Introduction
Marine biogeochemical model complexity has long been a subject of debate
(e.g. Anderson, 2005). Simpler models require fewer, often better understood
parameterizations, but omit processes which are known to be important. More
complex models explicitly include these processes, but require an increased
number of tuneable parameters, the ranges of which are often poorly
constrained by observations and/or poorly defined where they aggregate over a
range of species. No consensus exists within the scientific community, but
recent studies have shown that simple to moderately complex models still do
the best job of reproducing basic biogeochemical descriptors such as primary
production and carbon fluxes (Kwiatkowski et al., 2014; Friedrichs et al.,
2007; Ward et al., 2013; Xiao and Friedrichs, 2014). Further studies have
suggested that “Models of Intermediate Complexity for Ecosystem
assessments” are the most appropriate for fisheries management (Plagányi
et al., 2014).
Ultimately, models are tools, and the most appropriate tool should be chosen
for the task at hand; different scientific, societal, and managerial questions
will require models of different complexities. For instance, describing a
complex coastal environment will likely require the explicit inclusion of
processes which are less important when considering global-scale carbon
budgets. Furthermore, some users require more detailed information about the
marine environment than simple models can provide, necessitating the use of
more complex models.
This demand for detailed information applies to the North Sea, with users and
policy makers requiring information about topics including eutrophication and
nutrient ratios (Skogen et al., 2014), productivity in relation to fisheries
(Chassot et al., 2007), harmful and nuisance algal blooms (Blauw et al.,
2010; Kurekin et al., 2014), water clarity (Dupont and Aksnes, 2013; Capuzzo
et al., 2015), biodiversity (Brandsma et al., 2013), effects of climate
change (van der Molen et al., 2013; Wakelin et al., 2015a), effects of
trawling (Allen and Clarke, 2007; van der Molen et al., 2013), and impacts of
marine renewable energy generation (van der Molen et al., 2014, 2016). In particular, indicators of Good Environmental Status (GES)
are required in the context of the Marine Strategy Framework Directive (MSFD;
Borja et al., 2013). These include descriptors of food-web structure, trophic
status, and biodiversity, and elements of these can be assessed with various
modelling approaches (Piroddi et al., 2015; Hyder et al., 2015).
Ecosystem models are central to the delivery of marine ecosystem-based
management that is specified in existing legislation (MSFD – EU, 2008; CFP
– EU, 2013; WFD – EU, 2000). These models are important for the design of
management measures and to assess the social, economic, and environmental
performance of management in relation to targets (Defra, 2014; Sutherland et
al., 2006). This is done through improving our understanding of the links
between pressures (human and environmental) and the response of the system to
these pressures. However, ecosystem models are not used frequently in the UK
and Europe in support of policy and management (Hyder et al., 2015), despite
increasing use in the USA and Australia (Fulton and Link, 2014; Fulton et
al., 2007). For models to have a larger impact on policy development and
decision-making, modelling approaches need to be more transparent,
verifiable, and repeatable than they are at present, as any outputs can be
subject to legal challenge (Hyder et al., 2015). Poor uptake of ecosystem
models by decision-makers is due to a lack of confidence in and understanding
of models. This relates to how models are used, terminology, type of outputs,
treatment of uncertainty, required quality standards, and the presentation of
model products (Hyder et al., 2015). The use of ecosystem models will become
increasingly important as the complexity of marine legislation increases
(Boyes and Elliott, 2014). Hence, simple assessment of the skill of models in
predicting outcomes (validation – Mackinson, 2014), model comparisons (e.g.
Kwiatkowski et al., 2014), and the clear treatment of the uncertainty
associated with predictions (Thorpe et al., 2015; Gårdmark et al., 2013;
Stewart and Martell, 2015; Tebaldi and Knutti, 2007) are needed to increase
the confidence in and uptake of models (Hyder et al., 2015).
At the base of the marine food chain are phytoplankton, and phytoplankton
community structure is a fundamental consideration in any assessment of
marine biodiversity (Garmendia et al., 2013). Changes in community structure
can result from large-scale environmental changes such as temperature rises
or eutrophication, with different organisms favouring different conditions.
Some organisms that favour changed conditions may be harmful to human health
(Roselli and Basset, 2015; Bruggeman, 2009). Alternatively, top-down control
by benthic or pelagic grazers can change the size structure of phytoplankton
by selective removal of larger species, resulting for instance in an
increased proportion of pico-phytoplankton in areas with dense shellfish
aquaculture (Smaal et al., 2013). Phytoplankton vary in size by up to 9 orders of magnitude for cell volume (Finkel et al., 2010), with variations in
community structure reflected in the size and species of their predators, and
the number of links in the food chain (Ryther, 1969; Chavez et al., 2011).
Larger cells such as diatoms are consumed directly by copepod grazers, giving
a higher transfer of energy and ultimately impacting commercial fish stocks
(Jennings and Collingridge, 2015). As the physical structure of the North Sea
becomes increasingly well understood due to advances in hydrodynamic
modelling (van Leeuwen et al., 2015) and availability of long-term
observations (Greenwood et al., 2010; Núñez-Riboni and Akimova,
2015), the potential to predictively model plankton population structure and
distribution increases as well.
A common way to model plankton community structure is to take a phytoplankton
functional type (PFT) approach, such as is done in variants of the European
Regional Seas Ecosystem Model (ERSEM; Baretta et al., 1995). This approach
groups phytoplankton into a number of PFTs, based on their general function
within the ecosystem (Le Quéré et al., 2005). If information on
phytoplankton community structure is to be modelled and provided to users,
then it must be validated. Some studies have aimed to validate this against
observations (Lewis et al., 2006; Gregg and Casey, 2007; Lewis and Allen,
2009; Hirata et al., 2013), but commonly validation studies go no further
than total chlorophyll concentration (Edwards et al., 2012; de Mora et al.,
2013). This is largely because there is a lack of observations that contain
more detail about community structure against which to compare. Algorithms
for deriving phytoplankton community structure from remotely sensed satellite
ocean colour observations, either in the form of PFTs or phytoplankton size
classes (PSCs), are being developed (Brewin et al., 2011; Brito et al., 2014),
but have not yet reached maturity and are not yet widely available to the
general scientific community. Moreover, such remote-sensing products require
a similar level of validation (Brotas et al., 2013). In situ observations are
sparse, particularly in shelf seas, and the measured variables may not be
easily matched to model outputs, which do not always aggregate neatly over
species or size classes.
This study presents a novel set of in situ observations of phytoplankton
community structure in the North Sea using accessory pigment analysis
(Sherrard et al., 2006), noting that coastal seas are greatly
under-represented in the existing global collection of pigment data (Peloquin
et al., 2013). Pigment data were analysed so as to give the relative
distribution of different size classes, allowing a robust comparison with
outputs from ERSEM-type models. Two variants of ERSEM, run by the Centre for
Environment, Fisheries and Aquaculture Science (Cefas) and the Met Office,
both public bodies in the UK, were validated against these and other
observations. The aims of the study were to determine what these new
observations add to current scientific understanding of North Sea
biogeochemistry, assess the extent to which the models can reproduce the
observations, and discuss the implications for current and future user and
policy needs, observing strategies and model development.
Models
Two different physical–biogeochemical modelling systems were used in this
study: GETM-ERSEM-BFM, run by Cefas, and NEMO-ERSEM, run by the Met Office.
Each is described in turn below, followed by a discussion of their
differences and similarities. Existing configurations of each model were
used, with no attempt made to increase their similarity. Details of the model
configurations and forcing are given in Table 1, and the model domains and
bathymetries are shown in Fig. 1.
