Past model studies have projected a global decrease in marine net primary
production (NPP) over the 21st century, but these studies focused on the
multi-model mean rather than on the large inter-model differences. Here, we
analyze model-simulated changes in NPP for the 21st century under IPCC's
high-emission scenario RCP8.5. We use a suite of nine coupled carbon–climate
Earth system models with embedded marine ecosystem models and focus on the
spread between the different models and the underlying reasons. Globally, NPP
decreases in five out of the nine models over the course of the 21st century,
while three show no significant trend and one even simulates an increase. The
largest model spread occurs in the low latitudes (between 30
By producing organic matter, marine phytoplankton form the base of the marine
food web, control the amount of food available for higher trophic levels, and
drive the majority of the ocean's biogeochemical cycles, particularly that of
carbon. The net formation rate of organic carbon by phytoplankton, i.e., net
primary production (NPP), is a key determinant for the export of organic
carbon from the surface ocean, thereby governing how ocean biology impacts
the ocean–atmosphere exchange of CO
Several authors have analyzed trends in future NPP and the underlying drivers, using models
of strongly varying complexity and spatial resolution with regard to both the physical
and the ecosystem components and also investigating different climate change scenarios.
In the majority of these studies, global marine NPP was projected to decrease in response
to future climate change
However, a few studies produced contradicting results, i.e., they reported
increases in global NPP as climate change progresses over the 21st century
The past century provides very little experimental constraint on the impact
of long-term climate change on marine productivity, largely because of the
lack of long-term (
Far less work has been done regarding future trends in the biomass of specific plankton functional types (PFTs),
despite their importance in shaping ecosystem structure and function
While published studies emphasized the role of changes in bottom–up factors
in explaining the changes in NPP, top–down control by zooplankton grazing
may also drive future changes in total NPP or phytoplankton composition. This
mechanism is intriguing, since top–down control was recently identified as
one of the main drivers of phytoplankton competition during blooms in several
ecosystem models
Previous efforts in comparing different models with regard to future trends in NPP have analyzed the multi-model mean response and focused on identifying
regions of consistent changes and mechanisms among models
Reasons why models differ are seldom investigated in model comparison studies. In particular, it is often not readily clear whether the large spread in model projections is mainly caused by differences in the underlying ocean circulation model, by differences in the complexity of the ecosystem models or by differences in the parameterizations leading to differing sensitivities to, e.g., changes in temperature, nutrients and light. Such information is needed, however, in order to improve the existing models and eventually obtain reliable future projections.
In this work we go beyond the basic analysis of the multi-model mean and the
identification of regions of model consistency. Our aim is to identify where
models differ and by how much and then determine why they do so, i.e.,
to identify the underlying drivers of change. To this end, we use results from a
set of eight global marine ecosystem models coupled to or forced with nine coupled
carbon–climate Earth system models, which have simulated the future
evolution of marine NPP under the Intergovernmental Panel on Climate Change
(IPCC) Representative Concentration Pathways (RCP) 8.5
We use projections for the 2012–2100 period of nine model simulations for
the IPCC's RCP8.5 scenario from the “MARine Ecosystem Model
Intercomparison Project” (MAREMIP,
These criteria led us to use data from eight different marine ecosystem
models: diat-HadOCC, BEC, TOPAZ, PISCES, MEM, PELAGOS, REcoM2 and PlankTOM5.3
(Table
Overview of model simulations used in this work.
