Sea ice concentration products are used to understand the spatial characterization of sea ice cover and to calculate sea ice extent and area for time series analyses and process studies in the Arctic and Antarctic. Passive microwave data are particularly useful for sea ice studies because of the relatively high contrast in emissivities between open water and sea ice. This contrast is frequency-dependent; contrast increases with decreasing channel frequency. In most algorithms, atmospheric effects are assumed constant. The satellite-received radiation, expressed as a Brightness Temperature (T_b), is as follows (Cavalieri and Comiso 2000):

Tb = TbwCw + TbiCi   (Equation 1)

Where:

T_bw = brightness temperature of open water
T_bi = brightness temperature of sea ice
C_w = fraction of open water within instrument field-of-view
C_i = fraction of sea ice concentration within the instrument field-of-view

Sea ice concentration (C_I), corresponding to an observed T_b over a sea-ice-covered region (T_b), is derived as follows:

Ci = (Tb - Tbw) / (Tbi - Tbw)   (Equation 2)

The AMSR-E sea ice algorithms use this equation, but the channels and methods to derive sea ice concentrations are different. Values of brightness temperature of open water, brightness temperature of sea ice, and brightness temperature all include contributions from the intervening atmosphere. Brightness temperature of sea ice varies spatially because of spatial changes in emissivity and temperature of ice, while brightness temperature of open water is constant for open water within the ice pack.

Depending on the paramter being investigated, such as sea ice concentration or snow depth on sea ice, a different algorithm is used. Each algorithm is discussed in the following paragraphs.

Sea Ice Concentration

The 12.5 km sea ice concentration product is generated using the Enhanced NASA Team (NT2) algorithm described by Markus and Cavalieri (2000) for both the Arctic and the Antarctic. NT2 sea ice concentrations are calculated using the individual Level 2-A swaths rather than using gridded averaged brightness temperatures in order to make the atmospheric corrections on an orbit-by-orbit basis before obtaining daily average ice concentrations. However, previous AE_SI12 versions employed the Bootstrap Basic Algorithm (BBA) described by Comiso (1995) for the Antarctic. Now the BBA is only used in the sea ice concentration difference between the BBA and the NT2 (BBA-NT2) for both hemispheres.

The original NASA Team algorithm is based on techniques described in Cavalieri et al. (1984) and Gloersen and Cavalieri (1986). The NT2 algorithm (Markus and Cavalieri 2000) identifies two ice types for both the Arctic and the Antarctic:

In this algorithm, the primary source of error is attributed to conditions in the surface layer such as surface glaze and layering (Comiso et al. 1997), which can significantly affect the horizontally polarized 18.7 GHz brightness temperature (Matzler et al. 1984) leading to increased PR(18) values and thus an underestimation of sea ice concentration. The use of the horizontally polarized channels makes it imperative to resolve a third ice type to overcome the difficulty of surface effects on the emissivity of the horizontally polarized component.

Thus, one advantage of using the NT2 algorithm is that it uses the 89 GHz channels because the horizontally polarized 89 GHz data are less affected by surface conditions than the horizontally polarized 18.7 GHz data (Matzler et al. 1984), and the 89 GHz channels have successfully been used in sea ice concentration retrievals under clear atmospheric conditions. As a result, this algorithm can identify more complex ice types resulting from surface glaze and layering in the snow cover. The thin-ice bias is reduced through the use of the third ice type. The third ice type consists of surface glaze and layering of sea ice with snow cover. This type of ice primarily occurs in the Antarctic.

The Spectral Gradient Ratios (GR) and Polarization Ratios (PRs) are calculated from the observed and modeled brightness temperatures using the following two equations:

The Polarization Equation:

   (Equation 3)

The Spectral Gradient Ratio Equation:

   (Equation 4)

Where:
T_b is the brightness temperature at frequency ν for the polarized component p (vertical (V) or horizontal (H)).

The actual radiance ratios used in the NT2 algorithm are PR_R(18), PR_R(89), GR (89, 18, H-pol), and GR (89, 18, V-pol) where the subscript R refers to a rotation of axes. This rotation is done in the NT PR(18)-GR(37, 18, V-pol) domain, with the axes rotated through an angle, φ, until the ice-type lines (the FY-MY line for the Arctic and the A–B line for the Antarctic) are parallel to the GR axis. The rotated PR is defined by:

   (Equation 5)

and is independent of the two ice types. Next, we make use of the difference between two gradient ratios

   (Equation 6)

to resolve the ambiguity between pixels with true low sea ice concentration and pixels with significant scattering (defined as Ice Type C). Therefore, ΔGR serves as an indicator of the presence of Ice Type C. The higher the value of ΔGR the higher the amount of layering within a pixel. The retrieval of Ice Type C is applied to both, the Arctic and the Antarctic, oceans. Finally, a third parameter is defined to avoid the ambiguity between changes in sea ice concentration and changes in atmospheric conditions, because of the higher sensitivity of the 89-GHz channels to atmospheric variability compared to the lower frequency channels. This third parameter is the rotated PR(89), PR_R(89), computed from the PR(89)-GR(37V18V) domain analogous to the calculation of PR_R(18) but with a different angle.

