8.5 scenario (Riahi et al. 2011) with prescribed
CO2 emissions that reach 936 ppm by year 2100 (Jones
2013). One of the analyzed models, MRI-ESM1, did not
provide RCP8.5, so the esmRCP8.5 scenario, which explic-
itly computes CO2 levels from anthropogenic emissions
(Friedlingstein et al. 2014), was used instead for this model
only. We also computed averages over 50 years in the pre-
sent (1950–1999) and future (2050–2099), and found no
significant differences to the averages over 20 years.
Our selection of models was limited by the availability
of the variable ‘phyc’ (“total phytoplankton carbon concen-
tration”) at the time of writing, as well as the availability of
model output for the RCP8.5 scenario. In total, we selected
16 models with different resolutions (ocean grid varies
from 0.5° to 2°) and complexities in their biogeochemical
and ecological modules, as described in Table 1.
The marine biogeochemical routine for models
CanESM2, MIROC-ESM, and MRI-ESM1 is based on
the basic NPZD (Nutrient Phytoplankton Zooplankton
Detritus) structure with only one phytoplankton type and
one nutrient (nitrate). The complexity increases with MPI-
ESM, NorESM1, HadGEM2, and GISS-E2 via inclusion of
more nutrients (nitrate, silicate, iron) and additional types
of phytoplankton for HadGEM2 and GISS-E2. Finally,
IPSL-CM5, GFDL-ESM2, and CESM1-BGC are the most
ecologically complex models, with at least 2 types of phy-
toplankton, zooplankton types, more than 20 biogeochemi-
cal tracers, and inclusion of ballast in the last two models.
The variables used for our analysis based on availability in
each of the models are listed in Table S1.
2.2 Biome separation
We define a set of ecological biomes based on physical cri-
teria, closely following the definitions employed by Sarm-
iento et al. (2004a) and Marinov et al. (2013) and conceptu-
ally along the lines of Longhurst (1994). We use the sign
of the annual mean vertical velocity at 50 m (upwelling if
directed upward, downwelling if directed downward), the
extent of the sea-ice coverage, and the maximum mixed
layer depth based on monthly analyses to determine our
principal bio-geographical provinces or biomes (Fig. S1).
The mixed layer depth (MLD) is defined as the depth where
the density difference to the surface above is 0.125 kg/m3
(Levitus 1982).
The equatorial domain is defined as the region between
5°S to 5°N and is further divided into upwelling and down-
welling subregions. The low latitude upwelling biome
(LLU) contains upwelling regions between 5°S and 35°S
and 5°N and 30°N; it is primarily composed of high nutri-
ent, high productivity regions along western continental
margins. The subpolar biome is the region north of 30°N
and south of 35°S (outside of the LLU and marginal sea ice
biomes) again where there is upwelling. Lastly, the mar-
ginal sea ice biome contains areas (or grid cells) at least
10 % covered in sea ice on an annually averaged basis.
The subtropical region is the region outside of the equa-
torial domain defined by downwelling at 50 m depth. The
subtropical region is further divided into a permanently
stratified subtropical biome, where the maximum mixed
layer depth (based on monthly analyses) does not exceed
150 m and a seasonally stratified subtropical biome, where
the maximum mixed layer depth does exceed 150 m. When
calculating average ecological properties over each biome
we exclude marginal bodies of water (Mediterranean Sea,
Baltic Sea, Red Sea, Persian Gulf, Hudson Bay), but unlike
Sarmiento et al. (2004a), we take into account the entire
Arctic Ocean.
We used the vertical mass transport at 50 m depth to dis-
tinguish between upwelling and downwelling regions when
this data was available in CMIP5 model output. We calcu-
lated vertical velocity at 50 m depth from horizontal veloci-
ties via mass continuity when vertical mass transport data
was not provided, i.e. for models GFDL-ESM2G, MIROC-
2.3 Statistical analysis
It has previously been shown that the average over multi-
ple models typically matches observations more closely
than any single model alone (see a comprehensive review
by Knutti et al. 2010). However, given the wide ranges in
complexity and real-world applicability among the mod-
els studied here a weighting scheme is needed to avoid
unwanted bias in inter-model averages. An informed deter-
mination of the appropriate model weights should ideally
come from comparing the model simulations to real data.
However, this procedure may only improve the inter-
model mean by very little because available observations
are often biased themselves and model ranking is based
only on historical performance, which can differ signifi-
cantly from performance in future climate change scenarios
Consistent global responses of marine ecosystems
1 3
Table 1 Summary of the CMIP5 models that include phytoplankton biomass and primary production
The table includes: spatial resolution in the atmosphere and ocean, list of nutrient tracers, ecology subroutine, phytoplankton functional groups modeled, references, and weight we applied in
the inter-model averages (see Sect. 2)
Model Atm (levels, long/lat) Ocean (levels, long/lat) Nutrients Ecology module Phyto Refs. Weight
CanESM2 L35
N, (but also accounts for
Fe limitation)
NPZD based on Denman
and Peña (1999)
Phyto Zahariev et al. (2008) 1
P, N, Fe, Si MET Diatom, nanophyto,
Moore et al. (2004, 2006) 1
P, N, Fe, Si TOPAZ2 Large separated into
diatoms and non-diatom,
small cyanobacteria,
Dunne et al. (2013) 1
P, N, Fe, Si TOPAZ2 Large separated into
diatoms and non-diatom,
small cyanobacteria,
Dunne et al. (2013) 1
HadGEM2-CC L60
N, Fe, Si Diat-HadOCC (NPZD) Diatom, non-diatom Palmer and Totterdell
HadGEM2-ES L38
N, Fe, Si Diat-HadOCC (NPZD) Diatom, non-diatom Palmer and Totterdell
P, N, Fe, Si PISCES (from
Diatoms, nanophyto Aumont and Bopp (2006),
Seferian et al. (2013)
P, N, Fe, Si PISCES (from
Diatoms, nanophyto Aumont and Bopp (2006),
Seferian et al. (2013)
N NPZD-type
(Oschlies, 2001)
Phyto Watanabe et al. (2011) 0.5
N NPZD-type
(Oschlies, 2001)
Phyto Watanabe et al. (2011) 0.5
P, N, Fe, Si HAMOCC5.2 (NPZD) 1 (But separated into dia-
toms and calcifiers)
Ilyina et al. (2013) 0.5
P, N, Fe, Si HAMOCC5.2 (NPZD) Phyto (but separated into
diatoms and calcifiers)
Ilyina et al. (2013) 0.5
P, N NPZD (Oschiles 2001) Phyto Yukimoto et al. (2011) 1
NorESM1-ME L26
P, N, Fe, Si HAMOCC5.1 (NPZD) Phyto (but separated into
diatoms and calcifiers).
Assmann et al. (2010) 1
N, Fe, Si NOBM Diatoms, chlorophites,
cyanobacteria, coc-
Gregg (2008) 1
N, Fe, Si NOBM Diatoms, chlorophites,
cyanobacteria, coc-
Gregg (2008) 1
A. Cabré et al.
1 3
(Knutti et al. 2010). Given our imperfect understanding of
each model’s real-world applicability (Knutti and Sedlacek
2013), we decide to weight models based on their similari-
ties alone to avoid double counting. If two models carry the
same or very similar information from the point of view of
ocean biogeochemistry or physics, we give them a weight
of 0.5 instead of 1 (Table 1).
In order to assess how similar two models are for a given
variable, we calculated the difference between the two
models at each location and studied the frequency distri-
bution of the differences. In Fig. S2, we show the average
and standard deviation of the distribution of the differences
for sea surface temperature (SST), nitrate, and biological
production. The following pairs of models are very corre-
lated (low average and low std. deviation in the distribu-
tion of differences): IPSL-CM5A-LR and IPSL-CM5A-
MR; HadGEM2-ES and HadGEM2-CC; MIROC-ESM and
model pairs share the same ocean physics and biogeochem-
ical modules, and vary only in the atmospheric chemis-
try and/or resolution (for ocean or atmosphere). Note that
and GFDL-ESM2M are independent sets of models for
our purposes, since the ocean physical modules are fun-
damentally different, e.g. GFDL-ESM2G uses isopycnal
coordinates while GFDL-ESM2M uses regular depth coor-
dinates. In addition, we analyze all the historical ensembles
for the GFDL-ESM2M model to determine if the chosen
initial conditions (which vary among single model ensem-
bles) could be responsible for some of the inter-model dis-
persion. We find that the dispersion among single-model
ensembles is much lower than the dispersion among mod-
els, so we can safely use the first ensemble (called “r1i1p1”
in the CMIP5 runs) for all the models here.
Model agreement in future climate projections is com-
monly calculated as the percentage of models that agree on
the sign of the trend in a given grid cell, a technique some-
times improved by also including information on the signif-
icance of each individual model (Tebaldi et al. 2011). This
approach is problematic for us, however, because it does
not take into account inter-annual variability within indi-
vidual models and instead introduces imprecision due to
the small number of models compared. To overcome these
issues, we use the widely accepted statistical technique
known as bootstrapping (Efron and Tibshirani 1993). In our
bootstrap sampling, each realization is the weighted aver-
age over N models that are selected randomly with replace-
ment among the N available models. We represent inter-
annual variability by picking one of the 20 years in the
present and future randomly each time that we randomly
select a model. For each studied variable at each point in
the ocean, we create 1,000 realizations of the resulting
100-year trend and obtain the multi-model significance of
this trend using the percentage of realizations that predict
a trend above or below zero. Hence, we solve the discrete-
ness problem while simultaneously taking inter-annual var-
iation into account, all without needing to assume Gaussi-
anity in the final distribution. We show the resulting trend
maps for sea surface temperature (SST), stratification,
nitrate, and PP in Fig. 1 and maps for sea surface salinity,
iron, phytoplankton biomass, and POC export in Fig. S3.
3 Results
3.1 Physical mechanisms affecting biological production
biome by biome: consistency across models
Here, we study the linear correlations between 100-year
(1980–1999 to 2080–2099) physical and biogeochemical
changes among different models to understand the common
mechanisms affecting phytoplankton productivity. We also
look into mechanisms driving historical phytoplankton pro-
ductivity distributions.
We show correlations between different variables within
each biome and plot the best-fit linear regression line only
when the significance of the correlation is higher than
95 %. We limit our analysis to a qualitative overview since
models can be related (i.e. have similar underlying equa-
tions) and bias the meaning of the significance. We try to
diminish the inter-model bias by setting model weight to
0.5 for models that are known to be very similar, as was
described above (Sect. 2.3; Table 1).
In principle, we expect to find significant correlations
across physical and biogeochemical variables when the
same unique mechanism is dominant within all the models.
However, the strength of a given correlation is dampened
if different models describe the mechanism differently (or
they describe the mechanism in the same way but with dif-
ferent sensitivities), or if the biological response is a result
of multiple mechanisms, as in the case of light-nutrient
co-limitation of phytoplankton growth. In this last case, a
linear regression between phytoplankton growth and light
would not suffice, and multi-linear regressions are needed
instead. Since the CMIP5 models all describe ocean phys-
ics in very similar ways, we expect correlations between
different physical parameters (e.g. temperature and strati-
fication) to be highly significant. The connection between
physical triggers, supply of nutrients, and ultimately pro-
duction changes are harder to uncover, however, since
each model parameterizes biological processes quite dif-
ferently (Table 1). Additionally, correlations between two
small climate-driven changes should be better defined than
those between two historical values or two large non-lin-
ear changes, as small changes can usually be accurately
Consistent global responses of marine ecosystems
1 3
We chose the following list of physical variables for our
analysis based on standard knowledge: surface sea tem-
perature (SST), air surface temperature, surface sea salin-
ity (SSS), stratification (defined as the density difference
between 200 m and surface), mixed layer depth (yearly
average, winter and summer values), photosynthetic radia-
tion at surface (yearly average and annual maximum), per-
cent cloud coverage, P E (precipitation minus evapora-
tion), thermocline depth (at 10 °C), upwelling transport
between 50–150 m depths, ice coverage (yearly average,
winter and summer values), and surface winds strength.
We then sought physical connections to nitrate and iron
concentrations (yearly averaged in the surface, yearly max-
imum, yearly minimum, and yearly averaged top 100 m),
and primary production integrated to 100 m. Mixed layer
depth calculations shown follow Levitus (1982); using the
definition from de Boyer Montégut et al. (2004) led to the
same conclusions.
We found that consistently across models, SST and SSS
(which is often driven by P E changes) can be usually
correlated to watercolumn stratification, which is in turn
related to MLD and thermocline depth, which in turn can
be related to nutrient supply and finally PP. However, cloud
coverage and upwelling velocities could not consistently
Fig. 1 Multi-model averaged
maps. Multi-model averaged
maps for historical (average
over years 1980–1999) and cli-
mate change signal (difference
between years 2080–2099 and
1980–1999). Variables shown
are a sea surface temperature, b
stratification (defined as the dif-
ference between potential den-
sity at 200 m and the surface), c
nitrate concentration averaged
in the top 100 m, and d primary
production integrated over the
top 100 m. We show the aver-
age over 16 models (consistent
with global changes listed in
Table 3). Crossed masked areas
show significant trends at the
8–90 % level, and stripped
masked areas significance at
more than 90 % level, obtained
with bootstrapping technique
(described in Sect. 2). Note that
projections for temperature are
not color-centered at 0
A. Cabré et al.
1 3
explain common mechanisms driving ocean biology across
models (i.e. no correlations between these and other param-
eters across CMIP5 models). We found that the inclusion of
multiple physical drivers (multi-linear regression) did not
improve much the correlation between physics and biogeo-
chemistry in cases where a simple linear correlation did not
work, as for example to explain the link between physics
and nitrate supply in low latitude upwelling biomes. This
lack of improvement indicates that models describe such
mechanism differently, that even more drivers are neces-
sary, or that the biome definitions should be improved and
refined. However, most drivers of ocean biology were well
identified just with simple linear correlations, especially in
the 100-year trends correlation.
3.1.1 Equatorial biome (5°S–5°N upwelling
and downwelling zones)
The highest modeled biological productivity rates in the
ocean occur in the equatorial upwelling biome (blue dots in
Fig. 2a). Productivity is also relatively high in the equato-
rial downwelling biome (purple in Fig. 2a). The equatorial
biomes historically account for a large fraction of global
production (15 %, Fig. 2e), relative to their area (9.5 % of
total ocean area, Fig. 2c). See also Table 2.
Phytoplankton growth in the equatorial biome is primar-
ily macro-nutrient limited in both the upwelling (blue lines
in Fig. 3) and downwelling (purple in Fig. 3) regimes. This
is supported by the fact that models with higher historical
nitrate concentrations in the equatorial biome are also more
productive there (Fig. 3b). Hydrological balance is one
of the most important physical variables that controls the
stability of the water column in the equatorial biome and
consequently the supply of nutrients to the surface, relation
not improved when adding other drivers such as tempera-
ture or upwelling velocity. Models with lower net precipita-
tion minus evaporation (P E) rates are less stratified and
hence more productive here (Fig. 3a).
Climate change results in a productivity decrease of
1.0 PgC/year (from a historical value of 6.2 PgC/year) in
the equatorial biome, corresponding to 30 % of the overall
decrease in global productivity over the twenty-first cen-
tury (Table 2, Fig. 2f). The equatorial upwelling regions
(blue in Fig. 2d) expand slightly toward the western side
of the basins at the expense of the equatorial downwelling
regions (purple in Fig. 2d), as observed in previous mod-
eling studies (e.g., Sarmiento et al. 2004a; Marinov et al.
