Sea ice, permanent or seasonal, covers large areas of the ocean, especially in polar regions. Sea ice is a complex system with many important processes occurring. Being an interface between the atmosphere and the underlying medium, the sea ice surface temperature (in contrast to the sea surface temperature) has a noticeable diurnal cycle. Snow accumulation on the ice is accompanied by specific processes of snow–ice formation during the cold season and by snowmelt and the appearance of melt ponds during the warm season. Freezing of saline water results in brine droplets becoming trapped in the ice. This affects not only the ice thermal properties, but also the ice structure, due to the slow movement of the trapped droplets towards the ice bottom and the formation of channels. Finally, the ice-covered area is not a solid shield but a mixture of floes and polynyas that drift, being forced by the wind and ocean currents. Large-scale ice-covered areas strongly affect the properties of the atmospheric surface boundary layer over them. Night-time radiative cooling of the ice surface may lead to a very stable boundary layer and limited turbulent exchange between the surface and the atmosphere. Over regions containing a mixture of ice floes and polynyas, the turbulent fluxes are affected by form drag. Thus, it is very important to reproduce these processes over the sea ice correctly in numerical weather prediction models.

Simple parameterization schemes for sea ice are traditionally used in NWP applications. Information about the presence of sea ice cover is taken from observations (the analysis), and the sea ice thickness and sea ice temperature are modelled by a parameterization scheme. For parameterization of sea ice in NWP, two main approaches currently exist: sea ice schemes based on the solution of the heat diffusion equation with several ice layers but constant ice thickness, e.g. IFS-HRES (Integrated Forecasting System–High Resolution) by ECMWF (European Centre for Medium-Range Weather Forecasts) prior to version cy45r1 or HIRLAM (High Resolution Limited Area Model) , and bulk sea ice models with prognostic ice thickness and an assumed linear or polynomial shape of the temperature profile in the ice, e.g. COSMO (Consortium for Small-scale Modeling) by DWD (Deutscher Wetterdienst) . Snow on ice in these schemes is either not represented (e.g. IFS-HRES) or represented parametrically via changing the albedo from ice to snow during the melting period e.g. COSMO,. Simple sea ice schemes are used for operational forecasting. However, their performance has mainly been studied in general, with minor validations against observations and without comparisons with more advanced ice models.

More advanced ice models have been developed for ice forecasting applications and research purposes, for example CICE Community Ice CodE;, GELATO Global Experimental Leads and ice for ATmosphere and Ocean; and HIGHTSI HIGH resolution Thermodynamic Snow and Ice model;. They are applied in ocean modelling e.g. and in coupled ocean–ice–atmosphere systems for research purposes, climate simulations, and seasonal forecasting . In operational NWP they are applied in global NWP systems to provide medium-range weather forecasts, e.g. the UK Met Office Unified Model or IFS-ENS Integrated Forecasting System–ENSemble prediction system;. However, there are a number of reasons that advanced sea ice models are not widely used for short-range operational NWP. Firstly, the advanced ice models are computationally expensive. They parameterize in detail many processes that are important for the evolution of the sea ice itself, but are of secondary importance for the description of ice–atmosphere interactions. Secondly, their robustness and numerical stability during coupling with atmospheric models needs more studies within a framework of short-range operational NWP systems. Thirdly, they may require advanced methods of data assimilation for their initialization. In NWP research, in addition to coupled systems, advanced ice models may be used for the performance assessment of simple schemes.

The observations of sea ice properties that are currently used in NWP are very limited and often only indicate the presence of ice. One example is the sea ice concentration product provided by the Ocean and Sea Ice Satellite Application Facility (OSI SAF) . This product uses observations from passive microwave sensors and is included in the OSTIA (Operational Sea surface Temperature and sea Ice Analysis) product . For using the sea ice concentration observations in an NWP system, these data need to be projected onto an atmospheric model grid, which is usually done during the analysis step. Subsequently, the sea ice concentration governs the sea surface schemes: for the ice-covered part of a grid box, a simple ice model runs; for the ice-free part, the sea surface temperature is kept constant. Other observations of ice properties (e.g. sea ice microwave emissivity as discussed by ) are rarely used for assimilation in short-range NWP systems, and very few of them are used for validation. In ocean modelling systems, sea ice concentration and sea ice drift data from different sources, including remote sensing from passive microwave and visible channels, are used see, for example,. Acquisition of the sea ice depth data from active remote sensing is being developed , but the latency of these data is not yet acceptable for operational use in short-range forecasting. However, these data could also provide information about sea ice properties for research applications of NWP systems.