We describe the most important features of the ecosystem models in the following and give the full equations and parameters for the offline calculations shown in this work in the Appendix. The ocean ecosystem models used in this study are structurally similar, but they differ substantially in their details (see Table for an overview of the model structures). Within our selection, all models simulate at least two phytoplankton PFTs, usually representing diatoms and a nanophytoplankton type, and one zooplankton PFT. BEC and TOPAZ have an additional diazotrophic phytoplankton PFT. Moreover, TOPAZ differentiates between diatoms and other large phytoplankton depending on the availability of silicic acid. In PELAGOS, the nanophytoplankton type is further divided into flagellates and picophytoplankton. PlankTOM5.3 includes an explicit coccolithophore type, while in most other models coccolithophores are modeled implicitly as a fraction of nanophytoplankton. Regarding zooplankton PFTs, TOPAZ only has implicit zooplankton activity, diat-HadOCC, BEC, and REcoM2 have one zooplankton type, while PISCES and PlankTOM5.3 differentiate between micro- and mesozooplankton. MEM and PELAGOS have three zooplankton types, i.e., in addition to the micro- and mesozooplankton, they include predatory zooplankton in MEM and heterotrophic flagellates in PELAGOS. Finally, PELAGOS is the only model that includes heterotrophic bacteria explicitly.
Overview of ecosystem models used in this work, extended from
A change in NPP can be driven by (i) a change in the biomass-specific
rate of photosynthesis, (ii) changes in autotrophic respiration, or (iii) changes
in phytoplankton biomass through, e.g., zooplankton grazing, sinking and other loss processes of phytoplankton. However,
only PELAGOS and REcoM2 model photosynthesis (gross primary production, GPP) and autotrophic respiration separately. Rather,
most models calculate NPP directly as the product of the growth rate
In all models,
the growth rate of phytoplankton is parameterized using a multiplicative function of a maximum growth rate
The nutrient and light limitation factors have dimensionless values between 0
and 1, with higher values promoting higher growth. All models consider
limitation by multiple nutrients, with six out of the eight models applying
Liebig's law of the minimum
Comparison of temperature limitations in ecosystem models. Meso stands for for mesozooplankton, micro for microzooplankton, cocco for coccolithohores, and nano for nanophytoplankton.
For diat-HadOCC, the full model equations are not available; therefore we
cannot describe the light limitation. In all other models light limitation is
parameterized based on the work of
MEM uses the light limitation function from
Note that in most models, temperature and nutrient status influence also the light limitation, such that in addition to the direct effects of temperature and
nutrients on the growth rate, there is an additional indirect effect through light limitation
Since PELAGOS does not compute NPP directly and also uses a different
formulation for the growth limitation terms, it requires a separate analysis:
in this model, NPP is calculated for each phytoplankton type by subtracting
autotrophic respiration and other loss processes from its GPP, i.e.,
Regarding the loss terms for phytoplankton biomass, grazing is considered in all models. However, given the large diversity in the complexity
and parameterizations associated with the modeling of zooplankton, the role of grazing may differ substantially among the considered models
Grazing of zooplankton
Comparison of nutrient limitation of phytoplankton growth in ecosystem models.
Our analysis is based on monthly mean output for all surface ocean variables
for the period 2012–2100. In order to facilitate direct comparisons, we regridded
the model to a common 1
All models provided vertically (0–100 m) integrated NPP and biomass (in carbon units) of all PFTs. Primary production by diatoms and small phytoplankton was not available for PlankTOM5.3, MEM and PELAGOS and was estimated offline using the product of biomass and growth rate. The temperature limitations and growth rates were recalculated for all models except for PELAGOS and TOPAZ, where the growth rates were given in the model output. The nutrient and light limitation factors were included in the output of BEC, REcoM2 and TOPAZ, while they were recalculated from the monthly mean data for all other models using the original (not interpolated) data. The equations used for the recalculations are given in the Appendix. A comparison of recalculated and true values in the BEC model showed that the error in the recalculation is on the order of less than 10 %.
Comparison of prey dependence of grazing. For the full equations, see the Appendix.