GETM-ERSEM-BFM
GETM (General Estuarine Transport Model) is a public domain,
three-dimensional (3-D) finite difference hydrodynamical model (Burchard and
Bolding, 2002; available through http://www.getm.eu). It solves the 3-D
partial differential equations for conservation of mass, momentum, salt, and
heat. The ERSEM-BFM (European Regional Seas Ecosystem Model – Biogeochemical
Flux Model) version used here is a development of the model ERSEM III (see
Baretta et al., 1995; Ruardij and van Raaphorst, 1995; Ruardij et al., 1997,
2005; Vichi et al., 2003, 2004, 2007; van Leeuwen et al., 2013; van der Molen
et al., 2013, 2014, 2016), and describes the dynamics of the biogeochemical
fluxes within the pelagic and benthic environment. The ERSEM-BFM model
simulates the cycles of carbon, nitrogen, phosphorus, silicate, and oxygen,
and allows for variable internal nutrient ratios inside organisms, based on
external availability and physiological status. The model applies a
functional group approach and contains six phytoplankton groups, four
zooplankton groups, and five benthic groups, the latter comprising four
macrofauna and one meiofauna groups. Pelagic and benthic aerobic and
anaerobic bacteria are also included. The pelagic module includes a number of
processes in addition to those included in the oceanic version presented by
Vichi et al. (2007) to make it suitable for temperate shelf seas: (i) a
parameterization for diatoms allowing growth in spring, (ii) enhanced
transparent exopolymer particles (TEP) excretion by diatoms under nutrient
stress, (iii) the associated formation of macro-aggregates consisting of TEP
and diatoms, leading to enhanced sinking rates and a sufficient food supply
to the benthic system especially in the deeper offshore areas (Engel, 2000),
(iv) a Phaeocystis functional group for improved simulation of
primary production in coastal areas (Peperzak et al., 1998; Ruardij et al.,
2005), (v) a new resuspension module for inorganic fine SPM that responds to
combined currents and surface waves, and uses a concentration-dependent
settling velocity for improved simulation of the under-water light climate
(van der Molen et al., 2017), and (vi) resuspension of particulate organic
material, in proportion to the resuspended inorganic SPM and the relative
concentrations of organic and fine inorganic matter in the sea bed. The model
includes a three-layer benthic module comprising 53 state variables, which
enables it to resolve a high level of detail of benthic processes and
benthic–pelagic coupling. New features of the benthic model are: (i) benthic
diatoms, and (ii) active oxygen uptake of deposit feeders from the water
column. The first four additional pelagic processes listed above are related,
and based on detailed implementation of the dynamic model of phytoplankton
growth, explicit chlorophyll content, and acclimation of Geider et al. (1997).
In nutrient-enriched coastal zones, the competition between and seasonal
succession of PFTs is influenced strongly by differences in their
photosynthetic capability. The modelled photosynthesis and phytoplankton
carbon and chlorophyll content follows Geider et al. (1997) closely, by first
calculating light-saturated and nutrient-replete photosynthesis. In the
second stage, light-adapted chlorophyll content is calculated and light
limitation and nutrient limitation are applied. These result in changes in
the chlorophyll : carbon ratio, and growth. Photo-inhibition is included
explicitly in the chlorophyll calculations, and carbon uptake is calculated
before applying nutrient limitation. Under nutrient-limited conditions,
diatoms excrete the excess carbon as TEP, which is modelled as a separate
state variable. Phaeocystis cells, implemented as a simplified
version of the model of Ruardij et al. (2005), excrete TEP within the colony.
Implicit macro-aggregate sinking rates are calculated as a linear proportion
of a prescribed maximum sinking rate, governed by stickiness rates related to
the concentrations of diatoms, TEP, and the level of nutrient stress, and also
induce sinking of other PFTs. Because the initial slope
of the P–I curve αchl, the light-saturated carbon-specific
photosynthesis rate PmC, the saturation parameter for the
growth-irradiance curve KI, and the maximum chlorophyll : carbon
ratio θm are related (Geider et al., 1997) through
Maps of the model domain and bathymetry for (a) NEMO-ERSEM
and (b) GETM-ERSEM-BFM.
![]()
Parameters of the four coinciding phytoplankton functional types
related to inter-species competition.
NEMO-ERSEM
GETM-ERSEM-BFM
Method/parameter
Diatoms (P1)
Flagellates (P2)
Pico-phytoplankton (P3)
Dinoflagellates (P4)
Method/parameter
Diatoms (P1)
Flagellates (P2)
Pico-phytoplankton (P3)
Dinoflagellates (P4)
Doubling temperature
Q10
2
2
2
2
Q10
2
2
2
2
Maximum productivity at 10 ∘C
sum
2.5
2.7
3.3
1.5
sum
3.0
4.875
5.6
1.75
Respiration rate at 10 ∘C
srs
0.05
0.05
0.05
0.05
srs
0.125
0.1
0.1
0.125
Fraction of PP excreted as PLOC/PDET
pu_ae
0.05
0.2
0.2
0.05
pu_ae
0.05
0.1
0.1
0.05
Activity respiration rate
pu_ra
0.1
0.25
0.25
0.25
pu_ra
0.1
0.1
0.2
0.1
Half-value of SiO4-limitation
chP1sX
0.3
–
–
–
P_chPs
0.3
–
–
–
Minimum quota N
qnlP1cX … qnlP4cX
0.00687
0.00687
0.00687
0.00687
p_qnlc
0.00687
0.00687
0.00687
0.00687
Minimum quota P
qplP1cX … qplP4cX
0.0004288
0.0004288
0.0004288
0.0004288
p_qplc
0.0003931
0.0003931
0.0003931
0.0003931
Minimum quota Si
–
–
–
–
–
–
0.09
–
–
–
Multiplication factor for critical N : C ratio
xpcP1nX … xpcP4nX
1
1
1
1
Not included
–
–
–
–
Multiplication factor for critical P : C ratio
xpcP1pX … xpcP4pX
1
1
1
1
Not included
–
–
–
–
Multiplication factor for maximum quotum nitrate uptake
xqnP1X … xqnP4X
2
2
2
2
p_xqn
2
2
2
2
Multiplication factor for maximum quotum phosphate uptake
xqpP1X … xqpP4X
2
2
2
2
p_xqp
2
2
2
2
Multiplication factor for maximum quotum silicate uptake
–
–
–
–
–
p_xqs
1.5
–
–
–
Affinity for NO3
quP1n3X … quP4n3X
0.0025
0.0025
0.0025
0.0025
p_qun
0.15
0.215
1.29
0.0084
Affinity for NH4
quP1n4X … quP4n4X
0.01
0.01
0.02
0.01
Grouped with NO3
–
–
–
–
Affinity for P
qurP1pX … qurP4pX
0.0025
0.0025
0.0025
0.0025
p_qup
0.15
0.215
1.29
0.0084
Affinity for Si
qsP1cx
0.03
–
–
–
p_qus
0.1
0
0
0
Continued.
NEMO-ERSEM
GETM-ERSEM-BFM
Method/parameter
Diatoms (P1)
Flagellates (P2)
Pico-phytoplankton (P3)
Dinoflagellates (P4)
Method/parameter
Diatoms (P1)
Flagellates (P2)
Pico-phytoplankton (P3)
Dinoflagellates (P4)
Nutrient stress threshold for sinking
esNIP1X … esNIP4X
0.7
0.75
0.75
0.75
Different method: based on TEP production
–
–
–
–
Sinking by formation of macro-aggregates
–
–
–
–
–
–
Threshold process: in presence of sufficient TEP and diatoms
Sticking to macro-aggregates
Sticking to macro-aggregates
Sticking to macro-aggregates
Lysis rate
sdoP1X … sdoP4X
0.05
0.05
0.05
0.05 + 0.2
Max. by nutrient stress, p_sdmo
No lysis
0.025
0.15
0.001 + 0.0
Stress excretion
–
–
–
–
–
–
Extraction of TEP (carbohydrates)
–
–
–
Light susceptibility
P-I curve, alpha
2.98
2.98
2.98
2.98
P-I curve, KE
35
70
124
116
Maximum Chl to C ratio
phimP1X … phimP4X
0.035
0.035
0.035
0.035
p_qchlc
0.02
0.035
0.035
0.035
Minimum Chl to C ratio
phiP1HX … phiP4HX
0.025
0.025
0.025
0.025
p_qlPlc
0.0025
0.0035
0.0025
0.00225
Background phytoplankton sinking rate
m day-1
0
0
0
0
m day-1
0
0
0
-5
Maximum phytoplankton sinking rate
–
–
–
–
–
m day-1
90
90
90
90
αchl=PmCθmKI
and KI can be approximated by the light intensity at maximum
photosynthetic rate (KE), so KE was used to prescribe light
sensitivity. Values were selected to simulate observed successions in the
Dutch coastal zone (Table 2).
