The Los Alamos Sea Ice Model (CICE) is used by several Earth system models where sea ice boundary conditions are not necessary, given their global scope. However, regional and local implementations of sea ice models require boundary conditions describing the time changes of the sea ice and snow being exchanged across the boundaries of the model domain. The physical detail of these boundary conditions regarding, for example, the usage of different sea ice thickness categories or the vertical resolution of thermodynamic properties, must be considered when matching them with the requirements of the sea ice model. Available satellite products do not include all required data. Therefore, the most straightforward way of getting sea ice boundary conditions is from a larger-scale model. The main goal of our study is to describe and evaluate the implementation of time-varying sea ice boundaries in the CICE model using two regional coupled ocean–sea ice models, both covering a large part of the Barents Sea and areas around Svalbard: the Barents-2.5 km, implemented at the Norwegian Meteorological Institute (MET), and the Svalbard 4 km (S4K) model, implemented at the Norwegian Polar Institute (NPI). We use the TOPAZ4 model and a Pan-Arctic 4 km resolution model (A4) to generate the boundary conditions for the sea ice and the ocean. The Barents-2.5 km model is MET's main forecasting model for ocean state and sea ice in the Barents Sea. The S4K model covers a similar domain but it is used mainly for research purposes. Obtained results show significant improvements in the performance of the Barents-2.5 km model after the implementation of the time-varying boundary conditions. The performance of the S4K model in terms of sea ice and snow thickness is comparable to that of the TOPAZ4 system but with more accurate results regarding the oceanic component because of using ocean boundary conditions from the A4 model. The implementation of time-varying boundary conditions described in this study is similar regardless of the CICE versions used in different models. The main challenge remains the handling of data from larger models before its usage as boundary conditions for regional/local sea ice models, since mismatches between available model products from the former and specific requirements of the latter are expected, implying case-specific approaches and different assumptions. Ideally, model setups should be as similar as possible to allow a smoother transition from larger to smaller domains.
Global, Arctic or Antarctic wide applications of the Los Alamos Sea Ice Model (CICE) do not require any specific treatment regarding sea ice boundary conditions because the model domain is larger than the areas where sea ice may occur. However, this is not the case of regional implementations of the CICE or any other sea ice models. For such regional cases, the past and current versions of CICE include a simple way of dealing with open boundaries, restoring them every time step to the initial ice state or to some predefined value, using a relaxation timescale. In the words of Hunke et al. (2015), this implementation is only intended to “provide the hooks” for more sophisticated treatments. Therefore, the main goal of our study is to describe and evaluate the implementation of sea ice time-varying boundaries in the Los Alamos Sea Ice Model using two regional models: the Barents-2.5 km, implemented at the Norwegian Meteorological Institute (MET), and the Svalbard 4 km (S4K) model, implemented at the Norwegian Polar Institute (NPI). We have chosen to use these two models because the former is an operational forecasting system, using data assimilation and used for relatively short-term simulations (a few days); the latter is a research tool used for hindcast and forecast longer-term simulations (a few years), without data assimilation, and this allowed us to evaluate the time-varying boundary scheme for different types of models and simulations.
The use of sea ice models developed for large scales (like CICE) for small-scale forecasts was discussed by Hunke et al. (2020). On the scales of the Barents-2.5 km and S4K models, the use of a continuum hypothesis and the viscous plastic rheology is far from optimal. However, for coupled sea ice–ocean forecasts, good thermal and dynamical forcing and handling of ice–ocean fluxes are also very important for the usefulness and quality of the forecasts. Also, knowledge about the possibility of ice in an area might be more important for applications, such as navigation, than the specific details of the sea ice cover. Therefore, we think adding capability to handle open boundary conditions in the sea ice model can increase the usefulness of small-scale regional coupled model systems for many applications.
Examples of regional implementations of sea ice models may be found in, e.g.,
Smedsrud et al. (2006), Rousset et al. (2015) and Prakash et al. (2022).
Smedsrud et al. (2006) used the Regional Ocean Modeling System (ROMS,
We use The Regional Ocean Modeling System (ROMS,
The coupling between ROMS and CICE was implemented at the Norwegian
Meteorological Institute by using the Model Coupling Toolkit (MCT,
Exchange frequency between the models depends on synchronization time step and must be a common multiple of involved model time steps. In default setups, the models run concurrently on separate sets of compute cores, with a delayed exchange of fields, such that information calculated in one component is used in the other at the next coupling time interval. The coupled variables are declared in both ROMS and CICE and transferred both ways through MCT routines utilizing the underlying MPI library.
