Introduction
Modeling the biogeochemistry of coastal and shelf systems requires the
representation of a multitude of interacting processes, not only within the
water but also in the adjacent earth system components such as the atmosphere
(e.g., nitrogen (N) deposition), land (e.g., riverine inputs), sediment (e.g.,
diagenetic processes) and biochemical processes in water see.,
e.g.,. For being able to reproduce the large-scale
spatial and temporal distribution of biogeochemical variables in coastal
systems, a realistic representation of hydrodynamical processes is often
critically important, at least those relevant to the circulation patterns and
stratification dynamics: the former is needed to describe the spread of
nutrient-rich river plumes and exchange at the open ocean boundaries, and the
latter for being able to capture the vertical gradients in the light and
nutrient conditions for primary productivity. The representation of biological
processes and the two-way interactions between biological, chemical and
benthic compartments in models is particularly challenging, given the
complexity of physiological processes displayed by individual organisms,
e.g., regarding the regulation of their internal stoichiometries e.g.,
see and the differences in functional traits of species
constituting communities e.g., see.
Three-dimensional ecosystem models often describe the processes relevant to primary
production, e.g., the nutrient and light limitation of phytoplankton (B), using
heuristic formulations that have been shown to be inadequate in reproducing
patterns obtained in laboratory experiments. For instance, light limitation
is determined not only by the instantaneously available irradiance, but also
by the amount of light-harvesting apparatus, i.e., chlorophyll (Chl) pigments
maintained by the phytoplankton cells, which can change considerably through
a process referred to as photoacclimation. However, photoacclimation is often
completely ignored in 3-D model applications, or its effects are mimicked
heuristically, for instance, by describing the chlorophyll-to-carbon ratio as
a function of irradiance , which cannot
capture the dependence of chlorophyll synthesis on nutrient availability
e.g.,. Similarly, the interaction of
limitation by different nutrient elements is described by heuristic
formulations, dichotomously either by a product rule or a threshold function,
which, again, cannot reproduce complex patterns observed in laboratory
conditions, such as the asymmetric cellular N : C and P : C ratios
emerging under N- and P-limited conditions .
Such simplifications in the description of primary production processes, in
turn, potentially lead to flawed representations of nutrient cycling. Despite
the recently revived theoretical work on stoichiometric regulation and
photoacclimation
e.g.,,
an implementation of a model with a mechanistic description of the regulation
of phytoplankton composition at a full ecosystem scale in a coupled
physical–biological modeling framework remains lacking. In this study,
we therefore present a 3-D application of the Model for Adaptive Ecosystems
for Coastal Seas (hereafter MAECS), to the southern North Sea (SNS), for a
decadal hindcast simulation. MAECS features a photoacclimative autotrophic
growth model that has been recently introduced by , which
resolves the regulation of the stoichiometry and composition of autotrophs
employing an innovative suit of adaptive and optimality based approaches.
The SNS is part of a shallow shelf system (Fig. ). The
southeastern portion of the SNS, known as the German Bight surrounded by the
intertidal Wadden Sea, is especially characterized by steep gradients with respect to
both nutrients and turbidity. The latter is
largely determined by suspended particulate matter (SPM) concentrations
. These gradients are driven by a complex interplay of
riverine and atmospheric fluxes, complex topography, residual tidal currents,
density gradients, biological processing of organic matter (OM), benthic–pelagic
coupling and sedimentation–resuspension dynamics
.
A number of modeling studies previously addressed the biogeochemistry of the
North Sea, including the German Bight. In a majority of these studies, such
as ECOHAM-HAMSOM , NORWECOM ,
ECOSMO-HAMSOM , HAMOCC-MPIOM , ERSEM-NEMO
, ERSEM-POLCOMS
and ERSEM-BFM-GETM , large domains and
relatively coarse grids were employed (≥ 7 km). While showing good
skill in reproducing offshore dynamics, these models seemed to have a
relatively limited performance at the shallow, near-coast regions (when
reported). The BLOOM-Delft3D , however, is one of the rare
examples with a finer grid (down to 1 km at the Dutch coasts) – at the cost of
a relatively smaller domain, similar to ours. Although this model system
performs decently at both coastal and offshore areas, its performance within
the German Bight has not been fully assessed. Moreover, none of these models
provide elaborate descriptions of the stoichiometric regulation of
autotrophs, as mentioned above. Therefore, our new model system is expected
to fill two important gaps by
exemplifying for the first time, to the best of our knowledge, the implementation of
a highly complex phytoplankton growth model at an ecosystem scale, coupled to a hydrodynamic
model and other biogeochemical compartments, and gaining some first insight into the relevance of acclimation to the modeling of coastal
biogeochemistry; and
establishing the capacity to reproduce the biogeochemistry of the German Bight both at
coastal and offshore regions with a single parameterization and model setup.
For an 11-year hindcast simulation of the period 2000–2010, we show that the
model can adequately capture the spatiotemporal variability of the physical
and biogeochemical features of the SNS based on comparisons against various
data sources. Importantly, the model can reproduce the steep chlorophyll and
nutrient gradients prevalently observed across the Wadden Sea–German Bight
continuum. We show that the chlorophyll gradients are linked with nutrient,
and hence productivity, gradients and are further amplified by the acclimation capacity
of phytoplankton, and particularly by the high chlorophyll-to-carbon
ratios at the coastal regions.
Bathymetry of the model domain and the location of rivers considered
in this study. Gray lines display the model grid.
Methods
Observations
Observation data from Helgoland Roads, Sylt and 17 other monitoring stations
reflect surface measurements. Extensive analyses of the data from Helgoland
Roads have been previously performed by and from Sylt
by . Sparse measurements of temperature, salinity, dissolved
inorganic nitrogen (DIN), dissolved
inorganic phosphorus (DIP) and chlorophyll were obtained from the online
database of the International Council for the Exploration of the Sea (ICES,
www.ices.dk).
