The Multi-sensor Improved Sea Surface Temperature: continuing the GHRSST partnership and improving Arctic data project started in 2018. The objectives of this new project are on improving (i) satellite SST data distribution and international partnership activities, (ii) access to Arctic in situ SST data, (iii) the accuracy of satellite-derived SSTs in the Arctic, and (iv) our understanding of air-sea-ice interactions in the Arctic.

Much of the Arctic Seas were until recently ice-covered for all or most of the year, so studies of Arctic Sea Surface Temperature (SST) were not feasible. This has changed dramatically in recent years, owing to extreme seasonal sea ice melt-back and other climate impacts. In fact, this is now one of the most exciting areas of the world to study SST, in order to understand a variety of phenomena including heat exchange in the coupled air-sea system. However, satellite SST products in this region are presently very poorly validated, and are generally tuned to lower latitude in situ observations. The time is right to address these problems, given new observations that have been collected in recent years, in concert with the advantages that come from multiple passes of polar orbiting satellites at high latitude.
Improving our understanding of dynamical processes in the Arctic requires improving both the accuracy and characterization of observations used. To address this gap, this project proposes to collect, quality control, and re-distribute a large number of existing ‘research’ Arctic in situ datasets that have not been included in broadly used SST databases. These existing and new Arctic in situ data will be integrated into existing satellite-in situ matchup databases. As these matchup databases are the basis for most satellite SST algorithm development and uncertainty assessments, the addition of these new in situ data will greatly improve SST accuracy and uncertainty estimates in the Arctic. All satellite data, in situ data, and matchup databases developed in this project will be distributed publicly, to benefit this project, our international partners, and the wider scientific community.

Most researchers use global, gap-free, daily SST analyses, or Level-4 (L4) SSTs. L4 SSTs not only have large uncertainties in the Arctic but can vary by several degrees depending on which product is selected. The errors in L4 SSTs are partially due to the L4 analysis procedures (different handling of SSTs in the Marginal Ice Zone (MIZ), different bias corrections, different filtering of data, etc.) and partially due to errors in the orbital Level-2 (L2) satellite SSTs that are used to create the L4 SSTs. Improvements in L2 data accuracy, characterization and quality control, combined with better understanding of air-sea interactions and other processes, will feed through into better high-latitude L4 SST analyses that underpin so much Arctic research.

Area D - Board 400

The Sentinel-3A satellite has been launched in 2016, and the Sentinel-3B in 2018. The later is expected to reach its final orbit end of November 2018. The topography mission, using the SRAL altimeter onboard, has a primary aim to monitor ocean mesoscale features, but sea ice thickness monitoring is also one of the objectives. In addition to SRAL, Sentinel-3 also embarks two optical payloads: OLCI and SLSTR. The OLCI full resolution product has the same 300 m resolution as the SRAL product, which means that co-dated and co-localized data are available all along the track. Obviously, the OLCI images can only be used in sunlight condition which excludes the dark winter period in the Arctic. However, during Autumn and Spring, OLCI images enable to better understanding of what is measured with altimetry. This is particularly of interest for drifting sea ice, where a time difference between acquisitions will result in need to compensate ice movement when combining image and radar altimeter data. Furthermore, it is known that the current SRAL L1 processing is not fully optimized for sea ice purposes and OLCI data may help in this optimisation.

The Finnish ice service uses SAR images to produce daily ice chart to support shipping during winter in the Baltic Sea. However, near real time sea ice thickness based on altimetry and OLCI images are also relevant as a complement, if no SAR data is available for the current day. This paper aims at showing how SRAL and OLCI near real time data can be used by an operational ice service, also discussing blocking factors and how to mitigate them. An operational service requires to automatically download each day Sentinel-3 level 2 Marine SRAL and OLCI products from EUMETSAT CODA. For OLCI, the processing chain has to build RGB georeferenced images, restricted to the area of interest. The outputs can then be easily visualised with GIS tools by the ice analysts. The SRAL products are processed with the PySiral tool, developed by AWI, to compute the sea ice thickness. Finally, each time Sentinel-3 overflights the Baltic Sea, the ice analysts will be able to superimpose along-track sea ice thickness and the corresponding OLCI RGB image.

