Mercury (Hg) is a pollutant of global concern. Due to anthropogenic emissions, the atmospheric and surface ocean Hg burden has increased substantially since preindustrial times. Hg emitted into the atmosphere gets transported on a global scale and ultimately reaches the oceans. There it is transformed into highly toxic methylmercury (MeHg) that effectively accumulates in the food web. The international community has recognized this serious threat to human health and in 2017 regulated Hg use and emissions under the UN Minamata Convention on Mercury. Currently, the first effectiveness evaluation of the Minamata Convention is being prepared, and, in addition to observations, models play a major role in understanding environmental Hg pathways and in predicting the impact of policy decisions and external drivers (e.g., climate, emission, and land-use change) on Hg pollution. Yet, the available model capabilities are mainly limited to atmospheric models covering the Hg cycle from emission to deposition. With the presented model MERCY v2.0 we want to contribute to the currently ongoing effort to improve our understanding of Hg and MeHg transport, transformation, and bioaccumulation in the marine environment with the ultimate goal of linking anthropogenic Hg releases to MeHg in seafood.
Here, we present the equations and parameters implemented in the MERCY model
and evaluate the model performance for two European shelf seas, the
North and Baltic seas. With the model evaluation, we want to establish a set
of general quality criteria that can be used for evaluation of marine Hg
models. The evaluation is based on statistical criteria developed for the
performance evaluation of atmospheric chemistry transport models. We show
that the MERCY model can reproduce observed average concentrations of
individual Hg species in water (normalized mean bias: Hg
Mercury (Hg) is a global pollutant and a dangerous neurotoxin (AMAP/EMEP, 2019a). Since preindustrial times, the global Hg cycle has been significantly altered by anthropogenic emissions (Streets et al., 2019) resulting in a 3-fold pre-anthropogenic-to-present-day increase in the atmospheric and a substantial increase in oceanic Hg burden (Lehnherr, 2014; Amos et al., 2013). The major anthropogenic sources of Hg are emissions from coal-fired power plants, small-scale artisanal gold mining, and metal and cement production (Pirrone et al., 2010; AMAP/EMEP, 2013, 2019a, b). In addition, natural emissions and legacy re-emissions from previously deposited Hg (most of it of anthropogenic origin) also contribute significantly to the atmospheric Hg burden (Pirrone et al., 2010; Driscoll et al., 2013; Obrist, 2018). The atmospheric lifetime of Hg is estimated to be in the range of 0.6 to 1.0 years (Slemr et al., 2018), resulting in a global atmospheric distribution of Hg. Atmospheric Hg will eventually be deposited (Cohen et al., 2016; Jiskra et al., 2018). A large fraction is deposited directly into the ocean, but Hg deposited onto land can also be transported to the ocean via rivers and groundwater. In the aqueous phase, inorganic Hg can be methylated, forming the highly bioaccumulative monomethylmercury (MMHg) and/or dimethylmercury (DMHg). These MeHg compounds are readily accumulated in the food web and pose a risk to food safety and human health (Clarkson, 1990; Mason et al., 1996; Chen et al., 2012; Parks et al., 2013; Puty et al., 2019). Because of this, the international community, under the umbrella of the United Nations Environmental Programme (UNEP), signed the Minamata Convention on Mercury, which came into force in 2017. Under this convention, all participating 184 nations have agreed to assess Hg pollution under their jurisdiction, to minimize Hg usage and release of Hg compounds into the environment, and to regularly assess the impact of the reduction measures taken on the environmental Hg burden and distribution. In order to assess the impact of reduction measures, there is an urgent need to understand the Hg pathways from anthropogenic releases to top predators and humans, with specific attention to the marine ecosystem.
In this paper, we (1) introduce a newly developed numerical multi-compartment model for Hg cycling in the marine environment including accumulation in the marine food web (MERCY v2.0) and (2) evaluate the model performance to reproduce observed concentrations of, seasonality of, and variability in Hg species. For the latter, we apply performance criteria used for evaluation of atmospheric chemistry transport models and also for evaluation of marine Hg models. We use these criteria to (2.1) quantify the models' predictive capabilities based on our current understanding of Hg cycling, (2.2) identify the major sources of model error, and (2.3) quantify the constraints on model improvement based on current process understanding and measurement availability and uncertainty. With this study, we present an evaluation of our marine Hg model and a general framework that provide the basis for future intercomparison studies of marine Hg models.
