Sea Ice 2

Currently,  no satellite sensor  can measure the full range of sea ice thicknesses h_i from a few centimeters up to several meters. Many sensors can provide us information about some sea ice thickness range: e.g. SMOS L-band radiometer  (thickness <0.5m) and  CryoSat-2 radar altimeter (thickness > 1 m). Hence, in many applications the best option is to rely on the modeled ice thickness even if it may be subject to a large uncertainty.

We propose here an approach where we modify the modeled  h_i using satellite data to generate an ice thickness chart (ITC) with reduced uncertainty compared to the model data. At present, this is done in three different ways:

Enhance the spatial resolution of h_i  using SAR data.

Perform thin ice (<30 cm) area detection with radiometer (AMSR-2) data and also with SAR data.

Divide the marginal ice zone into ice concentration (IC) zones without thickness information if radiometer based IC<70%.

Thin ice (<30 cm) detection is performed using the radiometer polarization and spectral gradient ratios as discussed in Mäkynen and Similä (2015)*.  IC is retrieved from the JAXA AMSR2 L2 IC product with 10 km resolution.

The basic information source for ice thickness is provided by the TOPAZ4 system. It consists of an ocean model  and a dynamic-thermodynamic sea-ice model and is freely available through the Copernicus Marine Services. The ice thickness fields of TOPAZ4 are modified utilizing SENTINEL-1 SAR EW Mode imagery (HH and HV polarizations)  using factors which we call here as the scaling factors.

Our test area consists of the Kara Sea and the Barents Sea. A daily SAR mosaic is collected from the most recent Sentinel-1a (S-1a) images available over the region of interest. They are rectified, incidence angle corrected and sampled into the resolution of 500 m. Based on the CryoVex 2005 campaign results, we assume that the HH-polarization backscattering coefficient σ⁰_HH is linearly  proportional to the fraction of level ice area on the range [-20dB,-12dB].  Using this property we construct a scaling factor to modify the modeled ice thickness. The magnitudes of  the scaling factors vary from 0.5 (mainly level ice) to 1.4 (heavily ridged ice). Hence, we presume  ± 40 % uncertainty for modeled thickness values. σ⁰_HH values less than -20 dB are interpreted to represent either very thin ice or calm open water.  The values above -12 dB are regarded either as brash ice, thin ice with frost flowers or rough open water. 

The scaling factors are determined so that the total ice volume over the Kara Sea basin produced by TOPAZ4 and ITC are close to each other.

In the performance analysis our reference data consisted of weekly ice charts of the Arctic and Antarctic Research Institute (AARI).  The AARI charts use the WMO ice typing (stage of development), where an ice type represents a specific level ice thickness range. Deformed ice is not included in these AARI charts. Our TOPAZ4 based ITCs are on average close to the AARI charts, which suggests that  TOPAZ4 does not create enough deformed ice.

Our belief about the underestimation of ice thickness by the TOPAZ4 system became stronger when we run NEMO (Nucleus for European Modelling of the Ocean)  LIM (Louvain-la-Neuve Sea Ice Model) for the intercomparison between the models in winter 2014. A comparison between  TOPAZ and NEMO-LIM for winter 2015-16 data will be performed.

We are currently working on the S-1a based feature vectors to provide a more precise description of ice conditions than  σ⁰_HH alone.

*Mäkynen, M., Similä, M. 2015. Thin Ice Detection in the Barents and Kara Seas With AMSR-E and SSMIS Radiometer Data, TGARS, 53(9).

Leads, as large fracture in sea ice, are of potential interest for the navigation in the Arctic sea. Even though leads rapidly refreeze (as opposed to polynyas), they contribute to the Ocean-Atmosphere interactions. Beyond its role in the global climate, leads are used in the altimetry community to estimate the elevation difference with the top of sea ice surface. The estimated freeboard can be then converted to the sea ice thickness.

The Sentinel-1 (S-1) mission, as part of the Global Monitoring for Environment and Security (GMES/Copernicus) European initiative, is a two satellite constellation with the prime objectives of Land and Ocean monitoring. The goal of the mission is to provide C-Band SAR data continuity following the retirement of ERS-2 and the end of the Envisat mission. The Sentinel-1A satellite has been launched on 3^rd April 2014. Two altimetry missions observe also sea ice: Cryosat-2 (ESA mission) and SARAL/AltiKa (CNES/ISRO mission). Cryosat-2 carries a Ku-band altimeter (SIRAL) whereas SARAL carries a Ka-band altimeter (AltiKa). From these two types of sensors, we are able to generate lead maps.

On altimetry data, a neural network is used to classify the waveforms depending on their geometry. The leads are identified as peaky waveforms with a strong amplitude, named as “class #2”. A post-processing is then performed in order to remove any off-nadir detection.

The dual-polarized S-1 images acquired in the Extended mode are primarily used. The two HH and HV channels along with textural layers are used as input of a Bayesian scheme considering low resolution sea ice concentration from ancillary sources (e.g. from OSISAF). CMOD Geophysical Model Function is also integrated in the methodology to take into account the wind effect over the open sea surface. A maximum a posteriori (MAP) decision rule is employed to assign a label (open water or sea ice) to each SAR pixel. Morphological operators are finally used.

The aim of the study is to compare the lead detection obtained with SAR imagery and with altimetry in both Ku and Ka band, over Arctic during winter 2015. In this paper, we will present the automatic lead detection method based on S-1 images, the altimeter waveforms classification, and the comparison of the leads maps obtained on the 3 missions. This analysis aims at demonstrating the complementarity between SAR and altimetry data to monitor sea ice.