Area D - Board 401

Sea surface temperature (SST) information is requested by a wide range of users and is fundamental for climate monitoring, understanding of air-sea interactions, and for numerical weather predictions. Most users require the SST information as global gap-free analysis product (L4 data). Several L4 SST analysis products have been developed for various scientific and operational applications e.g. within ESA European Space Agency Climate Change Initiative (ESA-CCI). In most cases, the products are developed individually with specific user requirements, which guide the design of the given product. Different sea ice concentration (SIC) products have also been developed, usually based on different underlying satellite data e.g. EUMETSAT OSI-SAF and ESA-CCI Sea Ice (SICCI) products. Consequently, differences between products exist for both SST and SIC.
In this study, we analyze the differences and assess the consistency between different SIC and SST products focusing on the Arctic and the marginal ice zone. In the Polar Regions, sea ice information is used during the creation of the SST fields, which imposes some degree of consistency between SST and the SIC in the SST analysis. In this study, we have used climate data records from the ESA-CCI SST and Sea Ice L4 products together with the OSI-SAF SIC reanalysis and the Canadian Meteorological Center (CMC) L4 SST. The differences between products have been assessed and examined to improve the understanding of SST and SIC satellite products and the associated uncertainties. The largest SST and SIC discrepancies are as expected found in the marginal ice zone, where the spatial resolution of the SIC satellite data and the SST processing give rise to different representations. Unexpected inconsistencies between products can assist in identifying potential problems within individual products, highlighting areas for future improvements. This includes improvements of the products themselves, but also improvements of the relationship between SST and sea ice in L4 SST analysis products, which will further help facilitating the integration into atmospheric and ocean-sea ice models.

Area D - Board 402

Sea ice is an important component of the Arctic environment, reducing the heat transfer between ocean and atmosphere with its insulating snow cover and the absorbed solar radiation with its high albedo compared to the dark ocean. The Arctic sea ice retreat and thinning are significant climate change indicators and have severe consequences (e.g. loss of hunting and transportation routes across sea ice) for the indigenous people living and working in the Arctic. In addition, as sea ice retreats, a significant increase in ship traffic across the Arctic sea routes will require a reliable and high-resolution sea ice product in the near future.

The European Space Agency (ESA) is studying a range of potential future satellite missions to extend the Copernicus Sentinel fleet, that will greatly improve our understanding and monitoring of the polar oceans and sea ice. These missions respond to High Priority issues that are currently not addressed completely using the existing Copernicus fleet of satellites. The Copernicus Imaging Microwave Radiometer (CIMR) candidate mission is proposed to measure passive microwave emissions with an unprecedented spatial resolution compared to previous missions. In the polar regions, these measurements can be converted to sea ice concentration, multi-year ice fraction, thin sea ice thickness and snow depth on sea ice. CIMR responds to the EU Arctic Policy needs and user requirements set out by the European Commission including the high priority issue of losing all international capability at L- and C-band (e.g. AMSR2 and SMOS missions now exceeding their design lifetime). Additionally the Polar Ice and Snow Topographic (P-ICE) candidate mission is under study. This is designed to measure sea ice thickness and snow depth on sea ice using a multi-frequency SAR interferometric altimeter in a high inclination orbit (similar to CryoSat-2). Sentinel-1A and Sentinel-1B, two imaging radar satellites already in orbit, provide ice-charts and maps of sea ice drift that can be complemented by an L-band SAR mission (ROSE-L) that is also being studied as a High Priority Candidate Mission (HPCM).