The key question concerning Hg pollution is how changing Hg emissions and
other external stressors such as climate and land-use change impact MeHg
accumulation in seafood, which is an important global protein source for
human consumption (Pauly et al., 2002; Obrist et al., 2018). To anticipate
the natural Hg cycle and to identify the impact of human actions on the
system, it is necessary to develop multi-compartment chemistry transport
models (CTMs) including all relevant compartments: atmosphere,
soil/vegetation, rivers and oceans, sediments, and the marine ecosystem. The
need to incorporate all compartments into a single multi-compartment model
arises from the fact that Hg is non-degradable and constantly cycling
between environmental compartments, unlike most pollutants, which tend to
accumulate in a single compartment and/or degrade over time. For example,
atmospheric deposition of oxidized Hg is a major flux of Hg into the ocean,
but reduction reactions in the ocean and the high vapor pressure of
elemental Hg
Atmospheric Hg modeling is well established, and a large variety of global
(ECHMERIT: Jung et al., 2009; De Simone et al., 2014; GLEMOS: Travnikov and
Ilyin, 2009; Travnikov et al., 2009; GEM-MACH: Durnford et al., 2012; Kos et
al., 2013; Dastoor et al., 2015; GEOS-Chem: Holmes et al., 2010; Amos et
al., 2012; Song et al., 2015) and regional (CMAQ: Bullock et al., 2008; Bash, 2010; Zhu et al., 2015; DEHM: Christensen et al., 2004; WRF-Chem:
Gencarelli et al., 2017) atmospheric CTMs for Hg cycling have been
published. Due to this abundance, many model intercomparison and source
apportionment studies have improved our understanding of atmospheric Hg
transport and source–receptor relationships and have allowed us to predict future
atmospheric Hg levels and deposition fluxes (Bergan et al., 1999; Xu et al.,
2000; Petersen et al., 2001; Lee et al., 2001; Seigneur et al., 2001;
Bullock and Brehme, 2002; Dastoor et al., 2002; Hedgeock et al., 2004; Selin et
al., 2007; Travnikov et al., 2009; Bieser et al., 2014; Gencarelli et al.,
2017; Dastoor et al., 2015; Song et al., 2015; Cohen et al., 2016; Travnikov
et al., 2017; Bieser et al., 2017; Horowitz et al., 2017). These models and
studies are a keystone in informing policymakers to support the
implementation and effectiveness evaluation of the Minamata Convention
(
Compared to Hg modeling in the atmosphere, marine Hg modeling is still in its infancy, and only a limited number of models exist so far. The development of marine Hg models can be divided into four phases. At first, the ocean was implemented as a boundary for atmospheric CTMs, and nowadays most atmospheric CTMs implement some kind of surface ocean parameterization to explicitly include Hg air–sea exchange. One of the earliest marine Hg model developments were box models (Sunderland and Mason, 2007), followed by the addition of inorganic Hg redox chemistry and transport in a 2D slab ocean model coupled to the GEOS-Chem model (Selin et al., 2008; Strode et al., 2007; Soerensen et al., 2010). The aim of these early models was to improve air–sea exchange by including horizontal transport, redox chemistry, and river loads. Next came the development of the first marine 3D models. These models, still limited to the inorganic Hg cycle, were used to investigate marine Hg dynamics (Zhang et al., 2014a, b; Bieser and Schrum, 2016). In the next stage, several specialized marine Hg models were developed which were not based on 3D hydrodynamic models. Soerensen et al. (2016a) published a coupled physical–biogeochemical multi-box model including organic Hg chemistry to investigate the Hg budgets in the Baltic Sea. Focusing on bioaccumulation, Schartup et al. (2018) implemented Hg accumulation in a complex food web model and Sunderland et al. (2009, 2018) modeled the consumer exposure to MeHg in seafood. Finally, Pakhomova et al. (2018) developed a model with comprehensive Hg chemistry based on a hydrodynamic 1D model. Only in recent years has the development of comprehensive marine Hg models gained traction. So far, four marine Hg models based on numerical hydrodynamic 3D models have been published (Semeniuk and Dastoor, 2017; Zhang et al., 2019; Kawai et al., 2020; Rosati et al., 2022). All of these models include a complete marine Hg chemistry including MeHg. Yet, only Zhang et al. (2020) and Rosati et al. (2022) have also implemented Hg cycling into a biogeochemical model considering uptake to and release from marine biota, making this model the first hydrodynamic 3D Hg model to include the marine ecosystem.
Here we present our newly developed biogeochemical multi-compartment model for Hg cycling, MERCY v2.0, and evaluate its predictive capabilities and limitations using evaluation criteria applied for performance evaluation of atmospheric CTMs (Derwent et al., 2010; Thunis et al., 2012, 2013; Carnevale et al., 2014). We focus on the implementation of the marine Hg cycle including a comprehensive marine Hg chemistry and partitioning scheme as well as bioconcentration and biomagnification. We improve on the state of the art by introducing an experimental upper trophic layer that simulates Hg and MeHg accumulation in fish. To our knowledge, MERCY v2.0 includes all currently known processes controlling marine Hg cycling. The model is based purely on processes, reactions, and rates published in peer-reviewed literature, and no additional model tuning was performed.