2013). Table 2 shows the multi-model significance of this
tropical shift to upwelling biomes.
In this equatorial regime, twenty-first century climate
change results in temperature increases (Fig. 3c) and
salinity decreases (Fig. 3d) across all the models, the lat-
ter of which is associated with an intensified hydrological
cycle. Models that warm and freshen more experience
larger increases in stratification (Fig. 3c, d, also see labeled
the multi-linear regression for stratification change). As
expected, models that stratify more—or equivalently, mod-
els that undergo larger relative decreases in their maximum
MLD along with larger increases in precipitation rates—
exhibit greater relative decreases in nitrate (Fig. 3e) and
hence greater relative decreases in productivity (Fig. 3f).
Note that the biological responses to climate change
are similar in the upwelling and downwelling equatorial
biomes, suggesting that the mechanisms driving production
are not related to upwelling strength but to changes in over-
all watercolumn stratification.
3.1.2 Low‑latitude upwelling biomes
Productivity is relatively high in the low-latitude upwelling
(LLU) (dark green in Fig. 2a), but less so that in the equa-
torial biome. The low-latitude upwelling (LLU) region his-
torically accounts for 13 % of total ocean area (Fig. 2c) and
15 % of total global production (Fig. 2e).
Historically, models with higher sea surface tempera-
tures in the LLU regions are also more stratified (Fig. 4a).
Additionally, in the NH LLU biomes, models with lower
salinity due to a higher P E also have a more stratified
water column (Fig. 4b, also see multi-linear regression for
historical stratification).
Climate change results in a productivity decrease of
0.8 PgC/year (from 6.4 PgC/year historically) in this
biome, corresponding to 37 % of the overall decrease in
global productivity over the twenty-first century (Table 2,
Fig. 2f). The equatorial and LLU biomes together are
responsible for most of the global decrease in production
(67 %). The area of the low-latitude upwelling biomes
(dark green in Fig. 2d) does not change significantly.
Fig. 2 Evolution of biomes during the twenty-first century. For
each biome (see color legends) and model (see symbol legends),
from the South Pole to the North Pole, we show the historical (years
1980–1999) and climate change signal (difference between years
2080–2099 and 1980–1999) for: a, b productivity across biomes and
models, where each model’s productivity values were normalized to
50 PgC/year in order to eliminate model disagreement due only to
differences in total global production (model dispersion plotted in (a)
represents only inter-model differences in the distribution of produc-
tion across biomes), c, d percent historical area and relative change
in biome area, e, f percent historical contribution of each biome to
the global integrated production and contribution of each biome to
the global decrease in production (%) (for each model, the sum of
all the biome contributions adds to 100 %), g distribution of e-ratio
across biomes and models for the historical run, h contribution of
each biome to the global decrease in export (%), and i surface dia-
tom fraction across biomes and models for the historical run. The
global decreases in productivity and export for each model are listed
in Table 3. The bottom biome map is colored if at least 50 % of the
models agree at a given location. See Fig. S1 for all of the CMIP5
models biome maps
Consistent global responses of marine ecosystems
1 3
Following twenty-first century warming, all the mod-
els predict increases in sea surface temperature and hence
stratification in LLU biomes, which are greater for mod-
els that warm more (Fig. 4c). In contrast to the historical
case, changes in salinity do not play a significant role in
driving these changes in stratification. Ultimately, models
with a larger decrease in nitrate concentrations also show
a larger decrease in productivity in both NH and SH LLU
biomes (Fig. 4d), but the physical trigger that explains this
nitrate decrease is neither clear nor uniform across models
A. Cabré et al.
1 3
Table 2 Historical and climate changes in biome areas and productivities (biomes defined in Sect. 2)
For each biome (row), we list the multi-model average (and corresponding standard deviation) of the historical area (as a % of total ocean area for years 1980–1999), 100-year relative change
in area (% change from 1980–1999 to 2080–2099), historical productivity normalized to a global average production of 150 gC/year/m2 , 100-year change in productivity, biome contribution
to the historical, globally integrated production (%), biome contribution to the 100-year global decrease in integrated production (% of ΔPPglobal), biome contribution to the historical, globally
integrated export (%), biome contribution to the 100-year global decrease in integrated export (% of ΔExpglobal), historical integrated production over each biome, 100-year net change in inte-
grated production. Production is always integrated to 100 m depth
Area hist
(% global)
Rel. change
area (%)
PP hist
PP hist
(% globall)
% of Δ
Export hist
(% global)
% of Δ
PP hist
Ice-biome S 5.21 (1.17) 23.26 (14.52) 90.24 (48.09) 17.93 (15.61) 3.12 (1.55) 7.93 (9.98) 3.38 (1.29) 5.09 (3.46) 1.08 (0.52) 0.18 (0.14)
Subpolar S 9.80 (2.05) 0.87 (10.11) 177.49 (85.66) 7.51 (14.83) 12.46 (6.38) 4.77 (19.16) 10.73 (4.26) 0.05 (4.31) 4.19 (2.14) 0.05 (0.20)
Seas subtrop S 11.97 (2.40) 4.44 (7.12) 176.44 (35.87) 7.92 (11.60) 15.19 (4.80) 0.55 (13.67) 13.41 (3.58) 7.66 (5.52) 5.19 (1.62) 0.22 (0.31)
Perm subtrop S 19.26 (2.89) 10.34 (6.03) 119.67 (19.91) 10.46 (8.77) 16.49 (4.26) 0.03 (37.48) 16.88 (3.13) 14.39 (4.26) 7.12 (4.99) 0.41 (0.47)
Low-lat upwelling S 7.04 (1.08) 0.76 (6.97) 164.89 (39.28) 25.26 (18.22) 8.12 (1.85) 15.28 (11.28) 8.11 (1.33) 11.10 (3.62) 3.45 (2.24) 0.38 (0.26)
Equat. downwelling 3.54 (0.60) 10.08 (11.17) 180.22 (72.81) 37.24 (27.53) 4.76 (2.71) 12.08 (14.94) 5.55 (3.15) 10.05 (7.18) 2.04 (1.40) 0.32 (0.24)
Equat. upwelling 5.98 (0.63) 4.78 (3.96) 242.12 (69.46) 47.85 (39.10) 10.19 (2.83) 17.88 (20.51) 9.96 (1.78) 13.98 (6.47) 4.19 (2.36) 0.65 (0.65)
Low-lat upwelling N 6.32 (0.87) 0.84 (5.13) 155.68 (46.24) 24.80 (12.05) 6.95 (2.21) 21.93 (24.36) 7.07 (1.48) 13.01 (5.32) 2.99 (1.96) 0.40 (0.23)
Perm subtrop N 15.03 (1.75) 10.64 (5.75) 93.25 (33.34) 14.47 (7.36) 10.07 (4.09) 35.13 (39.83) 10.90 (2.23) 22.18 (9.44) 4.70 (3.83) 0.59 (0.33)
Seas subtrop N 5.32 (1.47) 19.10 (11.99) 126.27 (37.67) 27.09 (17.82) 4.79 (1.85) 13.89 (9.96) 4.46 (1.77) 8.55 (2.35) 1.78 (0.89) 0.30 (0.15)
Subpolar N 5.63 (0.57) 9.30 (3.79) 158.60 (43.57) 18.52 (16.49) 6.39 (2.11) 10.15 (11.11) 6.57 (1.88) 7.23 (4.37) 2.23 (0.78) 0.19 (0.12)
Ice-biome N 4.90 (0.82) 29.94 (10.13) 41.44 (15.51) 7.47 (7.26) 1.45 (0.61) 14.14 (26.68) 2.97 (2.13) 3.09 (5.17) 0.64 (0.47) 0.13 (0.14)
Consistent global responses of marine ecosystems
1 3
(i.e. no inter-model correlation between increased stratifi-
cation or other combination of studied physical drivers and
decreased nitrate was found).
3.1.3 Subtropical biomes
In the permanently stratified subtropics (orange in Fig. 2a),
downwelling and low mixing limit the supply of nutrients
and make these low productivity regions. In the seasonally
stratified subtropics, deeper winter mixed layers enhance
productivity over their permanently stratified counterparts
(brown in Fig. 2a).
In the historical run, the permanently stratified subtrop-
ics (orange in Fig. 2c) are the largest biomes in the ocean,
accounting for 19.3 % of the total ocean area in the SH
and 15 % in the NH on average (Table 2). Models with
larger permanently stratified biomes have smaller season-
ally stratified biomes, since the total subtropical area, con-
strained by the position of the westerlies, is relatively con-
sistent across models.
The subtropics historically account for 51.6 % of the
ocean area (35 % in the permanently stratified biomes and
13.9 % in the seasonally stratified biomes) and 46.6 % of
total oceanic production (Table 2, Fig. 2e). While 15.2 %
of the global production occurs in the southern season-
ally stratified subtropics, only 4.8 % of the global produc-
tion occurs in the northern counterpart. While 16.5 % of
the total production occurs in the southern permanently
stratified subtropics, only 10.1 % occurs in the northern
The total subtropical area (the sum of seasonally and
permanently stratified areas) slightly expands with climate
Fig. 3 Equatorial mechanisms (5°S–5°N). a, b Historical (years
1980–1999) relations between sets of two variables across CMIP5
models in the equatorial upwelling (blue) and downwelling (purple)
biomes, and cf 100-year changes in variables (difference between
years 2080–2099 and 1980–1999) plotted across CMIP5 models.
NO3 refers to nitrate averaged over the top 100 m, NO3surf refers to
surface nitrate, Prod refers to primary production integrated to 100 m,
P E refers to precipitation minus evaporation, and salinity refers to
sea surface salinity. Relative differences (e.g., ΔPP/PP) are calculated
relative to the 1980–1999 period. Legend numbers are the slopes of
the correlations, plotted as solid lines only when significant at a level
>95 %. Results for a multi-linear regression of stratification changes
against SST and SSS changes is shown below d
A. Cabré et al.
1 3
change among all the models studied, Fig. 2d, due to pole-
ward shifts in the subtropical-subpolar fronts, in turn driven
by the expansion of Hadley cells and the poleward shift of
westerly winds (Meijers et al. 2012). The subtropics expand
by up to 5 % of the original area in the NH and up to 10 %
in the SH, with higher inter-model disagreement in the SH
(see standard deviation in Table 2). The IPSL-CM5A mod-
els predict the biggest total subtropical expansion in both
hemispheres along with the strongest intensification of the
westerly winds.
The permanently stratified subtropical biomes are closer
to the equator, warmer, and more stably stratified than the
seasonally stratified subtropical biomes. To a good approxi-
mation, stratification is a function of surface temperature
and salinity (SST and SSS) in the permanently stratified
biomes (mostly SSS in the NH and SST in the SH). How-
ever, in the seasonally stratified subtropical biomes, sur-
face variables SST and SSS are not sufficient to explain
watercolumn stratification, as mixing is larger here and
stratification also depends on temperature and salinity
at depth. Within each type of subtropical biome, the NH
ones are always warmer and more stratified than their SH
counterparts (Fig. 5a). Warmer models supply less nitrate
to the surface ocean (Fig. 5a) and are also less productive
(Fig. 5b). This relationship holds throughout all of the sub-
tropical biomes, except for the NH permanently stratified
biome. For a given model, the warmer (or more stratified)
biomes require smaller supplies of nitrate to reach the same
levels of productivity because phytoplankton within these
more stably stratified biomes utilize nitrate more efficiently
(Sarmiento et al. 2004b).
Climate change results in a productivity decrease of
1.5 PgC/year (from 19.0 PgC/year historically), corre-
sponding to 49.5 % of the overall decrease in global pro-
ductivity over the twenty-first century (Fig. 2f). This
decrease occurs mostly within the NH (49 %), which
contains only 20 % of the total ocean area and accounts
for 15 % of total global historical production (Table 2).
Production in the southern subtropics remains virtually
unchanged, accounting for less than 1 % of the global
decrease in the multi-model average.
Over the twenty-first century, stratification intensifies
in the subtropics within all the models. These increases in
stratification are correlated across models with the degree
Fig. 4 Low-latitude upwelling
mechanisms. a, b Historical
(years 1980–1999) relations
between variables across
CMIP5 models in the low-
latitude upwelling biomes of the
NH (green) and SH (red), and
c, d 100-year changes in vari-
ables (difference between years
2080–2099 and 1980–1999).
NO3max refers to the annual
nitrate maximum value, which
occurs during winter. The
legend numbers are the slopes
of the correlations plotted as
solid lines only when the cor-
relation is significant at a level
>95 %. Results for a multi-
linear regression of historical
stratification against SST and
SSS shown below a
Consistent global responses of marine ecosystems
1 3
of warming experienced by the biome (Fig. 5d). In the
northern permanently stratified biome, changes in SSS
are anti-correlated with changes in SST across models
indicating increased precipitation, and both SST and SSS
contribute to increased stratification. In agreement with
an overall increase in water column stratification, the per-
manently stratified subtropical biomes (orange in Fig. 2d)
are predicted to expand significantly with warming across
all models in both hemispheres (mostly towards the East
Atlantic in the NH), while seasonally stratified subtropi-
cal biomes contract (red in Fig. 2d). Models with a larger
increase in stratification predict a larger expansion of the
permanently stratified biomes in both hemispheres, and
a larger retraction of the seasonally stratified biomes in
the NH only (Fig. 5c). In the SH, the seasonally strati-
fied biome expands slightly poleward due to shifts in the
subpolar-subtropical front, breaking the correlation with
stratification increase. Still, most of models predict an
overall retraction of the seasonally stratified biome in the
SH (Fig. 2d).
Models that stratify more show greater decreases (shoal-
ing) in MLD (figure not shown) and predict a larger reduc-
tion in the supply of surface nitrate (Fig. 5e) and ultimately
a larger decrease in PP (Fig. 5f). The seasonally stratified
biomes experience more pronounced MLD shoaling than
permanently stratified biomes due to their historically deep
MLDs, especially in the SH (as also observed by Sallee
et al. 2013).
This region shows some of the strongest and most con-
sistent correlations among variables, suggesting that (a) our
definition of subtropical biomes (including the separation
between the permanently and seasonally stratified domains)
is good, as each sub-region shows consistent mechanisms
which agree with our expectations, and (b) mechanisms
driving productivity in each subtropical sub-region are con-
sistent among models.
Fig. 5 Subtropical mechanisms. a, b Historical (years 1980–1999)
relations between variables across CMIP5 models in the subtropics,
and corresponding (cf) 100-year changes in variables (difference
between 2080–2099 and 1980–1999) plotted across CMIP5 models.
Mechanisms shown for the permanently stratified NH (light green)
and SH (yellow) subtropics, and the seasonally stratified NH (dark
green) and SH (red) subtropics. Legend numbers show the slopes in
the correlations plotted as solid lines only when the correlation is sig-
nificant at a level >95 %
A. Cabré et al.
1 3
3.1.4 Subpolar biome
The productivity in the subpolar biome is higher than in
the subtropical biome mainly due to upwelling of nutrients
(yellow in Fig. 2a). In the multi-model mean, the subpolar
biome historically accounts for 15 % of the total ocean area
(9.8 % in the SH and 5.6 % in the NH) and 19 % of the
total production (12.4 % in the SH and 6.4 % in the NH).
See Table 2.