In this study, we show how the different Sentinel HPCM will complement each other and highlight potential synergies in the retrieval of sea ice parameters with a focus on the Arctic Ocean. For instance, Sentinel-1 and CIMR are both capable of measuring the sea ice extent and to distinguish ice types but with significantly different coverage, spatial and temporal resolution. The P-ICE mission will provide estimates of sea ice thickness for ice > 0.5 m and is highly complementary to the proposed CIMR L- and C-band thin sea ice thickness products that are useful up to a thickness of ~0.5 m. ROSE-L has the potential to provide very high-resolution measurements of sea ice in the marginal ice zone that are complementary to daily revisit CIMR data in tactical navigation applications. We investigate ways to combine these datasets to exploit the advantages of the different missions and to provide a higher temporal and spatial resolution and enhanced-quality product.

Snow depth on sea ice is recognized as a crucial parameter to derive sea ice thickness. Hence, a more accurate and higher-resolution snow depth product (from CIMR) will in turn improve the sea ice thickness calculation (from P-ICE) significantly. Neural networks, as a means of Artificial Intelligence, are a powerful tool to handle complex data and sophisticated problems by self-learning non-linear relationships, which often occur in geophysical processes including the thermal emission and scattering of a microwave signal in a snow layer. In this study, we develop a neural network that estimates snow depth on sea ice, taking passive microwave data as input. To train the network, snow depth measurements from the Operation IceBridge mission are used. We show that the neural network approach improves the accuracy of snow depth compared to the classic technique using empirical regression coefficients.

Area D - Board 403

The dynamic ocean topography (DOT) is defined as the difference between the sea surface height and the geoid and enables studies about ocean surface currents and its variability. Satellite altimetry is an important part of the Global Geodetic Observing System (GGOS) providing precise information on sea level on different spatial and temporal scales. However, in polar regions, altimetry observations are affected by seasonally changing sea-ice cover and difficult ocean conditions leading to a fragmentary data sampling.
In order to overcome this problem an ocean model is used to fill in data gaps with the aim to obtain a homogeneous DOT representation enabling consistent investigations of ocean surface current changes. For this purpose, the global Finite Element Sea-ice Ocean Model (FESOM) is used. It is based on an unstructured grid and provides daily water elevations with high spatial resolution. Since it also contains a sea-ice component taking into account the main ice drift patterns, it is particularly suitable for polar regions such as the Greenland Sea.
The combination is done based on a Principal Component Analysis (PCA) after reducing both quantities by their constant and seasonal signals. In the main step, the most dominant spatial patterns of the modeled water heights as provided by the PCA are linked with the temporal variability of the estimated DOT from altimetry. At the end, the seasonal signal as well as the absolute reference from altimetry is added back to the data set.
This study describes the combination process and shows results for the area of Greenland Sea and Fram Strait based on several years of multi-mission altimetry data (e.g. ERS-2, ENVISAT) . In addition to DOT, surface current velocities are shown and validated.

Area D - Board 404

The drift of sea ice upon the polar oceans is a result of the balance between quadratic surface stresses at the air and ocean interfaces, the Coriolis acceleration, sea surface tilt and internal ice stresses. For ice concentrations less than 95% internal ice stresses are small and sea ice is said to be in 'free drift'. In this work we consider data from individual drifting buoys as well as Arctic wide gridded fields of wind, sea ice and ocean velocity to invert the momentum equation and estimate the associated quadratic drag parameters and therefore constrain the momentum budget in the ice covered seas.

In ice-covered regions the interface between atmosphere and ocean differs from its open ocean counterpart in many ways: the surface topography in contrast to the waves in the open ocean appears as frozen on time and length scales spanned normally by the wave spectra; the ice cover acts as an additional insulating layer both thermodynamically due to the low conductivity of ice and snow and mechanically due to the rigidity of ice floes; the sea ice seasonal evolution drives a buoyancy forcing at the ocean surface both via its energy (wind forcing, heat fluxes) and salinity modulation (brine release, freezing); the sea ice modifies turbulent fluxes of momentum, heat, freshwater, humidity, gas, and other tracers via due to its complex surface topography and two-phase nature with alternating open ocean (leads in the pack ice or of open ocean in the Marginal ice zone (MIZ)) and ice features (floes).