Continuous Scanfish and FerryBox measurements were performed within the
operation of the Coastal Observing System for Northern and Arctic Seas
COSYNA,.
Data collection, processing and quality
control of the Scanfish data are described by and of the
FerryBox data by . The satellite dataset used here is the
Ocean Colour Climate Change Initiative (CCI), version 3.1, European Space Agency (ESA),
available online at http://www.esa-oceancolour-cci.org/. Chlorophyll
estimates of the satellite product were bias-corrected according to the
product user guide :
Cbc=10log10(C)+δ, where Cbc, C and
δ are, respectively, the bias-corrected, raw and log10 bias
estimates for chlorophyll concentrations.
Model
The major processes taken into account by the model are the lower trophic
food web dynamics, phytoplankton ecophysiology and basic biogeochemical
transformations in the water, and the transformation of N and P species in
the benthos (Fig. and Sect. ). Physical
processes are resolved by the coupled 3-D hydrodynamical model, GETM
(General Estuarine Transport Model; Sect. ). Turbidity caused by suspended particulate matter,
nutrient loading by rivers and atmospheric nitrogen deposition were
considered as model forcing (Sect. ). The model grid and rivers
considered in this study are shown in Fig. .
Structure of the biogeochemical model. Model components (rectangles)
comprise the following. B is phytoplankton, Z is zooplankton, POM and DOM are particulate and
dissolved organic matter, DIM (N, P) is dissolved inorganic matter
(nitrogen and phosphorus), and bAP is the P adsorbed in iron–phosphorus complexes (see
Sect. and Appendix for further details). C, N and P
in small circles refer to carbon, nitrogen and phosphorus bound to each
component, respectively, whereas fLH and fC are the
allocation coefficients for light harvesting and carboxylation
(Sect. ). Boxes in dashed lines indicate model forcing.
Biogeochemical model
The pelagic module, the Model for Adaptive Ecosystems in Coastal Seas
(MAECS), is a lower-trophic-level model that resolves cycling of carbon,
nitrogen and phosphorus, and, importantly, acclimation processes
involved in phytoplankton growth. In MAECS, the acclimation of phytoplankton
is resolved by a scheme recently introduced by , which
describes the instantaneous or transient optimization of physiological
traits, x, by the extended optimality principle:
ddtx=δx⋅[∂VC∂x+∑i∂VC∂qi∂qi∂x],
where δx corresponds to the flexibility of traits
(Eq. A18), i expands to N and P, and the two terms in brackets
describe the direct effects of trait changes on the specific phytoplankton
growth rate VC (in units of cellular C) and the indirect effects
through changes in the Chl : C : N : P stoichiometry, expressed by the
quotas q, respectively. Specifically, three-levels of acclimative
regulations are considered see Fig. 2 in:
Machinery allocation: we describe the changes in allocations to light-harvesting, carbon-fixation and nutrient acquisition machineries,
as also in . These allocations correspond to the synthesis of cellular structures such as chloroplasts for absorbing
light, Rubisco enzyme involved in carboxylation process and proteins for gathering nutrient molecules; therefore, we track these
fractional allocations with two dynamic state variables, fLH and fC, that describe the allocations for light harvesting
and carboxylation, while the allocation for nutrient uptake, fV, is assumed to be the rest, 1-fLH-fC. Here, the flexibility term, set to δx=fx⋅(1-fx), regulates the speed of optimization as determined by the differential terms in Eq. ().
Nutrient affinity-processing optimality: we assume that there is a tradeoff between nutrient affinity
and processing, and the optimal affinity fractions for each nutrient, fiA, are instantaneously optimized,
such that dx/dt=0 and fiA are algebraically found by setting ∂VC∂x=0 .
Nutrient uptake activity: (down-) regulation of the uptake rate of nutrients,
which is often formulated as a linear function of nutrient quotas in traditional models, is in
our approach described by the instantaneously optimized uptake activity trait, ai. Assuming that energy expenditure
for taking up each nutrient depends on the metabolic needs, values of ai are found by scaling their marginal growth benefits (Eq. A17).
Driven by the variations of these physiological traits, Chl : C : N : P
stoichiometry varies continuously depending on ambient light and nutrient
conditions and on the metabolic demands of autotrophic cells. As a further
novel aspect of the acclimation model, multiple limitation is described as a
queuing function, which allows formulating the co-limitation strength as a
function of internal nitrogen reserve, qN, instead of prescribing
it to be either high as by a product rule or low as by a threshold (Liebig)
function . A detailed description of the phytoplankton
growth module can be found in . Equations and parameters of
the model are provided in Appendix .
Other components of the pelagic module are similar to standard descriptions
in state-of-the-art ecosystem models. Phytoplankton take up nutrients in the
form of dissolved inorganic matter (DIM). Losses of phytoplankton (B) and
zooplankton (Z) due to mortality are added to the particulate organic matter
(POM) pool, which degrades into dissolved organic matter (DOM), before
becoming again DIM and closing the cycle (Appendix ). As a
relevant aspect of the model, while the sedimentation speed of POM
(wPOM) is prescribed as a constant value, that of
phytoplankton, wB, is assumed to be modified by its nutrient (quota)
status. As decreased internal nutrient quotas likely affect the cells' ability
to regulate buoyancy and lead to faster migration towards deeper, potentially
nutrient-rich waters , we assume that maximum sinking rates
realized at fully depleted quotas converge to a small background value with
increasing quotas as has been observed especially for, but not limited to,
diatoms . Although the phytoplankton sinking
is often parameterized as a constant rate in 3-D modeling applications,
similar formulations of increasing sinking rates under nutrient stress have
also been used e.g.,.
