Operational ocean forecasting and reanalysis systems that integrate in situ measurements, remote sensing observations, modelling and data assimilation are fundamental tools for understanding the variability and dynamics of the physical and biogeochemical ocean state. Such systems are also essential for a better and more sustainable management of the oceans and marine ecosystems, supporting the development and understanding of human activities and the blue economy (von Schuckmann et al., 2016). In this context, the presentation of the underlying science, continuous evaluation and development of the forecast systems are required to provide the best possible forecast and reanalysis.

The presented model version, ECOSMO II(CHL), is adapted from the biogeochemical model ECOSMO (Schrum et al., 2006), later ECOSMO II (Daewel and Schrum, 2013; DS2013), and is currently used as the marine biogeochemical model for operational forecasts (https://doi.org/10.48670/moi-00003) of the Arctic Ocean (ARC MFC – Arctic Marine Forecasting Centre) under the umbrella of CMEMS (The European Copernicus Marine Environment Monitoring Service; https://marine.copernicus.eu/, last access: 11 May 2022). The biogeochemical forecast ECOSMO II(CHL) has been operational since April 2017, and the daily values of the selected variables can be retrieved from the CMEMS database. While based on the ECOSMO II version presented in DS2013, the transfer of the model to a different circulation model, region and model resolution necessitated an adjustment of model parameterizations and additional functionalities, which in turn required a series of new sensitivity tests.

ECOSMO II is an intermediate-complexity nutrient–phytoplankton–zooplankton–detritus (NPZD) type model describing the trophic interactions between three phytoplankton and two zooplankton components. It was shown to successfully simulate the seasonal and inter-annual ecosystem variability of primary and secondary production in the North and Baltic seas (Daewel and Schrum, 2013). In the framework of the ARC MFC forecasting system which covers the northern part of the Atlantic Ocean and the Arctic, its application and scientific scope was shifted to be used for the open-ocean and sea-ice-covered domains. Furthermore, when moving from one circulation model to another, biogeochemical models will behave differently as a result of differences in the physical model (Skogen and Moll, 2005). Both these changes require adjustments to the model formulation and parameters to give good results in the focus regions. ECOSMO II(CHL) most notably introduces chlorophyll a as a prognostic variable. Allowing a flexible chlorophyll-to-carbon ratio is more realistic and has been shown to be more stable when chlorophyll is assimilated (Ciavatta et al., 2011). This addition allows the direct assimilation of ocean colour observations into the forecasting and reanalysis systems. The description of the model changes, added components, and the evaluation of the ECOSMO II(CHL) results within the North Atlantic and Arctic form the main content of this paper.

The North Atlantic above 60∘ N, the focus in this paper, is a typical spring-bloom system (Longhurst, 1998; Rey, 2004). During winter, strong winds and cooling mix the water column several hundred metres and brings up nutrient-rich waters (Nilsen and Falck, 2006). Once the water column stratifies enough for the bloom to start, the diatoms dominate the system. When silicate is depleted, the smaller flagellates and dinoflagellates dominate the phytoplankton community (Rey, 2004). Sporadically there are also extensive coccolithophore blooms covering large areas (Baumann et al., 2000). The main species of mesozooplankton in this area, Calanus finmarchicus, overwinters at depth (Melle et al., 2004) and ascend to the surface at the onset of spring; therefore there is already some zooplankton biomass present at the time of the start of the spring. There is also a fall bloom present as seen from satellite observations. The areas closer to the Arctic, being covered by sea ice, have different dynamics. In sea-ice-covered regions, small blooms can occur in leads and under thin ice but the main bloom commences as the ice retracts (Dalpadado et al., 2020; Dong et al., 2020; Polyakov et al., 2020). Here, sea ice algae will make up some of the primary production (Gradinger, 2009), and other mesozooplankton, such as Calanus glacialis, specialized to the sea ice environment (Melle and Skjoldal, 1998), are also important. Close to sea ice and in coastal regions, early stratification can occur when sea or land ice melts resulting in a seasonal halocline. Water masses in the eastern part of the basin are relatively warm, saline water characterizing the North Atlantic Current (Orvik et al., 2001), while the western part of the basin has colder and fresher water masses of an Arctic or mixed origin (Fröb et al., 2018; Yashayaev et al., 2007).