Data exchange between ROMS and CICE through MCT (see text).
The Barents-2.5 km model is MET Norway's primary model for forecasting of
sea ice conditions in the northern regions. It consists of a fully coupled
ocean and sea ice model that covers the Barents Sea and areas around
Svalbard (Fig. 1). The modeling system employs the METROMS (
The S4K model has a slightly different domain than the
Barents-2.5 km model (Fig. 1) and lower horizontal (4 km) and vertical (35
sigma layers) resolution in the ocean, while the configuration of ice
thickness categories and vertical discretization is the same in both setups.
The domain covers a slightly different area to allow producing boundary
conditions for fjord models in Eastern Greenland. It is based on METROMS
coupled with an earlier “columnar” version of CICE, with a “column
package” for thermodynamics and biogeochemical processes developed as part
of the Accelerated Climate Model for Energy (ACME) project, close to
CICE6.0.0 alpha
(
Barents-2.5 km and S4K model domains. The insert in the bottom right corner represents Svalbard and the area where the various drifts (lines showing the start and end dates of each drift) of the N-ICE2015 expedition (Granskog et al., 2018) took place and along which sea ice and ocean data detailed in Table 5 were collected.
We describe the main code changes in Table 2. We defined a Boolean variable
(sea_ice_time_bry) that must be
set to
The new boundary variables match CICE variables. They have a prefix corresponding to the name of the corresponding variable in CICE (Table 2) followed by an underscore and the suffix “bry”. We separated the new variables into ice-category-dependent two- and three-dimensional (2-D and 3-D) and ice-category-independent variables (Table 2). The 2-D variables represent either surface sea ice properties or bulk properties of ice or snow. The 3-D variables represent properties that vary vertically in the ice or snow and are resolved as a function of the number of ice and snow layers defined for a simulation. The ice-category-dependent variables have a dimension used to store the values of different ice thickness categories, defined as a function of sea ice thickness. For details on CICE size thickness categories, see Hunke et al. (2015).
We allocate to the boundary variables the same dimensions allocated for the matching CICE variables, even though we need to track their values only along the open boundaries. This occupies more memory than necessary, with boundary variable “working” rectangular arrays being filled with zeros except for the boundary cells, but it simplifies the process of scattering variable values among different tiles in a parallel run, since we may reuse CICE data scattering routines. However, as described below, the boundary NetCDF files have only vector arrays and do not require “extra” space as the working arrays (see below).
The CICE file with more modifications for the time-varying boundary implementation is ice_forcing.F90 (Table 2). New routines were created to construct boundary file names, to read these files and to make the necessary time interpolations. Some specific file reading routines were implemented in ice_read_write.F90 given the format of boundary files (see below). These routines are called from ice_forcing.F90.
Boundary restoring takes place in file ice_restoring.F90, where the boundary values updated in ice_forcing.F90 are used to modify the corresponding CICE variables using a relaxation time defined in ice_in (trestore), along the “halo” cells (Hunke et al., 2015) located at the northern, southern, western and eastern limits of the model domain and their neighbor cells within the domain. These updates occur in the routine ice_HaloRestore that was modified from its original version. Snow and ice enthalpies are calculated from corresponding temperatures. In the tests carried out so far, we “relaxed” only the cells detailed above to follow exactly the way CICE deals with boundary conditions but a more complex treatment involving a larger relaxation zone may be considered.
Summary of main changes in the Los Alamos Sea Ice Model related
to the implementation of time-varying boundaries (
Wind-idealized experiments with the Barents-2.5 km model plotted
inside the TOPAZ4 model. The Barents-2.5 km model was run in its full state
except the wind forcing was idealized in the sense of constant wind in the
model
Minor adjustments were implemented for Barents-2.5 km to enhance reliability
for the operational system, particularly to blend mismatches between the
external and internal solutions. In ice_HaloRestore, the
first physical points as well as the halos are restored/nudged. Dynamical
variables uvel, vvel, divu, shear and strength are restored to the
neighboring interior point. Several technical additions address edge cases.