In the ALADIN–HIRLAM (Aire Limitée Adaptation dynamique Développement InterNational–High Resolution Limited Area Model) NWP system both the sea surface temperature and the sea ice surface temperature remain constant during the whole forecast. These variables are initialized from an external source (for example, from the global ECMWF model IFS) for each forecast cycle. This causes noticeable errors in near-surface air temperature forecasts over ice-covered and in ice-surrounded areas, especially for forecasts longer than 24 h.

This study presents the development of a simple sea ice parameterization scheme for the ALADIN–HIRLAM NWP system. The scheme solves the heat diffusion equation on a vertical grid within a sea ice slab of constant thickness. This level of simplification was chosen as a first step. In the scheme, provision is made to allow for coupling with the snow scheme after . Also, in the case of fractional ice cover, the form drag caused by ice floating over the water surface is taken into account following .

The simple sea ice model (parameterization scheme) is checked for sanity and its performance is assessed through a comparison with the off-line sea ice model HIGHTSI . The overall performance of the HARMONIE–AROME (HIRLAM–ALADIN Research on Mesoscale Operational NWP in Euromed–Application of Research to Operations at Mesoscale) configuration of the ALADIN–HIRLAM NWP system with the new sea ice parameterization scheme is evaluated against temperature and wind measurements from coastal meteorological (SYNOP) stations and ice surface temperature products derived from measurements by the MODIS (Moderate Resolution Imaging Spectroradiometer) and VIIRS (Visible Infrared Imaging Radiometer Suite) sensors. The scheme improves the forecast verification scores of the ALADIN–HIRLAM NWP system in coastal areas that are influenced by sea ice but overestimates the sea ice surface temperature in the Arctic, where the prescribed constant value of the ice thickness is too small. The experiments and results described in this paper enable a better understanding of the forecast errors and uncertainties and provide an advancement in the description of the interactions between sea ice and the atmosphere in NWP.

The paper is organized as follows. In Sect.  the scheme description and an overview of the physical equations are given. Numerical methods to solve the scheme equation are described in Appendix . Section  evaluates the performance of the new scheme by comparison with the thermodynamic sea ice model HIGHTSI, measurements from SYNOP stations, and observations from MODIS and VIIRS. In the final section a short summary of the obtained results is given and the perspectives for further developments are discussed. Fortran source code for the SICE scheme version 1.0-38h1 is provided in the Supplement.

The simple sea ice scheme (SICE, pronounced “ess ice”) is developed for the parameterization of sea ice in NWP to predict the surface temperature of a thick layer of sea ice. The ice thickness is prescribed. No ice melting or ice formation processes are included, and the heat flux from water to ice is neglected. Processes of snow–ice formation, which are discussed e.g. by , are not represented. The scheme describes only the processes in the ice slab but it can be coupled to a snow scheme that can provide the value of the heat flux on the lower boundary of the snow layer. The prognostic temperature profile in the ice is obtained from the solution of the heat diffusion equation using the heat balance equation and the temperature of water freezing as upper and lower boundary conditions, respectively: C∂T∂t=∂∂zλ∂T∂z-∂Q∂z0=F+λ∂T∂zz=0ifz=0,T=Tfrzifz=H, where t is time (s); z is depth (m); C is the volumetric heat capacity of ice (Wsm-3K-1); λ is the ice thermal conductivity (Wm-1K-1); Q is the solar radiation flux penetrating through the ice (Wm-2); T is the ice temperature (K); Tfrz is the freezing point of seawater (K); and H is the prescribed ice thickness (m). The term F in the second row of Eq. () represents the balance of incoming downward and upward heat fluxes: F=δHsnow[LW↓-εσTs4-ρacpcHVNTsΠs-TNΠN-LρacHVNqsat(Ts)-qN]+1-δHsnowGsnow, where δHsnow is the Kronecker delta, δHsnow=1ifHsnow=0,0ifHsnow≠0, and Ts is the ice surface temperature (K) (Ts≡T|z=0); LW↓ is the downward longwave radiation flux (Wm-2); ε is the surface emissivity; σ is the Stefan–Boltzmann constant (Wm-2K-4); ρa is the air density (kgm-3); cp is the air heat capacity with constant pressure (Wskg-1K-1); cH is the drag coefficient for heat; VN is the wind speed (ms-1); Π{s,N} is the value of the Exner function on the corresponding level; TN is the air temperature (K); L is the latent heat of sublimation (Wskg-1); qsat(Ts) is the saturation specific humidity near the ice surface (kgkg-1); qN is the specific humidity of air (kgkg-1); Gsnow is the heat flux from snow to ice (Wm-2); and Hsnow is the snow thickness (m). Index N denotes a variable at some level in the atmosphere (the lowest atmospheric model level if the scheme is included in an atmospheric model). The right-hand side of Eq. () is the sum of the longwave part of the radiative balance LW↓-εσTs4 and the turbulent fluxes of sensible H=ρacpcHVNTs/Πs-TN/ΠN and latent LE=LρacHVNqsat(Ts)-qN heat in the case of bare ice or the heat flux from snow to ice in the case when snow is present.