In the historical NH subpolar biome, warmer models
show higher evaporation, lower P E and hence higher
salinity, and less stratification (Fig. 6a, see labeled multi-lin-
ear regression for historical stratification). In the SH, strati-
fication cannot be well approximated as a function of sur-
face parameters (SST and SSS). Phytoplankton in the NH
subpolar region are primarily nitrate limited across models,
such that models with higher euphotic layer nitrate exhibit
higher productivity (Fig. 6b). Phytoplankton in the southern
subpolar region are iron and light co-limited. We find that
historically, models with higher (annual maximum) surface
iron exhibit higher PP irrespective of light (Fig. 6c).
Climate change results in a productivity decrease of
0.1 PgC/year (from a historical 6.4 PgC/year), corre-
sponding to 5 % of the overall decrease in global produc-
tivity over the twenty-first century (Fig. 2f, Table 2). This
decrease is dominated by the NH (10 %), but is partially
compensated by a productivity increase in the SH (5 %).
With twenty-first century climate warming, the subpolar
biome (yellow in Fig. 2d) expands poleward as the sea-ice
biome retracts (light green in Fig. 2d), especially into the
Barents Sea in the North Atlantic. Meanwhile, subtropical
expansion results in subpolar retraction on the equatorial
side. The net change in the subpolar biome area is positive
in both the NH (9.3 %) and the SH (0.9 %).
Fig. 6 Subpolar mechanisms. ac Historical (1980–1999) relations
between variables across CMIP5 models in the subpolar biomes
of the NH (green) and SH (red), and df 100-year changes in vari-
ables (difference between 2080–2099 and 1980–1999) plotted across
CMIP5 models. Legend numbers show the slopes in the correlations
plotted as solid lines only when the correlation is significant at a level
>95 %. Results for multi-linear regressions of stratification against
SST and SSS shown below a and d
Consistent global responses of marine ecosystems
1 3
As in all previously discussed biomes, models that
warm more yield greater increases in stratification in both
the NH and SH subpolar, with higher rates of increase
in the faster-warming NH (Fig. 6d). In the NH subpo-
lar biomes, models that warm more also yield greater
decreases in surface salinity, which are unrelated to
increases in P E, suggesting that melting of sea-ice con-
tributes to the freshening and stratification of the subpo-
lar NH across models (see multi-linear regression labeled
below Fig. 6d). In this biome, increased stratification then
affects the supply of nutrients and ultimately phytoplank-
ton production, such that models with larger increases in
stratification and larger decreases in nitrate experience
larger decreases in production (Fig. 6e, f).
In the southern subpolar biome, models disagree on the
main mechanism driving changes in production, which
explains the lack of inter-model correlations between
changes in stratification, nitrate, iron, and productivity.
In a companion paper, we study this important region in
detail by breaking it further in latitude bands (Leung et al.,
3.1.5 Sea‑ice biome
The least productive region is the marginal sea-ice biome
(light green in Fig. 2a), where pervasive ice coverage pre-
vents the growth of phytoplankton via light limitation.
Historically, models have simulated sea-ice coverage
more accurately in the NH, where the sea ice has greatly
diminished over the last 3 decades (1980–2010). Never-
theless, the new IPCC AR5 models under-estimate these
recent decreases, as did the earlier AR3 models (Stroeve
et al. 2012; Wang and Overland 2012; Massonnet et al.
2012). In the SH, observations show an overall increase in
sea ice extent over the past few decades (Zwally et al. 2002;
Turner et al. 2009), while almost all of the models predict a
decrease (Zunz et al. 2013). Poor model representation can
be explained by a lack of or discontinuity in observations
used to calibrate the simulations, as well as by strong and
poorly understood interannual variability in sea ice cover-
age (Zunz et al. 2013).
In the multi-model historical mean (1980–1999), the
marginal sea-ice biome accounts for 10 % of the area and
5 % of the total production (Table 2, Fig. 2d). Production
in the marginal sea-ice biome is greater in the SH (3.4 %)
than in the NH (1.1 %) even though both regions occupy
a similar area (around 5 % of total ocean area). This is
most likely because the Arctic is historically more stably
stratified and thus more nutrient-limited than the Southern
A key characteristic of these biomes is that surface
freshening plays the most important role in determining
water column stratification intensity. Accordingly, we find
that models with larger historical surface salinities are less
stratified in both NH and SH biomes (Fig. 7a). Another
interesting characteristic of sea-ice regions is that they are
light and nutrient co-limited. In the SH, the historical extent
of sea ice coverage depends on the air and sea surface tem-
perature, such that warmer models are less ice-covered
(Fig. 7b) and consequently less light limited, which boosts
production when iron concentrations are sufficiently high
(Fig. 7c). On the other hand, the NH sea-ice biome is pri-
marily nitrate-limited, such that models with higher histori-
cal nitrate (figure not shown) and iron concentrations are
more productive. In general, historical stratification inten-
sity determines the level of biologically available nutrients
here (figure not shown).
Climate change results in a productivity increase of
0.3 PgC/year (from a historical 1.7 PgC/year), correspond-
ing to 22 % of the absolute value of the overall decrease in
global productivity over the twenty-first century (Table 2,
Fig. 2f). The marginal sea-ice biome’s increase in produc-
tion thus compensates significantly for production losses in
the rest of the ocean.
The net warming in the NH sea-ice biome is larger than
in the SH sea-ice biome (Fig. 7d, f). The intense surface
warming in Arctic latitudes is a consistent fea-ture in tran-
sient climate model integrations (e.g., Manabe and Stouffer
1980, Manabe et al. 1991; Meehl et al. 2007) often referred
to as polar amplification. The weaker transient warming
response in high Southern polar latitudes has been linked
with deep-ocean mixing, strong ocean heat uptake and the
persistence of the vast Antarctic ice sheet.
Over the twenty-first century (1980–2100), all of the
CMIP5 models predict a retraction of the northern marginal
sea-ice biome by 20–60 % (Fig. 2d). In the SH, models
mostly agree on a projected retraction of the sea-ice biome
by up to 50 %, except for GFDL-ESM2G, which shows an
increase in area. Figure 2b shows that agreement among
models on the 100-year change is lower in the SH than in
the NH, possibly due to a better understanding of ocean cir-
culation responses to climate change and hydrological cycle
forcing in the NH (Zunz et al. 2013). Over the twenty-first
century, sea-ice biome retractions are well correlated with
temperature increases, such that models predicting stronger
warming yield greater decreases in sea-ice biome area
(Fig. 7d). In the north, the sea-ice biome retracts at a slower
net rate (3.5 % per 1 °C warming) than in the south (6.9 %
per 1 °C, slopes in Fig. 7d), likely because the ice in the SH
is highly seasonal and hence easier to melt. Still, because
of larger net warming in the NH due to the ice-albedo feed-
back, total fractional contraction of the sea ice biome over
100 years is larger in the NH (30 %) than in the SH (23 %)
(Table 2). Enhanced surface freshening (leading to lower
sea surface salinities) due to a combination of increased
precipitation and sea-ice melting rates plays an important
A. Cabré et al.
1 3
role in increasing stratification within the sea ice biomes,
such that models with greater decreases in salinity experi-
ence higher increases in stratification in both the NH and
SH biomes (Fig. 7e). In the SH, PP increases more in mod-
els projecting larger increases in air temperature (Fig. 7h),
which drive a faster shrinking of sea ice extent (Fig. 7f) and
Fig. 7 Sea-ice biome mechanisms. ac Historical (1980–1999) rela-
tions between variables across CMIP5 models in the marginal sea-ice
biomes of the NH (green) and SH (red), and di 100-year changes
in variables (difference between 2080–2099 and 1980–1999) plotted
across CMIP5 models. Legend numbers show the slopes in the cor-
relations plotted as solid lines only when the correlation is significant
at a level >95 %
Consistent global responses of marine ecosystems
1 3
thus a greater increase in phytoplankton light availability
(Fig. 7g). Increased iron does not correlate with increased
production here, suggesting that release of light limita-
tion is the common factor increasing production across
CMIP5 models despite the historical presence of iron-light
The Arctic, on the other hand, is light and nitrate co-
limited with ice coverage determining the extent of light
limitation. Recent studies by Popova et al. (2012) and
Vancoppenolle et al. (2013) investigating the phytoplank-
tonic response to climate change in the Arctic conclude
that while all the CMIP5 models simulate similar levels
of light limitation (depending on ice coverage over the
twenty-first century), they differ in nutrient utilization
parameterizations and as a result, disagree on whether light
or nitrate is the main limiting factor for primary production.
Here we find that over 100 years, release of light limitation
due to sea ice retreat is counter-balanced by an increase
in NO3 limitation due to increased stratification (Fig. 7i).
Interestingly, all the models show an initial increase in pro-
duction due to sea-ice retreat, which is replaced by a subse-
quent decrease in production when the effect of decreasing
nitrate takes over. In agreement with Popova et al. (2012),
we see decreases in nitrate (Fig. 7i), increases in produc-
tion, and increases in light availability (Fig. 7g) due to
melting of sea-ice. While most models (all but the IPSL-
CM5A ones) predict net increases in Arctic production by
the end of the twenty-first century, the different strengths of
Fig. 8 Change in phytoplank-
ton biomass, production, and
export. For each biome (see
color legends) and model (see
symbol legends): a relation
between phytoplankton biomass
averaged on the top 100 m and
primary production (PP) in the
historical run (1980–1999),
b relation between relative
changes in phytoplankton
biomass and production over
100 years (years 1980–1999 to
2080–2099), c relation between
POC export at 100 m and PP
in the historical run, d relation
between relative changes in
PP and POC export at 100 m
over 100 years. Circled colored
symbols refer to the NH biomes,
and dotted symbols refer to
simple models, defined here as
models with just one phyto-
plankton type. b, d 1–1 line
corresponding to no changes in
the export or e-ratio with cli-
mate change. In all panels, the
grey lines highlight approximate
correlations across biomes for a
given model (circles if the rela-
tion is disperse)
A. Cabré et al.
1 3
nutrient-light co-limitation among models result in no sig-
nificant multi-linear correlations between changes in light,
nitrate, and production across models.
3.2 Relationship between production, phytoplankton
biomass, and export across biomes: consistency
across models
Below we refer to models that have only one type of phyto-
plankton as “simple” models, and to models with multiple
phytoplankton types or phytoplankton functional groups
(PFTs) as being ecologically “complex”.
Within the simple models, primary production (inte-
grated to 100 m) is highly correlated with phytoplankton
biomass (averaged over the top 100 m) across biomes, sug-
gesting that phytoplankton biomass is primarily driven by
bottom-up ecological controls (see tight relation in Fig. 8a
for each model). For each model, prescribed grazing and
loss rates determine the slope of the primary production-
biomass quasi-linear relationship.
Among models with only one type of phytoplankton,
MPI-ESM and NorESM1-ME (with biogeochemical mod-
ule HAMOCC5) use a Holling type II shape to parameter-
ize grazing, while models MIROC-ESM, CanESM2, and
MRI-ESM1 (with simple NPZD biogeochemical modules)
use a Holling type III shape. The Holling type II scheme
assumes a hyperbolic predator intake rate with increasing
prey concentration following a Michaelis–Menten curve,
while the Holling type III scheme assumes a sigmoid pred-
ator intake rate, with accelerating predator intake at very
low prey concentrations and saturation at high prey concen-
trations. Additionally, all of these models include phyto-
plankton loss and aggregation terms as a linear or quadratic
function of phytoplankton concentration.
However, this bottom-up driven relationship between
PP and biomass may not be valid in models with at least
two types of phytoplankton because small phytoplankton,
which are abundant in oligotrophic biomes, are grazed
more efficiently than diatoms, which are abundant in
upwelling zones and high latitudes. The impact of top-
down controls on phytoplankton therefore varies across
biomes within these more complex models.
Among models with multiple PFTs, HadGEM2 models
apply a Holling type II curve to parameterize grazing, with
additional switches on predation rates depending on abun-
dances of small and large phytoplankton. GISS-E2 models
apply the Ivlev formulation to parameterize grazing (simi-
lar to Holling type II) with a grazing rate independent of
phytoplankton type. The HadGEM2 and GISS-E2 models
still show a tight correlation between PP and phytoplank-
ton, likely due to the simplicity of the grazing and loss
equations as well as the overall dominance of bottom-up
In the GFDL-ESM2 models (biogeochemical module
TOPAZ), grazing and losses are modeled altogether as
a function of phytoplankton to the power of 2 for small
phytoplankton and 4/3 for large phytoplankton in order
to ensure that grazers can keep up with small phytoplank-
ton but cannot keep up with diatoms, as observed in situ
(Dunne et al. 2005). IPSL-CM5A models include two types
of zooplankton and apply a Holling type II grazing scheme
with grazer preference, again in order to ensure that graz-
ers can keep up with small phytoplankton but not diatoms.
Finally, CESM1 applies a Holling type III grazing formula-
tion also with different predator intake rates for small and
large phytoplankton, again decoupling phytoplankton bio-
mass from PP and breaking the relationship expected for
bottom-up control alone.
Across these complex models, PP and phytoplankton
are not well correlated across biomes because areas domi-
nated by small phytoplankton (e.g. subtropics) undergo
further reductions in biomass due to enhanced grazing
rates, while the opposite is true for areas dominated by
large phytoplankton (e.g., the high-latitude subpolar and
sea-ice biomes). Additionally, the high PP values linked to
anomalously low phytoplankton concentrations in GFDL-
ESM2M (Fig. 8a) occur because there exists an especially
high relative proportion of small phytoplankton within this
model compared to the other models with more diatoms.
Within each ecological biome, the correlation between
PP and phytoplankton biomass across all the CMIP5 mod-
els is tight, suggesting that as a group, CMIP5 models are
bottom-up controlled.
Figure 8b plots the relative changes in phytoplankton
biomass and primary production with twenty-first century
climate change. In general, models with greater predicted
decreases in PP also exhibit greater predicted decreases
in biomass. The relative drop in production is larger than
the relative drop in biomass for most of the models (i.e.,
models align above the 1–1 line in Fig. 8b), as grazing also
diminishes in proportion to phytoplankton biomass. More-
over, the relative drop in production is tightly correlated
to the relative drop in biomass across biomes within most
simple models (models with one-type of phytoplankton)
and within HadGEM2 and GISS-E2 (on top of line labeled
‘Rest of models’ in Fig. 8b). CanESM2 shows the same
general relationship between PP and biomass changes but
with greater dispersion across ecological biomes than the
other simple models.
Within the ecologically complex GFDL-ESM2 and
IPSL-CM5A models, the tendency for decreases in pro-
duction to be larger than decreases in biomass (again, due
to decreased grazing rates at lower biomasses) is opposed
by the shift from large to small phytoplankton driven by
increasing nutrient limitation under more stratified future
conditions. This shift from large to small phytoplankton
Consistent global responses of marine ecosystems
1 3
increases grazing pressure and ultimately leads to a larger
relative drop in phytoplankton biomass than would have
occurred without a shift toward smaller phytoplankton.
These compensating effects thus result in a relative drop
in phytoplankton biomass equal to the relative drop in pro-
duction within IPSL-CM5A, and a relative drop in phy-
toplankton biomass greater than the relative drop in PP
within GFDL-ESM2M (Fig. 8b). We do not show results
for CESM1-BGC phytoplankton biomass averaged over
top 100 m in Fig. 8a, b because only surface output was
provided but results at the surface are consistent with a
complex model response.