The state of the sea ice is controlled by an interplay of dynamics and thermodynamics on all spatiotemporal scales represented by a myriad of processes such as ice growth and melt, mechanical strength of the ice, ridging, sea ice wave interactions, fast ice or leads formation. An observational gap remains at short time and length scales to resolve those faster processes. Over synoptic and slower time scales, the wind and ocean forcings, together with the internal forces in the sea ice control the sea ice motion and ultimately the total and regional ice volume contained in the polar oceans via redistribution of ice and export mostly out of the Fram Strait in the Arctic or via Ekman transport to the warmer Southern latitudes in the Antarctic.

With the advent of polar oceanography from altimeters in ice-covered regions, important new questions can now be addressed regarding the freshwater fluxes, surface currents and Eddy kinetic energy, as well as the spinning up or down of polar gyres. To improve further upon the resolution probed by conventional altimetry requires the joint measurements of surface winds, ice drift and ocean currents at sub-synoptic and Eddy resolving length scales that SKIM can achieve.

Area D - Board 405

Sea-ice concentration (SIC) is a key element in the vast majority of approaches to understand the changing sea-ice cover of our planet. This importance of sea-ice concentration derives both from the availability of a long, continuous record of the underlying passive-microwave satellite data, and from the central importance of sea-ice concentration for many physical processes connected to the sea-ice cover.

Although they all rely on the same underlying passive-microwave satellite data, available SIC Climate Data Records (CDRs) can differ widely on a day to day basis, because the underlying algorithms have different sensitivities to noise sources (including atmospheric contamination of the signal, varying emissivity of snow and sea ice, land spill-over contamination, etc…). These differences can impair our ability to understand the polar regions today, and fully characterize the fate of sea ice in a changing climate.

With the aim to understand and -as much as possible- reduce these uncertainties, the European Space Agency (ESA) conducts a dedicated research and development activity on the sea ice ECV (concentration, thickness, drift,...) within its Climate Change Initiative programme. For sea-ice concentration, ESA CCI complements the sustained efforts of the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF) to build long-term, stable, error-characterized Climate Data Records that approach the requirements identified by the Global Climate Observing System (GCOS), a body of the World Meteorological Organisation (WMO).

A first result from the ESA CCI Sea Ice project was a thorough algorithm intercomparison exercise, during which more than 30 published algorithms were evaluated against newly gathered ground-truth data. New SIC algorithms were then developed, that were implemented in three SIC CDRs released in early 2017. One was produced and released by the EUMETSAT OSI SAF project, and covers the 1979-2015 period with coarse-resolution sensors including the Scanning Multichannel Microwave Radiometer (SMMR), the Special Sensor Microwave/Imager (SSM/I), and the Special Sensor Microwave Imager / Sounder (SSMIS) instruments. It is now continuously extended using near-real-time observations from SSMIS. At the same time, two CDRs were also produced by ESA CCI. They provide higher resolution and accuracy by using the Advanced Microwave Scanning Radiometer - Earth Observing System (AMSR-E) (2002-2010) and AMSR2 (2012 onwards) instruments. Several evaluation efforts, against ship-based observations, high-resolution satellite imagery, and independent melt-pond observations are on-going.

In this contribution, we summarize the algorithm intercomparison results, the characteristics of the three CDRs, and their evaluation. We also present plans for future improvements of the algorithms, that include better uncertainty estimates, and use of the high-frequency (near 90 GHz) channels on board the SSM/I and SSMIS sensors from 1992 onwards. The high-frequency channels have the potential to provide higher-resolution SIC fields for a nearly 30 years period, but their interpretation is challenged by higher noise contribution from the atmosphere.