The benthic module describes only the dynamics of macronutrients N and P.
Degradation of OM to DIM is described as a one-step, first-order reaction.
Denitrification is described as a proportion of POM degradation, limited by
DIN and dissolved oxygen (DO) availability in benthos. As DO is not directly
modeled, it is estimated from temperature in order to mimic the seasonality
of the hypoxia-driven denitrification. The model accounts for the
sorption–desorption dynamics of phosphorus as an instantaneous process and also
as a function of temperature based on the correlation observed in the field
. Further details are provided in Appendix .
Hydrodynamic model and model coupling
The General Estuarine Transport Model was used to calculate various
hydrodynamic processes, as well as the transport of the biogeochemical
variables. A detailed description of GETM is provided by
and . GETM utilizes the turbulence
library of the General Ocean Turbulence Model (GOTM)
to resolve vertical
mixing of density and momentum profiles with a k-ε two equation
model . GETM was run in baroclinic mode, resolving the
3-D dynamics of temperature, salinity and currents and 2-D dynamics of sea
surface elevation and flooding–drying of cells at the Wadden Sea. Following
, we assumed the bottom roughness length to be constant
throughout the domain, and z0=10-3 m. We used 20 terrain-following
layers and a curvilinear grid of 144×98 horizontal cells, providing
a horizontal resolution of approximately 1.5 km at the southeast corner and
4.5 km at the northwest corner (Fig. ). The curvilinear grid
focuses on the German Bight and roughly follows the coastline
(Fig. ) for an optimal representation of along- and across-shore
processes. Similar gridding strategies were applied successfully in other
coastal setups with the GETM model . We
employed integration time steps of 5 and 360 s for the 2-D and 3-D
processes, respectively.
Integration of model forcing was realized through the Modular System for
Shelves and Coasts (MOSSCO, http://www.mossco.de), which, among others,
provides standardized data representations . Meteorological
forcing originated from an hourly resolution hindcast by the limited area model COSMO-CLM
. Boundary conditions for surface elevations are extracted
from an hourly resolution hindcast by TRIM-NP . For
temperature and salinity, daily climatologies from HAMSOM
are used, all of which are available through coastDat
(http://www.coastdat.de).
Two-way coupling of the biological model with GETM was achieved via the
Framework for Aquatic Biogeochemical Models FABM, .
The pelagic module is defined in the 3-D grid of the hydrodynamic model,
whereas the benthic module is defined in 0-D boxes for each water column
across the lateral grid of the model domain (Fig. ). Each benthic
box interacts with the bottommost pelagic box of the corresponding water
column in terms of a unidirectional flux of POM from the pelagic to the
benthic states and a bidirectional flux of DIM depending on the
concentration gradients.
For the integration of the source terms, a fourth-order explicit Runge–Kutta
scheme was used with an integration time step of 360 s, as for the 3-D
fields in GETM. The exchange between pelagic and benthic variables was integrated
with a first-order explicit scheme at a time step identical to that of the
biological model.
Model forcing and boundary conditions
Light extinction is described according to
I(z)=I0ae-zη1+I0(1-a)e-zη2-∫z0∑ikc,ici(z′)dz′,
where I0 is the photosynthetically available radiation (PAR) at the water
surface, and the first and second terms describe the attenuation at the red
and blue-green portions of the spectrum. We assume that the partitioning of
the two (a) and the attenuation length scale of the red light (η1)
are constant over space and time, as in , and that the
attenuation of blue-green light is due to SPM (as described by η2) and
organic matter (sum term). We chose a=0.58 and η1=0.35, which
correspond to Jerlov's type I water, thus clear-water conditions
, given that the attenuation by SPM and organic matter is
explicitly taken into account. For calculating attenuation due to SPM, a
daily climatology of SPM concentrations defined over the model domain was
utilized, such as in ECOHAM . The SPM field was constructed by
multiple linear regression of salinity, tidal current speed and depth for
each Julian day .
Then, η2, or the inverse of the SPM-caused attenuation coefficient, was calculated according to
1/η2=kSPM=ϵSPM⋅SPM,
where the attenuation for background turbidity Kw=0.16 m-1 and
specific attenuation coefficient for SPM
ϵSPM=0.02 m2 g-1 according to .
For calculating the attenuation due to organic matter in Eq. (),
phytoplankton, POC and DOC were considered (Table ).
Salinity (PSU) measured by FerryBox (a) and estimated by
the model (b) along the route shown in the inset. Note that the
lower range of salinity was truncated.
Freshwater and nutrient influxes were resolved for 11 major rivers along
the German, Dutch, Belgian and British coasts (Fig. ). For eight
of these rivers, and presented a
detailed quantitative analysis of nutrient fluxes. Besides the fluxes in
inorganic form based on direct measurements, fluxes in organic form have been
accounted for, first by calculating the total organic material concentration
by subtracting dissolved nutrient concentrations from total nitrogen and
total phosphorus, and then by assuming 30 % of the organic material to be in
particulate form i.e., POM;. Further, 20 % of POM is
assumed to describe phytoplankton biomass ,
the C : N : P ratio of which was assumed to be in Redfield proportions.