The model setup for the North Sea uses a spherical grid with a spatial
resolution of approximately 11 km and 25 layers in the vertical (Lenhart et
al., 2010; van der Molen et al., 2014, 2015). The model was forced with tidal
boundary conditions from a shelf-scale model, temperature and salinity
boundary conditions from a global hindcast (ECMWF-ORAS4; Balmaseda et al.,
2013; Mogensen et al., 2012), climatological nutrient boundary conditions,
observations-based river run-off, and riverine nutrient loads (the National
River Flow Archive (data available at
http://www.ceh.ac.uk/data/nrfa/index.html) for UK rivers, the Agence de
l'eau Loire-Bretagne, Agence de l'eau Seine-Normandie and IFREMER for French
rivers, the DONAR database for Netherlands rivers, ARGE Elbe, the
Niedersächsisches Landesamt für Ökologie, and the Bundesanstalt
für Gewässerkunde for German rivers, and the Institute for Marine
Research, Bergen, for Norwegian rivers; see also Lenhart et al., 2010), and
atmospheric forcing from the ECMWF ERA-40 and operational hindcast (ECMWF,
2006a, b).
NEMO-ERSEM
The hydrodynamic component of the Met Office modelling system is NEMO
(Nucleus for European Modelling of the Ocean; Madec, 2008). NEMO is an open
source community model originally developed for global ocean modelling (e.g.
Storkey et al., 2010), but which has also been recently developed for use in
shelf seas (O'Dea et al., 2012). The version used in this study (CO5; O'Dea
et al., 2017) is based on NEMO v3.4, and is a development of that described
in O'Dea et al. (2012) and Edwards et al. (2012). The main updates from O'Dea
et al. (2012) are an upgrade from NEMO v3.2 to v3.4, an increase in vertical
resolution from 33 to 51 levels and change of coordinate stretching function,
changes to the river inputs and Baltic boundary condition, a change of data
assimilation scheme from analysis correction to 3D-Var, and the use of bulk
formulae to calculate the input atmospheric fluxes rather than direct
forcing. These updates are described further below.
The version of ERSEM used is an alternative development of the original code
of Baretta et al. (1995), led by Plymouth Marine Laboratory (PML), and is
described in detail by Blackford et al. (2004) and Edwards et al. (2012). The
pelagic component includes four phytoplankton and three zooplankton
functional groups, and one bacterial group. The benthic component includes
aerobic and anaerobic bacteria, suspension feeders, bottom feeders, and the
meiobenthos. This version follows the photoacclimation model of Geider et
al. (1997) in an adapted form, in which nutrient limitation is applied before
the other calculations, leading to much lower estimates of excess carbon,
which is excreted as detritus. Photoinhibition is included as an additional
parameterization in the photoacclimation method. SPM is simulated as per
Sykes and Barciela (2012).
As part of the Forecasting Ocean Assimilation Model (FOAM; Blockley et al.,
2014) suite of models, NEMO-ERSEM is run operationally at the Met Office on a
daily basis, providing 5-day forecasts of physical and biogeochemical
variables for the Northwest European Shelf seas. Analyses and forecasts are
publicly available through the Copernicus Marine Environment Monitoring
Service (CMEMS; http://marine.copernicus.eu), which is the operational
service building on the MyOcean project. Physical and biogeochemical
reanalysis products (Wakelin et al., 2015b) are also available through CMEMS,
and results from the recent NEMO-ERSEM reanalysis were used in this study.
The model was run on the 7 km resolution Atlantic Meridional Margin (AMM7)
domain, covering the entire Northwest European Shelf seas, including the
North Sea. There are 51 vertical levels, using a hybrid σ–S coordinate
system with the stretching function of Siddorn and Furner (2013). This uses
terrain-following coordinates whilst ensuring a fixed surface resolution of
1 m. Remotely sensed and in situ observations of SST were assimilated using
a 3D-Var implementation of the NEMOVAR data assimilation scheme (Waters et
al., 2015; O'Dea et al., 2012). River inputs were taken from the E-HYPE model
(Donnelly et al., 2015) for flow values, and from the same climatology as in
Edwards et al. (2012) for nutrients and SPM. Lateral boundary conditions for
physical variables were taken from a reanalysis of the GloSea5 seasonal
forecasting system (MacLachlan et al., 2014) at the Atlantic boundaries, and
from the IOW-GETM model (Stips et al., 2004) at the Baltic boundary. For
biogeochemistry, lateral boundary conditions for nitrate, phosphate, and
silicate were taken from the World Ocean Atlas monthly climatology (Garcia et
al., 2010) at the Atlantic boundaries, and zero flux boundary conditions were
applied at the Baltic boundary. Zero flux boundary conditions were applied
for all other biogeochemical variables at all boundaries. Surface forcing was
from the ERA-Interim reanalysis (Dee et al., 2011). The NEMO-ERSEM reanalysis
covers the period January 1985 to July 2012, but for practical reasons was
run in three sections. The final section, which this study uses, started in
November 2003, with physics initial conditions taken from the corresponding
date of a non-assimilative hindcast of the entire reanalysis period.
Biogeochemical initial conditions were taken from a winter date of the run of
NEMO-ERSEM described in Edwards et al. (2012).
Comparison of the two models
Even though both models used versions of ERSEM, it is reasonable to expect
differences in the results. Such differences are inevitably a result of the
accumulation of differences between the models. It should be noted that both
models were run as usual, and no attempts were made to increase similarity.
It is recognized that this means definitive conclusions cannot be reached
here on the exact reasons behind differences in results, and this was not the
aim of this study. A preliminary discussion is provided here, with more
detailed follow-on experiments proposed in Sect. 5. To help understand the
differences in model behaviour, this section summarizes the main differences
between the two models. We focus on two types of differences: general level
differences (Table 1), and differences in phytoplankton parameters and
parameterizations (Table 2). For the sake of readability, and to limit
repetition, the following summary is kept at a fairly basic level; for
(numerical) detail the reader is referred to the tables.
The two hydrodynamics models were different, and in general used different
domains, resolutions, and forcing data. The NEMO-ERSEM model had a larger
domain (Fig. 1), at higher resolution, and used more advanced atmospheric
forcing. Moreover, in NEMO-ERSEM, SST was assimilated, while GETM-ERSEM-BFM
had no data assimilation. NEMO-ERSEM's river runoff originated from a model,
that of GETM-ERSEM-BFM from observations. The GETM-ERSEM-BFM model used time
series of riverine nutrient inputs whereas the NEMO-ERSEM model used a
climatology. The SPM model of NEMO-ERSEM contained explicit size fractions
and cohesive interactions, but was only forced by flow velocities, while that
of the GETM-ERSEM-BFM model was non-cohesive, with implicit size-related
behaviour and included resuspension by both currents and waves (van der Molen
et al., 2017). The models also used different initial conditions and spin-up
sequences.
Both ERSEM versions share a common origin, both use the same base nutrients
(N, P, Si, C), and are both based on a functional type approach. They share
four phytoplankton types, three zooplankton types, and a basic bacteria type.
Both have a three-layered benthic module, with similar nutrient regeneration
mechanisms.
GETM-ERSEM-BFM had a number of additional functional types compared to
NEMO-ERSEM: Phaeocystis colonies, benthic diatoms, carnivorous
zooplankton, filter feeder larvae, epibenthos, benthic predators, and benthic
and pelagic nitrifying bacteria. Furthermore, it used a CO2 module,
whereas in NEMO-ERSEM this was switched off.
The models used different methods for nutrient affinity, nutrient stress and
sinking, and light susceptibility. For nutrient affinity, GETM-ERSEM-BFM used
10–100 times higher affinity values for nutrient uptake. There are two ways
to measure phytoplankton nutrient uptake in experiments (Veldhuis and
Admiraal, 1987): (i) a short-duration experiment in which nutrients are added
to nutrient-deprived algal cultures and uptake rates into the internal
nutrient buffer are measured; and (ii) an experiment that lasts a full day in
which uptake rates needed for daily growth are measured. The parameters for
GETM-ERSEM-BFM were based on short-duration experiments, whereas those for
NEMO-ERSEM were based on long-duration experiments. The short-duration
parameterization allows for improved incorporation of the dependencies of
cell properties such as cell size and buffer capacity. These features were
needed to resolve the competition between diatoms and Phaeocystis
colonies during excessive spring blooms in the Dutch coastal zone, which
terminate through phosphate depletion. In GETM-ERSEM-BFM, nutrient stress of
pelagic diatoms leads to excretion of all (new fixated) organic C that cannot
be used for growth as carbohydrates (TEP). At high levels of diatoms, this
excretion leads to the simulation of the effect of macro-aggregate formation
through binding by these carbohydrates, through increases in the sinking rate
of live and dead particulate matter. NEMO-ERSEM used a more implicit approach
to sinking. For light susceptibility, both models used a
photosynthesis-irradiance (P–I) curve approach, but NEMO-ERSEM defined it
through the initial slope (alpha), whereas GETM-ERSEM-BFM defined it through
the light intensity at maximum photosynthetic rate (KE). For several elements where both models used the
same approach, parameter settings were different: maximum productivity,
respiration, excretion, minimum quota for P, lysis, and C : Chl ratios. For
these, there was typically more differentiation in settings between
phytoplankton functional types in GETM-ERSEM-BFM than in NEMO-ERSEM.