Our main objective with this paper is to provide the description of the latest updates in ECOSMO II(CHL) and its coupling to HYCOM. We will particularly focus on the description of the prognostic chlorophyll a formulation. We present the results from three experiments using ECOSMO II(CHL) adopting different parameter sets from DS2013 (the original parameter set tuned for the North and Baltic seas), CMEMS Arctic operational model prior to June 2021 and the current Arctic operational model parameterization. We applied these three parameter sets on a model set-up with a coarser grid than used for the operational simulations in order to allow for 2-decade-long simulations for each case. To document the performance of each of these parameter sets for the users of ECOSMO, we present a detailed objective analysis of the lower-trophic-level dynamics for the North Atlantic and the Arctic Ocean against local in situ observations, gridded climatology of nutrients and satellite data in Sect. 4. Following the evaluation we provide information on integrated quantities, such as annual primary production, inter-annual variability in phytoplankton production and seasonal succession of plankton functional types as a reference for the Nordic Seas and the Arctic. We will conclude by commenting on the future updates and implementations of ECOSMO II(CHL).

HYCOM-ECOSMO II(CHL) is a coupled physical–biological model (Fig. 1) where ocean physics are represented by the Hybrid Coordinate Ocean Model (HYCOM: Bleck, 2002) and the lower trophic marine biogeochemistry is resolved by ECOSMO II (Daewel and Schrum, 2013). The models are coupled online and the transport (advection and mixing) of biological state variables is handled as part of HYCOM's own native tracer-transport routines, thus both the physical and biological components use the same time stepping (20 min). The model is one-way coupled, and biology does not affect model physics. HYCOM, as a hybrid vertical coordinate model, can optionally combine the depth-level (z-level), topography-following and density-following (isopycnal) coordinates. In this study, we set vertical levels as the combination of z level for the upper ocean and the mixed layer and isopycnal layers below. The upper five layers are always kept in z levels ensuring a minimum vertical resolution which is important to resolve the light gradient in the upper ocean and thus representing the vertical variation in phytoplankton growth in a realistic manner. Isopycnal layers in the deep facilitates a good conservation of water masses and tracer distributions.

ECOSMO II(CHL) is an intermediate-complexity lower trophic level biogeochemical model which distinguishes four inorganic nutrients (nitrate, ammonium, phosphate and silicate) utilized by three types of phytoplankton (diatoms, flagellates and cyanobacteria). In this study, cyanobacteria are turned off, as they were parameterized to grow below a certain salinity threshold which was intended to represent the cyanobacteria in the Baltic Sea (Daewel and Schrum, 2013). Our area of concern is the high latitudes, specifically the area north of 60∘ N; thus the use of cyanobacteria falls short as a significant phytoplankton community for the region. Two types of zooplankton (micro- and meso-size classes) are parameterized based on their feeding preferences as herbivorous and omnivorous zooplankton and, as additional organic components, dissolved (DOM) and particulate (detritus) organic matter are included in the model. The model uses the molar Redfield ratio between C:N:Si:P components (106 : 6.625 : 6.625 : 1), and nutrients are tracked both in the water column and in the single sediment layer.