These future developments will lead to new releases of SIC CDRs by the EUMETSAT OSI SAF and ESA CCI+ at the horizon 2021.

Area D - Board 406

Polar sea ice is constantly on the move. Pushed by winds and ocean currents, it can move several tens of kilometers a day. Through deformation events (opening of leads and polynyas, formation of ridges, ...) and advection of existing deformations, sea-ice drift (aka motion) constantly shapes and relocates the sea-ice thickness distribution, opening, closing, or changing shipping routes, allowing or denying access to harbor. Sea-ice drift is also responsible for the export of sea ice towards lower latitudes where it melts and participates in the fresh water export from the polar regions.

Due to the remoteness of the vast sea-ice covered basins, and the short temporal and spatial scales at which sea-ice drift varies, an efficient way to monitor sea-ice drift at regional to global scale is from satellite data. The vast majority of algorithms to compute sea-ice motion are variations of the Maximum Cross Correlation (MCC) method. The generality of these motion tracking algorithms is such that they have been adopted successfully for different types of satellite imagery, among others brightness temperatures from Passive Microwave Radiometer (PMR) sensors, backscatter measurements from scatterometers and Synthetic Apperture Radars (SAR), and (cloud-free) visible and infrared images.

In this contribution, we focus on two families of sea-ice drift products, and propose an innovative approach to combine them, that increases accuracy, coverage, and usability of sea-ice motion information for the end users.

Since 2009, the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF, https://osisaf.met.no) produces daily sea-ice drift maps of the ice-covered oceans (Northern Hemisphere since 2009, and Southern Hemisphere since 2013). As of today, brightness temperatures from the Advanced Microwave Scanning Radiometer (AMSR2) and Special Sensor Microwave Imaging/Sounder (SSMIS), as well as backscatter from the Advanced Scatterometer (ASCAT) are processed by the Continuous Maximum Cross-Correlation (CMCC) algorithm to yield 48 hours motion vectors on a 62.5x62.5 km grid. In the official OSI SAF product, sea-ice motion vectors are processed after the satellite signal is aggregated on daily averaged maps, and the motion vectors thus lack accurate start and stop timing information. We present here initial results from computing motion vectors from individual overlapping satellite swaths, and explore the potential for sea-ice motion retrieval from using the Copernicus Microwave Imaging Radiometer (CIMR, https://cimr.eu).

Since 2007, the Technical University of Denmark (DTU) has produced near-real-time sea-ice drift data from overlapping pairs of SAR scenes from ENVISAT ASAR, Radarsat-2 and Sentinel-1 (both S1A and S1B satellites). The SAR scenes must be within 36 hours of each other. The SAR scenes are averaged to 300 m resolution for noise reduction, and the sea-ice drift vectors are found by cross correlating a circular sub-image of 10 km diameter in the first scene of the pair within an approximately 100x100 km area within the second scene. The location of the maximum cross correlation peak determines the motion vector, which is recorded in the output files. The product, run operationally as part of the European Union’s Copernicus Marine Environment Monitoring Service (CMEMS, https://marine.copernicus.eu), typically contains up to 200,000 sea-ice drift vectors per day, varying with season (ice-covered area) and with available number of overlapping SAR scene pairs. Conversely to the current OSI SAF approach, motion vectors are computed from individual overlapping scenes and are thus associated to accurate start and stop timing information.

On a daily basis, the collection of motion vectors resulting from the two products (the prototype “swath-to-swath” product from the OSI SAF, and that from CMEMS) are difficult to use for the non-expert user, because the location, timing, and duration of the vectors change across regions, and within one region on a day-to-day basis.