Historically, primary production and POC export at
100 m are also well correlated across biomes within most
models (Fig. 8c), meaning that POC export within simple
(one phytoplankton group) models is a constant fraction of
PP, i.e. the e-ratio (POC export at 100 m divided by pro-
duction integrated to 100 m) is fixed throughout the ocean.
Note however that this e-ratio (the regression slope in
Fig. 8c) varies across models. CanESM2 and MRI-ESM1
show high e-ratios compared to other ecologically simple
models such as NorESM1-ME, MPI-ESM and HadGEM2,
presumably because their single phytoplankton class is
modeled as more diatom-like (larger, faster sinking diatoms
result in higher POC export out of the euphotic layer).
The three most ecologically complex models (GFDL-
ESM2, CESM1-BGC, and IPSL-CM5A) include POC
concentrations that depend on the aggregation of dissolved
organic matter and depend on phytoplankton size such that
a larger pool of POC forms in the presence of large phy-
toplankton compared to small phytoplankton. Addition-
ally, CESM1-BGC and GFDL-ESM2 models include bal-
last material. HadGEM2, on the other hand, simply defines
POC as a portion of the total biomass with no distinction
between different phytoplankton sizes, leading to a well-
defined linear relationship between PP (which is linearly
related to biomass) and export (Fig. 8c).
Among the complex models, IPSL-CM5A and CESM1-
BGC have e-ratios closest to those of the simple models,
although the correlation between PP and export is more dis-
perse in the complex models because of the different model
parameterizations of zooplankton grazing and the different
relative contributions of small and large phytoplankton to
total productivity and export. GFDL-ESM2 models exhibit
low export rates relative to their high productivities due to
strong recycling of organic matter in the euphotic layer,
driven by a relative excess of small phytoplankton com-
pared to other models with greater diatom relative abun-
dances (e.g. HadGEM2-ES) or models which include a sin-
gle “diatom-like” phytoplankton type (e.g. CanESM2 with
high remineralization rate).
The findings above are consistent with Fig. 2g, which
shows historical e-ratios across biomes and models. The
biome-averaged e-ratio varies over an order of magnitude
among models from 0.05 for GFDL to 0.4 for CanESM2.
The e-ratio is approximately constant across biomes within
the simple models, which explains the tight correlation
between PP and export that we observed in Fig. 8c but fol-
lows latitudinal patterns for all models containing multiple
phytoplankton types (see Table S1) with higher e-ratios in
nutrient and diatom-rich high-latitude regions and lower
e-ratios in nutrient and diatom-poor (small phytoplankton-
rich) oligotrophic regions (Fig. 2i). The CanESM2 model,
with just one type of phytoplankton, also shows lower
e-ratios in lower latitudes as more organic matter is rem-
ineralized at the surface in more highly stratified waters,
which further reduces the number of particles transported
out of the euphotic layer.
We find that with twenty-first century climate change the
e-ratio does not change much in HadGEM2 or in simple
models with only one phytoplankton type, as the relative
changes in PP and export are identical for these models (see
dotted points in Fig. 8d, on top of the 1–1 line). In complex
models with multiple PFTs, the relative decrease in export
is greater than that in production, indicating a shift to lower
e-ratios and more oligotrophic conditions dominated by
small phytoplankton, particularly in low and mid-latitude
biomes. Hence, it is clearly the parameterization of PFTs
that decouples relative changes in export and primary pro-
duction over the twenty-first century.
3.3 North–south asymmetric response of biological
production to climate change
We next average all our physical and biogeochemical
indices over the northern (NH) and southern hemispheres
(SH) and explore north–south asymmetries in ecological
responses to climate change.
Historically, the NH (green symbols in Fig. 9a) is
warmer and more strongly stratified than the SH (red sym-
bols in Fig. 9a) across all models studied, an asymmetry
that arises from the subtropical and subpolar biomes in
particular. In 100 years (line ending with no symbol in
Fig. 9a), SH waters are predicted to become approximately
as warm as NH ones are at present, though less strongly
stratified because of their relative freshness. In the NH,
saltier models are less stratified, suggesting that salinity (or
freshening) is a crucial driver of ocean physics in the NH
(Fig. 9b). In accordance with nitrate being the main lim-
iting factor for phytoplankton growth, models with higher
historical nitrate are more productive in the NH (Fig. 9c).
In contrast, the SH’s behavior is more complex because of
strong iron-light co-limitation in high latitudes.
We next explore correlations between indices that repre-
sent north–south asymmetries, using the difference between
values in the NH and SH. Models with greater historical
A. Cabré et al.
1 3
north–south temperature and salinity asymmetries also
tend to have greater ocean stratification and precipitation
asymmetries, with larger SSTs, stronger stratification, and
higher rates of precipitation in the NH. Models with his-
torically higher north–south production differences also
exhibit higher north–south export differences. We did not,
however, find any robust connections between physics and
biogeochemistry in terms of inter-hemispheric asymmetry.
Throughout the twenty-first century, models that
undergo higher sea surface temperature increases also
undergo greater stratification increases. This relation holds
in the hemispheric averages (Fig. 9d). Also in the NH but
not the SH, models with greater rates of surface freshen-
ing stratify more (Fig. 9e), an effect most prominent in the
equatorial, subpolar, and sea-ice biomes. Similar NH-SH
asymmetric responses of SST to climate change have pre-
viously been observed in transient climate simulations,
as described in the introduction. Importantly, we find that
models which predict greater rates of warming also roughly
predict greater decreases in total production, especially in
Fig. 9 Inter-hemispheric asymmetry in ecological responses to cli-
mate. ad Historical (years 1980–1999) relations between variables
across CMIP5 models in the northern hemisphere (green) and south-
ern hemisphere (red) averages and eg 100-year changes in vari-
ables (difference between 2080–2099 and 1980–1999) plotted across
CMIP5 models. Legend numbers show the slopes in the correlations,
also plotted as solid lines when the correlation is significant at a level
>95 %. a Historical value plotted as a symbol and the consequent
100 year evolution towards higher SST and stratification
Consistent global responses of marine ecosystems
1 3
the SH, where a significant correlation exists across mod-
els. In the SH, for each degree of warming, production
decreases by about 6 % of its historical value (Fig. 9f). For
each model, relative decreases in production, biomass and
export are greater in the NH than in the SH, consistent with
stronger warming and stratification increases in the NH.
4 Discussion and conclusions
The 16 CMIP5 models analyzed here consistently predict
and confirm that over the twenty-first century, increases in
globally-averaged sea surface temperature combined with
surface freshening in the tropics stabilize the water column
and generally result in shallower mixed layer depths with
climate change, as summarized in Fig. 10b. These shal-
lower mixed layers lead to a reduced supply of nutrients
from the deep ocean, which in turn decreases phytoplank-
ton production, biomass, and ultimately export of organic
carbon to the deep ocean with potential positive feedbacks
to the climate and atmospheric CO2 concentrations. Macro-
nutrient limited low-latitude tropics and subtropics domi-
nate the global response of phytoplankton production to
climate change over the next 100 years across all the mod-
els (compare Fig. 10b with Fig. 10c).
These global results agree with a Coupled Model Inter-
comparison Project 3 (CMIP3) inter-comparison study
(Steinacher et al. 2010) that includes 4 models as well
as a recent study by Bopp et al. (2013) that includes 10
CMIP5 models. The GFDL models show some of the
smallest decreases in total production (0.5 PgC/year) and
their sensitivities to climate (defined here as ΔPP/ΔSST)
are the lowest (0.25 PgC/year per °C of warming for
GFDL-ESM2G and 0.28 PgC/year per °C of warming
Fig. 10 Multi-model averaged
response to climate change.
The different columns show
the multi-model average for:
absolute change in SST/10 (°C);
relative changes in stratifica-
tion, mixed layer depth (MLD),
surface iron (Fe), 100 m-depth-
averaged nitrate (NO3), top
100 m-integrated production,
POC export at 100 m, surface
phytoplankton biomass; the
absolute change in diatoms as
a fraction of total phytoplank-
ton biomass. The error bars
represent the standard deviation
among CMIP5 models. a
Relative difference between the
northern (green) and southern
(red) hemispheres, with respect
to the average of both hemi-
spheres, during the historical
period (years 1980–1999).
be Relative changes in these
variables averaged from years
1980–1999 to 2080–2099 for
different ocean regions, as
labeled. The regions are the
global ocean, the northern and
southern hemispheric means,
equatorial and low-latitude
upwelling (LLU) biome, north-
ern and southern subtropics
(includes both permanently and
seasonally stratified subtropics),
northern and southern subpolar,
and northern and southern
marginal sea-ice biomes. Note
that the y axis differs among
A. Cabré et al.
1 3
for GFDL-ESM2M), despite having the highest histori-
cal production rates (~61 PgC/year for GFDL-ESM2G
and ~83 PgC/year for GFDL-ESM2M) (Table 3). At the
other extreme, the model MPI-ESM-LR, with a historical
productivity of 58 PgC/year, exhibits the greatest decrease
in production (8.5 PgC/year) and the highest sensitivity
to increased warming, with a value of 3.4 PgC/year per
°C of warming. The predicted global decrease in produc-
tion across all CMIP5 models is in contrast with Schmittner
et al. (2008) and Sarmiento et al. (2004a), who saw
increases in projected global twenty-first century PP.
We divide the ocean into ecologically-relevant biomes
using upwelling velocities, sea-ice regimes, and maximum
mixed layer depths. We find that phytoplankton productiv-
ity responds to changes in physical mechanisms in a sur-
prisingly similar manner across models and biomes espe-
cially at low latitudes, indicating a common set of basic
modeled processes controlling productivity changes with
climate warming (Sect. 3.3). Throughout the twenty-first
century, models that undergo higher sea surface tempera-
ture increases also undergo greater stratification increases
(e.g. Figs. 3c, 4c, 5d, 6d), especially in already highly-strat-
ified regions, such as the low latitudes between 20°S and
20°N (Fig. 1b). Surface ocean freshening further enhances
stratification in the equatorial band and high latitudes (see
salinity patterns in Fig. S3a). This freshening is driven by
a larger increase in precipitation over evaporation in these
areas, in agreement with the “wet get wetter” theory (Held
and Soden 2006). Capotondi et al. (2012) detected simi-
lar spatial patterns in stratification increases (especially
severe in the tropics and NH) when they inter-compared the
CMIP3 models, despite using a milder scenario for projec-
tions (SRES-A2).
Models with higher increases in stratification experi-
ence greater reductions in mixed layer and thermocline
depths, and ultimately exhibit greater decreases in nutri-
ent concentrations and biological production across low
and mid-latitude regions (Figs. 3, 4, 5, 6). Biological pro-
duction between 40°S and 60°N is projected to decrease
most dramatically in zones of historically high production
(Fig. 1d) with spatial patterns of production change closely
following spatial patterns of nitrate (Fig. 1c) and stratifica-
tion change (Fig. 1b). The multi-model averaged decrease
in production is dominated by single nutrient (nitrate) mod-
and MRI-ESM1 (Fig. 1c). In the zonal average, produc-
tion decreases are maximal in the tropics with gradually
smaller decreases toward higher latitudes (Fig. 2b, Table 2).
Table 3 Global biological sensitivity to climate change across CMIP5 models
For each model, we list the globally-averaged change in sea surface temperature, the historical primary production (PP) integrated to 100 m
(years 1980–1999), the change in production over 100 years, the historical export of organic matter at 100 m, the change in export, the change
in diatom fraction (relative to the total phytoplankton biomass), the sensitivity of PP change to increased temperature, and the sensitivity of
export change to increased temperature. We also calculate the average and standard deviation of each of these variables across models (last row).
Changes always refer to differences between years (2080–2099) and (1980–1999)
Models ΔSST (°C) PP
Δ(diat/phyto) ΔPP/ΔSST
(PgC/year/ °C)
CanESM2 3.157 30.791 4.841 11.001 1.784 – 1.533 0.565
CESM1-BGC 2.484 54.722 1.803 8.105 0.771 – 0.726 0.31
GFDL-ESM2G 1.963 61.326 0.5 4.929 0.31 0.009 0.255 0.158
GFDL-ESM2M 1.909 82.848 0.543 7.892 0.587 0.007 0.284 0.307
HadGEM2-CC 3.063 35.173 5.01 5.635 0.751 0.004 1.636 0.245
HadGEM2-ES 3.262 34.76 5.061 5.59 0.764 0.005 1.552 0.234
MIROC-ESM 3.325 26.8 4.273 – – 1.285 –
3.519 26.857 4.653 – – 1.322 –
IPSL-CM5A-LR 3.435 35.14 3.337 7.295 1.296 0.971 0.377
IPSL-CM5A-MR 3.395 33.689 4.19 8.05 1.414 0.026 1.234 0.417
MPI-ESM-LR 2.492 58.114 8.572 8.462 1.297 – 3.44 0.521
MPI-ESM-MR 2.443 52.345 6.81 7.448 1.011 – 2.788 0.414
NorESM1-ME 2.206 40.076 3.266 8.004 0.679 – 1.481 0.308
MRI-ESM1 2.021 28.738 2.723 7.94 0.805 – 1.347 0.398
GISS-E2-H-CC 2.06 9.94 0.454 – 0.033 0.22 –
GISS-E2-R-CC 1.817 14.945 2.069 – 0.002 1.139 –
Average (SD) 2.507 (0.591) 39.569 (19.831) 3.096 (2.128) 5.759 (3.610) 0.683 (0.539) 0.005 (0.010) 1.175 (0.769) 0.262 (0.182)
Consistent global responses of marine ecosystems
1 3
Accordingly, most of the global decrease in biological pro-
duction occurs at low latitudes (Fig. 2f), although models
disagree on which specific biomes contribute the most.
Part of this disagreement stems from models with mul-
tiple nutrients, which predict more complex patterns of
biological production changes. For example, increased sup-
ply of iron in iron-limited areas such as the Eastern South
Pacific and North Pacific (Fig. S5) results in increased
production (Fig. S4) in models GFDL-ESM2M, GFDL-
ESM2G, IPSL-CM5A and CESM1-BGC. As extreme
cases, the two models GFDL-ESM2 predict that the global
decrease in production is driven by nitrate-driven produc-
tion decreases in the equatorial/LLU areas and NH per-
manently stratified subtropics (Fig. 2f), but that these
decreases are strongly compensated by iron-driven pro-
ductivity increases in the permanently-stratified subtropi-
cal East South Pacific and PP increases in the North Pacific
subpolar biome (Fig. S4 and Fig. S5). Note how the two
GFDL-ESM2 models agree with the rest of models on
regional contributions to the global decrease in POC export
as changes in PP are mostly due to small phytoplankton
populations, which contribute minimally to changes in
export (Fig. 2h).
Importantly, within iron-limited areas such as the
South Pacific and Southern Ocean, surface iron concen-
trations increase almost everywhere across most CMIP5
models throughout the twenty-first century (Fig. S6),
even though the continental input of dust is held clima-
tologically constant within all the models. One possible
cause of such increase is that climate warming-induced
stratification traps and concentrates aeolian iron in the
surface layers (Steinacher et al. 2010). However, accord-
ing to Misumi et al. (2014), most of the increase in iron
in the CESM1-BGC model over the twenty-first century
is due to changes in ocean circulation, a finding that
might also apply to the other models with explicit iron
In contrast to low latitudes, productivity at high lati-
tudes is predicted to increase with climate change due to
increases in light availability and increased supply of iron,
slightly compensating for the severe low-latitude losses in
PP (Fig. 10e). Here both the melting of sea-ice and the
shoaling of MLDs enhance light availability to phyto-
plankton, allowing productivity to increase across most
of the models despite cloud cover increases and nitrate
decreases in these same areas. Changes in iron and light
result in a banded productivity response in the Southern
Ocean with increases in the 40°S–50°S band and south
of 65°S and decreases in the 50°S–65°S band over the
twenty-first century. This complex ecological response is
discussed in detail in our follow-up paper (Leung et al.,
submitted). The southern subpolar biome defined here
includes two of these latitudinal bands with opposing PP
changes, canceling out to show almost zero change in pro-
duction (Fig. 10d).