The algorithm quantifies atmospheric effects by calculating brightness temperatures for each channel using a forward atmospheric radiative transfer model (Kummerow 1993) that incorporates four surface types: first-year ice, multiyear ice, Type C, and open water. Beginning with version 3, the Southern Hemisphere sea ice concentration algorithm no longer uses the Level-2A land flag and only uses the updated land mask for surface type classification. Thus, the 89 GHz channels together with a forward radiative transfer model provide ice concentrations under all atmospheric conditions. Figure 3 shows the general flow of the algorithm. First brightness temperatures are calculated for the four surface types and all weather conditions, currently 12. The response of the brightness temperatures to different weather conditions is calculated using the atmospheric radiative transfer model. The input data for the model consists of several things such as: the emissivities of the different surface types taken from Table 4-1 in Eppler et al. (1992) with modifications to achieve agreement between modeled and observed ratios, different cloud properties, specifically cloud base, cloud top, and cloud liquid water, taken from Fraser et al. (1975), and average atmospheric temperature and humidity profiles for summer and winter conditions taken from Antarctic research stations. Brightness temperatures are calculated for all possible ice concentration combinations in one percent increments, and the following ratios were calculate for each increment: PR_R(18), PR_R(89), and ΔGR. This creates a prism in which each element within this space contains a vector with the three ratios: PR_R(18), PR_R(89), and ΔGR. The subscript R refers to a rotation of axes in PR-GR space by the angle so that PR_R(18) and PR_R(89) are independent of ice types A and B in the Antarctic, and first-year and multiyear ice for the Arctic. For each pixel, the observed brightness temperatures are used to create a vector with the same ratios. The ice concentration for a pixel is determined where the difference between an observed and a modeled ratio is smallest.

While for the Antarctic the algorithm resolves fractions between Ice Types A, B, and C, and open water, in the Arctic we also resolve a thin ice type. Using a GR(37V19V) threshold of -0.02, we either resolve Ice Type C for pixels where GR(37V19V) is below this threshold, or thin ice for pixels where GR(37V19V) is above this threshold. A limitation, of course, is that mixtures of thin ice and thicker ice with layering cannot be resolved.

The algorithm eliminates spurious sea ice concentrations over open ocean by employing weather filters previously used for SSM/I (Gloersen and Cavalieri 1986; Cavalieri et al. 1995). Ice concentrations for image pixels with GR(37V19V) values greater than 0.05 and GR(2219) values greater than 0.045 are set to zero.

Although a land mask is applied to the ice concentration maps, land spillover still leads to erroneous ice concentrations along the coast lines adjacent to open water. This makes operational usage of these maps cumbersome. Therefore, a land spillover correction scheme is applied on the maps.

Sea Ice Concentration Difference

The BBA (Comiso 1995) is used in the calculation of sea ice concentration differences between the BBA and the NT2 (BBA-NT2) for both hemispheres. Comiso calculated tie points for consolidated ice (T_bi) and open water (T_bw) with two sets of channels: 18.7V GHz and 37V GHz (called V1937), and 37H GHz and 37V GHz (HV37). Using these channels for the retrieval of sea ice concentration ensures consistency with historical sea ice products, beginning with SMMR.

Snow Depth on Sea Ice

Snow depth over sea ice is reported as a 5-day running average, which is based on the current day and the previous four days. The 12.5 km snow depth on sea ice product is generated using the AMSR-E snow-depth-on-sea-ice algorithm described by Markus and Cavalieri (1998) for both the Arctic and the Antarctic. However, this algorithm is only valid over seasonal ice. The snow depth on sea ice is calculated using the spectral gradient ratio of the 18.7 GHz and 36.5 GHz vertical polarization channels:

hs = a1 + a2 GRV(ice)     (Equation 7)

Where:

GRV(ice) = [Tb(37V)-Tb(18V)-k1(1-C)] / [Tb(37V)+Tb(18V)-k2(1-C)]  (Equation 8)

Where:

The principal idea of the algorithm is similar to the AMSR-E snow-on-land algorithm (Kelly et al. 2003) utilizing the assumptions that scattering increases with increasing snow depth and that the scattering efficiency is greater at 36.5 GHz than at 18.7 GHz. For snow-free sea ice, the gradient ratio is close to zero and it becomes more and more negative as the snow depth and grain size increases. The correlation of regional in situ snow depth distributions and satellite-derived snow depth distributions is 0.81. The upper limit for snow depth retrievals is 50 cm, which is a result of the limited penetration depth at 18.7 and 36.5 GHz.