Maps of satellite-derived surface phytoplankton chlorophyll
distribution during the summer IBTS cruises of (a) 2010 and
(b) 2011, overlaid with the in situ observations in circles. White
areas are where no satellite data were available. Time series of
phytoplankton chlorophyll along the IBTS cruise track in (c) 2010
and (d) 2011 as assessed by continuous measurements of chlorophyll
fluorescence (solid black line), and sampling of surface water for
quantification of Chl a (open squares). Specific events along the tracks
are referenced with a letter.
![]()
Aggregating model PFTs to match observed PSCs
To allow validation of phytoplankton community structure from the models
against the IBTS observations, the four PFTs from NEMO-ERSEM and six PFTs
from GETM-ERSEM-BFM must be appropriately aggregated to match the observed
PSCs. Diatoms (both models), dinoflagellates (both models) and resuspended
benthic diatoms (GETM-ERSEM-BFM only) were considered to be
micro-phytoplankton. Flagellates (both models) and Phaeocystis
colonies (GETM-ERSEM-BFM only) were considered to be nano-phytoplankton. The
pico-phytoplankton PFT (both models) was directly mapped to the
pico-phytoplankton PSC. For consistency with the IBTS observations, the PFTs
and PSCs were expressed as fractions of total chlorophyll concentration,
rather than biomass.
Results
Observations
Each year, the IBTS cruise starts in early August in the Southern Bight of
the North Sea off the Thames Estuary (51.5∘ N) and proceeds
northwards via a series of longitudinal transects, with each transect taking
1–3 days, depending upon the width of the North Sea at each
point. The final transect between the Shetland Islands and Norway at
61∘ N was reached by early September for the 2010 and 2011 IBTS
cruises. The spatially averaged annual mean surface temperature for the North
Sea was 9.9 ∘C in 2010 and 10.0 ∘C in 2011, which were very
close to the long-term average of 10.0 ∘C for the 1985–2014 period.
Hence, the years surveyed represent near-average conditions for temperature.
A continuous recording of chlorophyll fluorescence showed good agreement with
the quantity of extracted Chl a determined by HPLC (r2=0.64 for 2010
and 0.65 for 2011). The number of same-day match-ups between in situ Chl a
and satellite-derived Chl a was low for both years, but a comparison of
8-day averaged surface Chl a from MERIS with in situ data showed an
excellent qualitative agreement for both years (Fig. 2). Satellite coverage
was more complete in 2011 than 2010. Time series plots and maps of the two
cruises showed a number of regularly occurring features that can be observed
at this time of year (labelled “A” to “J” in Fig. 2).
A zone of high Chl a (> 2 mg m-3) was observed with all methods
in the coastal waters of Belgium, The Netherlands, Germany, and Denmark. This
zone extended between points “A” and “B” for the map of 2010, and points
“F” and “H” for 2011. High chlorophyll values (> 2 mg m-3) were
observed in the outer Thames Estuary and close to the English east coast as
far north as the Humber Estuary, but the English coastal zone was not as
clearly demarked by high Chl a as the continental coast. The continuous
recording of the first 7–10 days of the IBTS thus alternated between
moderate Chl a (1 to 2 mg m-3) and high Chl a as the vessel
covered the southern North Sea between 51.5 and 55∘ N. An
exceptional bloom event with Chl a of over 6 mg m-3 was recorded at
location “G” in 2011, and was clearly visible in MERIS and MODIS images.
The central section of the North Sea between 55 and 58∘ N was
covered during the second and third weeks of the IBTS. This section showed
low Chl a values (< 1 mg m-3) across most of the zone (Fig. 2),
particularly in the region north of “I”, 56.5 to 58.5∘ N, 0 to
3∘ E, which was a large region with values < 0.5 mg m-3.
To the east, the Danish coastal waters (“B” and “H”) showed high Chl a.
The inshore English coast north of the Humber, and Scottish coastal waters,
are low in Chl a compared to those further south. A moderate Chl a bloom
was evident in the chlorophyll fluorescence trace, MERIS image and extracted
Chl a at position “C” in 2010, and a high Chl a patch was evident close
to the Scottish coast at Aberdeen at position “D”.
The northern North Sea was sampled in weeks three and four (from 28 to
29 August onwards) and was similar in 2010 and 2011. An arc of high Chl a
was detected from north of the Scottish mainland through the Orkneys and
Shetlands, e.g. from “D” to “E” in 2010, with particularly high values at
“E”. In 2011, high values were observed from the Orkneys through to north
of the Shetlands at “J”. The FerryBox chlorophyll fluorescence recorded a
further large bloom on 6 September 2011, but this event was not sampled for
pigments.
Maps of percentage surface PFT distribution during the summer IBTS
cruises of 2010 (upper maps, a–c) and 2011 (lower maps,
d–f) for pico-phytoplankton (a, d),
nano-phytoplankton (b, e) and micro-phytoplankton (c, f).
![]()
As described in Sect. 2.1, PFTs were determined on the basis of accessory
pigment composition. In general, pigment diversity was lower in coastal
waters and in the southern North Sea and reached peak diversity in the
stratified central North Sea. Fucoxanthin was the dominant accessory marker
pigment in the southern North Sea, and 19-hex was dominant in the northern
North Sea. Pico-phytoplankton were represented in this analysis by the marker
pigments zeaxanthin, Chl b and 19-hex; these pigments were rare in the
southern North Sea below a line from East Anglia to the Wadden Sea, and hence
pico-phytoplankton contribution was estimated in this region to be less than
10 % of total phytoplankton biomass (Fig. 3). The contribution of
pico-phytoplankton increased with increasing latitude so that the area with
highest contribution from the smallest PFT was found in both years to be
located north of 57∘ N and east of 0∘ E. Nano-phytoplankton
were represented by the pigments 19-hex, 19-but, alloxanthin, and Chl b. The
highest percentage contribution of nano-phytoplankton was found in both years
to be located in the central and northern North Sea, including the high
Chl a regions around the Shetland and Orkney islands. The largest PFT,
micro-phytoplankton, were represented by the pigments fucoxanthin and
peridinin. The distribution of this group showed highest abundance in the
southern North Sea high Chl a regions near the continental coast, but also
in location “G” (2011) and between “C” and “D” in 2010.
The combination of continuous underway logging with autonomous instruments,
high-precision pigment measurements at selected stations, and good satellite
Earth observation coverage allowed the patterns of surface phytoplankton
biomass and PFT distribution in the North Sea to be well understood.
Together, this provided a solid observational base with which to test
biogeochemical model accuracy.
Model validation – domain scale
This section presents validation of physical and biogeochemical model
variables against a range of observation-based products, in order to assess
the models' skill at broader scales than the IBTS observations measured.
Detailed validation of phytoplankton community structure against the IBTS
observations follows in Sect. 4.3. Since the focus of this study is on the
phytoplankton community structure in August 2010 and 2011, most of the
validation presented here is for these 2 months. For more general model
validation the reader is referred to Edwards et al. (2012) and Wakelin et
al. (2015b) for NEMO-ERSEM, and Lenhart et al. (2010), Aldridge et
al. (2012), van Leeuwen et al. (2013), van der Molen et al. (2013), and van
der Molen et al. (2016, 2017) for GETM-ERSEM-BFM in various configurations.