Sea ice dynamics within Arctic regions affect the surface energy and mass balances crucially. Dynamical processes contribute to ice growth through opening and closing of leads, and formation of pressure ridges. These processes change the morphology and surface roughness of the sea ice, which are important for summer melt pond and floe size distribution that influence the melt processes.

In this study we present a unique combination of two winter and spring sea ice drift datasets obtained during the N-ICE2015 (Norwegian young Sea ICE cruise) field campaign that took part from January to June 2015 in the region north of Svalbard: the in-situ data from the buoy array and the SAR derived sea ice velocities.

The buoy array consists of 42 buoys that were deployed in two consequent deployments, one in January/February and the second in April/May.  The buoys in both deployments were deployed at distances of 2 to 100 km apart and were tracking the sea ice drift for approximately 2 months. The positions of the buoys were triangulated to obtain the sea ice deformation information.

The sea ice drift data from the buoy array was used to validate the high resolution sea-ice drift dataset derived from a cross-correlation algorithm applied to satellite SAR data from RADARSAT-2, Sentinel-1, TerraSAR-X, and ALOS. We present an assessment of the algorithm quality in respect to the spatial resolution of used SAR platform and frequencies (C, X, L band).

The sea ice deformation calculated from the SAR data is also compared to the buoy-derived deformation. This allows us to look at the scaling effects on sea ice deformation; for example to determine what part of the sea ice deformation distribution is not captured by the SAR data due to the different spatial resolution and power-law scaling. We also investigate whether SAR data can be used to estimate the deformation part (long tail) of the sea ice thickness distribution.

Sea ice thickness retrieval using L band radiometry is one of the tools that enables measurements of thin sea ice on global scale. Current technology limits the resolution of the sea ice thickness product to ~40 km. By combining radiometer data with finer resolution auxiliary data the sea ice thickness product can be downscaled to a higher resolution and the uncertainty of the product can be improved.

The radiometers in SMOS (Soil Moisture and Ocean Salinity) and SMAP (Soil Moisture Active and Passive), although different in design, provide daily coverage of the polar regions. At the University of Hamburg this data is used in order to retrieve the sea ice thickness. The L-band brightness temperature (TB) is sensitive to level sea ice thickness up to ~0.5 m, beyond that thickness the signal saturates. One of the versions of the retrieval algorithm takes into account the heterogeneity of the ice thickness by introducing a distribution of sea ice thickness within the radiometer footprint. The auxiliary thickness distribution data originate mostly from field campaigns such as NASA's Operation Ice Bridge. However, the thickness distribution is a very local characteristic and varies in time and between geographical regions.

In the SMAP mission simultaneous radar acquisition with a resolution of 3 km are done within the radiometer footprint. When the radar back-scatter is related with sea ice thickness these data can be used to characterize the ice thickness distribution on sub-pixel scale. However, associating radar back-scatter with sea ice thickness is not straight forward since surface scattering depends on incident wavelength, dielectric properties of the medium as well as on surface roughness.

As a first step, we used data obtained during the SMOSice2014 campaign to characterize the surface roughness. The “Polar 5” research aircraft surveyed sea ice in the Barents Sea with an airborne laser scanner (ALS) scanning the ice surface in scan lines perpendicular to the flight direction. Simultaneously an aerial camera was taking photos of the ice surface, which were used to identify different ice types. For these ice types height distributions and power spectral density (PSD) of the surface roughness were derived from the scan lines of the ALS. The standard deviation of surface height and the type of auto-correlation function were then calculated from the PSDs – the two parameters essential for modeling of the surface scattering. Using these parameters, the radar back-scatter over different ice types was simulated with an analytical formula (“Stationary Phase Approximation”).

These results can then be used for the downscaling approach of the L-band TB sea ice thickness using radar data. The three available months of SMAP radar and radiometer data are used to demonstrate the principle. The radar images are used to classify the surface into open water, thin ice, and thick ice classes. Subsequently, the radiometer footprint is decomposed based on these ice classes. The TB value for the regions with the thick ice class is set to be at the saturation level and the retrieval of thickness is done only over the thin ice fraction of the footprint.

Our understanding of sea ice mass balance has been revolutionized thanks to satellite remote sensing observations that provide continuous monitoring of the ice pack, across basin scales. Recent results from ICESat, CryoSat-2 and IceBridge suggest interannual variability in Arctic sea ice thickness and volume, superimposed on a multi-decadal decline. The largest changes include a decline in the extent and thickness of the multiyear ice pack. A declining multiyear ice cover will precondition the pack for further loss, severely altering the mass and energy budgets of the Arctic Ocean. Further, regional sea ice thickness distributions are impacted by both the location and deformation characteristics of the multiyear ice cover. 

Satellite and airborne altimeters acquire unique measurements of sea ice elevation, from which ice thickness may be derived.  They also provide surface roughness and ice deformation information, at varying length-scales. The extent and location of multiyear ice across the Arctic basin may be accurately determined using contemporaneous scatterometer data. We examine sea ice thickness and surface roughness derived from high-resolution NASA Operation IceBridge airborne measurements, alongside complementary satellite radar and visible-band imagery, and scatterometry data. We assess the trends and map the interannual variability in the location, thickness and deformation characteristics of the multiyear ice pack. Quantifying why and how the multiyear ice cover is changing will improve predictions of the future Arctic climate system