Finally, no estuarine retention/enrichment was assumed, following
. All river data except for the river Eider were available
in daily resolution, however, with gaps. Short gaps (< 28 days) were filled
by linear interpolation. Loadings from the river Eider were calculated first
by merging the data measured at the stations on two upstream branches, Eider
and Treene, then by filling the short gaps (< 28 days) by linear
interpolation, replacing the larger gaps with daily climatology, and
extending from 2000 to 2003 by using the climatology as well. To describe DIN
deposition at the water surface, the sum of annual average atmospheric deposition
rates of oxidized and reduced nitrogen provided by EMEP
(European Monitoring
and Evaluation Programme, http://www.emep.int) were used. At the open
boundaries in the north and west of the model domain (Fig. ), all
state variables belonging to the phytoplankton and zooplankton compartments
are assumed to be at zero gradient. For DIM, DOM and POM, monthly values of
ECOHAM , interpolated to 5 m depth intervals, are used as
clamped boundary conditions.
Quantification of model performance
For the comparisons with the data at monitoring stations, sparse in situ
measurements from the ICES database, and with the satellite
dataset; Pearson correlation coefficients, ρ; and mean
normalized bias, B*=(〈S〉-〈O〉)/〈O〉, where 〈S〉 and 〈O〉, respectively,
are the average simulated and observed values, were calculated. For the DIN,
DIP and chlorophyll comparisons with the station and ICES data, these skill
scores are reported in a color-coded table, where the four color levels indicate
low (red: |B*|≥0.75 and ρ<0.25), moderately low (yellow:
0.5≥|B*|<0.75 and 0.25≤ρ<0.5), moderately high (green:
0.25≥|B*|<0.5 and 0.5≤ρ<0.75) and high (blue |B*|≤0.25
and ρ≥0.75) model performance. For the comparisons against the sparse
ICES and ESA-CCI data, correlation scores and model standard deviations
normalized to measured standard deviations are displayed as Taylor diagrams,
where the correlation score and the normalized standard deviation correspond
to the angle and distance to the center . For the
comparisons against the ICES and satellite data, only the middle 99 %
of simulated and measured values were considered (i.e., leaving out the first
and last 0.5 %).
For the ICES data, temporal matching was identified at daily resolution,
vertical matching was obtained by comparing the measurements within the upper
5 m from the sea surface and within the 5 m above the sea floor with the
model estimates at the topmost and bottommost layers, and finally
horizontal matching was obtained by calculating the average of the values from four
nearest cells surrounding the measurement location, inversely weighted by
their Cartesian distance. For the satellite data, the temporal matching was
obtained by averaging the data from both sources for the period 2008–2010
for particular seasons of the year and horizontal matching by performing a
two-dimensional linear interpolation of the satellite data to the model grid.
The extraction of the hourly model temporally matching to the Scanfish data was
based on the hourly binned average time for each cast (defined as a full
downward and upward undulation cycle), and 3-D spatial matching was obtained
by constructing an average vertical profile from the four closest cells to the
average coordinate of each cast. For facilitating the qualitative comparison
of the simulated chlorophyll and the Scanfish measurements of fluorescence,
which have different units and signal strengths, normalized anomalies were
used, according to pi^=(pi-〈p〉)/σp, where
pi^ and pi are the normalized anomaly and raw value of a given
data point, and 〈p〉 and σp are the mean and standard
deviation of all data points.
Results
Evaluation of model performance by in situ data
A comparison of simulated salinities with the FerryBox measurements along the
cruise between Cuxhaven (at the mouth of river Elbe) and Immingham (at the
mouth of river Humber) demonstrates that the model captures the horizontal
salinity distribution (Fig. ). In particular, the contrast between
the northwestern model domain characterized by the rapid flushing of the
coastal freshwater input and the southeastern model domain (i.e., German
Bight) characterized by a strong and permanent salinity gradient is well
captured. Confinement of the salinity front during winter towards the coast
and its seaward intrusion, especially during early spring, and the
smaller-scale modulations that appear to be controlled by the spring–neap cycle are
both reproduced by the model.
Comparison of simulated surface and bottom temperatures with those extracted
from the ICES dataset for the period 2006–2010 are provided in
Fig. . The high correlation scores and low bias attained for water
temperature and salinity suggest that the model can reproduce the seasonal
warming, spread of freshwater discharges and thermohaline stratification
dynamics. However, in a relatively small number of instances, surface
temperatures are underestimated and bottom temperatures are overestimated,
which indicates that not all stratification events were captured. Almost all
of these instances are found to be located either at the northeastern margin
(>4∘ E and >55∘ N) or at the northwestern corner
(<4∘ E and >54∘ N) of the model domain, i.e., close to the
open ocean boundary (Fig. ).
Comparison of modeled and measured (ICES) temperature (abbreviated
T in panels a, b, c, d) and salinity (S in a, b, e, f)
at the surface (left) and bottom (right) layers for the period 2006–2010.
The 2-D histograms show the number of occurrences of simulation–measurement pairs.
The normalized bias (B*), Pearson correlation coefficients (ρ) and
corresponding number of data points (n) are shown on top of the scatter plots.
Comparisons of surface chlorophyll, DIN and DIP concentrations estimated by
the model with the measurements in 19 stations scattered across the southern
North Sea are shown in Figs. –, and the
corresponding skill scores are listed in Table . Estimates of
average nutrient concentrations and the timing of their depletion and
regeneration in a majority of stations agree well with the observations, as
indicated by the frequency of high and moderately high scores
(Table ). Notably, at several stations (e.g., Sylt, T8, T36,
T26, T22, T11 and T12) the difference between the relative bias for DIP and DIN
(i.e., BDIP*-BDIN*) was relatively large (with 55 %
being the highest at T22), suggesting a tendency for underestimating the
DIN : DIP ratio, although this was not the case for the comparison against
ICES measurements (see below). Relative to the nutrients, the performance of the
model in estimating chlorophyll is lower, especially at the stations
located along the Dutch coast (Fig. , Table ).
However, for about half of the 10 stations where data are available, the model
performance is at moderate levels.