Prototyped during the ESA Climate Change Initiative (CCI) projects, the Ice Drift Analysis (IDA) system applies a variational approach to infer the spatio-temporal sea-ice drift field that provides the best possible match to a set of input sea-ice drift observations, while minimising the deviation from additional constraints. This best possible fit is achieved through minimisation of a cost-function by an efficient iterative gradient-based algorithm. IDA further provides per-pixel uncertainty ranges that are consistent with uncertainty ranges in the input observations and the additional constraints. This uncertainty propagation step employs the Jacobian that quantifies the sensitivity of the sea-ice drift observations with respect to the unknown sea-ice drift field. The sea-ice drift fields resulting from applying IDA have the definite advantage that they are on a structured space-time grid. Once it is developed further, IDA thus has the potential to be an operational framework for easy-to-use, error-characterized, global, daily analyses of sea-ice motion vectors, enabling synergetic use of SAR and PMR satellites.

Area D - Board 407

Polynyas are areas of persistent open water where we would expect to find sea ice. They occur in ice-covered ocean regions in the Arctic and Antarctic regions. In this study, we focus on the coastal polynya in Terra Nova Bay (TNB), Ross Sea, Antarctica. The polynya, oriented in east-west direction and bounded by the Drygalski Ice Tongue in the south and by the Campbell Ice Tongue in the north, is generated and maintained by persistent katabatic winds whose speed can reach up to 40 m/s pushing the bay ice offshore.
The area is mostly inaccessible during the 6-months long winter-time; hence, remote sensing provides an essential tool for gathering information. Actually, the key remote sensor is passive microwave radiometer (PMR) that provides observations independent of cloud coverage and with a very dense revisit time. However, one of the limiting factors that characterize PMRs is the coarse spatial resolution (tens of kilometers) that, considering the small width of several polynyas (here, “small” means 1–10 km), may significantly affect the extraction of finer-scale information. In addition, the poor spatial resolution leads also to contamination effects on the polynya edges (e.g.; signal mixtures of land and open water or thin and pack ice).
Within this context, synthetic aperture radar (SAR) sensors provide complementary information since they allow mapping the microwave reflectivity of the observed scene on a finer resolution scale (up to meters with imaged swaths of tens to hundreds of kilometers). Since they operate at microwave frequencies, their advantage is that they deliver data independently of cloud coverage and daylight – in contrast to optical (“visible range”) sensors which otherwise can operate at comparable spatial resolutions. Few studies analyzed the role of SAR imagery in providing local information on polynya, e.g.; boundary between ice-free and ice-covered area. Typically, single-polarization HH imagery is used that, resulting in a remarkable sea/ice contrast, performs best for this purpose.
In this study, dual-polarimetric coherent (HH-HV) SAR measurements are exploited to extract local information related to TNB polynya. A time series of 8 RadarSAT-2 HH-HV SAR scene collected in 2014-2015 both in summer- and winter-time is exploited to observe TNB polynya. Both polarimetric features, e.g.; polarimetric entropy and mean scattering angle, and intensity-related features, e.g.; polarization ratio and SPAN, are used to identify polynya extent and to better highlight ice-related features within the polynya area.
Experimental results clearly demonstrate that dual-polarimetric information allows: i) distinguishing ice-free polynya area from the surrounding iced sea; ii) emphasizing ice strips within the ice-free polynya area; iii) distinguishing land-ice from sea ice; iv) identifying icebergs and leads.