We investigate the model predictive capabilities, something we consider important before using the model to study budgets or global dynamics. This allows us to quantify our model uncertainty, which for other models has only been loosely constrained to be “orders of magnitude” (Kawai et al., 2020), and discuss the processes and parameters driving it. Set up on a high-resolution regional domain covering a wide range of marine regimes in a region with high primary productivity and a relative abundance of observations, we evaluate the ability of the model to reproduce observed concentrations of, seasonality of, and variability in individual marine Hg species. Using common practice from atmospheric Hg modeling, we establish a quantitative benchmark for the capability of the model to reproduce actual observations of marine Hg concentration and speciation. Based on this we discuss the major knowledge gaps and research questions that need to be tackled in order to improve our understanding of marine Hg cycling. Our ultimate goal is to improve capabilities to link changes in external stressors like anthropogenic emissions and climate change to MeHg accumulation in the marine food web by providing an independent model for marine Hg cycling and by fostering collaboration in the form of model intercomparison studies comparable to the efforts in atmospheric Hg modeling (Ryaboshapko et al., 2002; Bullock et al., 2008; Travnikov et al., 2017; Bieser et al., 2017). Finally, we want to identify and communicate the major needs for monitoring of Hg species in the marine environment.
The marine Hg chemistry scheme we develop for MERCY v2.0 is embedded into
GCOAST (Geesthacht Coupled cOAstal model SysTem), a modeling framework
coupling physical, chemical, and biological numerical models. It is an
update and overhaul of MERCY v1.0 (Bieser and Schrum, 2016), which featured
only inorganic Hg chemistry and no ecosystem interactions. As input, MERCY
uses hourly model output from four types of 3D hydrodynamic model
(atmospheric physics, atmospheric chemistry, marine physics, and marine
ecosystem) to drive the marine Hg speciation, transport, and bioaccumulation
model. While this approach requires a large amount of storage capacity, it
reduces the computational requirements and allows the model to be easily run
with input from alternative biogeophysical models. The external variables
used by MERCY are listed in Table 1. In brief, the models used in this work are as follows:
The regional weather and climate model COSMO-CLM (Rockel et al., 2008; Sørland et al., 2021)
provides meteorological variables used to calculate air–sea exchange
(temperature and wind speed) and photolytic reactions (surface shortwave
radiation). COSMO-CLM is nudged to the atmospheric reanalysis dataset ERA-Interim (Berrisford et al., 2011; Dee et al., 2011; Hersbach et al., 2020). The atmospheric chemistry transport model CMAQ-Hg (Byun and Schere, 2006;
Zhu et al., 2015; Bieser et al., 2016) is forced by COSMO-CLM meteorology
and used to calculate atmospheric transport, chemistry, particle
partitioning, and deposition for atmospheric trace gases. MERCY uses
atmospheric Hg concentrations and deposition fluxes from CMAQ-Hg. The physical hydrodynamic ice–ocean model HAMSOM (Backhaus, 1983;
Schrum and Backhaus, 1999) is directly coupled to the ecosystem
model ECOSMO, enabling it to represent the impact of the ecosystem on the
hydrodynamics (e.g., light attenuation by biota). In MERCY the physical
variables are used to calculate marine mercury transport as well as
temperature and salinity dependence of mercury cycling and speciation. The
HAMSOM advection scheme is used to transport all Hg state variables. The model setup is based on GEBCO bathymetry data (GEBCO Bathymetric Compilation Group). The marine end-to-end NPZD (nutrient, phytoplankton, zooplankton, detritus)
ecosystem model ECOSMO (Schrum et al., 2006; Daewel and Schrum, 2013; Daewel
et al., 2019) is a 3D-resolved food web model directly coupled with
HAMSOM. It includes nutrients (nitrogen, phosphorus, and silica) and a food
web based on a functional group approach with three phytoplankton species
(diatoms, flagellates, and cyanobacteria), two zooplankton species (herbivore
and omnivore), a macrobenthos group, and a pelagic fish group representing higher
trophic levels. Additionally, oxygen, biogenic opal, detritus, and dissolved
organic matter are considered, and the model includes a two-layer sediment
compartment to simulate sedimentation and resuspension. In MERCY detritus
and dissolved organic matter determine the partitioning of Hg and MeHg, and
factors such as light attenuation and oxygen concentration influence Hg
speciation. Moreover, concentrations of the various species of the model
food web are used to calculate bioconcentration and biomagnification of Hg
and MeHg.
All employed models and data are freely available (see “Code availability” and “Data availability” sections at
the end of the paper).