The full description of ECOSMO II is given in Daewel and Schrum (2013) (DS2013). In the following we provide a description of differences in the biogeochemical formulations in ECOSMO II(CHL) compared to DS2013. The most notable addition to DS2013 is the prognostic chlorophyll a for each phytoplankton type. The biological interaction (Rchlj) term of the introduced chlorophyll a for P1 and P2 (diatoms and flagellates respectively) is in similar fashion to that of RPj in DS2013, and the source terms are modified by the photoacclimation factor (ρchlj) which accounts for the variation in chlorophyll-to-biomass ratio resulting in increased chlorophyll production under low light conditions (Geider et al., 1997), hence 1Rchlj=ρchljσjϕPjCPj-∑i=12GiPjCZiChlPjCPj-mPjChlPj, where 2ρchlj=θPmaxϕPjCPjαPjI(x,y,z,t)Chl,3ϕPj=minα(I)βNβPβSi,4α(I)=tanh⁡(φI(x,y,z,t)),5βN=βNH4+βNO3,6βNH4=NH4/NH4+rNH4,7βNO3=NO3/NO3+rNO3exp⁡-γNH4,8βPO4=PO4/PO4+rPO4,9βSi=Si/Si+rSi,10GiPj=σi,Pjai,PjCPjri+Fi,11Fi=∑j=12ai,PjCPj, with j=1, 2 denoting the specific phytoplankton types and i=1, 2 the specific zooplankton types. C denotes carbon concentration specific to P (phytoplankton) and Z (zooplankton) in mg m-3, while Chl denotes chlorophyll a concentration in mg m-3. Photosynthetically active radiation (PAR) is given as I(x,y,z,t). DS2013 gives σj, ϕPj, φ,rNH4,NO3,PO4,Si, γ, Gi and mPj as the phytoplankton maximum growth rate, growth limitation, photosynthesis efficiency parameter, nutrient-specific half-saturation constant, NH4 inhibition parameter, zooplankton grazing rates and mortality rates respectively. σi,Pj denotes zooplankton specific grazing rate with ai,Pj and ri representing food preference coefficient and half-saturation constant respectively where Fi denote the total available food for the individual zooplankton. Silicate is not included in flagellate equations. Maximum Chl-to-C ratio (θPmax) is taken from Bagniewski et al. (2011), where they have tuned those parameters for the region south of Iceland. We note that their parameterization is N-based, while ECOSMO II(CHL) uses C-based parameters; thus we applied the conversion following the C:N Redfield ratio of 6.625 resulting in flagellates and diatoms with 0.048 and 0.037 mg Chl mg C-1 respectively. In relation to the addition of a prognostic chlorophyll a state variable, photosynthetically active radiation I(x,y,z,t) at depth undergoing attenuation was modified to have chlorophyll a in the exponential term: 12I(x,y,z,t)=Is(x,y)2exp⁡-kwz-kChl∫z0∑j=12ChlPj∂z, where Is(x,y) is the surface net solar radiation (W m-2) converted to PAR, and x,y identifies the models' horizontal grid points, with z the water depth in metres. kw and kChl are light extinction due to water (m-1) and chlorophyll a concentration (m2 mg Chl-1) respectively.

In addition to prognostic chlorophyll a state variables, phytoplankton and zooplankton loss terms now have an on/off switch regulated by a minimum concentration criterion preventing them from decreasing to very low concentrations. This allows them to recover and quickly respond to suitable growth conditions experienced in spring. The switch is applied to mortality and grazing terms for phytoplankton and chlorophyll a, and to mortality terms for zooplankton. The minimum concentration at which the loss terms are switched off are 0.1, 0.005 and 0.01 mg C m-3 for phytoplankton, chlorophyll a and zooplankton respectively.

Model simulations are configured on a relatively coarse grid that varies between 30 and 70 km where the highest resolutions are located in the mid-North Atlantic (Fig. 2). Although having finer resolution was previously shown to better represent nutrient dynamics for our domain (Samuelsen et al., 2015), the main purpose of our study is to introduce the required model structure for the North Atlantic and Arctic region and the experiments, which requires numerous tests and simulations in parallel. Therefore, we concluded that having a relatively coarse grid size fits better for our purposes.