The term Q in the first row of Eq. () describes the heat flux from solar radiation penetrating into the ice pack. This heat flux is calculated by using the Bouguer–Lambert law, with an approximation of radiation absorption in the thin layer of the ice following : Q(z)=δHsnow(1-α)SW↓i0⋅e-k⋅z where α is the ice albedo; SW↓ is the downward solar radiation flux (Wm-2); i0 is the fraction of radiation penetrating through the thin layer of the ice; and k is the extinction coefficient for the ice (m-1), which is parameterized according to values suggested by . The value i0 parameterizes the vertical inhomogeneity of the ice transparency and is dependent on depth. It is equal to 1 in the uppermost 0.1 m layer of ice and equal to 0.18 in the lower layers . Note that in the case of snow on ice, the remaining solar radiation that was not absorbed during penetration through the snowpack is assumed to be completely absorbed by the underlying ice surface.

The main prognostic variable of the SICE scheme is the temperature of the ice. Other parameters are either physical constants or taken from the external forcing. For calculation of the ice thermal conductivity and heat capacity we used the following formulations, which represent their dependency on the ice temperature and salinity : C=C0-TmltS-Tmlt0θ2L,λ=λbi-λbi-λbTmltS-Tmlt0θ, where, following , TmltS=273.15-0.0592S-9.37×10-6S2-5.33×10-7S3λbi=2λi+λa-2Vaλi-λa2λi+λa+2Vaλi-λaλiλi=1.1621.905-8.66×10-3θ+2.97×10-5θ2λb=1.1620.45+1.08×10-2θ+5.04×10-5θ2λa=0.03Va=0.025, and C0 is the volumetric heat capacity of the fresh ice (Wsm-3K-1); Tmlt(S) is a function of the melting point of the saline ice depending on the salinity (K); S is the ice salinity in parts per thousand; λ{i,b,bi,a} is the heat conductivity of fresh ice, brine, bubbly ice, and air, respectively (Wm-1); Va is the fractional volume of air in the sea ice; and θ is the ice temperature in ∘C.

In the case of bare ice (no snow), information about the ice albedo is needed to calculate the surface energy balance. The ice albedo strongly affects the temperature regime of the ice pack. The effects of some processes taking place on the ice surface, such as the effect of melt ponds, may be parameterized through the ice albedo even without their real physical description. In the SICE scheme several different parameterizations of ice albedo are available. In these parameterizations, albedo is defined as a constant value or as a function of the ice surface temperature. Numerical methods to solve Eqs. () and () are presented in Appendix .

The assumption of bare ice is the simplest possible approximation and may give reasonable results. However, such a simple parameterization describes processes on the ice surface covered by snow in a very approximate way. Snow upon the ice serves as an insulating layer with higher albedo and lower thermal conductivity than the underlying ice. For more physically correct simulations the ice scheme should reproduce the processes related to the evolution of snow on the ice surface. The form of the upper boundary condition presented by Eq. (), which contains the heat flux F from the snow layer to the ice layer, allows easy coupling with an external snow model to represent snow on ice. In our study, we used the snow module ISBA (Interactions Surface Biosphere Atmosphere) Explicit Snow ISBA ES, to represent snow processes. In the current version of SICE, when snowpack exists, it always covers the ice part of a grid cell as a layer of uniform thickness.