MERCY input variables and source models.
MERCY v2.0 implements all processes we identified as relevant to marine
(pelagic and benthic) Hg cycling into a 3D ocean-ecosystem model. MERCY is
based on basic principles describing Hg transport, transformation, and
bioaccumulation. It is set up on the same grid and domain as the coupled
ocean-ecosystem model HAMSOM-ECOSMO. Based on archived hourly HAMSOM-ECOSMO
output, it is effectively offline-coupled to the marine hydrodynamic and
ecosystem models. The HAMSOM-ECOSMO model has been shown to accurately
reproduce ecosystem dynamics in the coupled North Sea–Baltic Sea system. The
model equations and a model validation on the basis of nutrients are
presented in detail by Daewel and Schrum (2013), who showed that the model
can reasonably simulate ecosystem productivity in the North Sea and the
Baltic Sea on seasonal to decadal timescales. Using the same numerical
approximations as described in Daewel et al. (2019), the rate of change in the
concentration of Hg state variables over time
Partitioning
Bioconcentration
Biomagnification
Additional release from the biological matrix
Finally, the respective change in dissolved Hg concentrations
MERCY implements Hg using 35 variables (Table 2) representing different Hg species in the atmosphere, ocean, and sediment. For each model time step and each grid cell, the species are redistributed accounting for mass conservation based on physical, chemical, and biological processes. Figure 1 gives a graphical overview of transformations between Hg species in MERCY.
Schematic of the chemical mechanism in MERCY. Solid lines indicate
chemical reactions, fine dotted lines photolytic reactions, dash-dotted
lines instantaneous partitioning processes, and bold dotted lines
bioaccumulation and releases from biota into the dissolved phase. Colors
codes are white for elemental mercury, yellow for inorganic oxidized
mercury, pink for methylated mercury, and green for Hg in biota. The
physical state of each species is indicated by “g” for gaseous, “aq” for
dissolved, and “s” for solid. The upper row indicates Hg species in the
atmosphere, and the lower row indicates those in the sediment. All species and their
reactions are given in Tables 2 and 3. Note Reaction (R20) (reductive
methylation, Table 3) MMHg–DOM
Hg species in MERCY. Species can represent state variables in multiple models.
In this section, we present all chemical state variables and the
transformation processes in the model. A complete overview of all chemical
transformations and the respective reaction rates
Chemical reactions as implemented in the MERCY model.
Pseudo-first-order reaction rates
Hg redox chemistry is implemented with five reactions. Reduction (Hg
Additionally, we implemented Hg sulfur chemistry using oxygen
concentrations calculated by ECOSMO, whereas sulfur ions (S
The organic chemistry doubles the number of variables introduced for the
inorganic Hg chemistry mechanism (Fig. 1). In the model, we implemented
three sources for MMHg
Besides MMHg
In the sediments, we consider only two species: Hg
The speciation of Hg
Three-way partitioning is calculated as a function of Hg concentration,
particle load, and dissolved organic matter concentration (Eqs. 10–12). As
we could not obtain sorption and desorption rates and because our carbon
representation does not capture the number of O- and S-binding sites
available for Hg, we implemented partitioning based on partitioning
coefficients instead of a dynamic sorption/desorption process as described
in Eq. (4). We use a value of
The model assumes instantaneous equilibrium and redistributes Hg
Physical and biological constants used in MERCY v2.0.
The radiation available for photolytic reactions is determined from hourly
input fields using shortwave radiation reaching the surface as modeled by
the meteorological model COSMO-CLM (Table 1). As the reaction rates for Hg
photolysis are usually reported in relation to photolytically active
radiation (
Hg bioaccumulation has been implemented directly into the HAMSOM-ECOSMO
framework (Daewel and Schrum, 2013; Daewel et al., 2019). ECOSMO is based on
a functional group approach lumping species based on properties like
nutrient requirements (NO
In MERCY we consider bioaccumulation of inorganic Hg
Overview of the ECOSMO marine ecosystem nutrient and functional group model (Daewel et al., 2019).
Schematic overview of Hg
In MERCY dissolved Hg
For all non-phytoplankton species, we consider the active uptake
Flowchart of Hg bioaccumulation due to feeding following the ECOSMO end-to-end functional group approximation (Daewel et al., 2019). Rates for all depicted flows are given in Table 5.
Mercury accumulated by active
Overview of bioaccumulation parameters. External variables taken
from the ecosystem model ECOSMO such as mortality (
Following the sediment modeling concept by Daewel et al. (2019), we implemented a simple two-layer sediment system, where the first layer interacts with the lowest water column grid cell and the second layer represents a permanent sink.