Data for atmospheric forcing is retrieved from ECMWF ERA-Interim reanalysis with 6 h resolution (Dee et al., 2011). The variables used to force the ocean model are 10 m winds, air temperature at 2 m, dew-point temperature at 2 m, cloud coverage and total precipitation for the physical model and surface net solar radiation for the biogeochemical model. River runoff is modelled using a hydrological model, TRIP (Oki and Sud, 1998), resulting in a monthly climatology dataset, so the river runoff does not include any interannual variability. River runoff affects only salinity. Nutrient loads from the rivers are derived from the modelled dataset, GlobalNEWS (Mayorga et al., 2010; Seitzinger et al., 2010), and include nitrate, phosphate and silicate. Nutrient loads were scaled by the TRIP runoff volume resulting in monthly climatology loads.

The model physics was initialized in 1989 from a spin-up simulation that started in 1948 forced by the ECHAM6 atmospheric simulation (Schubert-Frisius and Feser, 2015). The biogeochemical model used inorganic nutrients (nitrate, phosphate and silicate) from the World Ocean Atlas 2013 (García et al., 2013) monthly climatology as the initial conditions; the biomass concentrations were initialized with uniform, low values. The same climatology was used for the relaxation of temperature, salinity, nitrate, silicate, phosphate and oxygen at the open boundaries. The simulation was conducted until the end of 2010. The results are evaluated starting with the year 1991.

In this study, we employ three sets of simulations (EXP1, EXP2 and EXP3) that use different phytoplankton growth rates, photosynthesis efficiency and zooplankton mortality rates (Table 1). EXP1 uses the DS2013 parameter set which was used for the North Sea and the Baltic Sea. Additionally, we introduce EXP2 (which uses the parameter set for the operational forecast model for ARC MFC prior to July 2021) and EXP3 (which uses the parameter set for the operational forecast model for ARC MFC currently online following July 2021). Since these parameter sets represent active use cases, the objective analysis of these experiments in the following sections provide the users of ECOSMO a reference on how the model performs with different set-ups and longer timescales. For the purpose of comparing these parameters, EXP2 and EXP3 can be considered as part of the same group against EXP1 such that in both EXP2 and EXP3, phytoplankton growth rates are set higher compared to EXP1. The reasoning behind this increase is a response to deep winter convective mixing and resulting light limitation on growth in the open ocean. To control excessive growth of phytoplankton in the following seasons, zooplankton grazing rates were increased. ECOSMO II has been used as an operational model for the Arctic since 2017, and its parameterization has been tested and improved various times, more than we can document here. Thus the parameter sets for EXP2 and EXP3 are provided here as milestones for ECOSMO II development. To document the development process of the ECOSMO, we present all the experiments representing DS2013 (EXP1), the operational model prior to July 2021 (EXP2) and the next-phase operational model (EXP3). The model evaluation is followed by an overview of the notable aspects of the simulated biogeochemistry of the North Atlantic and the Arctic oceans.

In addition to the 3D-HYCOM-ECOSMO II(CHL) simulations, we have also performed a 1D simulation using the General Ocean Turbulence Model (GOTM; Burchard et al., 2006) using the Framework for Aquatic Biogeochemical Models (FABM; Bruggeman and Bolding, 2014) as the online coupler at Station M (66∘ N, 2∘ E) in the Norwegian Sea to present the differences of ECOSMO II and ECOSMO II(CHL) both visually and statistically comparing versions of ECOSMO with and without explicit chlorophyll a. The details of this set-up are provided in Appendix A1. Station M is a long-term time-series station and is representative of the Norwegian Sea dynamics, and data from Station M are often used for the development of ECOSMO. The dynamics shown in Appendix A1 is representative for regions with similar plankton dynamics (e.g. Norwegian Sea, Barents Sea) and thus can be used as a showcase for the new chlorophyll a specific addition. Apart from the improvement of model chlorophyll a results, the addition of dynamic chlorophyll a establishes a higher level of functionality of ECOSMO such that phytoplankton functional types now have their unique carbon : chlorophyll a ratios and initial slope of P–I curves which enables better adaptability to different environments, and the model now has better integration with observation systems (e.g. remote sensing) and future improvements toward bio-optical modelling.