ISBA ES is a multi-layer snow scheme with prognostic snow water equivalent, snow heat content, and snow density. The number of layers may be defined by the user (the default value is three). The uppermost snow layer is always less than or equal to 0.05 m. The scheme explicitly describes the following processes: snow accumulation due to precipitation, heat redistribution, melting processes, and snowpack compaction. It also represents processes related to the meltwater within a snow layer. The heat diffusion and surface energy balance equations are solved numerically with implicit schemes. The snow module needs information about the atmospheric forcing and the temperature, heat conductivity, and thickness of the topmost layer of ice. It predicts the snow variables and provides the flux from the snowpack to the underlying medium. Thus, the coupling between snow and ice schemes is explicit. The snow surface albedo in ISBA ES is calculated through a simple aging scheme, which contains dry- and wet-snow albedo degradation formulations. In this aging scheme, the snow albedo may decrease during the degradation process from its maximum value of 0.85 to a minimum value of 0.5. When applying this scheme over sea ice, a snow albedo minimum value of 0.75 is used following and . The ISBA ES scheme parameterizes snow over land surfaces and contains no parameterizations of specific snow-over-ice processes, such as snow–ice formation or the evolution of melt ponds.

An atmospheric model may apply a tiling approach for better representation of the surface processes. This means that a model grid cell may contain a mixture of both seawater and ice. In such cases the ice and open water calculations are performed independently and the output flux to the atmosphere is represented by a weighted average of the fluxes from the water and ice parts. In this case, information about the ice concentration may be utilized to obtain the weighting coefficients. Ice concentration is estimated from satellite observations using the analysis procedure. This procedure contains a consistency check between the sea surface temperature and sea ice concentration fields .

Turbulent exchange between the sea ice and the atmosphere is a complex process that is influenced by the morphological features of the ice pack such as the presence of melt ponds, ice topography, ridges, and deformations. In the current version of the SICE scheme these complex features are not represented and the ice part of the grid cell is assumed to be a flat surface with uniform characteristics. When the sea ice concentration is less than 100 % (which means a mixture of open water and sea ice), one more factor influences the turbulent exchange with the atmosphere. This is the form drag, which is caused by the floes floating on the water with their upper edge higher than the water surface (ice obstacles). This subtle effect might be important for NWP systems that use ice concentration data to define the percentage of sea ice in a grid cell. Indeed, the roughness length of water is lower than that of ice, and simple weighted averaging according to the ice concentration values will lead to a decrease in the roughness length (compared to fully ice-covered area), while in nature it should increase. An accurate sea ice scheme should include a parameterization of the drag caused by ice obstacles (the form drag). Such schemes, discussed for example in , usually introduce an additional term in the weighted average, which depends on the ice fraction. The form drag was introduced into the SICE scheme in the following way: Cd,mean=ηCd,ice+(1-η)Cd,sea+Cd,f, where Cd,ice,sea,mean is the drag coefficient over ice, sea, and the mean drag coefficient over the grid cell, respectively, under neutral conditions; η is the fraction of sea ice in the grid cell; and Cd,f is the form drag. The form drag term is calculated by using a parameterization suggested by : Cd,f=7.68×10-3ln⁡(0.41/z0,w)ln⁡(10/z0,w)21-ηβη, where z0,w is the roughness length of the seawater surface and β is the tuning constant. Parameters of the SICE scheme are summarized in Table .

Technically, the SICE scheme was developed as a part of the externalized land and ocean modelling platform SURFEX . The externalized surface modelling platform SURFEX is a set of models used for the description of different types of surfaces: sea and inland water bodies, soil–vegetation, and urban environments. It assumes a tiling approach, distinguishing different surface types within one grid box of an atmospheric model. Each atmospheric model grid box contains some fraction of four different surface types (tiles): nature, urban, inland water, and sea. Fractions of these tiles are permanent model parameters obtained from land-use maps. For land-use mapping (physiography) SURFEX incorporates the 1 km resolution database ECOCLIMAPII . Over a sea tile, in turn, some fraction of sea ice may exist. This fraction is constant during the forecast run, but it changes at the moment of analysis (model initialization) according to the sea ice concentration estimated from observations. Thus, sea ice may be considered a sub-tile (or patch). The functionality of using main tiles is provided by SURFEX, but the possibility to use information about the fractional sea ice was introduced into SURFEX while implementing the SICE scheme. SICE utilizes the standard heat diffusion equation solver from the SURFEX suite. SICE, which will in the future contain a more advanced description of the sea ice, provides technical compatibility with the developing versions of SURFEX.

SURFEX provides diagnostic screen-level temperature, specific humidity, and 10 m wind speed from the predicted surface state and the atmospheric values (provided that the forcing is given at some upper level or at the lowest level of the host model) using the interpolation-like procedure of .