Note that stratification increases far less in the South-
ern Ocean subpolar compared to the northern subpolar
biome (Fig. 10d), partly owing to the intensification of the
Southern westerly winds, which increases mixing of sur-
face layers and opposes temperature and freshening driven
stratification. It is thus important to accurately predict
the long-term evolution of midlatitude westerlies, which
depends on a competition between the rate of greenhouse
gas emissions and the rate at which the ozone hole closes
(e.g., Polvani et al. 2011; Simpkins and Karpechko 2012).
In the RCP8.5 scenario, large greenhouse gas increases
dominate, producing an ongoing poleward shift of the
westerly jets in all seasons, particularly in the SH for all
the models (Swart and Fyfe 2012; Eyring et al. 2013). This
shift is predicted to have profound impacts on productiv-
ity in the SH subpolar region (Leung et al., submitted).
In the Arctic, the situation is more straightforward and
PP increases in most models are due to melting of sea ice
and consequent relief of light limitation, though the mag-
nitudes of changes here depend on the various degrees of
increased nitrate limitation among models (Fig. 7h). Note
that the northern subpolar region behaves on average like
a subtropical biome, undergoing a decrease in nitrate, pro-
duction, export and phytoplankton biomass (Fig. 10d) due
to increased stratification (Fig. 6).
Variability among models in predicted 100-year changes
is very similar in magnitude for physical and biogeochemi-
cal or ecological variables (Fig. 10b), as most of the inter-
model differences in biogeochemical changes are driven by
inter-model differences in physical changes rather than in
intrinsic model details. We found consistent links between
changes in physics and changes in biogeochemistry across
biomes in the equatorial biomes, subtropical biomes, north-
ern subpolar biome, and southern sea-ice biome. However,
we didn’t find a clear consistent link between changes in
physics and changes in nutrients and biology in the low
latitude upwelling biome, southern subpolar biome, and
northern sea-ice biome. Improving and refining the defi-
nition of these biomes might help improve these links,
although part of the disagreement might come from strong
differences in responses across models. The largest disper-
sions in projected biological production changes are found
in the sea-ice biomes, mostly due to less well-constrained
parameterizations of physical processes here (Fig. 10e).
Thus, in order to improve model agreement, accuracy, and
prediction regarding current and future phytoplankton pro-
duction, more energy should be focused on improving sim-
ulations of ocean–atmosphere dynamics and biogeochemi-
cal parameters within the high latitudes.
During the 1980–1999 historical period, there is less dis-
crepancy among CMIP5 models over the values of physical
A. Cabré et al.
1 3
variables compared to biogeochemical ones (compare st.
dev. bars in Fig. 10a) because physical processes are much
better understood, constrained, and modeled than biologi-
cal ones. For example, the overall geographical patterns
of primary productivity in the historical run are similar
across models (Fig. S4) and roughly agree with satellite-
derived distributions (e.g., Behrenfeld and Falkowski
1997; Behrenfeld et al. 2006), although magnitudes are
quite different among models as noted in Table 3. His-
torically, global production (integrated to 100 m) ranges
from ~10 PgC/year for GISS-E2-H-CC to 83 PgC/year
for GFDL-ESM2M (compared to an estimated 50 PgC/
year from satellite observations), while global POC export
at 100 m ranges from 5 PgC/year for GFDL-ESM2G to
11 PgC/year for CanESM2 (Table 3).
By analyzing all of the available historical ensemble
runs (each with slightly different initial conditions) for
GFDL-ESM2M, we found that inter-ensemble dispersion
in biogeochemical variables in this model is significantly
less than inter-model dispersion for the same variables.
This implies that across-model disagreement is likely not
due to different initial conditions (within ensemble differ-
ences), but is instead determined by differences in intrin-
sic model code (e.g., different equations for phytoplankton
growth among models as well as inherent differences in the
embedded physical parameterizations). The fact that most
of the wide among-model dispersion in biogeochemical
variables is not due to internal variability alone highlights
the critical need for further improvements in CMIP5 eco-
logical modules in order to bring them closer into agree-
ment. It also confirms the idea that we can indeed take
identically forced runs of different models as independent
realizations of a future global climate.
Historically, phytoplankton production and biomass are
well correlated across biomes in models with just one type
of phytoplankton. This suggests bottom-up control of phy-
toplankton populations, although regression slopes between
biomass and productivity across biomes vary among mod-
els due to differences in parameterizations of top-down
controls (Fig. 8a). Primary production and POC export at
100 m are also well correlated across biomes within models
with one phytoplankton type (Fig. 8c), meaning that POC
export is a constant proportion of PP (i.e., the e-ratio is
fixed). By contrast, phytoplankton biomass and POC export
become especially low compared to PP in low-productiv-
ity biomes (e.g. subtropics) across models with multiple
PFTs because of large relative abundances of small species,
which experience large grazing pressure (Calbet and Lan-
dry 2004) and contribute minimally to export compared to
diatoms. See Fig. S6 for phytoplankton biomass and POC
export across CMIP5 models.
A climate-driven decrease (increase) in production is
accompanied by a decrease (increase) in grazing, which
acts to offset a decrease (increase) in phytoplankton bio-
mass associated with lower production (Fig. 8b). How-
ever, grazing increases due to a shift to small phytoplank-
ton across complex models counteracting the previous
effect. We find that with climate change the e-ratio does
not change much in simple models with only one phyto-
plankton type, as relative changes in PP and export are vir-
tually identical within these models (see dotted points in
Fig. 8d, on top of the 1–1 line). CanESM2, the model with
the highest historical export rates due to its high reminer-
alization rate, shows the maximum decrease in export over
100 years (1.8 PgC/year) along with the maximum sensi-
tivity of export to warming (0.6 PgC/year per °C). In
models with multiple PFTs, relative decreases in export are
greater than in PP, indicating a shift to lower e-ratios and
more oligotrophic conditions particularly in low- and mid-
latitude biomes, in accordance with a decrease in the rela-
tive abundance of diatoms. The decoupling of production
and export changes in models with multiple PFTs suggests
that the incorporation and refinement of PFTs in ecological
modules is essential in creating an accurate representation
of export, e-ratios, and biological pump efficiency.
Grazing is the dominant loss term for phytoplankton in
the real ocean (Banse 1994). However, the biogeochemi-
cal modules in the CMIP5 model suite use widely differ-
ent parameterizations for grazing, creating large differences
in the relative contributions of bottom-up versus top-down
factors in the control of phytoplankton blooms (Hashioka
et al. 2013). Additionally, the response of net primary pro-
duction to climate change depends on the assumptions
made about the temperature dependence of metabolic pro-
cesses (Taucher and Oschlies 2011). It is thought that het-
erotrophic processes such as bacterial degradation increase
more quickly with temperature than autotrophic phyto-
plankton growth (e.g., Riebesell et al. 2009). IPSL-CM5A
is the only model that reflects this by using a higher Q10
coefficient for heterotrophic versus autotrophic processes
(e.g. Table 1b, Bopp et al. 2013 and references in Table 1).
We expect that inclusion of such a temperature depend-
ence increases remineralization and PP rates differentially,
resulting in larger and more pronounced changes in organic
matter export at 100 m compared to changes in PP. Indeed,
we note that in each of the biomes studied, the IPSL-CM5A
models systematically show smaller relative decreases in
PP compared to decreases in export production with cli-
mate warming (Fig. 8d). Thus, within the two IPSL-CM5A
models, the variable autotrophic/heterorophic Q10 coeffi-
cients together with the strong relative decrease in diatom
fraction, result in the strongest decrease in e-ratios across
the model suite within most biomes. We therefore sug-
gest that ecological modules that take variable temperature
dependencies of different processes into account will likely
predict smaller global decreases (or even increases) in PP,
Consistent global responses of marine ecosystems
1 3
but stronger decreases in e-ratios. Assuming that inclusion
of temperature dependence improves model accuracy, this
would mean a less efficient transfer of biological carbon to
depth with climate change than currently predicted by most
CMIP5 models.
One projected consequence of climate change common
to all the models is a poleward migration of the geographi-
cal boundaries corresponding to the different biomes,
resulting in expansion of subtropical, oligotrophic biomes
and contraction of marginal sea-ice biomes (Fig. 2d) in
agreement with Sarmiento et al. (2004a). This projected
subtropical expansion agrees in principle with satellite
observations that show low chlorophyll, oligotrophic prov-
inces getting larger over the past few decades (Polovina
et al. 2008; Irwin and Oliver 2009).
Historical biogeochemical and ecological conditions are
subject to north–south asymmetries owing to interhemi-
spheric asymmetries in physical drivers. Historical SST and
stratification are higher and more intense in the NH, while
MLD is shallower; consequently, historical production and
export are also lower in the NH (Fig. 10a). Furthermore,
faster warming and greater P E changes in the north lead
to greater increases in stratification of the NH compared to
the SH with climate change. As a result, all of the models
consistently predict stronger decreases in biomass as well
as primary and export production in the NH compared to
the SH (taller green than red bars in Fig. 10b). These main
findings support the results and inter-hemispheric asym-
metry mechanisms recently discussed in the context of one
model only (the National Center for Atmospheric Research
Community Climate System Model, version 3.0, or NCAR
CCSM-3) by Marinov et al. (2013).
In this paper, the significance of predicted trends is com-
puted from multi-model mean fields using a bootstrap-
ping technique and weighting based on similarity between
models. Previously, Steinacher et al. (2010) used regional
weights based on comparison to observations to compute
the significance of multi-model mean change fields in PP
and export among CMIP4 models. More recently, Anav
et al. (2013) ranked the CMIP5 models based on their
performance relative to observed integrated production,
MLD, and SST. However, these rankings differ across vari-
ables and regions, making their application difficult for the
present study. We emphasize the need for further work to
improve the metrics used to compare simulations and the
robustness tests used to assess multi-model significance. We
expect models to converge ever closer in the future, such
that more sophisticated robustness tests and comparison
techniques will be needed to calculate reliable and accurate
errors. We suggest that re-sampling techniques such as the
bootstrapping technique used here, combined with robust-
ness algorithms (Knutti and Sedlacek 2013) will produce an
easy-to-interpret significance in trends for future studies.
Acknowledgments A. C. and I. M. acknowledge support by NASA
ROSES grant NNX13AC92G and a University of Pennsylvania
research foundation grant.
Anav A et al (2013) Evaluating the land and ocean components of the
global carbon cycle in the CMIP5 earth system models. J Clim
26:6801–6843. doi:10.1175/jcli-d-12-00417.1
Arora VK et al (2013) Carbon-concentration and carbon-climate feed-
backs in CMIP5 earth system models. J Clim 26:5289–5314.
Assmann KM, Bentsen M, Segschneider J, Heinze C (2010) An isop-
ycnic ocean carbon cycle model. Geosci Model Dev 3:143–167
Aumont O, Bopp L (2006) Globalizing results from ocean in situ iron
fertilization studies. Global Biogeochem Cycles 20:15. doi:10.1
Banse K (1994) Grazing and zooplankton production as key controls
of phytoplankton production in the open ocean. Oceanography
Beaulieu C, Henson SA, Sarmiento JL, Dunne JP, Doney SC, Rykac-
zewski RR, Bopp L (2013) Factors challenging our ability to
detect long-term trends in ocean chlorophyll. Biogeosciences
10:2711–2724. doi:10.5194/bg-10-2711-2013
Behrenfeld MJ (2010) Abandoning Sverdrup’s critical depth hypoth-
esis on phytoplankton blooms. Ecology 91(4):977–989
Behrenfeld MJ, Falkowski PG (1997) Photosynthetic rates derived
from satellite-based chlorophyll concentration. Limnol Ocean-
ogr 42:1–20
Behrenfeld MJ et al (2006) Climate-driven trends in contempo-
rary ocean productivity. Nature 444:752–755. doi:10.1038/
Behrenfeld MJ, Doney SC, Lima I, Boss ES, Siegel DA (2013)
Annual cycles of ecological disturbance and recovery underly-
ing the subarctic Atlantic spring plankton bloom. Global Bio-
geochem Cycles 27(2):526–540
Bernardello R, Marinov I, Palter JB, Sarmiento JL, Galbraith ED,
Slater RD (2014) Response of the ocean natural carbon stor-
age to projected twenty-first-century climate change. J Clim
27:2033–2053. doi:10.1175/JCLI-D-13-00343.1
Bopp et al (2001) Potential impact of climate change on marine
export production. Global Biogeochem Cycles 15(1):81–99. doi
Bopp L, Aumont O, Cadule P, Alvain S, Gehlen M (2005) Response
of diatoms distribution to global warming and potential implica-
tions: a global model study. Geophys Res Lett 32. doi:10.1029/
Bopp L et al (2013) Multiple stressors of ocean ecosystems in the
21st century: projections with CMIP5 models. Biogeosciences
10:6225–6245. doi:10.5194/bg-10-6225-2013
Boyce DG, Dowd M, Lewis MR, Worma B (2014) Estimating global
chlorophyll changes over the past century. Prog Oceanogr
122:163–173. doi:10.1016/j.pocean.2014.01.004
Bryan K, Manabe S, Spelman MJ (1988) Interhemispheric asym-
metry in the transient-response of a coupled ocean atmos-
phere model to a CO2 forcing. J Phys Oceanogr 18:851–867.
Calbet A, Landry MR (2004) Phytoplankton growth, microzooplank-
ton grazing, and carbon cycling in marine systems. Limnol
Oceanogr 49:51–57
Capotondi A, Alexander MA, Bond NA, Curchitser EN, Scott JD
(2012) Enhanced upper ocean stratification with climate change
in the CMIP3 models. J Geophys Res Oceans 117. doi:10.1029
A. Cabré et al.
1 3
Curran MAJ, van Ommen TD, Morgan VI, Phillips KL, Palmer AS
(2003) Ice core evidence for Antarctic sea ice decline since the
1950s. Science 302:1203–1206
de Boyer Montégut C, Madec G, Fischer AS, Lazar A, Iudicone D
(2004) Mixed layer depth over the global ocean: an examination
of profile data and a profile-based climatology. J Geophys Res
109:C12003. doi:10.1029/2004JC002378
Denman KL, Peña MA (1999) A coupled 1-D biological/physical
model of the northeast suarctic Pacific Ocean with iron limita-
tion. Deep Sea Res Part II 46:2877–2908
Doney SC (2006) Oceanography—plankton in a warmer world.
Nature 444:695–696. doi:10.1038/444695a
Dufresne JL, Quaas J, Boucher O, Denvil S, Fairhead L (2005) Con-
trasts in the effects on climate of anthropogenic sulfate aero-
sols between the 20th and the 21st century. Geophys Res Lett
32:L21703. doi:10.1029/2005GL023619
Dunne JP, Armstrong RA, Gnanadesikan A, Sarmiento JL (2005)
Empirical and mechanistic models for the particle export ratio.