In this section, we present a selection of model results to provide an overview of the performance of ECOSMO II(CHL). While the model domain extends to the equatorial regions, our focus is on the Nordic Seas and the Arctic. We present the evaluation of the observable model output against in situ data with the relevant statistics. The focus of this assessment is on the key parameters of the chemical and biological fields on a regional scale where the subdomains defined for model assessment are given in Fig. 2. This approach allows for assessment of the local biogeochemical characteristics of the model. The purpose of this assessment is twofold: (1) to assess the model formulation and its parameterization as a regional hind-casting and forecasting tool, as a component of CMEMS, and (2) to introduce the model as a tool for scientific studies.

The extent of the subdomains depicted in Fig. 2 were defined by the geographical definitions of the regions and their environmental setting, as such the BARENTS region covers the shelf area south and east of Svalbard and border NOR. N at the opening to the Norwegian Sea where it is deeper and is highly influenced by the Atlantic inflow. The Norwegian Sea is divided into north and south to take into account for the differences in day length across the wide latitude range (20∘). The border between GREENLAND and NOR. N and S roughly locates the temperature changes of the different water masses in the region (Fig. 2). ARC regions were set to cover sea-ice-covered regions most of the year. The BERING STR. region was set to separate the boundary conditions from the rest of the domain. KARA and LAPTEV regions have naturally defined borders with the islands around them. SPG region is defined to represent the subpolar gyre region.

Primary production (PP) is the foundation for all marine biological production and the most frequently observed rate in BGC models. Still, there are only few observations of primary production available in the ocean as a whole, but the high Arctic is particularly poorly sampled (Matrai et al., 2013). Because the model does not have an explicit term for respiration, we can only extract gross primary production from the model, which is then compared to observations. The modelled gross annual primary production ranges from above 200 g C m-2 yr-1 in the southern part of the model domain to almost zero in the central Arctic and features a gradual decrease from 144.26 to 41.48 g C m-2 yr-1 from lower latitudes (SPG) towards the higher latitudes (Barents) respectively, with a sharp decrease to very low values (<6 g C m-2 yr-1) in the sea-ice-covered areas (Fig. 5) as a consequence of light limitation. Rey (1981) estimated the primary production in the Norwegian Coastal Current to range from 90–120 g C m-2 yr-1, which agrees well with the values from this model (Fig. 5), although the used coarse resolution model does not represent a very distinct coastal current. Previous studies have estimated the primary production in the Fram Strait from 50–80 g C m-2 yr-1 (Hop et al., 2006), while our model show values of 90–100 g C m-2 yr-1 in the Atlantic waters and up to 30–60 g C m-2 yr-1 on its western side. Lee et al. (2015) compared multiple Arctic models against in situ observations. Only a few of these observations were in the central Arctic, while the majority were located in the Chukchi Sea, which is very close to the zone where the model is relaxed to climatology. They found a median value of all Arctic observations of 246 mg C m-2 d-1 which corresponds to about 90 g C m-2 yr-1. The regional estimates of primary production were similar, but the shelf regions were the most productive. The model results for the regions surrounding the central Arctic Ocean fall in the range of this estimate, but observation base estimates for the central Arctic, although only few are available, are higher than the model results. From Lee et al. (2015) the primary production estimates from the central Arctic varied between 10 and 100 mg C m-2 d-1 (∼4–40 g C m-2 yr-1) while the model is below 1 g C m-2 yr-1. In the model formulation, the ice is blocking more light than what is realistic and ice leads cannot be resolved, so our estimate is expected to be low in ice-covered regions. It is known that both melt ponds and leads can act as windows into the ocean, facilitating blooms (Assmy et al., 2017). The light below the ice will be improved in future versions of the model system. In situ observations in the Arctic range up to more than 5000 mg C m-2 d-1. The model does not reproduce the extremes in primary production, but the mean values are overall consistent with available observations.