SURFEX is incorporated into the ALADIN–HIRLAM NWP system to parameterize the underlying surface processes. The ALADIN–HIRLAM NWP system includes the atmospheric model configuration HARMONIE–AROME , which is a version of the non-hydrostatic limited-area atmospheric model AROME . A variety of sub-grid-scale physical processes are taken into account by the model parameterization schemes. In the ALADIN–HIRLAM NWP system, boundary conditions and some initial conditions are taken from larger-scale models, such as IFS or HIRLAM. The ALADIN–HIRLAM NWP system contains a data assimilation system that uses the three-dimensional variational analysis (3D-Var) method for upper air. Data assimilation of the surface variables uses the optimal interpolation method for snow depth and screen-level temperature and relative humidity. In the configuration of the system used in this study, variables in the soil are initialized according to the optimal interpolation method described in .

HARMONIE–AROME performs short-term cycles to produce forecasts. Each cycle contains the data assimilation procedure and the model forecast. The background fields for the data assimilation are fields of prognostic variables at the end of the previous model forecast. In the configuration of HARMONIE–AROME used in this study, the length of the cycle was 3 h. Starting from 00:00 and 12:00 UTC analysis times, longer forecasts (up to 48 h) are performed. In each cycle, the seawater surface temperature and the fraction of sea ice are kept constant during the forecast being interpolated bilinearly from the host model IFS-HRES, with extrapolation by the nearest neighbour method in specific areas such as fjords. The same is done with the initial value of the sea ice surface temperature. In turn, IFS-HRES uses OSTIA data for the sea surface temperature and ice fraction. For the ice surface temperature, IFS-HRES runs its own simple ice model.

Prior to implementation of the SICE scheme into the ALADIN–HIRLAM NWP system (through SURFEX), sea ice was accounted for in a very crude way in the HARMONIE–AROME configuration. The sea ice surface temperature was initialized by values modelled by IFS-HRES. The ice surface temperature remained constant (equal to its initial value) through the whole forecasting period (similar to the sea surface temperature). This introduced large errors, mainly due to the absence of a diurnal cycle over the ice surface. When the SICE scheme is used, its prognostic variables are updated during each cycle as described in the following. If in the grid cell in question the ice cover exists in the background field, the prognostic variables of SICE are kept unchanged (SICE runs freely). Otherwise, in the situation in which new ice is observed according to OSTIA, the initial (analysed) values of the prognostic SICE variables are obtained via extrapolation from the nearest grid cells of the background field in which the ice exists.

The main objective of the SICE scheme is to reproduce the evolution of ice surface temperature because this variable provides an interface between the atmosphere and the underlying surface. Observations of sea ice surface temperature in the area of interest, which may be used to evaluate the performance of the model and/or parameterization scheme, are limited. Preliminarily, the modelling results from SICE were compared with the results of the well-tested sea ice model HIGHTSI as an overall technical sanity check of SICE and to better understand its limitations and weaknesses. For the next step, coupled experiments with SICE were performed with the HARMONIE–AROME model configuration of the ALADIN–HIRLAM NWP system for the spring period. Since the end of October 2015, SICE has been run operationally within the ALADIN–HIRLAM NWP system version 38h1.2 by the Norwegian Meteorological Institute for the AROME Arctic domain (in June 2017 the operational system was updated to the ALADIN–HIRLAM NWP system version 40h1.1). Here we present the results of comparisons of SICE with HIGHTSI, the results of coupled experiments for the spring period, and 1-year period validations of operational runs. Results of spring experiments are verified against the screen-level temperature and 10 m wind speed observations from SYNOP stations and compared with the sea ice surface temperature products from MODIS and VIIRS. The operational runs are compared with MODIS and VIIRS observations only.

A simple thermodynamic sea ice scheme, SICE, was developed to represent sea ice processes in NWP. In this scheme, the temperature profile in the ice is predicted by solving the heat diffusion equation in the slab of ice with a prescribed thickness. The scheme design allows for explicit coupling with a snow scheme via the fluxes and temperature at the snow–ice interface. Also, the scheme includes the form drag for momentum flux in the surface layer due to ice obstacles in the case of fractional ice cover.

The scheme was preliminarily tested by comparing it with the sea ice model HIGHTSI in off-line mode. In the off-line experiments, when the snow module was switched off in both schemes, the difference in the simulated ice surface temperature between SICE and HIGHTSI was small (the difference standard deviation is equal to 1.04 ∘C) when the ice thickness modelled by HIGHTSI was approximately equal to that prescribed in SICE. Due to high variability of the snow surface temperature, the difference in surface temperature between the two schemes was larger when the snow module was included, with a difference standard deviation of 1.99 ∘C. From this comparison the sensitivity of the results to the prescribed value of the ice thickness was noted.