Sedimentation occurs due to the settling of Hg bound to particles and
detritus. The sedimentation flux
Resuspension
Hg
Air–sea exchange of elemental Hg
In the above equations (Eqs. 30–36),
As a basis for the presented model development, we build upon the setup used for the earlier-published inorganic marine Hg model MERCY (Bieser and Schrum, 2016). All processes are implemented as stand-alone routines which are called from a main driver function containing several time loops (Fig. 5). Data for the wet cells (pelagic) are stored in vector form to reduce overhead, and data for sediments (benthic) and the lowest atmospheric layer are stored in 2D fields. Input data (Table 1) are read directly during run time from binary ECOSMO output as hourly mean values. This approach was chosen because there is no feedback from the Hg chemistry on the physical and biological models and because it allows us to reduce the computational costs of running the marine Hg model. All output files are created with daily mean values and saved in netCDF format using the IO-API interface (Byun and Schere, 2006; IO-API). The model is set up in a way that it runs for a single year using the last output time step of the previous year as an initial condition. For this initial model evaluation, we run MERCY for 17 years from 2000 to 2016.
Schematic overview of the MERCY model routines and main time loop.
We determine the model performance in reproducing observed concentrations and dynamics (e.g., variability and seasonality) of individual Hg species. Based on this analysis, we identify the processes and parameters responsible for the model error. The model is not specifically calibrated to the area of application, the North and Baltic seas. It is built on the current understanding of mercury cycling in the ocean and should be generally applicable. Major factors that need to be considered before applying the MERCY model to other regions are (1) partitioning coefficients for organic material (OM) as the type of OM varies regionally, (2) the parameterization for biogenic reduction as the values presented here are based on cyanobacteria in the Baltic Sea, (3) the uptake and release rates for bioaccumulation which might not be representative of other regions, and (4) the ecosystem model used that is needed to drive MERCY.
Because there are no established quality criteria for marine models, we use
criteria commonly used for evaluation of atmospheric CTMs (Derwent et
al., 2010; Thunis et al., 2012, 2013; Carnevale et al., 2014). We start by
comparing the observed and predicted means (Eq. 37) using daily model
averages in the
Equation (37) gives the mean.
Equation (38) gives the normalized mean bias.
Equation (39) gives the normalized centered root-mean-square error.
Equation (40) gives the standard deviation.
Equation (41) gives the normalized mean standard deviation.
Equation (42) gives the correlation coefficient (
Equation (43) gives the root-mean-square error.
Equation (44) gives the factor of 2.
Secondly, we use the more technical model quality objective (MQO) as defined
by Carnevale et al. (2014). The MQO (Eq. 45) relates the root-mean-square
error (Eq. 46) to the root-mean-square uncertainty (Eq. 47). The MQO can be
interpreted as follows: for
Equation (45) gives the model quality objective.
Equation (46) gives the root-mean-square error.
Equation (47) gives the root-mean-square uncertainty.
Equation (48) gives the model performance criterion for NMB.
Equation (49) gives the model performance criterion for NMSD.
Equation (50) gives the model performance criterion for RMSE.
Here, we evaluate the model for the North and Baltic seas in northern Europe
(model domain shown in Fig. 6). This area was chosen for model evaluation
as it covers a large range of different physical and biological conditions:
the Baltic Sea (Fig. 6; marine regions 8–15) is an enclosed shelf sea with
a surface area of 377 000 km
The North Sea has a surface area of 575 000 km
Due to their close vicinity to the coast and national monitoring programs, there are a comparably large number of Hg observations available for both the North Sea and the Baltic Sea. However, the data on Hg are still sparse in some areas, especially regarding Hg speciation, which is a major obstacle for model evaluation.
To generate the necessary forcing data (Table 1) to run the MERCY model, we
used the four models described in Sect. 2.1. For the atmosphere, COSMO-CLM
was run on a regional domain for Europe driven by ERA-Interim re-analysis
data (Berrisford et al., 2011; Dee et al., 2011; Hersbach et al., 2020). The atmospheric model domain covers the
entire European landmass, including northern Africa and western Russia, with a
resolution of
As initial conditions, we interpolated observations in water, biota, and sediment using a traditional kriging methodology to produce realistic initial starting conditions (mostly the pronounced vertical gradient) and minimize the spin-up time required (Cressie, 1990). The observational Hg data were retrieved from the database of the German Federal Maritime and Hydrographic Agency (MARENET, 2020). We ran the model using initial conditions multiplied by factors of 0.5 and 2.0 and tested the time necessary for the two runs to converge. For our model domain, which is a relatively small and in parts enclosed shelf sea area, the model runs started to converge after only a few years in the water column but took several years for Hg in sediments and biota (especially at higher trophic levels). For this study, we used a spin-up time of 30 years to reach realistic initial conditions for the production runs.