The algorithm is applicable to dry snow conditions only. At the onset of melt, the emissivity of both the 18.7 GHz and the 36.5 GHz channels approach unity (that of a blackbody) and the gradient ratio approaches zero initially before becoming positive. Thus, snow depth is indeterminate under wet snow conditions. Snow, which is wet during the day, frequently refreezes during the night. This refreezing results in very large grain sizes (Colbeck 1982), which leads to a reduced emissivity at 36.5 GHz relative to 18.7 GHz, thereby decreasing GRV(ice) and thus leads to an overestimate of snow depth. These thaw-freeze events cause large temporal variations in the snow depth retrievals. This temporal information is used in the algorithm to flag the snow depths as unretrievable from those periods with large fluctuations.

As in situ grain size measurements are even less frequently collected than snow depth measurements, the influence of grain size variations could not be incorporated into the algorithm. Because of the uncertainties in grain size and density variations as well as sporadic weather effects, AMSR-E daily snow depth products are five-day running averages similar to the AMSR-E snow depth on land product.

Figure 4 summarizes the snow depth algorithm. Snow depths are retrieved for the entire Southern Ocean, but only for the seasonal sea ice zones in the Arctic because the retrieval of Arctic snow depth is complicated by the presence of multiyear ice, which has a signature similar to snow cover on first-year ice. Both multiyear ice and deep snow on top of first-year ice results in increasingly negative values for the spectral GR; therefore, the algorithm only retrieves snow depth in the seasonal sea ice zones and in regions where the value of GR(37V19V) is greater than -0.02. This threshold corresponds to multiyear ice concentration of less than 20 percent. Where GR(37V19V) is less than -0.02, the algorithm flags pixels as multiyear ice. Because of the higher sensitivity of snow depth retrievals to ice concentration less than 20 percent, the algorithm limits snow depth retrievals to ice concentration between 20 -100 percent. Ice concentrations less than 20 percent appear almost exclusively near the ice edge.

Processing Steps

Figure 5 summarizes the input products and algorithms used to create gridded AMSR-E 12.5 km sea ice products.

The AMSR-E/Aqua L2A Global Swath Spatially-Resampled Brightness Temperatures are binned into 12.5 km grid cells using a drop-in-the-bucket method where the grid cell that contains the center of the observation footprint is given the whole weight of the observation. With this procedure, the number of observations is always in whole numbers. All valid brightness temperature observations within the extent of the polar grids are binned into grid cells, including land observations.

After input Level-2A brightness temperatures are binned into 12.5 km grid cells, the ascending, descending, and daily data are averaged. The daily average is not simply an average of ascending and descending orbits because a given pixel could have, for example, three measurements from ascending orbits and two from descending orbits. Instead, the daily average is of all the observations for that grid cell. For example, if A = ascending and B = descending:

((A1 + A2) / 2 + (B1 + B2 + B3) / 3 ) / 2  (Equation 9)

is not equal to:

(A1 + A2 + B1 + B2 + B3) / 5  (Equation 10)

However, this biases day-time ascending orbits over night-time descending.

Sea Ice and Snow Melt Masks

Beginning with algorithm version V12, the sea ice and snow melt masks used for the Northern Hemisphere snow depth on sea ice parameter are stored within this data set. The masks from the most recent HDF file are used to generate each new HDF file. If the masks used are more than seven days old, the metadata science quality flag is set to Suspect. 

The multiyear mask field in the HDF file refers to the multiyear sea ice mask. A default mask is used to initialize the multiyear mask on the first day of mission and also to reset the mask on the first processing date in October of each year.

The variability 5-day field in the HDF file refers to the snow melt mask. The snow melt mask corresponds to the running 5-day snow melt product, and its date must match that of the multiyear mask. The snow melt mask is reset at the same time as the multiyear mask.

Land Masks

The AMSR-E sea ice product utilizes a 12.5 km Northern Hemisphere land mask (amsr_gsfc_12n.hdf) and a 12.5 km Southern Hemisphere land mask (amsr_nic_12s.hdf). The North land mask is identical to gsfc_12n.msk and was last updated in 1997. The South land mask includes an updated ice shelf definition created by the National Ice Center Science Department in June 2011 and an updated shoreline developed from ENVISAT and RADARSAT imagery from October 2009 to April 2010.

A 1-byte integer array is included in each HDF file.

amsr_gsfc_12n.hdf: 608 columns x 896 rows
Values are 0 (water) and 1 (land).
amsr_nic_12s.hdf: 632 columns x 664 rows
Values are 0 (water), 1 (land), and 2 (coast).

Version History

Changes to the Version 3 (V15-stage1) algorithm for these data include: 

• Use of the most recent version (Version 3) of the AMSR-E/Aqua L2A Global Swath Spatially-Resampled Brightness Temperatures data as input
• An improved Antarctic land mask
• No longer uses the Level-2A land flag and only uses the updated land mask for surface type classification in the Southern Hemisphere sea ice concentration algorithm
• Inclusion of ISO lineage metadata

Refer to the AMSR-E Data Versions Web page for a summary of changes since the start of mission

Refer to the AMSR-E Instrument Description document.