Additional grid variables are extrapolated to halo cells (ice_grid.F90). Halo cells are no longer zeroed during multiprocessor
communications (ice_boundary.F90). Boundary values are
restored before both thermodynamics and dynamics (in CICE_RunMod.F90), which is necessary for prescribing boundary values (i.e., when
trestore
In the S4K model, the only exception in the boundary restoring process is with uvel and vvel, which are restored as any other boundary variable when there is sea ice outside the domain, else internal velocities are assumed in line with Rousset et al. (2015). This is to guarantee that the sea ice motion inside the model domain is properly affected by larger-scale drift trends in “long-term” simulations (several months).
Our approach differs from that described by Rousset et al. (2015) for the lateral boundary conditions in the Louvain-La-Neuve sea ice model (LIM3.6) in that we restore tracer boundary values irrespective of the velocity direction across the boundaries. Moreover, we do not fill the boundaries with ice thickness categories following a statistical law – categories are filled depending on their availability in the boundary data. In any case, specific changes can be easily made in the code to test different settings.
The main challenge with the boundary data is the matching between available
model output for a larger domain and the data needs of CICE. In the examples
provided here, we used data from TOPAZ4 as explained above. The available
outputs relevant for CICE boundaries include daily values for ice
concentration, ice and snow thickness, and ice east–west and south–north
velocities. There are no data for ice or snow internal or surface
temperatures, or for ice salinity. There are no data of any kind of ice
thickness categories. Therefore, we had to make some assumptions. These will
have to be defined for each application depending on available boundary
data. In our case, we proceeded as follows:
TOPAZ values located along the boundaries of our domains were linearly
interpolated to our grids. Ice-category-dependent variables were stored in boundary files assuming the
same number of categories used in our runs (five). For each grid point, all
values were set to zero, except for the category where available “bulk”
ice thickness belonged. Surface (skin) snow or ice temperatures (in the absence of snow) were set to
air temperatures taken from the atmospheric forcing files, when air
temperature was Inner snow and ice temperatures were obtained by linearly interpolating
between the surface temperature and the freezing water temperature. The same
temperature trend was assumed for snow and ice. Therefore, when snow was
present, its height was taken into account as the thickness of each ice
layer. Inner ice salinities were calculated to match multiyear and first-year ice
(MYI and FYI, respectively) profiles described in the literature (Gerland et
al., 1999). We assumed that when ice thickness was
Examples of boundary files may be found at
The data used to evaluate the Barents-2.5 km model can be found in Table 3.
For this model system, the focus was purely on remote sensing of sea ice
concentration. AMSR2 (
Operational simulations with the Barents-2.5 km model, plotted
inside the TOPAZ4 model. These plots are taken directly from the operational
model at MET and illustrate the effects of time-varying boundary conditions
in the operational model. Sea ice concentration and surface water
temperature fields (in the open water areas) are shown for three different
dates at 00:00 UTC. Panel
The data assimilation applied in the operational Barents-2.5 km model is the
combined optimal interpolation and nudging (COIN; Wang et al., 2013). It was
originally developed for assimilating sea ice concentration in a two-level
sea ice model within ROMS and is now further developed for the
multi-category CICE model in METROMS
(
Datasets used for Barents-2.5 km model evaluation. The listed references include links to the repositories where data and details on sampling and data processing can be found.
Ice concentration values and their categorization used for the ice charts and Barents-2.5 km model validation.
Datasets used for model evaluation are listed in Table 5, with links or
citations to the various data sources. These include ocean, sea ice and snow
data. We used satellite products and in situ data collected during the N-ICE2015
expedition (Granskog et al., 2018 and Fig. 1). Therefore, more detailed
comparisons between observations and model results are given for 2015. We
also compare TOPAZ4 reanalysis (
We used satellite data of sea ice concentrations from regional high-resolution sea ice charts for the Svalbard region (the same mentioned above
for the Barents-2.5 km model) and for sea ice and snow thickness from the
European radar altimeter CryoSat-2, generated at the Alfred Wegener Institute
(AWI) for the winter period (October–April) (Hendricks and Ricker, 2020).
We also used CryoSat2-SMOS weekly Arctic sea ice thickness data (Ricker et
al., 2017,
Sea ice plus snow thickness were collected during the N-ICE2015 expedition with a helicopter-borne electromagnetic induction sounding (HEM) (King et al., 2016) and a ground-based electromagnetic instrument (EM31) (Rösel et al., 2016a) with footprints of approximately 50 and 3–5 m, respectively (Haas et al., 2009). Snow thickness was measured with a Magnaprobe with a footprint of approximately 0.2 m (Rösel, 2016b).
