Snow 2: Sea Ice

We evaluate Operation IceBridge (OIB) ‘quick-look’ (QL) snow depth on sea ice retrievals using in situ measurements taken over immobile first-year ice (FYI) and multi-year ice (MYI) during March of 2014. Good agreement was found over undeformed FYI (-4.5 cm mean bias) with reduced agreement over deformed FYI (-6.6 cm mean bias). Over MYI, the mean bias was -5.7 cm but 54% of retrievals were discarded by the OIB retrieval process as compared to only 10% over FYI. Footprint scale analysis revealed a root mean square error (RMSE) of 6.2 cm over undeformed FYI with RMSE of 10.5 cm and 17.5 cm in the more complex deformed FYI and MYI environments. Correlation analysis was used to demonstrate contrasting retrieval uncertainty associated with spatial aggregation and ice surface roughness.

Satellite remote sensing is an important and sometimes the only tool for monitoring the sea ice state, particularly in the difficult-to-access polar regions. The core of remote sensing algorithms is the relation between the measured radiative characteristics and the retrieved physical properties of the surface.

The snow grain size, the soot contaminants concentration, the melt pond fraction (during the summer melt) and the surface albedo are among the key characteristics of the underlying surfaces in polar regions for various problems regarding the climate changes, quantifying the ice melting process, particularly for understanding and predicting the expected drastic changes in the Arctic environment.

Several methods and algorithms have been suggested for the retrieval of the sea ice albedo and the fractions of snow-covered ice, white bare ice, melt ponds, or open water. The main approach used in these works is based on the a priorisuggestions about the reflection spectrum of each surface component.

We present another technique to retrieve the characteristics of the sea ice surface and its albedo. The algorithms are based on models of the optical properties of melting sea ice and snow as scattering media. Derived relationships between the physical characteristics of the surface and its spectral reflectance form the base of the algorithms.

In the cold period snow covers a larger part of the surface. The reflectance is determined by the two main parameters: the effective snow grain size and the concentration of soot contamination. The algorithm to retrieve these parameters will be presented. It is named SGSP (Snow Grain Size and Pollution) and currently uses the MODIS data on the spectral snow reflectance in the visible and IR range.

In summer, because of melting, melt water forms ponds in the depressions and leaves the drained parts of the ice surface. The algorithm to investigate the sea ice state during summer is named MPD (Melt Pond Detector). It considers the summer ice as consisting of white ice (high reflective) and melt ponds. The total reflection is determined by the optical characteristics of white ice, melt ponds, and the surface fractions of these components. The MPD input is the MERIS spectral data in the visible and near IR range.

Both algorithms (SGSP and MPD) retrieve the surface characteristics in a iterative procedure that includes the simulation of the ice reflectance, the simulation of the radiation transfer through the atmosphere, the comparison of the simulated reflectance factors with those measured by a satellite sensor at the top of the atmosphere, and the correction of the ice characteristics with the Newton-Raphson method. Both algorithms use the simple optical models of ice and snow, take into account the bidirectional reflection and include the atmospheric correction. Both algorithms have been successfully verified with the aerial, sheep, and in situ measurements carried out in more than 10 campaigns during about two last decades.

The algorithms are implemented as Linux console applications. Their outputs are maps of the snow and sea ice characteristics and the pixel spectral albedo. The produced maps allow monitoring the long-time global changes of the Arctic ice state.

Currently these packages are working in the MODIS and MERIS processing chains in the University of Bremen. These retrievals provide the historical data on the process of Arctic ice melting, which is so important for understanding the climate changes in the Arctic.

It is planned to implement the codes for processing the data of OLCI on Sentinel-3.

Parts of these studies have been supported by the 6^th and 7^th Framework Programmes of the EU (projects DAMOCLES and SIDARUS).