Observations (gray dots) and model estimates (lines) of surface
chlorophyll, DIN and DIP concentrations at the stations located along the
coasts of the German Bight, operated by the Alfred Wegener Institute (Helgoland
and Sylt), Landesamt für Landwirtschaft, Umwelt und ländliche
Räume des Landes Schleswig-Holstein (S. Amrum, Norderelbe), and
Niedersächsischer Landesbetrieb für Wasserwirtschaft, Küsten- und
Naturschutz (Norderney). The normalized bias (B*), Pearson correlation
coefficients (ρ) and corresponding number of data points (n) are
shown on top of each panel.
As in Fig. , but for the stations located along the coasts of the Netherlands, operated by Rijkswaterstaat.
As in Fig. , but for the offshore monitoring stations
operated by the Bundesamt für Seeschifffahrt und Hydrographie.
The comparison of model results with the DIN, DIP and chlorophyll measurements
available at the ICES database at the surface and bottom layers for the
entire simulation period indicates a negligible normalized mean bias (≤12 %) and correlation coefficients at around 0.6–0.7 for nutrients and
about 60 % overestimation and correlation coefficients of about 0.3 for
chlorophyll (Fig. , Table ). The modeled
variability for all three biogeochemical state variables is within an
approximately 50 % envelope of the observed variability
(Fig. ).
Comparison of simulated and measured (ICES) DIN (a, b, c, d), DIP (a, b, e, f) and chlorophyll (a, b, g, h) at the
surface (left) and bottom (right) layers for the period 2000–2010. The 2-D
histograms show the number of occurrences of simulation–measurement pairs.
The normalized bias (B*), Pearson correlation coefficients (ρ) and
corresponding number of data points (n) are shown on top of the scatter plots.
Skill scores obtained at each station (B* is the normalized bias,
ρ is the Pearson correlation coefficients and n is the number of matching data
points) against ICES and ESA-CCI data shown in
Figs. – and Fig. partially (for
the averages of months 1–3, 10–12 and 4–9). Colors
indicate skill level, with red being low, yellow being moderately low, green being moderately high and blue being high (see Sect. ).
For an assessment of the accuracy of the simulated vertical distributions,
water density (expressed as σT) and fluorescence captured by a
Scanfish cruise (Heincke Cruise HE331) during 13–19 July 2010 were compared
to those estimated by the model (chlorophyll for fluorescence) averaged over
the same time period (Fig. ). This period was characterized
by significant thermal summer stratification reaching deep into the near
coastal regions of the German Bight. Thus, σT reflects two major
mechanisms that control the distribution of phytoplankton: the first is the
characterization of the
vertical gradients by denser water at the bottom layers, which
is mainly driven by thermal stratification as suggested by temperature
profiles (not shown). The second is the characterization of the horizontal gradients by
lighter water at the coasts, driven by low salinity due to the freshwater
flux from the rivers. The model can accurately reproduce both vertical and
horizontal density gradients, although some discrepancies exist, such as
the slightly underestimated depth of the pycnocline and steepness of lateral
gradients at around the coastal section. Fluorescence measurements along the
Scanfish track in July 2010 indicate frequent occurrences of subsurface
chlorophyll maxima (Fig. ). These are in some cases in the
form of higher concentrations below the pycnocline but in some others appear
as thin layers at around the pycnocline. While the deep chlorophyll maxima
are prevalently found in stratified offshore regions, the well-mixed
shallower regions mostly show homogeneously distributed high chlorophyll
concentrations throughout the water column due to higher dissipation rates
. The MAECS simulation agrees qualitatively well with these
patterns and captures the spatial variability of the observed vertical
chlorophyll distribution (Fig. ).
(a, b) σT measured by Scanfish and estimated by
the model; (c, d) normalized anomalies of fluorescence measured by
Scanfish and chlorophyll concentrations estimated by the model. The track of the
cruise, which took place between 13 and 19 July 2010, is shown in
(e).
Coastal gradients
Temperature stratification is one of the key drivers of biogeochemical
processes through its determining role on the resource environment, i.e.,
light and nutrient availability experienced by the primary producers. The
comparison against the Scanfish transect (Fig. ) showed that the
physical model has the potential to realistically capture the density
stratification. Using the temperature difference between surface and bottom
layers as an indicator of temperature stratification
, and using monthly averages across
all simulated years (2000–2010), the areal extent and seasonality of
stratification within the SNS is shown in Fig. . This analysis
suggests that a large portion of the model domain deeper than ∼ 30 m
becomes stratified from April to September, with a maximum areal coverage and
intensity (slightly above 8 K) in July.
Average temperature difference (K) between the surface and bottom
layers, averaged throughout 2000–2010 for each month. Gray lines show the
isobaths.
Simulated climatological concentrations of DIN and DIP display steep coastal
gradients along the coasts of the German Bight (Fig. ), both
during the non-growing season (months 1–3 and 10–12) and the growing season
(months 4–9). Within the ROFI region of freshwater
influence, of the Rhine, nitrogen concentrations decrease about 5
fold (from ≥ 48 mmolN m-3 to 8–16 mmol N-3) within a few
grid cells, corresponding to about 10–15 km distance. In the German Bight,
the non-growing season is similarly characterized by a thin stripe of high
nutrient concentrations along the coast, whereas during the growing season
especially phosphorus becomes depleted outside a confined zone of the Elbe
plume. At the offshore areas, nutrient concentrations during the growing
season are considerably lower than those during the non-growing season,
driven by the phytoplankton growth both directly by nutrient uptake and, for
the case of nitrogen, also indirectly by fueling denitrification in the
sediment. The DIN : DIP ratio in the offshore regions is close to the
Redfield molar ratio of 16:1 throughout the year, reflecting oceanic
conditions, while much higher at the coastal areas, particularly during the
non-growing season, reflecting the high N : P content of the continental
rivers . This transition from high coastal to low offshore
N : P ratios is qualitatively consistent with observations
e.g.,.