Changes for all properties are computed by first averaging
the data for two 20-year periods, i.e., 2012–2031 and 2081–2100, and
then taking the difference. For the growth limitation factors, we show the ratio
changes, i.e., for any limitation factor
Taylor diagram showing the model–data correspondence for NPP (red),
surface chlorophyll (light blue), NO
Most of the models analyzed in this study have been evaluated individually in
their respective documenting publications (see references in Table
Model skill in representing global NPP, measured in Spearman's rank
correlation, normalized standard deviation (NSD) and bias. The NPP data
are from
On a global scale, the model-simulated nitrate fields correlate reasonably
well with the observations, with all models showing correlations between 0.62
and 0.85 and normalized standard deviations (NSD) between 0.86 and 1.10.
However, the bias is rather large, with values between
The correlations for chlorophyll are mostly between 0.5 and 0.72; however,
the normalized standard deviations are rather low (most models have NSD
values
Model skill in representing surface nutrients, measured in Pearson
correlation, normalized standard deviation (NSD) and bias. Nutrient data
from
Least well simulated is the distribution of NPP. The correlations are
relatively low (0.18–0.69), the range of normalized
standard deviation is as large as that of silicic acid (0.78 to 1.49), and in some of the models,
the bias is very large (
However, global correlations in nutrients and NPP are strongly influenced by the globally dominant gradient between the Southern Ocean and the low latitudes. While this gradient is generally well reproduced by the models, the model skill in reproducing the regional nutrient and NPP patterns is considerably lower (not shown).
Starting from very different levels, the models simulate global NPP to change under the RCP8.5 scenario anywhere from
Projected trends in annual mean integrated net primary production
(NPP) for the 2012–2100 period under RCP8.5, shown both in Gt C yr
The regional pattern of the multi-model median change in NPP (Fig.
To illustrate these inter-model differences, we show the IQR (Fig.
NPP changes (total and in percentage of the global value) in different regions. The Pacific upwelling region is shown in
Fig.
The changes in NPP in the different models can be driven either by changes in
the growth rates (bottom–up) or phytoplankton biomass (top–down control; see
Sect.
Spatial patterns of multi-model annual mean integrated net primary production (NPP)
for
We focus next on the drivers affecting the growth rates, i.e., the bottom–up factors temperature, light, and nutrients, and afterwards discuss the factors affecting phytoplankton biomass, i.e., the top–down control, and do so from a global perspective. We then extend the analysis to the level of individual phytoplankton PFTs, which is best done on the regional scale, across which the responses are relatively homogeneous in contrast to the global scale.
Figure
In the low latitudes, sea surface temperature is projected to warm by about
2–3
In contrast to the large changes in temperature, the PAR at the surface changes little globally, with the
important exception of the high latitudes (Fig.
The iron concentrations are projected to change in a latitudinally relatively
uniform manner with changes between
First-order Taylor decomposition of the surface NPP changes between
2012–2032 and 2080–2100 in
Zonal mean of projected sea surface temperature change, photosynthetically active radiation (PAR) change and change in surface Fe and NO
Zonal mean of the relative change in temperature, nutrient and light limitation and growth rate. For each model only the values for
the phytoplankton PFT with the strongest temperature (or light or nutrient limitation factor or growth rate) response is shown.
We calculate the relative change as
In nearly all models, the magnitude of the nutrient limitation term is determined solely by the most limiting nutrient
(Liebig limitation, see Sect.
Changes in relative diatom nutrient limitation (calculated as the 2081–2100 average divided by the 2012–2031 average) in all models that use Liebig limitation (smallest individual nutrient limitation term determines total nutrient limitation). The colors indicate changes in the nutrient limitation value, with positive values indicating an increase in nutrient limitation factor which is equivalent to lower nutrient limitation and an increase in growth. The hatching indicates the limiting nutrient. A change in limiting nutrient during the simulation period is shown with dots. REcoM2 does not simulate the Arctic; these missing values are shown in white.
As half of the models use specified N : P Redfield ratios instead of modeling an explicit PO
In summary, the changes in nutrients and temperature emerge as the most
important determinants for the changes in the growth rates, with light
generally playing a lesser role, except for the very high latitudes,
particularly the Arctic. The changes in the bottom–up factors determine the
changes in phytoplankton growth rate, which are shown in
Fig.