Area D - Board 408

Svalbard is probably the most data-rich region in the Arctic in terms of in situ datasets; however, these datasets are spatially unevenly distributed and there are still massive data gaps. To address these gaps, our presentation mainly focuses on Earth Observation activities of the Svalbard Integrated Arctic Earth Observing System (SIOS), an international collaboration of 25 institutes from 10 nations. One of SIOS’ tasks is to integrate the most pressing remote sensing products, along with associated validation data using in situ measurements from Svalbard, via our common data access point. This presentation will highlight efforts from SIOS members in the operationalisation of remote sensing services. Most of the geospatial products developed by SIOS members are derived from the Copernicus Sentinels. We will cover two selected applications of Sentinels: understanding permafrost movement and estimation of glacier velocity over Svalbard using interferometric synthetic aperture radar (InSAR). Both of these applications are of wider interest for the remote sensing community in Svalbard to address broad Earth system science questions. (1) The observed warming of Svalbard causes thawing of permafrost, resulting in a deeper active layer. Extensive permafrost thawing will likely lead to increased risks of landslides and a potential hazard to the infrastructure in Svalbard. (2) Svalbard-wide glacier velocity products are essential for understanding the response of glaciers to the warming climate. SIOS remote sensing services emphasise the mutual benefits to field scientists and satellite owners. Permafrost movement and glacier velocity derivation include extensive remote sensing applications supported by ground validation by field scientists. Therefore, SIOS is planning to make these remote sensing products available for the wider research community by collaborating with members from different nations. International cooperation such as SIOS can provide a broader picture of Svalbard by optimising the integration of satellite data with in situ data provided by research infrastructures distributed in Svalbard. The limited funding available to individual institutions to conduct dedicated cal/val studies in Svalbard highlights the need for an international collaboration like SIOS, which optimises the use of existing infrastructure and data. Finally, our presentation will provide a glimpse of our upcoming plans and challenges to provide an operational remote sensing service.

Area D - Board 409

Monitoring and forecasting the Arctic environment requires an extensive network of observations. Many such observations are best acquired by satellites. However, space infrastructure such as satellites and their instruments are expensive, and it is thus important to be able to base instrument design and selection on potential quantified impact on improving relevant analysis/observations and forecasts.

Quantative Network Design (QND) is a computationally efficient approach that evaluates a set of observations (network) in terms of its constraint on a target quantity, i.e. a quantity of interest. This evaluation is performed in a modelling system that is capable of simulating counterparts of the observations and of the target quantity from a set of unknowns in the system (control variables). It proceeds in two steps: In the first step, the observational information is used to infer the uncertainty in the posterior control vector, C(x), that is consistent with the observational uncertainty, C(d), and the prior uncertainty of the control vector, C(x0) (Equation (1)). In the second step the uncertainty in the control vector is mapped onto an uncertainty in the target quantity, σ(y) (Equation (2)). Both steps use appropriate sensitivities (linearisations/response functions) of the modelling system. The first step, an inversion step, is formalized by

C(x)-¹ = M'T C(d)-¹ M' + C(x0)-¹ .................... (1)

and the second step by

σ(y)² = N' C(x) N'T + σ(y_mod)² ...................... (2)

where T denotes the transpose operation [sorry for the bad readability: sub- and superscripts (except for numbers) are not supported by the editor for this abstract], M′ (N′) denotes the sensitivity of the vector of observations (the target quantity) with respect to the control vector, as simulated by the modelling system, and σ(y_mod) is the residual uncertainty in the simulation of the target quantity that remains even for a perfect control vector. The approach represents a network through observational locations and times and the observational uncertainty, but does not require real observations. Consequently, it can evaluate hypothetical networks/space missions based on assumed instrumental specifications and space-time coverage. Observations can range from point-scale (in situ) to gridded data sets or level 1 satellite data, if appropriate forward models/observation operators (Kaminski and Mathieu, 2017) are available.

We report on an activity funded through the European Space Agency's Support to Science Element as a contribution to the Year of Polar Prediction, which constructed the Arctic Mission Benefit Analysis (ArcMBA) system. The system exploits the fact that model sensitivities at observational times and locations as well as the target quantities can be pre-computed, so that the actual assessment of a data set requires only matrix multiplications and inversions (Equation (1) and Equation (2)). This means the assessment can be performed so fast that the ArcMBA system could be used as an interactive tool to assist decision makers, for example, in a meeting. Currently the system uses pre-computed model sensitivities M′ for observations of sea ice thickness, sea ice freeboard, radar freeboard, laser freeboard, and snow depth along with precomputed model sensitivities N′ for forecasts of snow and ice volumes for three regions along the Northern Sea Route. These sensitivities were computed with the Max-Planck-Institute Ocean Model (MPIOM) in a global configuration with high resolution over the Arctic (Niederdrenk, 2013). The control vector is composed of physical constants in the model's process representations as well as initial- and boundary conditions.