However, some statistical assessment has been performed here for SST,
chlorophyll, and SPM over the period March 2010 to October 2011. Two seasons
have been defined for this assessment: the growing season and winter. The
growing season is defined as March to October, and is averaged over 2010 and
2011. Winter is defined as November 2010 to February 2011. Statistics have
been calculated in observation space by performing a bilinear interpolation
of the daily mean model fields to the observation locations. Calculations
have been performed for log10(chlorophyll) rather than for chlorophyll
in order to provide a more Gaussian distribution (Campbell, 1995).
Taylor plots (Taylor, 2001) of SST, log10(chlorophyll), and SPM are shown
in Fig. 4. SST is a good match for the observations in both the growing
season and in winter, although lower correlations are found for both models
in August 2010 and 2011 than for the whole seasons. Slightly better
statistics are obtained for NEMO-ERSEM than for GETM-ERSEM-BFM, reflecting
the assimilation of SST data into NEMO-ERSEM. The statistics for
log10(chlorophyll) differ more between models and between seasons, and
the models are not as good a match for the observations than with SST, as is
common in physical–biogeochemical models. With SPM, the two models show large
differences in variability. GETM-ERSEM-BFM has a much higher standard
deviation than the observations in both seasons, especially the growing
season, whilst the standard deviation of NEMO-ERSEM is too low all year
round.
Taylor plot of SST (purple), log10(chlorophyll) (green), and
SPM (orange) for the growing season (circles), winter (squares), August 2010
(upwards triangles), and August 2011 (downwards triangles). Filled symbols
are GETM-ERSEM-BFM, unfilled symbols are NEMO-ERSEM. Validation has been
performed in observation space, validating the models against OSTIA for SST,
and OC5 ocean colour data for log10(chlorophyll) and SPM. A perfect
model would plot at 1.0 on the x axis, marked with a black star.
![]()
Maps of mean SST for August 2010 and 2011, the months during which most of
the IBTS observations were collected, are plotted in Fig. 5, from
GETM-ERSEM-BFM, NEMO-ERSEM, and OSTIA. There is a great deal of similarity
between NEMO-ERSEM and OSTIA, which is unsurprising since NEMO-ERSEM
assimilates SST data, but both NEMO-ERSEM and GETM-ERSEM-BFM are able to
simulate the spatial features seen in OSTIA, as well as the inter-annual
variability between 2010 and 2011. Boundary effects can be seen in
GETM-ERSEM-BFM, which has a smaller domain.
Maps of mean sea surface salinity (SSS) for August 2010 from GETM-ERSEM-BFM
and NEMO-ERSEM are plotted in Fig. 6. Overlaid in circles are the in situ SSS
observations from the 2010 IBTS cruise. Both models show a good qualitative
match for the observations in most regions. The only area which differs
substantially between the models is the Norwegian Trench and surrounding area
in the northeast North Sea. In GETM-ERSEM-BFM the Norwegian coastal current
disperses erroneously, spreading freshwater into the North Sea. This is not
seen in the observations, and is an issue of model resolution:
finer-resolution configurations of GETM do not suffer from this. The
coarseness of the resolution also accounts for the Rhine freshwater plume
being wider than observed.
Maps of monthly mean SST for August 2010 (a–c) and
August 2011 (d–f): observational data (a, d),
GETM-ERSEM-BFM (b, e), and NEMO-ERSEM (c, f).
![]()
Maps of monthly mean SSS for August 2010 from
(a) GETM-ERSEM-BFM and (b) NEMO-ERSEM. The in situ IBTS
observations from August 2010 are overlaid in circles.
![]()
Maps for sea surface chlorophyll concentration are plotted in Fig. 7, from
the two models and the OC5 products. Daily ocean colour coverage is
incomplete due to cloud cover, so the observations plotted here are simply a
composite of all observations available during the month, rather than a true
monthly mean. To ensure a fair comparison, the daily mean model fields were
bilinearly interpolated to observation locations, and equivalent composites
plotted rather than the true model mean. Van der Molen et al. (2017)
presented a comparison of sub-sampled model results, accounting for cloud
cover, of SPM with the true model mean, which suggested noticeable
differences in winter, but only small differences in summer. The match
between the models and the observations is not as good for chlorophyll as for
SST, but both models were still able to capture some of the observed
features. In the central and northern North Sea, which has the lowest
chlorophyll concentrations, values were generally under-estimated by
GETM-ERSEM-BFM and over-estimated by NEMO-ERSEM. High coastal chlorophyll
values were better simulated by GETM-ERSEM-BFM, whilst the Norwegian Trench
is better represented by NEMO-ERSEM. GETM-ERSEM-BFM has more spatial
variability than NEMO-ERSEM, despite having a lower model resolution. As with
SST, there is notable inter-annual variability in the observations, with
higher chlorophyll concentrations in 2011 than 2010. Both models captured
this variability, although it is less evident in GETM-ERSEM-BFM, and
over-pronounced in NEMO-ERSEM.
Composites of sea surface chlorophyll at ocean colour observation
points for August 2010 (a–c) and August 2011 (d–f):
satellite observations (a, d), GETM-ERSEM-BFM (b, e), and
NEMO-ERSEM (c, f).
![]()
The models are similarly compared to the OC5 SPM products in Fig. 8.
NEMO-ERSEM and GETM-ERSEM-BFM both under-estimate SPM in the central and
northern North Sea, with NEMO-ERSEM also under-representing the plume of SPM
off southeast England. Overall, the two models give very different results
for SPM, and the reasons for and potential consequences of this are discussed
in Sect. 5.
Composites of sea surface SPM at ocean colour observation points for
August 2010 (a–c) and August 2011 (d–f): satellite
observations (a, d), GETM-ERSEM-BFM (b, e), and
NEMO-ERSEM (c, f).
![]()
Maps of mean surface nitrate, phosphate, and silicate for each model for
August 2010 are shown in Fig. 9, alongside the corresponding World Ocean
Atlas climatology fields. Only 2010 is plotted because very similar patterns
are seen in the models for both years, and the climatologies do not include
inter-annual variability. It should also be noted that the climatologies are
of relatively coarse 1∘ resolution, so provide only a basic
representation of the North Sea, but are the only source of data with full
spatial coverage available for such a comparison. The climatologies have been
used as boundary conditions by both models, so are not strictly independent,
but values within the North Sea domain have not been used as input to the
models. For nitrate, the main limiting nutrient in the North Sea,
GETM-ERSEM-BFM shows high coastal concentrations, and very low concentrations
elsewhere. NEMO-ERSEM has a similar pattern, but with a much less extreme
range of values. Likewise for phosphate and silicate, NEMO-ERSEM and
GETM-ERSEM-BFM show differing distributions to each other, and match some of
the climatological features but not others. Overlaid on the maps are in situ
surface nutrient observations sampled on the 2010 IBTS cruise. These show
near-depletion of nitrate and phosphate across most of the North Sea. The
depletion of nitrate is captured well by GETM-ERSEM-BFM, but is not seen to
the same extent in either NEMO-ERSEM or the climatological World Ocean Atlas
fields. The in situ observations also show greater depletion of phosphate
than either of the models or the climatology, but the models are a better
match for the silicate observations.
Maps of monthly mean sea surface nitrate (a–c),
phosphate (d–f), and silicate (g–i) for August 2010:
World Ocean Atlas climatology (a, d, g), GETM-ERSEM-BFM (b, e, h), and NEMO-ERSEM (c, f, i). The in situ IBTS surface
observations from August 2010 are overlaid in circles.
![]()
Statistical comparison of log10(chlorophyll) and phytoplankton
size classes against IBTS observations.