The chosen domain, including only the North Sea and Baltic Sea, has only a
very small open boundary: the English Channel in the southwest, which forms
a narrow connection to the Atlantic Ocean, and the wider opening in the
northern channel. The North Sea in the north of the domain, which receives most of the Atlantic inflow, is connected to the open Atlantic Ocean at
the shelf break. This region is characterized by an outflow in the eastern
part and inflow in the western part. At the open boundaries, we prescribe
constant Hg concentrations using 1.0 pM Hg
River loads are taken from OSPAR and HELCOM reports and the Norwegian
Tilførsel program (Green et al., 2011; HELCOM, 2007, 2011). We implemented
rivers using monthly load data in the North Sea and annual values for the
Baltic Sea as described in Bieser and Schrum (2016). The annual inflow of Hg
through rivers is 1100 kg a
Dry and wet Hg deposition is read in as hourly totals from CMAQ netCDF
output files. The deposited Hg
For the model performance, we start by evaluating total Hg (Hg
Annual averages:
The available Hg
In the North Sea we use 435 measurements of Hg
For the evaluation of Hg
Bioaccumulation in the marine biota is evaluated by comparing their total Hg
and MeHg content to measured concentrations in biota in the Baltic Sea (Nfon
et al., 2009). For evaluation of fish total Hg, we use Hg
Figure 8 compares the frequency distribution of 435 Hg
Hg
In the less dynamic open North Sea, the model performs better (FAC2
Frequency distribution of observed (red) and model (blue)
Hg
In summary, for Hg
Regional model performance for Hg
In the Baltic Sea, model performance for Hg
Hg
Surface transect of the Hg
Spatially aggregated observed and modeled
Hg
For a more detailed analysis, we separate the Baltic Sea into three
regions: (1) the western part, which includes the Belt, Arkona, and Bornholm
seas; (2) the Gotland Sea in the central Baltic; and (3) the northern part
which includes the Bothnian Sea and Bothnian Bay. Moreover, we evaluate the
oxic surface/intermediate waters and the deep anoxic waters in the Gotland
area separately (Table 7). It is seen that the model is able to reproduce
surface concentrations in the western and central areas with a bias close to
zero. The model bias is larger in the deep basins, but model performance is
still comparable to the North Sea. Here, the low vertical resolution in the
model setup below 100 m will play a role. In the northern part, the model
strongly overestimates Hg
Figure 10 depicts the seasonality and Fig. 11 three vertical profiles in the Gotland Sea. It is seen how quickly Hg concentrations can change in this region and, depending on physical drivers, how different the seasonality of vertical mixing can be. At location A (Gotland Deep) Hg concentrations are around 1.5 pM for most of the year with a strong surface depletion (1 pM) during August and September. At location C, located at the opposite side of Gotland, the seasonality is reverse with the highest concentrations (1.2–1.4 pM) during August and September and much lower concentrations (0.9–1.1 pM) throughout the rest of the year.
In summary, our conclusion is similar to that of the North Sea, i.e., that
better data on Hg inputs from rivers and a better resolution of the physical
processes in the domain seem the most promising options for improving model
performance. Especially in the Bothnian Bay, Hg cycling seems to be strongly
influenced by terrestrial organic matter. In the central Baltic, we found
that typically used
Vertical Hg profiles in the central Baltic Sea observations (red) (Soerensen et al., 2018) and model values (blue) for the three central Baltic deep basins given in Fig. 9.
Finally, as the deep basins of the Baltic Sea are anoxic, in this area sulfur chemistry becomes relevant (Reactions R6–R9, Table 3). The effect of adding HgS and HgS–DOM to the chemistry scheme leads to particulate Hg–POM transforming into dissolved HgS species. The effect of this is two-fold: (1) firstly, Hg that is scavenged from the stratified surface layer by detritus (biological pump) accumulates directly at the boundary between oxic and anoxic waters. (2) Secondly, as eventually all inorganic Hg is transformed into HgS species, particle settling stops being a sink and Hg persists in the water column, whereas Hg is effectively transported to the sediment in model runs without sulfur chemistry. This leads to Hg concentrations being constant in the anoxic layer with higher values found only directly at the seafloor. Comparing to observations, we find that the model with sulfur chemistry is better able to capture the observed Hg distribution (Soerensen et al., 2018).
In the marine environment, elemental Hg the reducible fraction of Hg the parameterization of biologically induced reduction processes; the modeled photon flux and wavelength-dependent extinction in water
impacting photolytic reduction; air–sea exchange parameterizations, especially during high wind speeds.