Datasets used for S4K model evaluation. The listed references include links to the repositories where data and details on sampling and data processing can be found. CTD – conductivity–temperature–depth; MSS90L – ocean microstructure profiler; HEM – helicopter-borne electromagnetic induction sounding; EM31 – ground-based electromagnetic instrument.
Simulations carried out with the Barents-2.5 km model are short term, in
accordance with its operational nature. Model evaluation was based on
idealized simulations and on operational simulations and focused on sea ice
concentration, which is the main variable of interest for this model. In the
case of the S4K model,
Model experiments with idealized wind forcing have been conducted with the
Barents-2.5 km model in order to visually showcase the effects of using
time-varying boundary conditions. The model was initialized from TOPAZ4
fields at 1 September 2019 and it ran until 20 September 2019. One run without the
time-varying boundaries (just like the operational model ran before) and one
with the boundaries extracted from TOPAZ4 results for the same period. All
aspects of the model run, except the wind forcing, were realistic. The wind
forcing was idealized to be purely in the model
The model was initialized from TOPAZ4 fields and ran from January 2014 until
July 2015. Results were analyzed only from October 2014 after some spin-up
time. Model output was compared with observations of ocean and sea ice
variables measured in situ during the N-ICE2015 expedition (Granskog et al.,
2018). Here, we focus only on the evaluation of hydrographical properties
with depth and on temperature–salinity diagrams. The satellite data were used
mainly for evaluation of sea ice concentration and sea ice and snow
thickness (Table 5). Comparisons were also made with TOPAZ4 results since it
is an operational system in use by the Copernicus Marine Service (
TOPAZ4
The idealized simulations (results available at
Salinity and temperature median bias
Results from these simulations are available at
Temperature–salinity diagrams for observations collected during
the N-ICE2015 expedition (Granskog et al., 2018).
Figure 4a shows the root mean square error (RMSE) of the predicted sea ice concentration from March 2019 to April 2021 in the operational Barents-2.5 km model calculated against AMRS2 and Svalbard ice chart observations, which tracked the performance of the operational Barents-2.5 km in the early 2 years. The vertical red line indicates the time when applying the time-varying boundaries, and the vertical green line shows the time when applying the data assimilation (see Sect. 2.3.1). Before the time-varying boundaries, the RMSE was generally between 0.2 and 0.4 (before mid-August 2019). Due to the large error in the open boundaries, the initial conditions had to be re-initialized in late August and September, which is seen in the abrupt decrease of the RMSE. However, the RMSE increased rapidly after each re-initialization. After implementing the time-varying boundaries in October 2019, the average RMSE is generally below 0.25, much lower than in the previous period. To further analyze the effect of the time-varying boundaries, we computed Taylor diagrams (Taylor, 2001), using the MATLAB PeterRochford-SkillMetricsToolbox-d7ea0d3. The improvement in model performance was negligible when the daily total sea ice extent was considered (Fig. 4b). However, a large improvement is apparent when spatially resolved data are compared (Fig. 4c), with higher correlation coefficient and lower RMSE for the simulation with time-varying boundaries. Moreover, the model standard deviation becomes very close to that of the data. Altogether, this shows that the model accuracy improved and that ice concentration variability is better captured.
Dinessen and Hackett (2016) CMEMS (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_002)
We present first results for ocean variables and then for sea ice variables. In both cases, we compare S4K with TOPAZ4 results and with observations (see Sect. 2.3.2).
Extreme median salinity and temperature biases are
CryoSat2-SMOS
Sea ice concentration and sea ice plus snow thickness from satellite products, TOPAZ4 and S4K show similar patterns (Figs. 8 and 9). In Figs. 8e–f and 9d, we plot S4K fields within a rectangle defined by a dashed line and “surrounded” by TOPAZ4 fields to evaluate the transition from TOPAZ4 forcing to the S4K fields. Boundary effects resulting from forcing S4K with TOPAZ4 sea ice data are not visible in the sea ice concentration plots (Fig. 8e and f) and they are quite smooth in the sea ice and snow thickness plots (Fig. 9d), with the exception of thinner ice along the northeast boundary in January 2015 (Fig. 9d). In some occasions, S4K predicts thin ice south eastwards of Greenland to a larger extent than observed in satellite data and protruding from the ice flowing along Greenland and out of the Fram Strait (Figs. 8f and 9d). This is neither visible in the satellite data nor in TOPAZ4 results (Figs. 8 and 9).