Year
Model
Variable
Bias
RMSE
Correlation
No. observations
2010
GETM-ERSEM-BFM
log10(chlorophyll) (log10 (mg m-3))
0.369
0.597
0.334
46
Micro (fraction)
0.069
0.285
0.010
Nano (fraction)
-0.006
0.170
0.116
Pico (fraction)
-0.062
0.182
-0.265
NEMO-ERSEM
log10(chlorophyll) (log10 (mg m-3))
0.017
0.339
0.446
Micro (fraction)
-0.385
0.434
0.160
Nano (fraction)
0.146
0.165
0.604
Pico (fraction)
0.240
0.292
-0.369
2011
GETM-ERSEM-BFM
log10(chlorophyll) (log10 (mg m-3))
0.178
0.543
0.343
39
Micro (fraction)
0.138
0.261
0.265
Nano (fraction)
-0.076
0.134
0.387
Pico (fraction)
-0.062
0.178
-0.061
NEMO-ERSEM
log10(chlorophyll) (log10 (mg m-3))
0.067
0.389
0.157
Micro (fraction)
-0.270
0.382
-0.416
Nano (fraction)
0.063
0.121
0.359
Pico (fraction)
0.207
0.299
-0.530
2010 + 2011
GETM-ERSEM-BFM
log10(chlorophyll) (log10 (mg m-3))
0.282
0.573
0.320
85
Micro (fraction)
0.101
0.274
0.097
Nano (fraction)
-0.039
0.154
0.178
Pico (fraction)
-0.062
0.180
-0.163
NEMO-ERSEM
log10(chlorophyll) (log10 (mg m-3))
0.040
0.363
0.358
Micro (fraction)
-0.332
0.411
-0.210
Nano (fraction)
0.108
0.146
0.401
Pico (fraction)
0.225
0.295
-0.452
Model validation against PFT observations
This section presents validation of the models against the chlorophyll and
PFT observations collected on the IBTS cruises. Hereafter, Micro is used to
refer to the fraction of total chlorophyll represented by the
micro-phytoplankton size class, and similarly Nano and Pico. Model PFTs were
aggregated as described in Sect. 3.4. The most northerly of the IBTS
observations were located outside the GETM-ERSEM-BFM model domain; these have
been excluded from the assessment to ensure the same observations were used
to validate both models.
Maps of mean surface model PFTs for August 2010, as fractions of total
chlorophyll, are plotted in Fig. 10. These show very different distributions
for NEMO-ERSEM and GETM-ERSEM-BFM, much more so than the difference in total
chlorophyll might suggest. In particular, NEMO-ERSEM shows dominance by
pico-phytoplankton in the southern North Sea, similar fractions of
pico-phytoplankton and flagellates in the rest of the domain, and generally
low concentrations of diatoms and dinoflagellates. In contrast,
GETM-ERSEM-BFM shows dominance by diatoms in coastal regions, and by
pico-phytoplankton in the centre of the domain, with generally lower
fractions of the remaining PFTs. The two PFTs unique to GETM-ERSEM-BFM,
resuspended benthic diatoms and Phaeocystis colonies, only show
notable concentrations in certain coastal areas. The reasons for the
differences between the two models are discussed in Sect. 5.
Maps of monthly mean sea surface PFT fractions for August 2010 from
NEMO-ERSEM (a–d) and GETM-ERSEM-BFM (e–j). PFT fractions
have been calculated as the proportion of the total sea surface chlorophyll
concentration.
![]()
The bias, root mean square error (RMSE), and correlation of modelled versus
observed total chlorophyll concentration are shown in Table 3. GETM-ERSEM-BFM
has a slight bias towards too-high chlorophyll, whilst the bias for
NEMO-ERSEM is near-zero. GETM-ERSEM-BFM chlorophyll values are typically
higher than those from NEMO-ERSEM, but also show a greater range. These
features are consistent between the two years. However, whilst GETM-ERSEM-BFM
has a similar correlation value for both years, the correlation for
NEMO-ERSEM is much higher in 2010 than 2011. It should be noted though that
these statistics are based on a relatively small number of points, so any
conclusions drawn from this comparison are not guaranteed to be robust,
particularly given the domain-scale spatial variability (see Sect. 4.2 and
Fig. 7).
A comparison of phytoplankton community structure in the models and IBTS
observations has been made by aggregating the model PFTs into the three
observed PSCs, as described in Sect. 3.4. The bias, RMSE, and correlation of
the modelled versus observed total PSCs are shown in Table 3. Bias and RMSE
are generally lower for GETM-ERSEM-BFM than for NEMO-ERSEM, particularly for
Micro and Pico. NEMO-ERSEM has higher absolute correlations, but for Micro
and Pico these tend to be negative, suggesting that NEMO-ERSEM is getting
Nano approximately correct, but Micro and Pico are inversely distributed
compared with the observations.
The distribution of relative PSC fractions with total chlorophyll is plotted
for each data set in Fig. 11. Consistent with results from previous studies
(e.g. Devred et al., 2011), as observed chlorophyll increases, Micro tends to
increase, and Nano and Pico decrease. This pattern is also seen to some
extent in GETM-ERSEM-BFM, but less so in NEMO-ERSEM (and only in 2010),
although NEMO-ERSEM has a smaller range of chlorophyll concentrations. In the
IBTS data there is a clear overall dominance of Micro. This is well
reproduced by GETM-ERSEM-BFM, but the opposite is found in NEMO-ERSEM. The
exception to this is a group of observations at low chlorophyll
concentrations, most notably in 2011, in which Micro is least abundant,
better matching typical NEMO-ERSEM results. These observations were all taken
in the central North Sea, and this behaviour is discussed further in Sect. 5.
To explore the model behaviour further, and allow comparison with other works
such as de Mora et al. (2016), histograms of the distribution of relative PSC
fractions with total chlorophyll are plotted in Fig. 12, from each model grid
point in the North Sea. These have used the mean model fields for August 2010
and August 2011. With this extended number of model points, a clear
relationship is seen for GETM-ERSEM-BFM, with Micro increasing with total
chlorophyll, and Pico decreasing. This matches the trend seen in the IBTS
observations, as well as previous studies (e.g. Brewin et al., 2010; Devred
et al., 2011). For NEMO-ERSEM, the range of chlorophyll concentrations
remains small, making any relationship difficult to assess. When there are
higher chlorophyll values a similar pattern of increasing Micro and
decreasing Pico is seen, but there are too few points to draw a robust
conclusion on the model relationship.
Three variables which always sum to one can be displayed in a single space,
barycentric coordinates, using a ternary plot (e.g. Jupp et al., 2012).
Phytoplankton community structure is plotted this way in Fig. 13. The
observations form a distinct line in this space, from the centre of the plot
to the corner representing dominance by Micro. At lower chlorophyll
concentrations (not shown in Fig. 13, but consistent with Fig. 11) there are
roughly equal fractions of Micro, Nano, and Pico. As chlorophyll concentration
increases, Nano and Pico decrease in roughly equal amounts, with Micro
increasing accordingly. The fact that the observations form a line in this
space shows Nano and Pico to change in roughly equal proportions when Micro
changes with chlorophyll, which differs to some extent from other studies
such as Brewin et al. (2010). GETM-ERSEM-BFM displays a similar pattern, with
values in the same area of the plot as the observations, although with a less
distinct relationship. NEMO-ERSEM values show a very different distribution
however. There is some overlap with the observations in 2011, but otherwise a
much less Micro-dominated regime is evident.
Phytoplankton size class (PSC) distribution as a function of
chlorophyll concentration, plotted at each IBTS observation location in
2010 (a, c, e) and 2011 (b, d, f) from the IBTS
observations (a, b), GETM-ERSEM-BFM (c, d), and
NEMO-ERSEM (e, f).
![]()
The ternary plot can also be used as a colour key to produce a map of
phytoplankton community structure, as in Fig. 14. This plots the August mean
community structure for each model and each year, overlaid with the IBTS
point observations in circles. Plotting the community structure in such a
fashion demonstrates that whilst GETM-ERSEM-BFM and NEMO-ERSEM give very
different results in terms of the magnitudes of the PSC fractions, there are
nonetheless some broadly consistent features in terms of spatial patterns,
which are also evident to some extent in the observations. For instance, both
models (although NEMO-ERSEM less so in 2010) show a distinct split in
community structure between the southern and northern North Sea, and around
bathymetric features such as Dogger Bank and coastlines. Such a split can be
clearly seen in the 2011 observations, which show very little variation
throughout the central North Sea, but is less clear in the 2010 observations.
A difference between the years in the community structure in the central
North Sea is also seen in NEMO-ERSEM, and to a lesser extent GETM-ERSEM-BFM,
although the direction of change in the models is from Pico-dominated to
Micro-dominated, the opposite of that in the observations. Although in most
cases the community structure in the models does not match that of the
observations, GETM-ERSEM-BFM is a very good match for the observations in the
southern North Sea, an area particularly dominated by diatoms in the model.