We applied the system to evaluate real and hypothetical remote sensing products. The real products were monthly sea ice thickness, sea ice freeboard, and radar freeboard, all derived from CryoSat-2 by the Alfred Wegener Institute. These real products were complemented by two hypothetical monthly laser freeboard products (2 cm and 20 cm accuracy, respectively), as well as two hypothetical monthly snow depth products (2 cm and 15 cm accuracy, respectively). Target quantities are four-week forecasts of snow and ice volumes over three target regions along the Northern Sea Route.

As an example, Figure 1 shows the posterior uncertainty in snow and ice volumes over the three target regions, when the CryoSat-2 sea ice thickness product (Ricker et al., 2014) is used alone, in combination with the hypothetical snow depth product with 15 cm accuracy, or in combination with the hypothetical snow depth product with 2 cm accuracy. To sharpen the contrast between the observational scenarios, σ(y_mod) (which acts as an offset) is set to zero. Comparison of the top and middle panels shows the added value of the hypothetical snow product, not only for the snow volume forecast but also for the sea ice volume forecast. Sea ice volume is sensitive not only to initial sea ice thickness but also to initial ice concentration and snow depth, which are both constrained by the snow depth product. Sea ice volume is also sensitive to some of the process parameters that are constrained by snow depth, notably the ice strength (pstar). Comparison of the middle and top panels shows the added value of a higher accuracy in the snow depth product. Increasing the accuracy from 15 cm to 2 cm results in a reduction in uncertainty of the sea ice volume forecast for the East Siberian Sea target region from to 63 km3 to 24 km3.

The ArcMBA can be extended to cover further EO products and further target variables. In the setup used here, the model can simulate a range of sea ice-ocean variables in addition to those considered in the present study (e.g. ice drift, mixed layer depth, freshwater/sea surface salinity, sea surface temperature, circulation). Switching to a more comprehensive model configuration would enable the investigation of yet further variables. For example the model can be operated with its biogeochemistry module HAMOCC activated (Ilyina et al., 2013) or in a mode coupled to an atmospheric general circulation model and thus enable the analysis of biogeochemical products/target quantities or 2 Arctic mid-latitude linkages. The extension of ArcMBA by a terrestrial biosphere component is planned within the H2020 project KEPLER, which will allow joint assessment of ocean and land observations. We will discuss potential observational scenarios to be evaluated within KEPLER.

Ilyina, T., Six, K. D., Segschneider, J., Maier-Reimer, E., Li, H., and NezRiboni, I.: Global ocean biogeochemistry model HAMOCC: Model architecture and performance as component of the MPI-Earth system model in different CMIP5 experimental realizations, Journal of Advances in Modeling Earth Systems, 5, 287-315, https://doi.org/10.1029/2012MS000178, URL http://dx.doi.org/10.1029/2012MS000178, 2013.
Kaminski, T. and Mathieu, P.-P.: Reviews and syntheses: Flying the satellite into your model: on the role of observation operators in constraining models of the Earth system and the carbon cycle, Biogeosciences, 14, 2343-2357, https://doi.org/10.5194/bg-14-2343-2017, URL http://www.biogeosciences.net/14/2343/2017/, 2017.
Niederdrenk, A.: The Arctic hydrologic cycle and its variability in a regional coupled climate model, PhD Thesis, University Hamburg, pp. 1-186, 2013.
Ricker, R., Hendricks, S., Helm, V., Skourup, H., and Davidson, M.: Sensitivity of CryoSat-2 Arctic sea-ice freeboard and thickness on radar-waveform interpretation, The Cryosphere, 8, 1607-1622, https://doi.org/10.5194/tc-8-1607-2014, 2014.

Area D - Board 410