Sea ice and snow thickness results from S4K model are generally lower than those from satellite products and TOPAZ4 results for the overlapping areas (Fig. 9). However, sea ice and snow thickness frequency histograms based on EM31 data (Table 5) overlap more with S4K than to TOPAZ4 (Fig. 10a and b). A similar comparison based on HEM data shows similar trends (Fig. 10c and d). Despite the much smaller footprint of both the EM31 (3–5 m) and HEM data (50 m) (see Sect. 2.3.2) compared with the model resolution (12.5 km for TOPAZ4 and 4 km for S4K), the observed sea ice and snow thickness ranges are much larger than those predicted by both models. Regarding snow thickness based on Magnaprobe data, both models have a negative bias (Fig. 10e and f) and larger in absolute value than that for sea ice and snow thickness.
Observed (blue) and modeled (brown) frequency distributions of
snow and ice thickness
Here, we show only a limited number of results due to space constraints.
However, monthly averaged map plots of sea ice concentration and sea ice
plus snow thickness, from the satellite products listed in Table 5 and from
TOPAZ4 and S4K for the period August 2014–July 2015) may be found at
The implementation of time-varying boundaries both in the Barents-2.5 km and the S4K models, resulted in a generally smooth transition between the fields of TOPAZ4, providing the boundary conditions, and the fields of the former two models. The performance of the operational Barents-2.5 km improved significantly with the usage of time-varying sea ice boundaries. This upgraded performance was also a large contributor to the Barents-2.5 km operational forecasts being more widely adopted in downstream applications like drift models and vessel icing models and as support for a specific ship salvage operation near Svalbard. There is a large demand for more realistic operational forecasts to support search and rescue, oil spill and other similar scenarios in the Barents Sea. The implementation of a more realistic boundary treatment for sea ice is a central step to achieve a wider usage of the operational fields.
Notwithstanding these results, we still can see some “seams” between the TOPAZ4 fields and those of the other two models. For example, some ice and snow thickness “artifacts” are visible in the S4K model results, especially on the northeastern border of its domain (Fig. 10d). These “artifacts” may arise from drift differences inside the domain and at the boundaries. Such artifacts were already noted in the Barents-2.5 km model (refer to Sect. 3.1.1). Another problem is the different horizontal spatial resolutions of TOPAZ4 (12.5 km) and the models described herein (2.5 and 4 km). Perhaps the more likely explanation is the mismatch between available TOPAZ4 sea ice fields and those required by CICE (refer to Sect. 2.2.2, model boundary data details). Recall from Sect. 2.2.2 that extensive assumptions had to be made in order to fit the limited TOPAZ4 data for all the boundary variables required by CICE. In fact, experiments (not shown) done with a higher-resolution model (500 m horizontal resolution) implemented with CICE, nested in the Barents-2.5 km model, and using exactly the same sea ice data of the larger model, did not show any seam but instead, a near-perfect transition between both domains. This shows the importance of coordinating the storage of adequate outputs from larger models with the “needs” of regional models. The ideal output from a larger model should include the variables listed in Table 2 (corresponding to the variables defined to store boundary values) and use the same sea ice thickness categories of the nested model and the same number of sea ice and snow layers.
In the tests carried out so far, we “relaxed” only the halo zone (more specifically, the grid cells surrounding the domain) and their neighbor cells to follow exactly the way CICE deals with boundary conditions. The default value in CICE for the thickness of this zone is one cell. In fact, this halo zone includes not only the domain boundaries but also the boundaries of all blocks of cells used in a parallel simulation. However, the boundary code affects only the cells surrounding the domain. A more complex treatment involving a broader relaxation zone with more than one cell thickness may be considered, but it is out of the scope of the present study.
The S4K model has a smaller ocean temperature and salinity bias than that of TOPAZ4, in the region north of Svalbard, where the N-ICE2015 expedition took place (Granskog et al., 2018). Observed biases are larger at the depth range where Atlantic Water and modified Atlantic Water are found (Meyer et al., 2017). There is a better fit between TOPAZ4 results and satellite data than those of S4K, which may partly result from the data assimilation process of the former. “Spurious” thin sea ice predicted by S4K southeastwards of Greenland (see Sect. 3.2.2 and Fig. 8f) results from the placement of the front between the inflowing Atlantic Water and the outflowing Polar Surface Water (e.g., Våge et al., 2018). In the S4K model, this front is not close enough to east Greenland on some occasions, allowing very cold surface water to spread towards Svalbard, with production of some thin sea ice.