Due to the fast exchange between atmosphere and water, Hg
Comparison of modeled and observed Hg
Comparison of modeled and observed Hg
The observed annual average Hg
We acknowledge that the redox chemistry used is based on measurements in
the Baltic Sea (Kuss et al., 2015). Thus, it needs to be investigated
whether it shows equally good performance for other marine regions. We find
that the model performs similarly well throughout the year with the largest
bias during summer, when the dynamics driving biological and photolytic
reduction lead to a higher variability in Hg
Seasonal breakdown of Hg
Figure 13 depicts the seasonality of a mean Hg
Annual profile of mean Hg0 concentration in the
Baltic Sea
Due to the complexity of the analytical methods and the extremely low
environmental levels of observed concentrations, MeHg observations in the marine environment are rare.
Additionally, they are the most uncertain observations. Here, to calculate
the MQO, we assume an uncertainty of 50 %. We evaluate the model
predictive capabilities in reproducing (1) MeHg concentrations and (2) the
methylated Hg fraction
Evaluating the relative
Observed (Kuss et al., 2017; Soerensen et al., 2018) and
modeled frequency distribution of the methylated Hg fraction
Evaluation of seasonally and vertically clustered
Vertical MeHg profiles for Baltic deep basins. Negative oxygen
concentrations indicate sulfide concentrations.
The model can reproduce the seasonality and vertical gradient of the methylated fraction. On the one hand photolytic demethylation leads to lower MeHg concentrations in the surface ocean during summer. On the other hand, biological activity leads to increased MeHg formation in spring and summer. We find that a biologically induced methylation parameterized with biomass or phytoplankton concentration leads to spring becoming the season with the most effective net methylation. By linking biological methylation to the remineralization of organic carbon, we introduce a temperature dependency that shifts this towards summer (Fig. 16) (Eq. 9, Sect. 2.3.1). Yet, the model still overestimates methylation in spring and underestimates methylation in summer. For a more detailed analysis, we look at surface layer MeHg concentrations on four specific dates. Figure 17 depicts MeHg measurements for 21 March and 1 August of the years 2014 and 2015. In March MeHg concentrations are between 40 and 300 fM and in August between 10 and 200 fM with pronounced spatial gradients. This “spottiness” of the MeHg concentrations partially explains the large random error in the model. Moreover, while the general patterns are similar, methylation shows a significant interannual variability (Fig. 17).
Seasonality of the biologically induced methylation reaction using different parameterizations (Reaction R12, Table 2).
Methylmercury concentrations in the surface ocean on
Overall, the model reproduces 53 % of MeHg values within a factor of 2. We
find that the model performance (MQO
Figure 18 depicts annual average Hg loads in the different ecosystem biota
species. The North Sea exhibits higher Hg loads in biota, which can be
explained by the high Hg load from rivers, especially the Elbe and Scheldt; the
lack of permanent sedimentation; and the earlier onset and higher overall
primary production, which increases the effectiveness of the active uptake
pathway. The average amount of Hg in biota ranges from 1 % to 5 % of the
Hg
Annual average Hg
Seasonality of modeled
As the last step of the model evaluation, we compare Hg
Next, we evaluate the model capabilities to reproduce Hg content in fish.
For this, we compare the modeled bioaccumulation in the functional
ecosystem group representing fish to herring. This pelagic species
corresponds best to the fish functional group implemented into ECOSMO
(Daewel et al., 2019). The analysis is based on 1166 measurements of Hg
in fish muscle tissue. We use the same conversion factors as for zooplankton
to convert the model carbon dry weight to wet weight total biomass (1 ng g the Swedish west coast, a stripe from Gothenburg to Oslo; the southern Baltic Proper, which includes the Bornholm Sea and the
southern Gotland Sea; the northern Baltic Proper, which includes most of the Gotland Sea; the Bothnian Sea; the Bothnian Bay.
It is not possible to compare the caught fish to an individual model grid
cell and time step. Therefore, we compare them to observed average Hg
Modeled and observed frequency distribution of Hg in fish in the Baltic Sea regions (Soerensen and Faxneld, 2020).
In this paper, we present the regional-scale 3D high-resolution biogeochemical multi-media Hg model MERCY v2.0. The numerical model combines hydrodynamic models for the atmosphere and ocean, including a marine ecosystem model. MERCY includes a comprehensive marine Hg scheme to calculate transport, transformation, and bioaccumulation. The schemes for chemistry, partitioning, and bioaccumulation are based on literature values, and no domain-specific model tuning has been done. We would like to emphasize that MERCY is suitable for any marine region or even for global application. The major factors when applying the MERCY model to other regions are (1) partitioning coefficients to organic material (OM) as the type of OM varies regionally; (2) the parameterization for biogenic reduction as the values presented here are based on cyanobacteria in the Baltic Sea; and (3) the ecosystem model, as trophic dynamics and phytoplankton uptake rates can vary widely between regions. To our knowledge, it is the first model capable of linking atmospheric Hg emissions to MeHg accumulation at higher trophic levels. The intention of this initial model publication is the detailed presentation of the model and first results, focusing on model performance evaluation and the identification of the processes and parameters responsible for the model error. A more comprehensive analysis of the dynamics of and variability in Hg speciation, partitioning, and bioaccumulation is required for future studies. While our model performs more realistically than earlier models for marine Hg cycling, there are still large uncertainties, especially regarding methylation.