SICE was evaluated in a coupled framework within the HARMONIE–AROME configuration of the ALADIN–HIRLAM NWP system to assess the scheme's performance and to study possible errors. Since the end of October 2015, SICE has been running operationally by the Norwegian Meteorological Institute within the ALADIN–HIRLAM NWP system version 38h1.2 for the AROME Arctic domain. For the spring period, SICE was carefully evaluated from the coupled experiments; for the rest of the year, the operational archives of the Norwegian Meteorological Institute were used. The ice fraction field for both the SICE experiments and in operational runs was provided by the OSTIA product via the lower boundary conditions from the ECMWF model, IFS-HRES. In the reference experiment, the sea ice surface temperature was taken from the IFS-HRES model and remained constant during the forecasting cycle. Coupled experiments were performed for different configurations of the SICE scheme: the snow-free experiment, the experiment with snow on top of the ice, and the experiment with the form drag term included. A separate experiment in which the ice thickness in SICE is initialized from the model climatology was also performed. Data from coastal SYNOP stations in the Svalbard and Gulf of Bothnia regions were used for validation in spring experiments. Comparisons against ice surface temperature products from MODIS and VIIRS allowed for a study of the performance of SICE on the large scale over an entire year.

Verification of coupled experiments against measurements from coastal SYNOP stations showed that the impact of SICE on the 2 m temperature scores was positive without the snow model, but with the snow model no clear positive impact was seen. For the mean sea level pressure verification scores, a minor positive impact was seen for all SICE experiments. In the SICE experiments without the form drag compared to the reference experiment (which contains no ice fraction representation), a positive 10 m wind speed bias was noted. This bias was reduced after accounting for the form drag in SICE. However, our conclusion about the impact of the form drag is still preliminary. Also, the form drag term strongly depends on the ice fraction value; thus ice concentration observations of better quality than low-resolution passive microwave data used in OSTIA are desirable. For example, in it is shown that using high-resolution passive microwave data from the Advanced Microwave Scanning Radiometer 2 (AMSR-2) leads to a substantial decrease in model errors.

Comparisons of the model experiments with the satellite ice surface temperature products over the Arctic domain showed that the ice surface temperature forecast has smaller errors in the reference experiment than in the experiments with SICE (with no snow scheme included) for the first 12 h of forecast. This happens because the prescribed ice thickness in the sea ice parameterization used by IFS-HRES is tuned to reproduce large-scale fields rather than local effects, and it is larger than in SICE. However, general patterns of the ice surface temperature field are well captured by SICE; after 24 h of the forecast, the predicted ice surface temperature in the experiments with SICE shows a smaller root mean square error than in the reference experiment. Considering both the forecast bias and standard deviation of errors, the best results were obtained from the SICE scheme experiments with the ice thickness prescribed from the model climatology provided by the TOPAZ4 reanalysis .

Assessment of the SICE performance throughout the year using data from the operational archive showed that forecast errors with SICE are smaller than without SICE during autumn, winter, and early spring. In late spring and summer, errors with and without SICE are similar. This happens because the ice surface temperature during this season is close to melting point.

The numerical experiments allow us to conclude that SICE can improve forecasts of the HARMONIE–AROME configuration of the ALADIN–HIRLAM NWP system in ice-surrounded areas, especially for forecasts longer than 24 h. At the moment we recommend its use without snow parameterization for a trouble-proof result. The prescribed ice thickness is an important parameter, and since no estimates of the ice thickness from observations or other sources are used in the current version of the ALADIN–HIRLAM NWP system, it should be tuned.

Of course, in the future the scheme itself should reproduce the spatial and temporal inhomogeneity of the ice thickness. Further development will be focused on the physical processes that control the evolution of sea ice, such as ice freezing and melting. Additionally, possibilities to improve the parameterization of snow on sea ice will be studied. More tests on the parameterization of the form drag are planned. The performance of the scheme will be more carefully evaluated with more experiments, for more regions, more seasons, and using more data. For example, summer periods in the Arctic region, when ice is melting, need more validations using SYNOP observations, more remote sensing observations, and observational campaign results if possible. Melt ponds affect the atmosphere mainly through changing radiation fluxes, but they may also influence the modelling results of a whole NWP system, since they lead to higher uncertainty in the ice concentration observations coming from passive microwave remote sensing. The initialization of the ice parameterization scheme and model error corrections (especially for the snow module) using observations are also of high importance. The possibilities to use more observations and to develop methods to assimilate them, as well as to improve the methods of using existing observations, should be carefully studied.