The characteristic feature of the Arctic sea ice in summer is the presence of melt ponds, which affect albedo, mass balance and heat balance of the ice. 96% of total annual heat input through sea ice occurs from May to September when the melt ponds are present (Arndt and Nicolaus, 2014). The increased amount of energy passing through the sea ice affects also eco-systems and geochemical processes in and beneath the sea ice. Due to the large role the melt ponds play in the energy balance of the region, a reliable remote sensing melt pond fraction dataset is required for further assessments of the state of Arctic climate. An algorithm Melt Pond Detector (MPD) to retrieve melt pond fraction (MPF) and sea ice albedo from MEdium Resolution Imaging Spectrometer (MERIS) data presented by Zege et al. (2015) is validated against aerial, ship borne and in situ campaign data. The results show the best correlation for landfast and multiyear ice of high ice concentrations. For broadband albedo R² is equal to 0.85, with the RMS being equal to 0.068, for the melt pond fraction: R² is equal to 0.36 with the RMS being equal to 0.065.

Further, sea ice albedo and MPF are compared to surface air temperature NCEP reanalysis data, to MPF values retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) (Rösel et al., 2012). Moreover, albedo and MPF trends are examined. Weekly averages of MPF for 2007 and 2011 show different MPF dynamics while summer sea ice minimum is similar for both years. The gridded daily MPF and albedo products compare well to independent reanalysis temperature data and show melt onset when the temperature gets above zero. However MPD shows an offset at low MPFs of about 10% due to unscreened high clouds. Weekly averaged trends show pronounced dynamics of both, MPF and albedo: a negative MPF trend in the East Siberian Sea and a positive MPF trend around Queen Elizabeth Islands. To decrease the cloud contamination of the MPD product, we use (1) the MODIS cloud fraction for additional cloud screening of the gridded melt pond product, and a (2) Bayesian cloud detection scheme for MERIS developed within this work. It is designed to derive pixel-accurate cloud masks over ice and snow, which is the most difficult cloud screening scenario, and reproduces masks from AATSR with good accuracy.

The MPD dataset is compared to an independent melt pond product from ICDC based on MODIS data and a static neural network classification scheme (Rösel et al., 2012). We find a remarkably good agreement for daily maps of the central Arctic (RMSD = 0.04 and R = 0.95) after we compensate for a positive 0.08 bias of the relative melt pond fraction in the ICDC dataset.

 We investigate the open water fraction, retrieved by the same neural network from MODIS data, and find a linear dependency on MPF. This could be explained by the spectral unmixing algorithm used for the ICDC dataset being unable to distinguish dark melt ponds or darker (melting) ice from open water. We also find that MPD retrieves higher values in regions of higher open water fraction by comparing the MPD product to high resolution satellite images from the Global Fiducials Library. Therefore, we conclude that both melt pond products are influenced by open water.

Arndt, S. and Nicolaus, M.: Seasonal cycle and long-term trend of solar energy fluxes through Arctic sea ice, Cryosph., 8(6), 2219–2233, doi:10.5194/tc-8-2219-2014, 2014.

Rösel, A., Kaleschke, L. and Birnbaum, G.: Melt ponds on Arctic sea ice determined from MODIS satellite data using an artificial neural network, Cryosph., 6(2), 431–446, doi:10.5194/tc-6-431-2012, 2012.

Zege, E. P., Malinka, A. V., Katsev, I. L., Prikhach, A. S., Heygster, G., Istomina, L. G., Birnbaum, G. and Schwarz, P.: Algorithm to retrieve the melt pond fraction and the spectral albedo of Arctic summer ice from satellite data, Remote Sens. Environ., 163, 153-164, doi:10.1016/j.rse.2015.03.012, 2015.

The SMOS mission provides daily coverage of the polar regions, and SMOS L-band brightness temperature (TB) measurements have been used to retrieve thin sea ice thickness up to about one meter. In addition, there has been an attempt to retrieve snow thickness over thick Arctic multi year ice, which is a crucial parameter for the freeboard-based estimation of (thick) sea ice thickness from lidar and radar altimetry, for example. Furthermore, information on both sea ice and snow thickness is needed to estimate the energy and mass balance of the polar oceans.