Model performance evaluation of Hg
n/a: not applicable.
We evaluated model performance for key Hg species based on a simulation for
the North and Baltic seas for the years 2000 to 2016. We chose these regions
due to the availability of observations. Moreover, the two regions cover a
range of regimes, have high primary productivity, and are relevant to
fisheries. Unlike atmospheric Hg modeling, there is no precedent or
scientific consensus defining the state-of-the-art requirements and
limitations of reproducing concentrations of different marine Hg species.
Considering the inherent uncertainty in a comparison of model values and
observed concentrations (e.g., measurement error, sampling error, error in
the hydrodynamic models, the uncertainty in reaction rates, and unknown
processes), we define model values within a factor of 2 of the observations
as a reasonable agreement. Moreover, we used a statistical model quality
objective (MQO
A detailed model performance evaluation for the North and Baltic seas
demonstrates that the model can reproduce concentrations and seasonality of
single Hg species to a degree that validates the model predictive
capabilities. For Hg
We summarize that the improvement of the model performance for Hg
The model performed best for elemental Hg
Evaluation of MeHg resulted in the methylated fraction
Finally, we evaluate the model's ability to reproduce Hg in biota. Our model provides Hg and MeHg loads in phytoplankton, zooplankton, and fish which are inside of the observed range. We find that the modeled phytoplankton concentrations vary within the observed maximum and minimum loads. Zooplankton changes at the trophic level over the course of the year due to changes in diet. As expected, the model predicts the highest MeHg loads in fish, making up 90 % of the total Hg in fish due to its high transfer efficiency. Most parameters used for bioaccumulation are highly uncertain, and there is ample room for improvement in this part of the model. We hypothesize that the ecosystem model which is focused on correctly reproducing carbon fluxes, needs improvements regarding functional traits relevant to bioaccumulation such as size, shape, or feeding behavior.
The presented model allows hypothesis testing within a consistent physical–biological–biogeochemical framework based on basic principles. We are currently working on a model version that allows for seamless coupling with different hydrodynamic ocean and marine ecosystem models to increase the applicability of the model. The model performance is here only cursorily evaluated to limit the length of the paper. For the future, we plan to investigate the sources of model uncertainty and sensitivity in order to identify the insufficient understanding of the processes and find out the imprecise or unknown parameters, especially concerning methylation and biological uptake. Finally, we want to employ and promote the MERCY model as a tool for hypothesis testing and prediction within a consistent physical–biological–biogeochemical framework based on basic principles. This will enable researchers to (1) improve our understanding of the natural variability from seasonal to decadal timescales; (2) investigate forcing dynamics, leading to MeHg accumulation in seafood; and (3) estimate the impact of anthropogenic and natural drivers in support of the Minamata Convention on mercury.
The MERCY v2.0 source code is freely available at
COSMO-CLM v4.0 is freely available at
CMAQ v4.7.1 is an active open-source development project of the US Environmental Protection Agency (EPA)
that consists of a suite of programs for conducting air quality model
simulations. The model is freely available at
HAMSOM-ECOSMO_2e2 v1.0 is freely available at
The code is also available from the Helmholtz Centre Hereon Git
repository:
The data used in this article are available as follows:
ERA-Interim ( GEBCO 2022 gridded bathymetry ( Mercury observations in water and sediments of the North and Baltic seas – MARENET ( Mercury observations in fish (
The authors contributed as follows.
The contact author has declared that none of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We want to thank all data providers for their invaluable input, without which we would not have been able to develop the MERCY model. Special thanks to Lars-Eric Heimbürger-Boavida for fruitful discussions on marine Hg cycling and Franz Slemr for sharing his vast knowledge on atmospheric chemistry and Hg cycling and for proofreading the final paper.
This research has been supported by Horizon 2020 (GMOS-Train (grant no. 860497) and ERA-PLANET (grant no. 689443)), the Swedish Research Council Formas (grant no. 2021-00942) and the Helmholtz Association of German Research Centers, Program Oriented Funding (POV-IV).The article processing charges for this open-access publication were covered by the Helmholtz-Zentrum Hereon.
This paper was edited by Andrew Yool and reviewed by Yanxu Zhang and one anonymous referee.