Glob Biogeochem Cycles 19(4). doi:10.1029/2004GB002390
Dunne JP et al (2013) GFDL’s ESM2 global coupled climate-carbon
earth system models. Part II: carbon system formulation and
baseline simulation characteristics. J Clim 26:2247–2267.
Efron B, Tibshirani RJ (1993) An introduction to the bootstrap, 1993.
Chapman & Hall, New York
Eyring V et al (2013) Long-term ozone changes and associated cli-
mate impacts in CMIP5 simulations. J Geophys Res Atmos
118:5029–5060. doi:10.1002/jgrd.50316
Finkel ZV, Katz ME, Wright JD, Schofield OME, Falkowski PG
(2005) Climatically driven macroevolutionary patterns in the
size of marine diatoms over the cenozoic. Proc Natl Acad Sci
USA 102:8927–8932
Finkel ZV, Beardall J, Flynn KJ, Quigg A, Rees TAV, Raven JA
(2010) Phytoplankton in a changing world: cell size and ele-
mental stoichiometry. J Plankton Res 32:119–137
Flato GM, Boer GJ (2001) Warming asymmetry in climate change
simulations. Geophys Res Lett 28:195–198. doi:10.1029/200
Friedlingstein P et al (2006) Climate-carbon cycle feedback analy-
sis: results from the C4MIP model intercomparison. J Clim
19:3337–3353. doi:10.1175/JCLI3800.1
Friedlingstein P, Meinshausen M, Arora VK, Jones CD, Anav A, Lid-
dicoat SK, Knutti R (2014) Uncertainties in CMIP5 climate
projections due to carbon cycle feedbacks. J Clim 27:511–526.
Friedman AR, Hwang Y-T, Chiang JCH, Frierson DMW (2013)
Interhemispheric temperature asymmetry over the twenti-
eth century and in future projections. J Clim 26:5419–5433.
Frierson DMW, Hwang Y (2012) Extratropical influence on ITCZ
shifts in slab ocean simulations of global warming. J Clim
25:720–733. doi:10.1175/JCLI-D-11-00116.1
Frölicher TL, Joos F, Plattner GK, Steinacher M, Doney SC (2009)
Natural variability and anthropogenic trends in oceanic oxygen
in a coupled carbon cycle-climate model ensemble. Glob Bio-
geochem Cycles 23/GB1003
Fung IY, Doney SC, Lindsay K, John J (2005) Evolution of car-
bon sinks in a changing climate. Proc Natl Acad Sci USA
102(32):11201–11206. doi:10.1073/pnas.0504949102
Gregg WW (2008) Assimilation of SeaWiFS ocean chlorophyll
data into a three-dimensional global ocean model. J Mar Syst
69:205–225. doi:10.1016/j.jmarsys.2006.02.015
Gregg WW, Casey NW, McClain CR (2005) Recent trends in global
ocean chlorophyll. Geophys Res Lett 32. doi:10.1029/200
Hashioka T et al (2013) Phytoplankton competition during the spring
bloom in four plankton functional type models. Biogeosciences
10:6833–6850. doi:10.5194/bg-10-6833-2013
Held I, Soden B (2006) Robust responses of the hydrological cycle to
global warming. J Clim 19:5686–5699
Henson SA et al (2010) Detection of anthropogenic climate change in
satellite records of ocean chlorophyll and productivity. Biogeo-
sciences 7:621–640
Hutchinson DK, England MH, Santoso A, Hogg AM (2013) Inter-
hemispheric asymmetry in transient global warming: the
role of Drake Passage. Geophys Res Lett 40:1587–1593.
Ilyina T, Six KD, Segschneider J, Maier-Reimer E, Li HM, Nunez-
Riboni I (2013) Global ocean biogeochemistry model HAM-
OCC: model architecture and performance as component of
the MPI-Earth system model in different CMIP5 experimental
realizations. J Adv Model Earth Syst 5:287–315. doi:10.1029/
Irwin AJ, Oliver MJ (2009) Are ocean deserts getting larger? Geophys
Res Lett 36. doi:10.1029/2009gl039883
Jones C (2013) Twenty-first-century compatible CO2 emissions and
airborne fraction simulated by CMIP5 earth system models
under four representative concentration pathways. J Clim 26.
Kang SM, Frierson DMW, Held IM (2009) The tropical response to
extratropical thermal forcing in an idealized GCM: the impor-
tance of radiative feedbacks and convective parameterization. J
Atmos Sci 66. doi:10.1175/2009JAS2924.1
Knutti R, Sedlacek J (2013) Robustness and uncertainties in the new
CMIP5 climate model projections. Nat Clim Change 3:369–
373. doi:10.1038/nclimate1716
Knutti R, Furrer R, Tebaldi C, Cermak J, Meehl GA (2010) Chal-
lenges in combining projections from multiple climate models.
J Clim 23:2739–2758. doi:10.1175/2009jcli3361.1
Le Quéré C et al (2007) Saturation of the Southern Ocean CO2
sink due to recent climate change. Science 316:1735–1738.
Levitus S (1982) Climatological atlas of the world ocean. NOAA/ERL
GFDL Professional Paper 13, NTIS pB83-184093, Princeton,
173 pp
Levitus S, Antonov JI, Boyer TP, Stephens C (2000) Warming
of the world ocean. Science 287:2225–2229. doi:10.1126/
Levitus S, Antonov JI, Boyer TP, Locarnini RA, Garcia HE, Mis-
honov AV (2009) Global ocean heat content 1955–2008 in light
of recently revealed instrumentation problems. Geophys Res
Lett 36. doi:10.1029/2008gl037155
Longhurst A (1994) Ecological geography of the ocean. Academic
Press, San Diego 176 pp
Lovenduski NS, Ito T (2009) The future evolution of the Southern
Ocean CO2 sink. J Mar Res 67:597–617
Manabe S, Stouffer RJ (1980) Sensitivity of a global climate model to
an increase of CO2 concentration in the atmosphere. J Geophys
Res 85:5529–5554
Manabe S, Stouffer RJ, Spelman MJ, Bryan K (1991) Transient
responses of a coupled ocean atmosphere model to gradual
changes of atmospheric CO2. 1. Annual mean response. J Clim
4:785–818. doi:10.1175/1520-0442(1991)004
Marinov I, Follows M, Gnanadesikan A, Sarmiento JL, Slater RD (2008)
How does ocean biology affect atmospheric pCO2? Theory and
models. J Geophys Res Oceans 113. doi:10.1029/2007jc004598
Marinov I, Doney SC, Lima ID (2010) Response of ocean phytoplank-
ton community structure to climate change over the 21st cen-
tury: partitioning the effects of nutrients, temperature and light.
Biogeosciences 7:3941–3959. doi:10.5194/bg-7-3941-2010
Consistent global responses of marine ecosystems
1 3
Marinov I, Doney SC, Lima ID, Lindsay K, Moore JK, Mahowald N
(2013) North–south asymmetry in the modeled phytoplankton
community response to climate change over the 21st century.
Glob Biogeochem Cycles 27. doi:10.1002/2013GB004599
Martinez E, Antoine D, D’Ortenzio F, Gentili B (2009) Climate-
driven basin-scale decadal oscillations of oceanic phytoplank-
ton. Science 326:1253–1256. doi:10.1126/science.1177012
Martinez E, Antoine D, D’Ortenzio F, Montegut CD (2011) Phyto-
plankton spring and fall blooms in the North Atlantic in the
1980s and 2000s. J Geophys Res Oceans 116:11. doi:10.1029
Massonnet F, Fichefet T, Goosse H, Bitz CM, Philippon-Berthier G,
Holland MM, Barriat PY (2012) Constraining projections of
summer Arctic sea ice. Cryosphere 6:1383–1394. doi:10.5194
Meehl GA et al (2007) The WCRP CMIP3 multimodel dataset—
a new era in climate change research. Bull Am Meteorol Soc
Meijers AJS, Shuckburgh E, Bruneau N, Sallee JB, Bracegirdle TJ,
Wang Z (2012) Representation of the Antarctic circumpolar
current in the CMIP5 climate models and future changes under
warming scenarios. J Geophys Res Oceans 117. doi:10.1029/20
Misumi K, Lindsay K, Moore JK, Doney SC, Bryan FO, Tsumune D,
Yoshida Y (2014) The iron budget in ocean surface waters in
the 20th and 21st centuries: projections by the community earth
system model version 1. Biogeosciences 11:33–55. doi:10.5194
Moore JK, Doney SC, Lindsay K (2004) Upper ocean ecosystem
dynamics and iron cycling in a global three-dimensional model.
Glob Biogeochem Cycles 18. doi:10.1029/2004gb002220
Moore JK, Doney SC, Lindsay K, Mahowald N, Michaels AF (2006)
Nitrogen fixation amplifies the ocean biogeochemical response
to decadal timescale variations in mineral dust deposition. Tel-
lus B 58:560–572. doi:10.1111/j.1600-0889.2006.00209.x
Moore CM et al (2013) Processes and patterns of oceanic nutrient
limitation. Nat Geosci 6:701–710. doi:10.1038/ngeo1765
Nakićenović N, Swart R et al (2000) IPCC SRES special report on
emissions scenarios: a special report of Working Group III of
the Intergovernmental Panel on Climate Change. Cambridge
University Press. ISBN:0-521-80081-1, 978-052180081-5.
pb:0-521-80493-0, 978-052180493-6
Palmer JR, Totterdell IJ (2001) Production and export in a global
ocean ecosystem model. Deep Sea Res Part I Oceanogr Res Pap
48:1169–1198. doi:10.1016/s0967-0637(00)00080-7
Polovina JJ, Howell EA, Abecassis M (2008) Ocean’s least produc-
tive waters are expanding. Geophys Res Lett 35. doi:10.1029/
Polvani LM, Previdi M, Deser C (2011) Large cancellation, due to
ozone recovery, of future Southern Hemisphere atmospheric cir-
culation trends. Geophys Res Lett 38. doi:10.1029/2011gl046712
Popova EE et al (2012) What controls primary production in the Arc-
tic Ocean? Results from an intercomparison of five general cir-
culation models with biogeochemistry. J Geophys Res Oceans
117. doi:10.1029/2011jc007112
Riahi K et al (2011) RCP 8.5—a scenario of comparatively
high greenhouse gas emissions. Clim Change 109:33–57.
Riebesell U, Kortzinger A, Oschlies A (2009) Sensitivities of marine
carbon fluxes to ocean change. Proc Natl Acad Sci USA
106(49):20602–20609. doi:10.1073/pnas.0813291106
Russell JL, Stouffer RJ, Dixon KW (2006) Intercomparison of the
Southern Ocean circulations in IPCC coupled model control
simulations. J Clim 19:4560–4575. doi:10.1175/jcli3869.1
Sallee JB, Shuckburgh E, Bruneau N, Meijers AJS, Bracegirdle TJ,
Wang Z (2013) Assessment of Southern Ocean mixed-layer depth
in CMIP5 models: historical bias and forcing response. J Geo-
phys Res: Oceans 118(4):1845–1862. doi:10.1002/jgrc.20157
Sarmiento JL et al (2004a) Response of ocean ecosystems to cli-
mate warming. Glob Biogeochem Cycles 18. doi:10.1029/200
Sarmiento JL, Dunne J, Amstrong RA (2004b) Do we now understand
the ocean’s biological pump? US JGOFS News 12(4)
Schmittner A, Oschlies A, Matthews HD, Galbraith ED (2008) Future
changes in climate, ocean circulation, ecosystems, and biogeo-
chemical cycling simulated for a business-as-usual CO2 emis-
sion scenario until year 4000AD. Glob Biogeochem Cycles 22.
Seferian R et al (2013) Skill assessment of three earth system mod-
els with common marine biogeochemistry. Clim Dyn 40:2549–
2573. doi:10.1007/s00382-012-1362-8
Serreze MC, Holland MM, Stroeve J (2007) Perspectives on the
Arctic’s shrinking sea-ice cover. Science 315:1533–1536.
Simpkins GR, Karpechko AY (2012) Sensitivity of the southern
annular mode to greenhouse gas emission scenarios. Clim Dyn
38(3–4):563–572. doi:10.1007/s00382-011-1121-2
Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB,
Tignor M, Miller HL (2007) Contribution of Working Group I
to the fourth assessment report of the Intergovernmental Panel
on Climate Change. Cambridge University Press, Cambridge
Steinacher M et al (2010) Projected 21st century decrease in
marine productivity: a multi-model analysis. Biogeosciences
Stocker TF et al (2013) IPCC, 2013: Climate change 2013: the
physical science basis. Contribution of Working Group I to
the fifth assessment report of the Intergovernmental Panel on
Climate Change. Cambridge University Press, Cambridge,
1535 pp
Stroeve JC, Kattsov V, Barrett A, Serreze M, Pavlova T, Holland M,
Meier WN (2012) Trends in Arctic sea ice extent from CMIP5,
CMIP3 and observations. Geophys Res Lett 39:7. doi:10.1029/
Strong C, Davis RE (2007) Winter jet stream trends over the North-
ern Hemisphere Quarterly. J R Meteorol Soc 133:2109–2115.
Swart NC, Fyfe JC (2012) Observed and simulated changes in the
Southern Hemisphere surface westerly wind-stress. Geophys
Res Lett 39:L16711. doi:10.1029/2012GL052810
Taucher J, Oschlies A (2011) Can we predict the direction of marine
primary production change under global warming? Geophys
Res Lett 38. doi:10.1029/2010GL045934
Taylor KE, Stouffer RJ, Meehl GA (2012) An overview of CMIP5
and the experiment design. Bull Am Meteorol Soc 93:485–498.
Tebaldi C, Arblaster JM, Knutti R (2011) Mapping model agreement
on future climate projections. Geophys Res Lett 38. doi:10.102
Thompson DWJ, Solomon S (2002) Interpretation of recent Southern
Hemisphere climate change. Science 296:895–899. doi:10.1126/
Toggweiler JR (2009) Shifting westerlies. Science 323:1434–1435.
Turner J et al (2009) Non-annular atmospheric circulation change
induced by stratospheric ozone depletion and its role in the
recent increase of Antarctic sea ice extent. Geophys Res Lett
36. doi:10.1029/2009gl037524
Vancoppenolle M, Bopp L, Madec G, Dunne J, Ilyina T, Halloran
PR, Steiner N (2013) Future Arctic Ocean primary productiv-
ity from CMIP5 simulations: uncertain outcome, but consist-
ent mechanisms. Global Biogeochem Cycles 27:605–619.
A. Cabré et al.
1 3
Wang MY, Overland JE (2012) A sea ice free summer Arctic within
30 years: an update from CMIP5 models. Geophys Res Lett 39.
Watanabe S et al (2011) MIROC-ESM 2010: model description and
basic results of CMIP5-20c3 m experiments. Geosci Model Dev
4:845–872. doi:10.5194/gmd-4-845-2011
Xu Y, Ramanathan V (2012) Latitudinally asymmetric response of
global surface temperature: implications for regional climate
change. Geophys Res Lett 39. doi:10.1029/2012gl052116
Yoder JA, Kennelly MA, Doney SC, Lima ID (2010) Are trends
in SeaWiFS chlorophyll time-series unusual relative to his-
toric variability. Acta Oceanol Sin 29:1–4. doi:10.1007/
Yoshimori M, Broccoli AJ (2009) On the link between Hadley circu-
lation changes and radiative feedback processes. Geophys Res
Lett 36:L20703. doi:10.1029/2009GL040488
Yukimoto S et al (2011) Meteorological Research Institute-Earth Sys-
tem Model Version 1 (MRI-ESM1)—model description, mete-
orological reports of the Meteorological Research Institute, 64
Zahariev K, Christian JR, Denman KL (2008) Preindustrial, his-
torical, and fertilization simulations using a global ocean car-
bon model with new parameterizations of iron limitation,
calcification, and N-2 fixation. Prog Oceanogr 77:56–82.