DIN (a, b) and DIP (c, d) concentrations at the
surface layer, averaged over the non-growing (months 1–3 and 10–12, left)
and growing seasons (months 4–9, right) for the entire simulation period
(2000–2010). Concentrations at the bottom layer are almost identical for
months 1–3 and 10–12 and similar for months 4–9. Gray lines show the
isobaths. Note the different color scales used for each panel and that the scale
used for DIN is 16 times that of DIP, such that the identical coloring for DIN
and DIP for the same season indicates a Redfield ratio of
16:1.
Both the satellite (ESA-CCI) images and our model estimates, averaged again
for the non-growing and growing season, suggest steep coastal gradients in
chlorophyll concentrations (Fig. ) similar to the nutrient
gradients shown above (Fig. ). The large-scale agreement in
coastal gradients results in high correlation coefficients
(Fig. , Table ). The normalized mean bias is small for
the non-growing season but relatively high and positive (i.e.,
overestimation) for the growing season. Higher model estimates at the lower
range (0–10 mgChl m-3) are responsible for this positive bias, which
is particularly the case during the first half of the growing season, where
the bias is highest (Table ).
Comparison of satellite (ESA-CCI; a, b) and MAECS
(c,d) estimates of surface chlorophyll concentrations averaged over
2008–2010 and for the non-growing (months 1–3 and 10–12, left) and growing
seasons (months 4–9, right). The 2-D histograms (e, f) show the number
of occurrences of simulation–satellite data pairs. Gray lines in
(a)–(d) show the isobaths. The normalized bias (B*),
Pearson correlation coefficients (ρ) and corresponding number of data
points (n) are shown on top of the scatter plots.
Our simulation results indicate significant spatiotemporal variability in
the Chl : C ratio, even when the seasonal averages are considered, i.e.,
omitting short-term variability (Fig. ). The Chl : C ratio is
generally higher at the coasts than offshore. Higher Chl : C ratios during
the non-growing (months 10–12 and 1–3) season similarly reflect light
limitation due to low amounts of incoming shortwave radiation at the water
surface. The simulated spatiotemporal differences in Chl : C ratios reach
to about 3 fold between different seasons of the year and between
offshore and coastal areas. The latter suggests that the differential
acclimative state of phytoplankton cells amplifies the steepness of the
chlorophyll gradients across the coastal transition shown in
Fig. .
Chlorophyll : C ratio in phytoplankton, averaged over the
non-growing (a) and growing season (b) of 2010. Gray lines
in (a)–(d) show the isobaths.
For gaining a better understanding of the relevance of acclimation in
capturing the coastal gradients, we considered a simplified, non-acclimative
version of the model in which the resource utilization traits were fixed (see
Appendix for a detailed description) and two alternative
parameterizations regarding the allocations to the light-harvesting, nutrient
acquisition and carboxylation machineries (which are state variables in the
full model): the first one with equal (balanced) allocation coefficients
(=0.333) and the second one by assigning the spatiotemporal averages of the
state variables integrated by the full (reference) model. The results of these
two parameterizations were almost identical, so hereafter we will refer to
them as the “fixed” model in short, without specifying the particular
parameterization.
The annual average coastal phytoplankton concentrations estimated by the
acclimative model are much higher than those estimated by the fixed model,
with no significant difference between the surface and water-column-averaged
values (Fig. a, b). In the offshore areas, the estimates of
the acclimative model are higher than those of the fixed model at the surface
(Fig. a), but slightly lower when water column averages are
considered (Fig. b), indicating that the phytoplankton growth
occurs mostly at the bottom layers in the fixed-trait model, which is
consistent with the daily vertical profiles in the fixed-trait model
(Fig. c). Importantly, these results suggest that a coastal
gradient in phytoplankton concentrations is predicted by a non-acclimative
model, which is presumably driven by the nutrient gradients
(Fig. ), but a much stronger gradient emerges when the acclimation
processes are resolved. Specific to this example, towards the coast,
phytoplankton adapt to the deteriorating light climate (Fig. ) and
increasing nutrient availability by investing more in the light-harvesting
machinery, as indicated by the increasing Chl : C ratios
(Fig. ), and thereby achieving higher coastal production rates
than in the case where their physiology is fixed. As a result of increasing Chl : C
ratios towards the coast, the chlorophyll concentrations display an even
stronger gradient than that of the biomass: at the surface layer, the increase of
biomass concentrations towards the coast is about 3.5 fold (from about 10 to
35 mmolC m-3), while that of chlorophyll is about 7 fold (from
about 2 to 14 mg m-3) along the transect shown in
Fig. .
Annual average phytoplankton carbon (and for R, chlorophyll)
concentrations in 2010 (a) at the surface layer; (b) averaged over the water column; and obtained with the following models:
R is the reference (acclimative)
model, F-bal is the fixed-physiology model with balanced investments and
F-avg(R) is
fixed-physiology model with allocation parameters as average trait values
produced by R.
Discussion
In order to assess the performance of the new model system presented for the
first time in this study, we employed several independent observation sources
and types: FerryBox measurements to assess the horizontal distribution of
salinity (Fig. ); sparse in situ measurements from the ICES
dataset for an overall evaluation of the physical and biogeochemical model
(Figs. , ); measurements from 19 monitoring
stations for evaluating the estimates for DIN, DIP and chlorophyll at
specific locations (Figs. –); Scanfish
measurements for evaluating the vertical density and chlorophyll profiles
(Fig. ); and finally the satellite observations for
evaluating the model skill regarding the horizontal distribution of
climatological chlorophyll concentrations (Fig. ) and
attenuation of light (Fig. ).