Silicate in this region is near-depleted in GETM-ERSEM-BFM (see Fig. 9), but
abundant in NEMO-ERSEM. In GETM-ERSEM-BFM, distinct blue patches can be seen
off East Anglia, South Dorset, and the German Bight, which are mostly areas
where Phaeocystis colonies dominate in the model.
2-D histograms of the distribution of each phytoplankton size class
(PSC) with chlorophyll, from NEMO-ERSEM (a, c, e) and
GETM-ERSEM-BFM (b, d, f). The PSC fractions have been calculated at
each surface model grid point in the monthly means for August 2010 and
August 2011. Colours represent the number of occurrences per bin.
![]()
Summary and discussion
This study has presented a new set of in situ phytoplankton pigment
observations for the North Sea, processed to give information on
phytoplankton community structure. Two physical–biogeochemical models, the
biogeochemical components of which are different variants of ERSEM, were then
validated against these and other observations. Both models were a good match
for SST observations, and a reasonable match for chlorophyll observations,
but gave contrasting results for SPM. Furthermore, the two models displayed
very different phytoplankton community structures. GETM-ERSEM-BFM was able to
reproduce many of the features of the observations, particularly in the
southern North Sea, whereas NEMO-ERSEM was a poor match for the observations,
except at the lowest chlorophyll concentrations. Nonetheless, both models
shared some similarities with each other and the observations in terms of
spatial features and inter-annual variability.
Ternary plots of phytoplankton community structure, showing the
phytoplankton size class distribution at IBTS observation locations in
(a) 2010 and (b) 2011, from the IBTS observations,
GETM-ERSEM-BFM, and NEMO-ERSEM.
![]()
Maps of modelled surface phytoplankton community structure for
August 2010 (a, b) and August 2011 (c, d) from
NEMO-ERSEM (a, c) and GETM-ERSEM-BFM (b, d). The IBTS
observations, sampled in August and early September of each year, are
overlaid in circles.
![]()
The distribution of total phytoplankton biomass across the North Sea during
summer of both years showed a high degree of consistency between three
different observational methods: satellite remote sensing, high-frequency
continuous measurement of chlorophyll fluorescence, and Chl a
quantification at discrete stations. A similar set of spatial features can be
observed in 2010 and 2011, which can be explained by the underlying
hydrodynamics (van Leeuwen et al., 2015). The central, strongly stratified
region of the North Sea has very low nutrient concentrations and
correspondingly low Chl a. In areas where vertical mixing, riverine input,
or horizontal advection bring nutrients into the upper water column,
phytoplankton biomass is elevated. As well as observing the total quantity of
phytoplankton, deriving the composition of size and functional types is
important for a better understanding of ecosystem function and energy flows
to higher trophic levels (Chavez et al., 2011). Accessory pigments have been
widely used in biological oceanography to investigate community composition,
but caution must be applied when interpreting results, and support from other
methods should be used where possible (Schlüter et al., 2014). The
original equations used by Hirata et al. (2008, 2011) to convert pigments to
pico-, nano-, and micro-phytoplankton size classes underestimated the
fraction of pico-phytoplankton compared to flow cytometric observations, and
were modified by increasing the contribution of Chl b and 19-hex to this
class. Results in both years showed a consistent pattern of decreasing
micro-phytoplankton abundance with distance from the coast, and with
increasing latitude, and this is supported by previous pigment-based studies.
Work in the German Bight has also shown a change from a coastal,
diatom-dominated community to a more diverse, small-celled community further
offshore (Brandsma et al., 2013; Wollschläger et al., 2015).
Both NEMO-ERSEM and GETM-ERSEM-BFM have shown the ability to reproduce the
physics and broad-scale biogeochemistry of the North Sea. However, results
are more varied when considering specific aspects such as the phytoplankton
community structure in August. In some ways this is to be expected, as this
study has used existing model versions which have not been previously
validated against or tuned to such observations. Furthermore, August is a
challenging month to model in the North Sea, as evidenced by the reduced SST
skill for this month compared with the seasonal average. This is because
simulating the details of the stratification, nutrient concentrations and
therefore phytoplankton concentrations is dependent on having successfully
simulated processes in previous months, as well as the processes seen during
August. This is more important than at other times of year, because, being at
the height of summer, there are no strong temporal gradients driving the
response of the system, as in spring and autumn. As a result, the internal
biogeochemical dynamics can play out most freely (both in reality and in the
model). Nonetheless, this kind of specific information is in increasing
demand, and if results are to be provided to users then they should be
understood and validated.
NEMO-ERSEM and GETM-ERSEM-BFM gave very different representations of SPM
concentrations, which impacts ecosystem functioning through light limitation
(see also the light susceptibility parameters, Table 2). NEMO-ERSEM uses the
two-size class SPM model described by Sykes and Barciela (2012). This was
implemented by Sykes and Barciela (2012) in the POLCOMS (Proudman
Oceanographic Laboratory Coastal Ocean Modelling System; Holt and James,
2001; Holt et al., 2005) physical framework, in which it gave skilful results
compared with observations. However, the model has not yet received the same
degree of tuning and development since being implemented in NEMO-ERSEM, which
may explain the consistent under-estimation of SPM found in this study.
Furthermore, the high vertical resolution of NEMO-ERSEM means that the
settling velocities must sometimes be artificially limited when used by the
SPM model, in order to avoid breaking the Courant–Friedrichs–Lewy (CFL)
condition (Courant et al., 1928), thus reducing resuspension. Changes to the
settling parameters would be expected to lead to improvements. GETM-ERSEM-BFM
uses an alternative SPM model (van der Molen et al., 2017), which only has
one size class but includes resuspension by waves as well as currents, and
which was developed within the GETM-ERSEM-BFM framework. This generally
matches spatial distributions of SPM better, but often has concentrations
which are extremely high or low compared with satellite data, leading to a
degradation in some error statistics.
The starkest contrast between the model results presented in this study is in
the simulated phytoplankton community structures, which differ far more than
might be expected given the corresponding total chlorophyll concentrations.
GETM-ERSEM-BFM gave a wider range of combinations of biomass in the three
size classes resolved by the observations than NEMO-ERSEM did. This was also
reflected in more spatial variability and stronger spatial gradients, which
resulted in a better match of the coastal to offshore change in phytoplankton
community structure evident in the observations, in which diatoms are
particularly important. The limited biomass in the two additional PFTs in
GETM-ERSEM-BFM (benthic diatoms and Phaeocystis colonies, Fig. 10)
suggests that these were not the primary cause of this difference in
response. Two mechanisms are likely to play important roles in causing the
differences: (1) the higher coastal nutrient concentrations in combination
with the different nutrient affinity settings in GETM-ERSEM-BFM, allowing
diatoms to out-compete other types, in contrast with identical nutrient
affinity settings in NEMO-ERSEM; (2) the coincidence of diatoms with areas of
high SPM concentrations in GETM-ERSEM-BFM (mostly absent in NEMO-ERSEM) in
combination with greater light susceptibility of diatoms (again contrasting
with uniform values in NEMO-ERSEM), giving them competitive advantage.
Similar, but more subtle effects will modulate the response of the other PFTs
in GETM-ERSEM-BFM. Overall, the more uniform parameter settings of NEMO-ERSEM
promote a more uniform response of the PFTs, as is evident in the results.
Further work, with dedicated experiments performed in a configuration
designed for such a comparison, could be carried out to further investigate
these differences, and this is discussed below.
GETM-ERSEM-BFM provided a better match for the IBTS observations of
phytoplankton community structure than NEMO-ERSEM did. The exception to this
was in the low-chlorophyll waters of the central North Sea, the region of the
domain with the weakest currents and largest residence times. Here the
community structure of the observations more closely resembled that typical
of NEMO-ERSEM. There are indications that recent versions of NEMO-ERSEM,
applied to the global ocean rather than to the Northwest European Shelf
seas, perform better at reproducing observed community structures (de Mora et
al., 2016). Together with the results presented here, this suggests that
NEMO-ERSEM may be more representative of an open ocean environment, whereas
the settings in GETM-ERSEM-BFM are better suited to the complex coastal
environment of the North Sea.
Whilst there were large contrasts in the corresponding ratios of PSCs between
all three data sets, there was more agreement between the data sets about the
spatial patterns of community structure. For instance, each had contrasting
structures between the southern and central North Sea, and in coastal areas.