Zunz V, Goosse H, Massonnet F (2013) How does internal variability
influence the ability of CMIP5 models to reproduce the recent
trend in Southern Ocean sea ice extent? Cryosphere 7:451–468.
Zwally HJ, Comiso JC, Parkinson CL, Cavalieri DJ, Gloersen P
(2002) Variability of Antarctic sea ice 1979–1998. J Geophys
Res Oceans 107. doi:10.1029/2000jc000733
... POC accounts for the majority of the biological carbon pump. The magnitude of the POC export flux out of the euphotic zone has been estimated to be between 8-24% of the annual NPP rate (Laws et al., 2000;Schlitzer, 2002;Dunne et al., 2005;Henson et al., 2011;Siegel et al., 2014;Cabréet al., 2015;DeVries and Weber, 2017). Importantly, this POC flux decreases with depth as particles fragment, decompose, and remineralize through microbial respiration (e.g., Martin et al., 1987). ...
... -12 GtC yr -1 (The magnitude of the POC export flux out of the euphotic zone 1 ) 2.31 ± 0.60 GtC yr -1 (DOC export flux at 74m depth; (Roshan and DeVries, 2017) 12% POC export decrease by 2100* (Cabréet al., 2015) DOC export projected to become more relevant in oligotrophic areas (Roshan and DeVries, 2017) Phytoplankton >1 1 GtC ( Bar-On et al., 2018) 50 GtC yr -1 (Net primary production; (Carr et al., 2006) 8% primary production (PP) decrease by 2100 driven by low-latitude decreases, despite a PP increase in polar areas* (Cabréet al., 2015) ...
... Anthropogenic changes include both historical (^) and projected (*) as indicated. Laws et al., 2000;Schlitzer, 2002;Dunne et al., 2005;Henson et al., 2011;Siegel et al., 2014;Cabréet al., 2015;DeVries and Weber, 2017. The most recent estimate puts this rate of export of POC out of the euphotic layer at 9.1 ± 0.2 GtC yr -1 (DeVries and Weber, 2017). ...
Full-text available
The ocean is gaining prominence in climate change policy circles as a tool for addressing the climate crisis. Blue carbon, the carbon captured and stored by marine and coastal ecosystems and species, offers potential as a “nature-based solution” to climate change. The protection and restoration of specific ocean ecosystems can form part of a climate response within climate mitigation policies such as Nationally Determined Contributions under the United Nations Framework Convention on Climate Change. For mitigation policies that seek to implement management actions that drawdown carbon, ecosystem sequestration and emissions must be measurable across temporal and spatial scales, and management must be practical leading to improved sequestration and avoided emissions. However, some blue carbon interventions may not be suitable as a climate mitigation response and better suited for other policy instruments such as those targeted toward biodiversity conservation. This paper gives context to numerous blue carbon sequestration pathways, quantifying their potential to sequester carbon from the atmosphere, and comparing these sequestration pathways to point-source emissions reductions. The applicability of blue carbon is then discussed in terms of multiple international policy frameworks, to help individuals and institutions utilize the appropriate framework to reach ocean conservation and climate mitigation goals.
... These blooms are driven by the changes in the physical forcing primarily associated with the southwest (summer) and northeast (winter) monsoon [3][4][5][6] . The phytoplankton blooms regulate the food availability for higher trophic levels, making primary production central to both the aquatic food web and the Indian Ocean rim population that is dependent on marine fisheries for their livelihood [7][8][9][10][11] . Recent decades have observed warming of the earth's climate unequivocally, with the oceans accounting for approximately 93% of this increased energy uptake 12,13 . ...
Full-text available
Continuous remote-sensed daily fields of ocean color now span over two decades; however, it still remains a challenge to examine the ocean ecosystem processes, e.g., phenology, at temporal frequencies of less than a month. This is due to the presence of significantly large gaps in satellite data caused by clouds, sun-glint, and hardware failure; thus, making gap-filling a prerequisite. Commonly used techniques of gap-filling are limited to single value imputation, thus ignoring the error estimates. Though convenient for datasets with fewer missing pixels, these techniques introduce potential biases in datasets having a higher percentage of gaps, such as in the tropical Indian Ocean during the summer monsoon, the satellite coverage is reduced up to 40% due to the seasonally varying cloud cover. In this study, we fill the missing values in the tropical Indian Ocean with a set of plausible values (here, 10,000) using the classical Monte-Carlo method and prepare 10,000 gap-filled datasets of ocean color. Using the Monte-Carlo method for gap-filling provides the advantage to estimate the phenological indicators with an uncertainty range, to indicate the likelihood of estimates. Quantification of uncertainty arising due to missing values is critical to address the importance of underlying datasets and hence, motivating future observations.
... Earth system modelling efforts have strived to include as many multiple stressors as possible, with the physiological effects of nutrient limitation, deoxygenation, and warming increasingly represented in the biogeochemical models used in e.g. Coupled Model Intercomparison Project (CMIP) activities [42][43][44][45][46], although ocean acidification effects remain underrepresented. It was recently noted that with the most recent CMIP6 models, climate sensitivity has increased (resulting in stronger changes to ocean biogeochemistry with forcing), but that ocean biological carbon cycle indicators such as primary and export fluxes have become less sensitive [43]. ...
Plastic pollution can both chemically and physically impede marine biota. But it can also provide novel substrates for colonization, and its leachate might stimulate phytoplankton growth. Plastic contains carbon, which is released into the environment upon breakdown. All of these mechanisms have been proposed to contribute global impacts on open ocean carbon cycling and climate from ubiquitous plastic pollution. Laboratory studies produce compelling data showing both stimulation and inhibition of primary producers and disruption of predatory lifecycles at individual scale, but global carbon cycle impacts remain mostly unquantified. Preliminary modelling estimates ecosystem alterations and direct carbon release due to plastic pollution will remain vastly less disruptive to global carbon cycling than the direct damage wrought by fossil fuel carbon emissions. But when considered by mass, carbon in the form of bulky, persistent plastic particles may be disproportionally more influential on biogeochemical cycling than carbon as a gas in the atmosphere or as a dissolved component of seawater. Thus, future research should pay particular attention to the optical and other physical effects of marine plastic pollution on Earth system and ecological function, and resulting impacts on oxygen and nutrient cycling. Improved understanding of the breakdown of plastics in the marine environment should also be considered high-priority, as any potential perturbation of biological carbon cycling by plastic pollution is climate-relevant on centennial timescales and longer.
... For instance, NPP drives the vitality of marine ecosystems, biogeochemical cycling and the biological carbon pump. Several modelling studies have used Earth system models (ESMs) to project the evolution of marine NPP over the 21st century under different global warming scenarios (Bopp et al., 2001;Steinacher et al., 2010;Cabré et al., 2015;Laufkötter et al., 2015;Kwiatkowski et al., 2020;Tagliabue et al., 2021). Many of these studies have suggested decreases in global NPP in response to future climate change. ...
Full-text available
The impact of anthropogenic climate change on marine net primary production (NPP) is a reason for concern because changing NPP will have widespread consequences for marine ecosystems and their associated services. Projections by the current generation of Earth system models have suggested decreases in global NPP in response to future climate change, albeit with very large uncertainties. Here, we make use of two versions of the Institut Pierre-Simon Laplace Climate Model (IPSL-CM) that simulate divergent NPP responses to similar high-emission scenarios in the 21st century and identify nitrogen fixation as the main driver of these divergent NPP responses. Differences in the way N fixation is parameterised in the marine biogeochemical component PISCES (Pelagic Interactions Scheme for Carbon and Ecosystem Studies) of the IPSL-CM versions lead to N-fixation rates that are either stable or double over the course of the 21st century, resulting in decreasing or increasing global NPP, respectively. An evaluation of these two model versions does not help constrain future NPP projection uncertainties. However, the use of a more comprehensive version of PISCES, with variable nitrogen-to-phosphorus ratios as well as a revised parameterisation of the temperature sensitivity of N fixation, suggests only moderate changes in globally averaged N fixation in the 21st century. This leads to decreasing global NPP, in line with the model-mean changes of a recent multi-model intercomparison. Lastly, despite contrasting trends in NPP, all our model versions simulate similar and significant reductions in planktonic biomass. This suggests that projected plankton biomass may be a more robust indicator than NPP of the potential impact of anthropogenic climate change on marine ecosystems across models.
... important link in the microbial loop and trophic web of the present BC in the SBB (Moser et al., 2016). In a future warmer scenario, studies using climate and niche models indicate an expansion of oligotrophic regions and replacing larger phytoplankton with smaller organisms (Cabréet al., 2015;Sǒlićet al., 2018). The picoplankton will probably play an even more important role in the carbon cycle in these situations. ...
Full-text available
Despite the increased number of paleoceanographic studies in the SW Atlantic in recent years, the mechanisms controlling marine productivity and terrestrial material delivery to the South Brazil Bight remain unresolved. Because of its wide continental shelf and abrupt change in coastline orientation, this region is under the influence of several environmental forcings, causing the region to have large variability in primary production. This study investigated terrestrial organic matter (OM) sources and marine OM sources in the South Brazil Bight, as well as the main controls on marine productivity and terrestrial OM export. We analyzed OM geochemical (bulk and molecular) proxies in sediment samples from a core (NAP 63-1) retrieved from the SW Atlantic slope (24.8°S, 44.3°W, 840-m water depth). The organic proxies were classified into "terrestrial-source" and "marine-source" groups based on a cluster analysis. The two sources presented different stratigraphical profiles, indicating distinct mechanisms governing their delivery. Bulk proxies indicate the predominance of marine OM, although terrestrial input also affected the total OM deposition. The highest marine productivity, observed between 50 and 39 ka BP, was driven by the combined effects of the South Atlantic Central Water upwelling promoted by Brazil Current eddies and fluvial nutrient inputs from the adjacent coast. After the last deglaciation, decreased phytoplankton productivity and increased archaeal productivity suggest a stronger oligotrophic tropical water presence. The highest terrestrial OM accumulation occurred between 30 and 20 ka BP, with its temporal evolution controlled mainly by continental moisture evolution. Sea level fluctuations affected the distance between the coastline and the sampling site. In contrast, continental moisture affected the phytogeography, changing from lowlands covered by grasses and Frontiers in Marine Science saltmarshes to a landscape dominated by mangroves and the Atlantic Forest. Our results suggest how the OM cycle in the South Brazil Bight may respond to warmer and dryer climate conditions.
... A more stable water column effectively inhibits mixing, compromising vertical exchanges and decreasing the input of nutrient in the euphotic layer. This has been consistently shown in earth system model projections (Bopp et al., 2001;Cabré et al., 2015;Fu et al., 2016). The magnitude of these changes remains uncertain (Bopp et al., 2013;Krumhardt et al., 2017;Kwiatkowski et al., 2020) but, in most cases, will induce a reduction of primary productivity, particularly at low latitudes (Steinacher et al., 2010;Bopp et al., 2013;Moore et al., 2018). ...
Full-text available
Increased Net Primary Productivity (NPP) around small islands have been documented worldwide. Despite having been known for decades, the interactions between physical and biogeochemical processes behind this phenomenon – that takes the name of Island Mass Effect –remain unclear. In this paper we review the physical processes involved while proposing a method to identify the prevailing mechanisms by analyzing their imprint on NPP and Sea Surface Temperature (SST). These mechanisms can be quite different, but all enhance vertical exchanges, increasing the input of nutrients in the euphotic layer and favoring biological productivity. Nutrient-rich deeper waters are brought up to the surface through upwelling and mixing, leaving a cold imprint on the SST as well. Here we analyze satellite data of SST and NPP around small islands and archipelagos to catalog the physical mechanisms that favor the Island Mass Effect, with the aid of oceanic and atmospheric reanalysis. The multiplicity of these processes and the convolution of their interactions highlight the complexity of the physical forcing on the biomass production and the uniqueness of each island. However, analysis from 19 small islands throughout the tropics shows that two kinds of SST patterns emerge, depending on the size and altitude of the island. Around islands with considerable elevation and greatest diameters, cold/warm anomalies, most likely corresponding to upwelling/downwelling zones, emerge. This signal can be mainly ascribed to oceanic and atmospheric forcing. Around small islands, on the other hand, warm anomalies do not appear and only local cooling, associated with current-island interactions, is found. In the vicinity of a single island, more than one process responsible for the increased nutrient input into the euphotic layer might coexist, the prevailing one varying along the year and depending on the strength and direction of the incoming atmospheric and oceanic flow.
... The biological carbon pump is a key mechanism driving the carbon and nutrient cycles at higher latitudes. How primary productivity (e.g., Leung et al., 2015) and carbon export (Cabré et al., 2015;Moore et al., 2018) will vary spatially and seasonally remains unclear, especially in ice-covered areas; this uncertainty is further complicated by knowledge gaps related to the cycling of nutrients within the seasonally varying mixed layer (e.g., Fourquez et al., 2020;Mdutyana et al., 2020) and potential shifts in phytoplankton dynamics (Deppeler and Davidson, 2017). The microbial carbon pump also contributes to carbon sequestration and food-web fluxes (Jiao et al., 2010), yet the response of nutrient recycling and the MCP to Southern Ocean warming and acidification is uncertain, as is how interactions between the biological carbon pump and microbial carbon pumps are likely to change (Jiao et al., 2010;Legendre et al., 2015). ...
Technical Report
Full-text available
The Southern Ocean plays a central role in the Earth System by connecting the Earth’s ocean basins, and it is a crucial link between the deep ocean, surface ocean and atmosphere. Hence, the ongoing changes in the Southern Ocean impact global climate, rates of sea level rise, biogeochemical cycles and ecological systems. Yet, understanding of the causes and consequences of these changes is limited by the short and incomplete nature of observations. To address this issue, sustained, integrated and multidisciplinary observations are needed. Due to the size of the Southern Ocean, this requires international agreement on the priority observations to be collected, and also internationally coordinated data management and delivery. The Southern Ocean Observing System (SOOS) was initiated in 2011 to support these efforts. In the last decade, SOOS has enhanced regional coordination and observing system capabilities through network development, data curation and publication, development of data discovery and coordination tools, and providing strong advocacy mechanisms for the Southern Ocean community. Significant data gaps remain in observations of the ice-affected ocean, sea ice habitats, the ocean at depths >2000 m, the air-ocean-ice interface, biogeochemical and biological variables, and for seasons other than summer. This Science and Implementation Plan articulates the scientific priorities for SOOS through the identification of these key gaps in the observational network and by identifying the priorities in addressing these gaps. This Plan covers the five year period 2021-2025, with emphasis on the capabilities required to support data collection and delivery, and the objectives and actions that SOOS will implement. Five Science Themes have been identified, each encompassing a number of Key Science Challenges. These Themes and Challenges incorporate many scientific drivers that are cross-disciplinary, reflecting the highly-interconnected nature of the Southern Ocean, and Theme 5 is cross-cutting and highlights a number of linkages amongst Themes 1-4. The Themes provide a framework for enhancing the coordination of international data collection and delivery efforts that will contribute to understanding and quantifying the state and variability of: Theme 1: Southern Ocean cryosphere Theme 2: Southern Ocean circulation Theme 3: Southern Ocean carbon and biogeochemical cycles Theme 4: Southern Ocean ecosystems and biodiversity Theme 5: Southern Ocean-sea ice-atmosphere fluxes Addressing the data gaps across these inherently interconnected Themes sustainably and systematically requires parallel advances in coordination networks, cyberinfrastructure and data management tools, observational platform and sensor technology, and development of internationally agreed sampling and analytical standards and data requirements of key variables. In recognition of this, SOOS has also identified a number of Foundational Capabilities that will need to be developed or expanded.