The physical model can provide a realistic description of the hydrodynamical
processes foremost relevant for modeling the biogeochemistry of the system.
Horizontal circulation patterns are captured as evidenced by the salinity
distribution being in agreement with the observations (Figs. ,
). The density structure of the system during summer, driven by a
complex interplay between the salinity gradients, heat fluxes at the surface
and tidal stirring, is realistically captured, although the pycnocline depth
seems to be underestimated (Fig. ). Accordingly, temperature
estimations match well with the observations, although there are cases where
the stratification events are not reproduced by the model
(Fig. ), most of which are found to be within the western
portion of the model domain. The areal extent and seasonality of
stratification (Fig. ) are in agreement with those reported by
earlier studies . For nutrient
concentrations, a relative bias of ≤ 12 % and correlation coefficients
between 0.58 and 0.72 correspond to a high and moderately high model skill,
respectively (Table ). For the pointwise comparisons of
chlorophyll, model skill was moderate for the sparse measurements included in
the ICES database, and for the stations in the German Bight
(Table ), but mostly low for the stations within the western
portion of the model domain. The comparison of climatological averages of the
simulated chlorophyll with those of the satellite observations resulted in
high correlations for all seasons and a low to moderately low bias, except
during the early growing season (Table ).
The model captures the subsurface chlorophyll maxima occurring in the deeper
parts of the model domain (Fig. ). This phenomenon has been
previously documented in the southern North Sea
. Former 3-D modeling studies, such as that of
, apart from capturing the presence of a deep
chlorophyll maximum, did not reproduce the rich variability revealed by the
observations. Our comparative analysis shows that the formation and
maintenance of such structures are critically dependent on the
parameterization of the sinking rate of phytoplankton (Fig. a) and
underwater light climate (Fig. b). The sinking speed of phytoplankton
in MAECS is inversely related to the nutrient quota of the cells, which
mimics the internal buoyancy regulation ability of algae depending on
internal nutrient reserves (see Appendix ) but also indirectly
emulates chemotactic migration as is typical for dinoflagellates
. This quota dependency results in considerable spatial and seasonal
variability in sinking rates (Fig. ).
The critical dependence of the formation and
maintenance of vertical chlorophyll structures on the functional
representation of sinking underlines the relevance of a consistent
description of the intracellular regulation of nutrient storages. The latter,
in turn, is determined by the metabolic needs, such as the intensity of light
limitation, and hence, investments in the synthesis of pigmentary material
. Indeed, the non-acclimative (fixed-trait) version of the
model (Appendix ) predicts qualitatively different vertical
profiles of phytoplankton biomass (Fig. c), although the sinking
parameterization in that simplified version is identical to that in the fully
acclimative version. The non-acclimative model version might be tuned to
match the observed vertical distributions of phytoplankton; however, this
would probably be at the expense of compromised performance in some other
respects, such as the horizontal gradients, or timing and amplitude of
chlorophyll blooms.
We conclude from the extensive model performance assessment that the model
reproduces the main physical and biogeochemical characteristics of the
southern North Sea, especially within the German Bight, where the model
resolution is finest (Fig. ), and the influence of fluxes at the
open boundaries is relatively small, given the predominantly
counterclockwise circulation pattern . The process of
performance assessment also helped in identifying the possibilities for further
model refinement. For instance, a comparison with the satellite observations
revealed that the light attenuation in the offshore areas is overestimated by
the model, primarily because of the contributions by the climatological SPM
forcing (Fig. ). A likely consequence of the overestimated
attenuation is an underestimation of the depth of primary production (e.g.,
Fig. b), and this may, in turn, explain the overestimated
chlorophyll concentrations in the offshore areas during the growing season
(Fig. b, d). Another source of error regarding the SPM-caused
turbidity is the fact that, at specific coastal sites, such as at the Noordwijk-10
station, the measured SPM concentrations
show considerable interannual variations that can obviously not be
represented by the climatological SPM forcing (Fig. ), which may
explain the particularly low correlation coefficient (0.14) obtained at this
station for chlorophyll. A better representation of the SPM-caused turbidity
might be achieved by an explicit description of the SPM dynamics e.g.,
as in. Coupling the biogeochemical model with such an SPM model
would then also allow the description of the two-way interactions, i.e., not
only light limitation , but also the acceleration of
sinking of SPM by the production of transparent exopolymer particles
. At the stations within the ROFIs of major
rivers, such as the Norderelbe and S. Amrum (Fig. ) and
Noordwijk-2 and Noordwijk-10 (Fig. ), the skill scores are relatively
low (Table ). These stations, especially Noordwijk-2 and
Noordwijk-10, are located where the concentrations change dramatically within
10–15 km (e.g., Fig. ). Accordingly, a slight error by the
physical model in predicting the salinity front, e.g., because of an
inadequate representation of the tidal dynamics, might result in considerable
deviation of the estimated concentration of biogeochemical variables from the
measurements. The relatively coarser model resolution around the Dutch coast
might therefore explain the consistently lower skill scores obtained at the
Dutch stations. Identifying such potential inadequacies of the physical model
requires further investigation, such as an assessment of the tidal
constituents at the tidal gauges e.g.,. Another
potential source of error for the mismatches within the ROFIs is the
potential flaws in the description of riverine loadings, such as assuming
that the non-dissolved fractions of the total nitrogen and total phosphorus
are entirely in labile form (Sect. ). Although the earlier
replenishment of phosphorus relative to nitrogen in the coastal sites is
often reproduced (e.g., Sylt, Noordwijk-2, Noordwijk-10, Terschelling-4),
some delays occur in stations such as Norderney, which probably reflects the
oversimplification of the benthic processes with respect to the description
of oxygen-driven iron–phosphorus complexation kinetics
(Appendix ), which have been suggested to be the main driver for
the phenomena in the coastal areas
.