Furthermore, inter-annual variability in the central North Sea was clearly
evident in the observations, and also each of the models. This can be
compared to differences in SST between the two years and the two models,
suggesting a set of physical drivers which the models were able to capture.
Across much of the North Sea the SST was cooler in 2011 than 2010, as seen in
Fig. 5. Typical mixed layer depths were also deeper in the models in 2011
(not shown), likely in response to increased wind speeds. This deepening of
the mixed layer cooled the SST and brought more nutrients to the surface,
increasing production and chlorophyll, as seen in Fig. 7. In NEMO-ERSEM,
which showed the most inter-annual variability in phytoplankton community
structure, all PFTs increased in chlorophyll, but dinoflagellates increased
the most in percentage terms, shifting the community structure more towards
Micro-dominance. This could be because the changes in temperature and prey
availability favoured smaller predators more than larger ones, and an
increase in the ratio of nitrate to ammonium best suited larger
phytoplankton. This implies that even if models are currently unable to
accurately represent the exact community structures, they can still be used
to assess the distribution of different habitats, and when and where these
may change.
Due to differences in locations, timescales, and data sets, a robust comparison
with results from other PFT modelling studies in the literature cannot be
made at this stage. Nonetheless, a consideration of the types of results
obtained using different modelling approaches is of value. Lewis et
al. (2006) and Lewis and Allen (2009) both validated a similar version of the
PML-developed ERSEM, coupled with the POLCOMS hydrodynamic model, against in
situ observations of PFTs in the Northwest European Shelf. Lewis et
al. (2006) validated against Continuous Plankton Recorder (CPR; Richardson et
al., 2006) data, and found that whilst the model reproduced the main seasonal
features of plankton succession, diatoms consistently bloomed too early, and
there were some spatial differences, especially in the North Sea. This would
seem consistent with the low diatoms seen in NEMO-ERSEM in August in this
study. Lewis and Allen (2009) validated against a station in the Western
English Channel, and found a poor match for phytoplankton variables when
assessment was performed in observation space, as has been done here, but a
better match when assessing more general seasonal variability. On a global
open-ocean scale, Gregg and Casey (2007) validated the NASA Ocean
Biogeochemical Model (NOBM) against both in situ and remotely sensed
estimates of PFTs, with a focus on coccolithophores, and described their
results as “mixed”, noting contrasting results elsewhere in the literature.
Hirata et al. (2013) validated the MEM-OU model against remote sensing
estimates of large and small phytoplankton, and found good agreement at basin
scales, but which reduced at smaller scales. De Mora et al. (2016) compared
PFT distributions with chlorophyll from the most recent ERSEM version of
Butenschön et al. (2016) to those obtained using various remote sensing
algorithms applied to the model data, and found that the model displayed similar
properties at global scales. These studies (not exhaustive) all used models
with a relatively small fixed number of specifically parameterized PFTs. An
alternative approach is to initialize the model with a large number of
randomly parameterized PFTs, with the best-suited PFTs naturally dominating
(Follows et al., 2007). In a global setting this approach has been found to
successfully reproduce large-scale patterns of phytoplankton diversity in
terms of organism size (Follows et al., 2007) and number of co-existing
species (Barton et al., 2010). The ability of such a model to reproduce the
short-term variability of a complex coastal environment such as the North Sea
would make for an interesting future inter-comparison. In general, results in
the literature suggest some success of different approaches at reproducing
large-scale patterns of phytoplankton community structure, but with more
detailed skill yet to be properly demonstrated.
Careful thought needs to be given therefore to what products and information
can be offered to users, which address user and policy requirements with a
sufficient level of skill (Hyder et al., 2015). Continual model developments
will be required. A comprehensive review of the challenges faced is given by
Holt et al. (2014), and results from this current study should further inform
future model development. This will be particularly relevant in the context
of the UK Shelf Seas Biogeochemistry (SSB) research programme
(http://www.uk-ssb.org), in which Cefas and the Met Office are both
participants. One of the aims of the SSB programme is to create a common
version of ERSEM to be used by the UK research community, by combining
features of the two versions of ERSEM used in this study. An initial combined
version is described by Butenschön et al. (2016), and this will be
further developed within the SSB programme. The version of Butenschön et
al. (2016) provides a major update to that of Blackford et al. (2004), which
forms the basis of the NEMO-ERSEM version used in this study, and initial
applications (Butenschön et al., 2016; de Mora et al., 2016; Ciavatta et
al., 2016) have shown different phytoplankton community structures to that
obtained in this work. It is clear from this study that the details of the
model components and parameterizations can lead to very different results,
and validation against a range of data, using a range of methods, is vital
throughout the model development cycle.
The assessment presented in this study suggests that the biogeochemical model
parameterizations are important in controlling the phytoplankton community
structure. However, due to the many differences in physical modelling
environment and experimental configurations, it cannot be ruled out that
differences in the physics are responsible for the differences in
phytoplankton community structure, and previous studies have found these to
be important. For instance, Sinha et al. (2010) coupled a single marine
biogeochemical model with two different global physical ocean models, and
found contrasting phytoplankton community structures due to differences in
mixing. A number of further experiments could be performed to investigate the
differences in PFT response between the models, and develop improvements.
These can make use of the two ERSEM versions used in this study and the new
SSB-ERSEM, along with NEMO, GETM, and GETM's 1-D counterpart GOTM (General
Ocean Turbulence Model). SSB-ERSEM is compatible with the Framework for
Aquatic Biogeochemical Models (FABM; Bruggeman and Bolding, 2014), allowing
it to be coupled with either GOTM, GETM, or NEMO. Building on this framework
could allow the different ERSEM versions to be run with identical
hydrodynamics in 1-D and 3-D, and similarly allow individual ERSEM versions
to be run with different hydrodynamics. This would help identify the
differences in results introduced by different components of the system.
Within this experimental framework, controlled differences to physical and
biogeochemical model parameters could be made to investigate further. These
would allow very efficient testing. However, the current field observations
(quasi-synoptic spatial distribution for 1 month of the year) do not allow
a fully robust model comparison. An observational time series resolving the
seasonal cycle at one or more locations would be needed for this exercise.
Recent developments of algorithms to derive PFTs from remotely sensed data
(Nair et al., 2008; Hirata et al., 2013) could benefit all these potential
strands of further work.
A further development will be the assimilation of biogeochemical data. Ocean
colour data assimilation is being increasingly utilized by the reanalysis and
forecasting community (Gehlen et al., 2015), and has already been
successfully demonstrated for ERSEM (Ciavatta et al., 2011, 2014, 2016). A
suitable ocean colour assimilation scheme for operational purposes is being
developed as a collaboration between the Met Office and PML, to be
implemented in the SSB ERSEM version and run operationally as part of CMEMS.
This will also give the opportunity to take advantage of the advent of remote-sensing PFT/PSC products, incorporating such data into the assimilation and
routine validation.
Information on the marine environment can come from three sources: in situ
observations, remote-sensing data, and models. These three sources are
inter-linked and all are vital; sufficient scientific understanding of the
North Sea and other environments cannot be obtained if any of these three
sources are removed. Models provide complete 3-D spatial and temporal
coverage, can be used to simulate a range of hypotheses, and are relatively
inexpensive. However, as demonstrated in this study, observations are
necessary for the validation and development of models, and model data cannot
be relied upon in isolation. Remote-sensing data provide considerably greater
observational coverage than in situ measurements, but this coverage is still
limited to the sea surface and cloud-free conditions, and empirical
algorithms based on in situ data are used in the construction of remote
sensing products. These satellite data must be comprehensively ground-truthed
against in situ observations if they are to be used with confidence.
Continuing in situ observations are therefore required to under-pin model and
remote sensing data, as well as to provide unique insights into the marine
environment. In turn, modelling studies can be used to help inform sampling
strategies of future observing programmes, to help provide value for money
without sacrificing accurate scientific understanding.
Finally, it remains to answer the question at the heart of this paper: “Can
ERSEM-type models simulate phytoplankton community structure?” The evidence
from this study suggests that ERSEM-type models have the potential to
accurately simulate phytoplankton community structure, but certain model
formulations and parameterizations are required to do so, and these two ERSEM
versions do not reliably do so at this stage. Appropriate model development,
informed by detailed validation studies, appears to be a major but achievable
challenge, and will help facilitate the wider application of marine
biogeochemical modelling to wide-ranging user and policy needs.