Possible reasons for the simulated phytoplankton biomass decrease between 2000 and 2100 are (1) changes in circulation or mixing leading to a stronger lateral or vertical loss of biomass, (2) increased aggregation or mortality of phytoplankton if explicitly modeled or (3) a higher grazing pressure. Unfortunately, none of these fluxes have been stored by most models. Further, recalculated values are not precise enough to analyze the difference between NPP and loss processes. Therefore, we cannot quantitatively differentiate our analysis into the changes in grazing loss, aggregation and physical biomass loss across all models. We nevertheless try to shine some light on this critically important issue by using qualitative arguments and the partial information we have from those models that were able to provide the phytoplankton grazing loss.
We hypothesize that the loss of biomass caused by physical transport does not
significantly increase, as all models show an increase in stratification over
the next century. Furthermore, phytoplankton aggregation (and mortality)
depends exponentially (linearly) on biomass but are temperature independent,
so neither aggregation nor mortality losses can increase at lower biomass
levels, eliminating this set of processes as well. This leaves us with
increased grazing pressure as the most likely driver of the simulated biomass
loss in the low and intermediate latitudes and the high northern latitudes.
This hypothesis is supported by the fact that in all five models for which
the grazing fluxes were available the fraction of grazed NPP increases
throughout the 21st century north of 50
Fraction of NPP that is grazed (grazing / NPP) normalized to the 2012–2031 average at the surface of the low and high northern latitudes
(
To better understand the potential drivers of this increase in the grazing
pressure, we analyze the fraction of NPP that is grazed by zooplankton, given
by
Climate change affects the ratio between grazing and NPP via temperature and
also via changes in nutrient and light limitation. Furthermore, the
grazing : NPP ratio is affected by changes in zooplankton biomass, i.e.,
increases in total grazing and zooplankton mortality indirectly play a role.
In the models where the same temperature function for both phytoplankton
growth and zooplankton grazing is used (i.e.,
Lacking the model output from the three-dimensional models, we use a one-box model to explore the sensitivity of the grazing : NPP ratio to
changes in temperature and nutrient limitation instead. To this end we consider only one phytoplankton and one zooplankton group, using a simplified
form of the equations and parameterizations of the BEC model. We did not include further phytoplankton loss terms like aggregation or mortality and used a
quadratic temperature-independent mortality as loss process for zooplankton.
We performed a spin-up until the model reached an equilibrium state under conditions
representative of the low latitudes (temperature limitation of 0.8 corresponding to
about 27
To test the sensitivity of grazing pressure to nutrient changes, we enhanced
nutrient limitation by 30 % (nutrient limitation factor decreases from
0.1 to 0.07) over 10 years, while keeping temperature constant at
27
In the following we will refine our analysis to include differences in PFT responses and focus on two example regions:
the low latitudes (30
All models analyzed in this study except one agree that NPP decreases between
the 2012–2031 and the 2081–2100 periods in the low latitudes
(Fig.
Decomposition of annual mean area-averaged low-latitude surface NPP
changes between 2012–2032 and 2080–2100 (red bar,
in mol C m
Relative change in temperature limitation factor (red), light limitation factor (yellow), nutrient limitation factor (orange), growth rate (green), biomass (light blue) and NPP (purple) for nanophytoplankton (full), diatoms (hatched) and coccolithophores (dotted) at the surface of the low latitudes for all models where the full equations were available. An increase in limitation factor denotes weaker limitation, which leads to stronger growth. The relative change in a variable is the ratio between the 2081–2100 average and the 2012–2031 average. A value of 1 means no change, 1.2 corresponds to a 20 % increase, 0.8 corresponds to a 20 % decrease. The product of the relative change in temperature, light and nutrient limitation results approximately in the relative change in growth rate. See main text for further details.