For our SMOS-based retrieval of ice or snow parameters, we first model 1.4 GHz TBs for different ice conditions and different values of the retrieval parameter (here: snow thickness) at horizontal and at vertical polarization for the SMOS incidence angles θ=0...60°and compare these modeled TB(θ)-curves with the observations. For modeling the TBs, we use an emission model that is based on the radiative transfer model developed by Burke et al. (1979). We have modified the Burke model to account for higher order reflection terms and apply it to an ice column that is described by one snow layer on top of eight ice layers. In the model, the TB of (snow-covered) sea ice is then a function of the ice and snow layer thicknesses and the permittivities of these media. The permittivities of ice and snow are described by empirical functions, which relate the permittivities to ice temperature, salinity, and type (first year or multi year ice) or to snow density, respectively. Thus, in order to retrieve snow thickness, we have to make assumptions on the remaining variables, i.e. vertical temperature and salinity distributions, ice type, snow density, and ice thickness. Finally, each grid cell of our retrieved map is assigned the snow thickness that is associated with the minimum value of the least squares differences between the TB(θ)-curves as modeled for the given ice conditions and the TB(θ)-curve as observed by SMOS.

Here, we present our new SMOS-based Arctic snow thickness maps, which are produced using different input data to run the emission model. Namely, these are ice surface temperature from MODIS thermal imagery, ice thickness class from CryoSat-2 altimeter measurements, snow density from a climatology, ice type from OSISAF, and ice concentration from ASI SSM/I data. Additionally, we assume average temperature profiles within the snow and ice as obtained from ice buoy data for different ice surface and ice and snow thickness conditions. Regarding the ice salinity, we use average profiles as found in the literature for first year and multi year ice, respectively. Furthermore, we use information on the statistical distribution of snow thicknesses on scales comparable to the SMOS footprint of about 35-50 km.

The resulting maps, representing 3-4 day averages, are compared with airborne ice and snow thickness measurements provided by NASA's Operation IceBridge missions from springs 2011 to 2015. When comparing individual grid cells, SMOS and IceBridge snow thicknesses may vary notably, also due to the very different footprints. However, in general, SMOS and IceBridge-based snow thickness distributions show similar spatial patterns of large-scale areas of thin snow and areas predominantly covered by thicker snow.

Current observations of the state of Arctic sea ice indicate a trend towards a younger, thinner, and more mobile pack that exhibits significant inter-annual variability. Radar returns from altimeters are impacted by the morphology of snow and ice features on the surface as well as the characteristics of radar pulse penetration through the snow pack. Together these contribute to uncertainty in the procedures for deriving sea ice freeboard, and hence thickness, from radar altimeter data. Here we seek to better constrain snow depth estimates and quantify the uncertainty resulting from ice surface morphology, with respect to ice type. The goal is to obtain an enhanced understanding of sources of uncertainty in altimeter-derived snow depth and sea ice thickness products by assessing both instrumental effects and effects related to the sea ice environment.

We demonstrate how improvements in the design and processing of FMCW snow radar data affect snow depth.  FMCW snow radar instruments deployed as part of airborne campaigns such as NASA’s Operation IceBridge mission and the Naval Research Laboratory’s DISTANCE project can detect the air/snow and snow/ice interfaces.   However sidelobes in the radar echograms have impacted data collected to date. New techniques have been employed to remove coherent noise and the impact of sidelobes through deconvolution. We demonstrate that the reprocessed data provide an improvement over initial data versions and can be used for more reliable derivation of snow depth.  We quantify the effect of ice morphology on the derived snow depth using a nested approach that includes forward modeling, airborne snow radar data, and a comprehensive set of in situ snow measurements collected in two-dimensional grids during the 2014 and 2015 sea ice field seasons.