For the Norwegian and Barents seas, the modelled primary production shows distinct seasonal patterns with almost negligible productivity between November–April due to low light availability (Fig. 6). During the onset of the spring bloom, production is notably at its highest during May–June followed by a gradual decrease towards late fall. Regional differences in primary production are also evident in year-round time series, where the Norwegian Sea primary productivity is significantly higher than the Barents Sea productivity. The southern part of the Norwegian Sea (NorwegianS) has a notably earlier (∼2 weeks) bloom compared to the northern counterpart.

The simulated seasonal evolution of primary production reflects the growth of plankton functional types, with diatoms (compared to flagellates) being the dominant type in the Nordic Seas spring bloom (Fig. 7). A relatively minor flagellate bloom follows a few weeks after that of diatoms. Zooplankton biomass increase from May–June in response to phytoplankton growth and is maintained until the end of the year. Note that ECOSMO II(CHL) allows zooplankton to feed on detritus, which contribute to zooplankton-sustaining growth beyond the seasons of phytoplankton activity. Towards the lower latitudes, south of 45∘ N, flagellates maintain a similar annually integrated productivity (∼1.5 vs. ∼1.0 g C m-2) to that of diatoms (Fig. 5c–e). Mesozooplankton are the dominant grazer in all regions (Fig. 5d–f). Similar to primary production, the NorwegianN and NorwegianS functional type biomasses are higher compared to Barents functional type biomasses with daily 200 m averaged biomasses reaching 75–100 mg C m-3 for diatoms and mesozooplankton in the Norwegian Sea, and ∼50 mg C m-3 for the Barents Sea respectively. For both Barents and Norwegian seas, flagellate and microzooplankton biomasses do not exceed ∼25 mg C m-3 during their highest productive seasons (Fig. 7).

The model predicts regionally high annually averaged inorganic nutrient concentrations for the subpolar gyre compared to the Norwegian and Barents seas which is also reflected in monthly and regionally averaged concentrations (Fig. 8) with relatively lower concentrations in the coastal regions of the Nordic Seas compared to their offshore regions. The model also predicts a contrast between nutrient-specific regions of high concentrations. Nitrate concentrations are higher at the lower latitudes, whereas phosphate and silicate are higher towards the higher latitudes. These features generally agree with the features of WOA2013 data (Fig. 8). As the model is relaxed towards the climatology at the Bering Strait through a sponge layer in the model domain, the overall high nutrient concentrations near the Bering Strait and especially the high silicate concentrations at the Siberian coast due to the higher Si/N ratio of Pacific origin water masses compared to the Atlantic water masses and the addition of high Si/N ratio river discharge is reflected in the modelled annual averages. As mentioned earlier, the model does not allow light to penetrate sea ice. For this reason, the model overestimates surface inorganic nutrients compared to climatology below the sea ice as these nutrients are not consumed by primary production but are only affected by transport and remineralization. Overall, the model performs well in terms of N/P molar ratios (NO3/PO4; Fig. A4). Both model and climatology suggest a higher N/P ratio for the Nordic Seas and lower latitudes (∼12–16). At the northern and southern Barents Sea, the climatology has a lower N/P ratio (<7) but has a high ratio at the ice-edge region (>17). In contrast, the model predicts a more regular N/P distribution with a gradual decrease from 16 to 12 from lower to higher latitudes at the Barents Sea.