Figure
The PlankTOM5.3 trend is caused by an increase in coccolithophore NPP
(
All models simulate an increase in surface NPP in the Southern Ocean south of
50
Decomposition of Southern Ocean (50–90
In seven out of eight models, surface ocean warming is the most important
driver of the increase in phytoplankton growth for both diatoms and
nanophytoplankton. All but the CNRM–PISCES and PELAGOS model show a relief
from nutrient stress for all phytoplankton types, i.e., an increase in the
nutrient limitation factor (1–15 % increase), although these models
remain iron limited throughout the 21st century. Diatoms respond more
strongly to changes in nutrient concentrations than nanophytoplankton in all
models except for PlankTOM5.3. In addition, in many models a stronger
top–down control of nanophytoplankton than diatoms becomes apparent,
indicated by differences in biomass changes despite similar growth rate
changes. Only in MEM and PlankTOM5.3 do diatoms seem to be more strongly top–down
controlled. In PELAGOS the diatom fraction is almost 100 % south of
50
Our finding of the key role of temperature in defining the response of NPP to
future climate change contrasts with the conclusion of the majority of the
past studies, which attributed the decrease in NPP to a decrease in nutrient
availability, particularly in the low latitudes
The importance of warming which we have identified for future NPP is more in line with
another group of studies, where global NPP was projected to increase with
climate change and a temperature-driven increase in metabolic rates was
identified as the cause
Finally,
While we emphasize here the role of temperature in the models, our
understanding of how temperature controls the most important ecological and
biogeochemical processes in real marine ecosystems is not well established.
Most models base their parameterizations of temperature effects on laboratory
studies that show – within favorable thermal ranges – an exponential
increase in growth with increasing temperature
Seven out of nine models in our study show a global decrease in the relative
abundance of diatoms with decreases in low latitudes but increases in the
Southern Ocean, confirming results reported by
The spread in globally integrated NPP projections in our study is 45 %, with the PlankTOM5.3 model causing 25 % of it alone. Given this wide spread in NPP projections, we attempt to identify the different sources of uncertainty in the following and then investigate whether there is a way to narrow the uncertainty of the projections using emergent constraints.
The biogeochemical and biological parameterizations that contribute the largest uncertainties are
initial nutrient concentrations: models (except PlankTOM5.3) agree on similar decreases in nutrient concentration in the low latitudes, despite disparities with regard to the identification of the most limiting nutrient. However,
the differences in relative nutrient limitation change are very large ( the relative importance of iron versus nitrate limitation and projections for iron concentrations:
increases in iron availability allow the small global increase in
nanophytoplankton NPP in REcoM2 and attenuate or even outbalance the
low-latitude NPP decrease in BEC and TOPAZ. This is of particular relevance
in the equatorial upwelling region in the Pacific (see Fig. different the relative importance of the response of top–down controls versus that of the microbial loop, potentially related to different the fact that there is no agreement with regard to the direction of change in light limitation in the Southern Ocean, reflecting the wide range in
projected sea-ice changes and other factors influencing surface light such as cloud cover. However, light limitation currently only introduces a minor uncertainty compared to the nutrient and temperature effects, at least for
surface NPP.
In order to reduce the spread in NPP projections, we also need to understand
how much of the uncertainty arises from the underlying physical forcing
(transport, mixing, and temperature) and how much is caused by the different
ecosystem parameterizations. Unfortunately, the design of our study does not
allow for a clear distinction between uncertainty from physical forcing and
from the use of different ecosystems, as this would require us to compare
projections from different Earth system models using the same ecosystem model
with projections of one Earth system model coupled to different ecosystem
models. Results from
Relative changes between 2012–2032 and 2080–2100 in annual mean
temperature limitation factor (red), light limitation factor (yellow),
nutrient limitation factor (orange), growth rate (green), biomass (light
blue) and NPP (purple) for nanophytoplankton (full), diatoms (hatched) and
coccolithophores (dotted) at the surface of the Southern Ocean
(50–90
The concept of emergent constraints
We did not find any significant correlation between model skill and NPP
changes, neither on regional nor on global scales, and the relation is weak at
best between globally integrated NPP and the absolute change in NPP
(Fig.