The model predicts steep coastal gradients in nutrient concentrations
(Fig. ), which is in line with prior observations
e.g.,. The maintenance of these gradients
during winter is explained by the limited horizontal mixing due to the
density gradients caused by the freshwater influx from the land
and the trapping of this nutrient-rich freshwater
at the coast due to the alongshore currents in the study system driven by
predominantly westerly winds and the coriolis forcing
. During the warmer seasons when the offshore
waters are stratified, owed to the presence of horizontal salinity gradients,
a mechanism similar to the estuarine circulation was
suggested to further promote these gradients along the Wadden Sea, as well as in
regions far from river inputs
. Coastal waters remaining
nutrient-replete during the growing season lead to high phytoplankton
concentrations (Fig. ), despite the higher turbidities at the
coastal waters (Fig. ).
The comparison of the present model with earlier attempts is neither in the scope
of this study nor possible without a dedicated benchmarking effort, using
standardized forcing data and skill performance assessment datasets and
methodology e.g., as in. Even a qualitative
comparison is difficult, given that spatial and temporal binning of the data,
frequently employed in model validation (such as in our Fig. ),
can dramatically impact the skill scores and that pointwise comparisons with
sparse observation datasets (such as in our Figs. and
) are rarely performed . However, the skill of
the presented model in estimating the chlorophyll concentrations in the SNS
can argued to be at least comparable to that of the recent modeling
applications for a relevant region e.g.,noting that all these
studies had larger model domains and were evaluated for different time
intervals. This is
noteworthy, given that phytoplankton is represented by a single species in
our model, whereas in other modeling approaches several species or groups
are resolved. The inclusion of multiple functional types is motivated by the
spatial and seasonal variability in the phytoplankton composition observed in
the field: coastal areas of the SNS are dominated by diatoms throughout the
year in some sites e.g., and during spring
in some others, later replaced by Phaeocystis during summer
e.g.,, whereas the offshore areas are often dominated
by dinoflagellates especially during summer
. These phytoplankton groups differ from
each other in a number of traits, including the physiological traits that
determine their ability to access the (mineral and light) resources and build
biomass. For instance, in an experimental work, two diatom species were shown
to have on average more than 3-fold-higher Chl : C ratios than those of
two dinoflagellate species , therefore making them more
tolerant to the light-limited conditions of the turbid, coastal waters. In
the presented approach, the cellular composition of the single, but
acclimative, phytoplankton group dynamically approaches towards (for some
traits, instantaneously adopts) the physiological state of the ideal resource
competitor in a given environment, which, in nature, happens through various
processes – from the plastic response of the individual cells to the species
sorting at the community level. In a traditional, plankton functional type
model, on the other hand, the species with the most suitable traits would
become the most dominant among others, while the proximity of the
physiological traits to the theoretical optima, and thus the overall
productivity,
would be determined by the resolution of physiological traits as represented
by the defined clones.
The worst-case scenario is when there is only one
non-acclimative group, as illustrated in our experiment: at the turbid but
nutrient-rich coastal areas, prioritization of the light harvesting over the
nutrient acquisition machinery, as evidenced by the higher Chl : C ratios
predicted by the acclimative model (Fig. ), leads to better
fitness and thus higher phytoplankton concentrations in comparison to their
non-acclimative equivalents (Fig. ). Moreover, because of the
high Chl : C ratios at the coastal areas (Fig. ), chlorophyll
concentrations display even steeper gradients than the phytoplankton
concentrations (Fig. ). The transitional Chl : C pattern
suggested by our model has been previously identified based on monitoring
data by . The Chl : C ratios ranging between
0.01–0.1 gChl gC-1 at the coastal stations and
0.002–0.02 gChl gC-1 at the offshore stations reported by
envelop our estimated seasonal average values
of 0.045 and 0.015 within the respective regions. According to the simulation
results, Chl : C ratios also differ considerably between the non-growing
and growing season, with higher values during winter, due to low light
availability. A similar seasonal amplitude in Chl : C has been found by
for the English Channel, with higher ratios during
winter.
As mentioned above, the physiological composition is not the only relevant trait
for determining the community composition in the study system. Diatoms are
fast growers and are defended against the efficient microzooplankton grazers, but
this comes at the cost of silicate requirement for their growth
and higher sedimentation losses .
Phaeocystis spp. are slow growers, but, by forming large colonies, they
are well defended against zooplankton . Finally, the
dinoflagellates, also despite being slow growers, are mobile
and mostly have access to alternative nutrient sources
through their phagotrophic abilities . The representation of
zooplankton with a single group may also be an oversimplification, as the
microzooplankton and mesozooplankton have considerably different growth rates
and functional responses to prey availabilities
. Moreover, effects of temperature on mesozooplankton
occur through phenological shifts e.g., that might have
a determining role on the maximum chlorophyll concentrations
, which can probably be only partially reflected by the
simple Q10 rule we applied for grazing rates (Appendix ).
None of these ecophysiological aspects were taken into account in our model,
and this may explain some of the discrepancies between the simulated and
observed chlorophyll concentrations. In future work, inclusion of few other
phytoplankton groups, each being acclimative, and one additional zooplankton
group is foreseen. While the consideration of other phytoplankton traits
should be straightforward, the inclusion of phagotrophy as an additional
physiological allocation trait represented by a state variable is possible
e.g., as in but would require the re-derivation of
the model equations.