As a final note, we emphasize here the compatibility of the changes described in this study with the most recent versions of the Los Alamos Sea Ice Model (CICE and ICEPACK), since the files changed and listed in Table 1 are similar to those of the most recent versions.
We implemented time-varying sea ice boundaries in the Los Alamos Sea Ice Model (CICE). This implementation was tested using two regional coupled ocean–sea ice models, both covering a large part of the Barents Sea and areas around Svalbard: the Barents-2.5 km, an operational forecast model, and the S4K, a model used for research purposes. Sea ice boundary conditions were obtained in both cases from TOPAZ4 – a well-tested and documented assimilative coupled ocean and sea ice model covering the Arctic and North Atlantic oceans. Obtained results show significant improvements in the performance of the Barents-2.5 km model after the implementation of the time-varying boundary conditions. The performance of the S4K model in terms of sea ice and snow thickness is comparable to that of the TOPAZ4 system. The implementation of time-varying boundary conditions described in this study is similar regardless of the CICE versions used in different models. The main challenge remains the handling of data from larger models before its usage as boundary conditions for regional/local sea ice models, since mismatches between available model products from the former and specific requirements of the latter are expected, implying case-specific approaches and different assumptions. Ideally, model setups should be as similar as possible to allow a smoother transition from larger to smaller domains.
The software code used in this study for the Barents-2.5 km model may be
found at
The ocean modeling code is a ROMS branch. Code licensing may be found at
The software code used in this study for the SA4 model may be found at
Results from the Barents-2.5 km model may be found at
Graphical sea ice and snow results from the TOPAZ4 and S4K simulations may
be found at
PD made the first version of software changes related to the implementation of time-varying boundaries in the CICE code and ran the simulations with the S4K model. JB, YG and NS implemented, tested, and adapted those changes in the Barents-2.5 km model and ran the simulations shown in the paper with this operational model. DS performed software development and implemented and tuned the S4K model. PB processed and helped analyze S4K model results. JA and AM contributed to the analysis of the S4K model results. AS provided the boundary conditions from TOPAZ4. KW prepared the AMSR2 sea ice concentration and its standard deviation and performed the data assimilation. JBD led and performed the implementation of the CICE–ROMS coupling in METROMS and contributed to discussions of the OBC implementation in Barents-2.5 km model. All authors contributed to the writing of the paper.
The contact author has declared that neither they nor their co-authors have any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We acknowledge the usage of CryoSat-2 satellite products from the AlfredWegener Institute (AWI), publicly available under a Creative Commons Attribution 4.0 International (CC BY 4.0) license, and the usage of CryoSat2-SMOS satellite products. We thank the US Department of Energy's (DOE) Earth System Modeling Program for allowing the use of this version of the CICE model, which includes updated biogeochemistry parameterizations within a “column package” developed as part of the Accelerated Climate Model for Energy (ACME) project. Tuning and model validation work performed under the DOE Regional and Global Climate Modeling Program also contributed to these results.
This work has been supported by the Fram Centre Arctic Ocean flagship project “Mesoscale physical and biogeochemical modeling of the ocean and sea-ice in the Arctic Ocean” (project reference 66200), the Norwegian Metacenter for Computational Science application “NN9300K – Ecosystem modeling of the Arctic Ocean around Svalbard”, the Norwegian “Nansen Legacy” project (grant no. 276730), the European Union's Horizon 2020 research and innovation program under grant agreement no. 869154 via project FACE-IT (The future of Arctic coastal ecosystems – Identifying transitions in fjord systems and adjacent coastal areas), the “Arktis 2030” program, and the project “Vrog havvarsling for arktisk miljberedskap” funded by the Norwegian Ministry of Climate and Environment, as well as the European Union's Horizon 2020 research and innovation program under grant agreement no. 101003826 via project CRiceS (Climate Relevant interactions and feedbacks: the key role of sea ice and Snow in the polar and global climate system). The production of the merged CryoSat2-SMOS sea ice thickness data was funded by the ESA project SMOS & CryoSat-2 Sea Ice Data Product Processing and Dissemination Service, and data from DATE to DATE were obtained from AWI.
This paper was edited by Christopher Horvat and reviewed by two anonymous referees.