Here we present the evaluation of each model experiment against satellite chlorophyll a. Since the parameters in EXP2 and EXP3 are used in open-ocean operational models, their performance in representing satellite chlorophyll a is vital for the assimilation of chlorophyll a in the operational model. The comparisons of co-located surface in situ, model and satellite data are given in Fig. 9, and their statistics are summarized in Table 4. The purpose of comparing the satellite data to both in situ and model data is to evaluate the satellite product itself for the region, as satellite products are prone to uncertainties based on the algorithms used and are related to differences in absorption and backscattering properties of phytoplankton and concentrations of coloured-dissolved organic matter (CDOM) and minerals (Dierssen, 2010). Thus, in the absence of in situ data, we have a better understanding when model and satellite data are compared. Table 5 summarizes model statistics against satellite data, which is both independent of the in situ samples, and, due to the volume of satellite data, the statistics here are based on a much more extensive dataset compared to the statistics in Table 4.

For the three regions of interest, the satellite data have a negative bias against the in situ data (Table 4). The percentage bias is minor for the Barents region (-5.32 %), but for the Norwegian Sea, the biases are -21.34 % and -16.11 % for the north and south respectively. The NSDs range between 0.51–0.65 mg m-3 suggesting that satellite chlorophyll a underrepresents the amplitude of the in situ observed variability of chlorophyll a. Satellite chlorophyll a RMSEs range between 0.6–0.8 mg m-3.

EXP1 has higher chlorophyll a concentration compared to EXP2 and EXP3. This is visually evident when model and satellite data are plotted against in situ data (Fig. 9) where EXP2 and EXP3 generally form clusters distinct from EXP1. EXP1 chlorophyll a are mainly located at the right side of the 1-to-1 line suggesting a positive bias against the in situ data evident in Table 4 with 39.75 % for the Barents Sea and 178.25 % and 140.56 % for the north and south Norwegian Sea respectively. The RMSEs and NSDs are also higher compared to EXP2 and EXP3. Relatively, EXP1 is the least representative of the in situ chlorophyll a data among the experiments. EXP2 and EXP3 overestimate chlorophyll a for the Norwegian Sea with percentage biases ranging between 23.5 %–50.76 %, and over-representing the amplitude of variability with NSDs ranging between 1.59–2.68 mg m-3. For the Barents Sea, while EXP2 and EXP3 have negative biases and RMSEs around ∼1 mg m-3, their NSDs show that they correctly estimate the amplitude of variability. With a much larger number of data points, the model error statistics computed from satellite data are similar (Table 5) with EXP1 resulting in the highest chlorophyll a values and statistically performing the worst compared to EXP2 and EXP3. Notably, EXP2 and EXP3 biases are much lower compared to the statistics against in situ data with the exception of NorwegianN, as well as performing with fewer errors overall. Satellite data also increase the regional coverage of the statistical analyses where SPG region statistics show that EXP2 and EXP3 outperform the EXP1 statistics (Table 5). The consistent higher bias of EXP1 compared EXP2 and EXP3 can be explained by its higher photosynthesis efficiency (Table 1). EXP1 shows a very fast primary production response to light availability during the spring bloom period with notably higher chlorophyll a concentrations compared to the observations (results not shown) evident in the high percentage biases, whereas chlorophyll a concentrations in EXP2 and EXP3 are closer to observed values during spring bloom. Originally, the ECOSMO II parameterization was set for the North Sea and the Baltic Sea with different light conditions. In the open ocean such a high response curve leads to an overestimate of the bloom. However, winter convective mixing is very deep in the Nordic Seas; thus the light is a limiting factor on growth. To overcome deep mixing and prevent a late spring bloom, the phytoplankton were allowed to have very high growth rates for EXP2, and relatively fewer higher growth parameters were set for EXP3. Statistically and visually (Fig. 9), both EXP2 and EXP3 are very similar, with EXP3 performing slightly better statistically.