Relationship between the change in NPP and the 2012–2031 average NPP for
all models. Change in NPP has been calculated as the sum of the differences between the
2012–2031 average and the 2081–2100 average for each grid cell (open dots). We additionally show
the negative absolute differences of the changes (full dots), calculated by taking the sum of the negative absolute differences
between the 2012–2031 average and the 2081–2100 average for each grid cell.
Each color represents a model,
In this work we present a multi-model comparison of nine model simulations
with regard to NPP and its underlying drivers. We show projected changes in
global NPP between
One major difficulty faced in this study is the limited availability of model output variables related to ecosystem growth and loss rates, particularly limitation factors and grazing rates. The changes in growth rate, temperature limitation, and light and nutrient limitation reported in this work have been recalculated in six out of nine models using surface monthly mean fields. The obtained results are therefore an approximation of the original values. We have compared recalculated values with original values in the models where the limitation factors were given, and we estimate the error to be less than 10 %. We conclude that, while the absolute values reported might be inaccurate, the relative importance of nutrient vs. temperature limitation shown in this work is correct. Furthermore, we can discuss only surface NPP changes. For the models where three-dimensional limitation factors were available (BEC, REcoM2), we compared our results for the surface with the 100 m average, and we can confirm that the same mechanisms that govern the surface changes also hold for the 100 m average. In addition, the changes in surface NPP correlate with the changes in integrated NPP in all models, except for the Arctic Ocean. It therefore seems likely that our surface drivers also describe the changes in integrated NPP. To ease future studies of NPP changes, we recommend the inclusion of mixed layer averages of growth rate, light and nutrient limitation and grazing fluxes in the standard model output of future model intercomparison projects. The availability of changes in growth rates could prevent common misinterpretations of drivers by analyzing univariate correlations with only one of several possible drivers.
To reduce the uncertainty in NPP projections, the representation of present-day nutrient concentrations and resulting limitation patterns should be further improved. Particularly a bias in present-day nutrient concentration strongly affects relative changes in nutrient limitation and therefore NPP projections. Furthermore, given the importance of top–down control shown in this work, we need a better understanding of zooplankton mortality and further potential drivers of zooplankton biomass like phenological or trophic mismatches, diseases or changes in predation from higher trophic levels. Finally, a better understanding of the temperature dependency of all key ecological or biogeochemical processes is needed. In particular, this includes the determination of the different temperature response functions for the different PFTs and trophic levels.
In the following, we give the equations and parameters governing NPP in all models except for diat-HadOCC, where the full equations are currently not available. We abbreviate nanophytoplankton with “nano”, diatoms with “diat”, zooplankton with “zoo”, and meso- and microzooplankton with “meso” and “micro”, respectively.
Symbols used in the model equations.
BEC parameters.
TOPAZ parameters.
(for nanophytoplankton, diatoms and microzooplankton)
PISCES parameters.
MEM parameters.
Grazing of zooplankton type
PELAGOS parameters.
PlankTOM5.3 parameters.
REcoM2 parameters.
C. Laufkötter and the research leading to these results have received funding from the European Community's Seventh Framework Programme (FP7 2007–2013) under grant agreements no. 238366 (Greencycles II) and 264879 (CarboChange). M. Vogt and N. Gruber acknowledge funding by ETH Zürich. S. C. Doney and I. D. Lima acknowledge support from NSF (AGS-1048827). We thank the climate modeling groups for calculating and providing their model output. We also acknowledge the World Climate Research Programme's Working Group on Coupled Modeling, which is responsible for CMIP. For CMIP the US Department of Energy's Program for Climate Model Diagnosis and Intercomparison provided coordinating support and led the development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. We thank T. Frölicher and C. O'Brien for fruitful discussions. This is a contribution from the MAREMIP project. Edited by: C. P. Slomp