... Combined, they capture the tendency of each model to vary with the observations and the magnitude of the discrepancies between them (Stow et al., 2009). The analysis here complements prior studies to provide model assessment for the tropical Indian Ocean (as in Allen et al., 2007;Stow et al., 2009;Bopp et al., 2013;Ilyina et al., 2013;Séférian et al., 2013;Cabré et al., 2015;Rickard and Behrens, 2016;Bao and Li, 2016;Mohan and Bhaskaran, 2020;Jacobs et al., 2021). Table 1 lists the r and RMSE for each model. ...
Full-text available
Marine ecosystems are expected to be increasingly affected by climate change, impacting their physical and biogeochemical environment. Changes in primary production, temperatures and hence species distribution, may lead to critical consequences for fishery exploitation. Therefore, future projections are essential to develop sustainable strategies and climate change adaptation plans for fisheries, and fishery-dependent societies. In this study, we focus on the Agulhas Bank, a broad extension of the continental shelf of the South African coast, along which flows the western boundary Agulhas Current. The Agulhas Bank is known for being biologically productive and is an important nursery ground for many commercially exploited fish species, including the chokka squid fishery, a vital source of income for many people in the Eastern Cape Province. Squid catches manifest strong interannual fluctuations, at times causing fishery crashes. Additional impacts due to climate change will have significant socio-economic consequences for this all-important fishery. To investigate future variations of the physical and biogeochemical environment on the Agulhas Bank, we used the global ocean model NEMO-MEDUSA, forced by the high emissions scenario RCP8.5. Our simulations show a significant increase in sea surface temperature and bottom temperature, but limited changes in primary production. Projections highlight an increase in current velocity on the Agulhas Bank throughout the course of this century, induced by an onshore shift of the Agulhas current. This current shift may pose a threat to squid recruitment success as a large fraction of squid paralarvae may be removed from their shelf feeding grounds and lost to the greater ocean via the Agulhas current. The results further show that planktonic food for the paralarvae is less likely to become the main limiting factor in the future, while increasing temperatures may affect growth rates and spawning success.
Full-text available
Picocyanobacteria (< 2 µm in diameter) are significant contributors to total phytoplankton biomass. Due to the high diversity within this group, their seasonal dynamics and relationship with environmental parameters, especially in brackish waters, are largely unknown. In this study, the abundance and community composition of phycoerythrin rich picocyanobacteria (PE-SYN) and phycocyanin rich picocyanobacteria (PC-SYN) were monitored at a coastal (K-station) and at an offshore station (LMO; ~ 10 km from land) in the Baltic Sea over three years (2018–2020). Cell abundances of picocyanobacteria correlated positively to temperature and negatively to nitrate (NO3) concentration. While PE-SYN abundance correlated to the presence of nitrogen fixers, PC-SYN abundance was linked to stratification/shallow waters. The picocyanobacterial targeted amplicon sequencing revealed an unprecedented diversity of 2169 picocyanobacterial amplicons sequence variants (ASVs). A unique assemblage of distinct picocyanobacterial clades across seasons was identified. Clade A/B dominated the picocyanobacterial community, except during summer when low NO3, high phosphate (PO4) concentrations and warm temperatures promoted S5.2 dominance. This study, providing multiyear data, links picocyanobacterial populations to environmental parameters. The difference in the response of the two functional groups and clades underscore the need for further high-resolution studies to understand their role in the ecosystem.
Full-text available
Marine net primary production (NPP) is remarkably high given the typical vertical separation of 50–150 m between the depth zones of light and nutrient sufficiency, respectively. Here we present evidence that many autotrophs bridge this gap through downward and upward migration, thereby facilitating biological nutrient pumping and high rates of oceanic NPP. Our model suggests that phytoplankton vertical migration (PVM) fuels up to 40% (>28 tg yr⁻¹ N) of new production and directly contributes 25% of total oceanic NPP (herein estimated at 56 PgC yr⁻¹). Confidence in these estimates is supported by good reproduction of seasonal, vertical and geographic variations in NPP. In contrast to common predictions, a sensitivity study of the PVM model indicates higher NPP under global warming when enhanced stratification reduces physical nutrient transport into the surface ocean. Our findings suggest that PVM is a key mechanism driving marine biogeochemistry and therefore requires consideration in global carbon budgets.
Full-text available
The carbon cycle is a major forcing component in the global climate system. Modelling studies, aiming to explain recent and past climatic changes and to project future ones, increasingly include the interaction between the physical and biogeochemical systems. Their ocean components are generally z-coordinate models that are conceptually easy to use but that employ a vertical coordinate that is alien to the real ocean structure. Here, we present first results from a newly-developed isopycnic carbon cycle model and demonstrate the viability of using an isopycnic physical component for this purpose. As expected, the model represents well the interior ocean transport of biogeochemical tracers and produces realistic tracer distributions. Difficulties in employing a purely isopycnic coordinate lie mainly in the treatment of the surface boundary layer which is often represented by a bulk mixed layer. The most significant adjustments of the ocean biogeochemistry model HAMOCC, for use with an isopycnic coordinate, were in the representation of upper ocean biological production. We present a series of sensitivity studies exploring the effect of changes in biogeochemical and physical processes on export production and nutrient distribution. Apart from giving us pointers for further model development, they highlight the importance of preformed nutrient distributions in the Southern Ocean for global nutrient distributions. The sensitivity studies show that iron limitation for biological particle production, the treatment of light penetration for biological production, and the role of diapycnal mixing result in significant changes of nutrient distributions and liniting factors of biological production.
Full-text available
We investigated the simulated iron budget in ocean surface waters in the 1990s and 2090s using the Community Earth System Model version 1 and the Representative Concentration Pathway 8.5 future CO2 emission scenario. We assumed that exogenous iron inputs did not change during the whole simulation period; thus, iron budget changes were attributed solely to changes in ocean circulation and mixing in response to projected global warming, and the resulting impacts on marine biogeochemistry. The model simulated the major features of ocean circulation and dissolved iron distribution for the present climate. Detailed iron budget analysis revealed that roughly 70% of the iron supplied to surface waters in high-nutrient, low-chlorophyll (HNLC) regions is contributed by ocean circulation and mixing processes, but the dominant supply mechanism differed by region: upwelling in the eastern equatorial Pacific and vertical mixing in the Southern Ocean. For the 2090s, our model projected an increased iron supply to HNLC waters, even though enhanced stratification was predicted to reduce iron entrainment from deeper waters. This unexpected result is attributed largely to changes in gyre-scale circulations that intensified the advective supply of iron to HNLC waters. The simulated primary and export production in the 2090s decreased globally by 6 and 13%, respectively, whereas in the HNLC regions, they increased by 11 and 6%, respectively. Roughly half of the elevated production could be attributed to the intensified iron supply. The projected ocean circulation and mixing changes are consistent with recent observations of responses to the warming climate and with other Coupled Model Intercomparison Project model projections. We conclude that future ocean circulation has the potential to increase iron supply to HNLC waters and will potentially buffer future reductions in ocean productivity.
Full-text available
A global three-dimensional marine ecosystem model with several key phytoplankton functional groups, multiple limiting nutrients, explicit iron cycling, and a mineral ballast/organic matter parameterization is run within a global ocean circulation model. The coupled biogeochemistry/ ecosystem/circulation (BEC) model reproduces known basin-scale patterns of primary and export production, biogenic silica production, calcification, chlorophyll, macronutrient and dissolved iron concentrations. The model captures observed high nitrate, low chlorophyll (HNLC) conditions in the Southern Ocean, subarctic and equatorial Pacific. Spatial distributions of nitrogen fixation are in general agreement with field data, with total N-fixation of 55 Tg N. Diazotrophs directly account for a small fraction of primary production (0.5%) but indirectly support 10% of primary production and 8% of sinking particulate organic carbon (POC) export. Diatoms disproportionately contribute to export of POC out of surface waters, but CaCO3 from the coccolithophores is the key driver of POC flux to the deep ocean in the model. An iron source from shallow ocean sediments is found critical in preventing iron limitation in shelf regions, most notably in the Arctic Ocean, but has a relatively localized impact. In contrast, global-scale primary production, export production, and nitrogen fixation are all sensitive to variations in atmospheric mineral dust inputs. The residence time for dissolved iron in the upper ocean is estimated to be a few years to a decade. Most of the iron utilized by phytoplankton is from subsurface sources supplied by mixing, entrainment, and ocean circulation. However, owing to the short residence time of iron in the upper ocean, this subsurface iron pool is critically dependent on continual replenishment from atmospheric dust deposition and, to a lesser extent, lateral transport from shelf regions.
Full-text available
The Meteorological Research Institute (MRI) of Japan developed the Earth System Model MRI�ESM1 to enable us to simulate both the climate system and global material transport, along with their interaction. Its core component, the atmosphere–ocean coupled global climate model MRI�CGCM3, represents a substantial advance from the previous model, MRI�CGCM2.3, which made important contributions to the fourth assessment report of the Intergovernmental Panel on Climate Change. The global atmospheric model MRI�AGCM3, used as the atmospheric component of MRI�CGCM3, incorporates various new physical parameterizations, including a cumulus convection scheme, a high�accuracy radiation scheme, a two�moment bulk cloud model that explicitly represents aerosol effects on clouds, and a new, sophisticated land�surface model, into the dynamics framework by a conservative semi�Lagrange method. MRI.COM3, also newly developed at MRI, is used for the global ocean�ice component of MRI�CGCM3. We adopted for MRI.COM3 a tripolar grid coordinate system, in which the North Pole is not a singular point, because MRI.COM3 supports general orthogonal curvilinear coordinates. The sea�ice model has also been updated; it now represents the sub�grid ice�thickness distribution by thickness categories, and incorporates ice rheology dynamics in addition to detailed thermodynamics. The MASINGAR mk�2 aerosol model takes into account five kinds of atmospheric aerosols, sulfate, black and organic carbon, mineral dust, and sea salt. The MRI�CCM2 atmospheric chemistry climate model (ozone model) is used to treat chemical reactions and the transport of atmospheric species associated with both tropospheric and stratospheric ozone. To represent the global carbon cycle, terrestrial ecosystem carbon cycle and ocean biogeochemical carbon cycle processes are incorporated into the land�surface model and the ocean model, respectively. The Scup coupler developed at MRI is used to integrate each component model, the atmospheric, ocean, aerosol, and ozone models, into MRI�ESM1. This flexible coupler can couple models with different resolutions and grid coordinates with variable coupling intervals. This advantage not only leads to efficient execution of the earth system model but also allows the efficient and independent development of the component models.
As a part of Arctic Ocean Intercomparison Project, results from five coupled physical and biological ocean models were compared for the Arctic domain, defined here as north of 66.6°N. The global and regional (Arctic Ocean (AO)–only) models included in the intercomparison show similar features in terms of the distribution of present-day water column–integrated primary production and are broadly in agreement with in situ and satellite-derived data. However, the physical factors controlling this distribution differ between the models. The intercomparison between models finds substantial variation in the depth of winter mixing, one of the main mechanisms supplying inorganic nutrients over the majority of the AO. Although all models manifest similar level of light limitation owing to general agreement on the ice distribution, the amount of nutrients available for plankton utilization is different between models. Thus the participating models disagree on a fundamental question: which factor, light or nutrients, controls present-day Arctic productivity. These differences between models may not be detrimental in determining present-day AO primary production since both light and nutrient limitation are tightly coupled to the presence of sea ice. Essentially, as long as at least one of the two limiting factors is reproduced correctly, simulated total primary production will be close to that observed. However, if the retreat of Arctic sea ice continues into the future as expected, a decoupling between sea ice and nutrient limitation will occur, and the predictive capabilities of the models may potentially diminish unless more effort is spent on verifying the mechanisms of nutrient supply. Our study once again emphasizes the importance of a realistic representation of ocean physics, in particular vertical mixing, as a necessary foundation for ecosystem modeling and predictions.
We investigated the mechanisms of phytoplankton competition during the spring bloom, one of the most dramatic seasonal events in lower-trophic level ecosystems, in four state-of-the-art Plankton Functional Type (PFTs) models: PISCES, NEMURO, PlankTOM5 and CCSM-BEC. In particular, we investigated the relative importance of different ecophysiological processes on the determination of the community structure, focusing both on the bottom-up and the top-down controls. The models reasonably reproduced the observed global distribution and seasonal variation of phytoplankton biomass. The fraction of diatoms with respect to the total phytoplankton biomass increases with the magnitude of the spring bloom in all models. However, the governing mechanisms differ between models, despite the fact that current PFT models represent ecophysiological processes using the same types of parameterizations. The increasing trend in the percentage of diatoms with increasing bloom magnitude is mainly caused by a stronger nutrient dependence of photosynthesis for diatoms compared to nanophytoplankton (bottom-up control). The difference in the maximum photosynthesis rate plays an important role in NEMURO and PlankTOM5 and determines the absolute values of the percentage of diatoms during the bloom. In CCSM-BEC, the light dependency of photosynthesis plays an important role in the North Atlantic and the Southern Ocean. The grazing pressure by zooplankton (top-down control), however, strongly contributes to the dominance of diatoms in PISCES and CCSM-BEC. The regional differences in the percentage of diatoms in PlankTOM5 are mainly determined by top-down control. These differences in the mechanisms suggest that the response of marine ecosystems to climate change could significantly differ among models, even if the present-day ecosystem is reproduced to a similar degree of confidence. For further understanding of plankton competition and for the prediction of future change in marine ecosystems, it is important to understand the relative differences in each physiological rate and life history rate in the bottom-up and the top-down controls between PFTs.
We investigate the impact of century-scale climate changes on the Southern Ocean CO2 sink using an idealized ocean general circulation and biogeochemical model. The simulations are executed under both constant and changing wind stress, freshwater fluxes, and atmospheric pCO(2), so as to separately analyze changes in natural and anthropogenic CO2 fluxes under increasing wind stress and stratification. We find that the Southern Ocean sink for total contemporary CO2 is weaker under increased wind stress and stratification by 2100, relative to a control run with no change in physical forcing, although the results are sensitive to the magnitude of the imposed physical changes and the rate of increase of atmospheric pCO(2). The air-sea fluxes of both natural and anthropogenic CO2 are sensitive to the surface concentration of dissolved inorganic carbon (DIC) which responds to perturbations in wind stress and stratification differently. Spatially averaged surface DIC scales linearly with wind stress, primarily driven by changes in the Ekman transport. In contrast, changes in the stratification cause non-linear and more complex responses in spatially averaged surface DIC. involving shifts in the location of isopycnal outcrop for deep and thermocline waters. Thus, it is likely that both wind stress and stratification changes will influence the strength of the Southern Ocean CO2 sink in the coming century.