The statistical analysis performed against satellite chlorophyll a highlights the use of satellite data as an independent dataset for model evaluation, and its domain-wide (though limited to surface) coverage allows for a more composite evaluation of the model as a whole. Satellite data is acquired in near-real time and thus presents a valuable opportunity for an operational model validation, whereas model validation with in situ data has significant delays (though it is very valuable for hind-cast evaluation). Recent additions to satellite datasets such as the phytoplankton functional types (e.g. https://doi.org/10.48670/moi-00099) further detail the use of satellite data for models. As an operational model, ECOSMO II(CHL) works well with satellite data with the inclusion of explicit chlorophyll a variables for each phytoplankton functional type (PFT). ECOSMO II(CHL) not only better resolves surface chlorophyll a with its light-dependent dynamic carbon : chlorophyll a ratios (see Sect. A1), PFT specific parameters such as initial slope of P–I curves add further details to model adaptability to varying environmental conditions (Fig. A1d) compared to an average constant ratio common to all PFTs. PFT-specific model configurations further synergize with satellite PFT observations in the context of operational biogeochemical modelling. Future iterations of ECOSMO should also include such kinds of evaluations. An important addition to explicit chlorophyll a variables is the inclusion of the initial slope of P–I curves to light limitation on growth. In this study, light limitation was approached in a PFT- and chlorophyll-a-independent fashion (Eq. 4). Future versions of ECOSMO should adopt ways (e.g. Evans and Parslow, 1985) to include the PFT-specific P–I curve slopes to take full advantage of the explicit chlorophyll a variable. This would allow PFTs to differentiate their niche light conditions for production and further allow better integration with the bio-optical modelling of the marine environment.

We note that the statistical analysis results against satellite chlorophyll a contradict the statistics against in situ data (Sect. 4.3), as EXP1 performed better in some cases such as percentage bias for the Barents Sea and south Norwegian Sea. However, in the analysis against satellite chlorophyll a, EXP1 was statistically outperformed by EXP2 and EXP3. The first possible cause of this difference may be that the in situ data and satellite data are different datasets such that they cover different locations and seasons and use different size of data points. In situ data were restricted in both the overall number of data points and the seasons as most of the data were from late spring and onwards, whereas satellite data also cover earlier parts of the year under favourable weather conditions. As a result, model statistics may have a seasonal bias towards the timing of the in situ sampling. Second, satellite data cover only the surface of the water column where in situ chlorophyll a were well below the penetration depth of the satellites which might affect the statistics.

In this paper we present the mathematical description of ECOSMO II(CHL) which is used as the biogeochemical model for the operational forecasting of the Arctic Ocean. We document ECOSMO II(CHL) model performance by objectively analysing the model inorganic nutrients and chlorophyll a against available data spanning from climatology to in situ and satellite chlorophyll a data. We compare three experiments with different parameters representing the original implementation of ECOSMO II, CMEMS Arctic operational ECOSMO II(CHL) for the years 2016–2021 and the current (since June 2021) operational ECOSMO II(CHL). Through presenting the model description and its evaluation, we document the performance of ECOSMO for each of these use cases for the users of the model. While each set-up performs better for some variables or datasets, the qualitative and quantitative evaluation of the model results of inorganic nutrients, chlorophyll a and primary production for each case demonstrated that the model is consistent with the large-scale climatological nutrient variability and is capable of representing regional and seasonal changes. The model primary production agrees with previous measurements. ECOSMO II(CHL) benefits from the use of an explicit definition of phytoplankton functional type chlorophyll a implementation, i.e. the use of phytoplankton-specific dynamic chlorophyll a-to-carbon ratios in reference to a fixed ratio in the original model, with improved surface estimations of chlorophyll a, and gains added value towards improving model evaluation opportunities using satellite observations and phytoplankton-functional-type-specific additions to model structure. In its current state, ECOSMO II(CHL) with its intermediate-complexity definition of the North Atlantic and Arctic Ocean ecosystem structure including a sediment layer is a capable modelling tool for both scientific and operational use. The modelling structure presented in this study, ECOSMO II(CHL), including the physical model, HYCOM, forms the basis of the modelling framework that the future updates will build on.