The input data for the sea ice CDR variables, the NASA team sea ice variables, and the Bootstrap sea ice variables are described in Table 13.

Overview

NSIDC processes the input brightness temperatures into two different intermediate sea ice concentrations using two GSFC-developed algorithms: NASA Team (Cavalieri et al. 1984) and Bootstrap (Comiso 1986). These intermediate NSIDC NASA Team and Bootstrap sea ice concentrations are used in the NOAA/NSIDC CDR algorithm described in further detail in the CDR Algorithm section of this document.

The passive microwave channels employed for the sea ice concentration product are the 19 GHz, 22 GHz, and 37 GHz frequencies. Table 14 lists the channels used for each algorithm and the channels used for the weather filters. For a complete description of the weather filters, see the Climate Algorithm Theoretical Basis Document (C-ATBD): Passive Microwave Sea Ice Concentration document (Meier, Savoie, and Mallory 2011).
 

Since this data set uses multiple sensors over time, the sea ice algorithms are intercalibrated at the product (concentration) level by NASA GSFC. Thus, the brightness temperature source is less important because the intercalibration adjustment includes any necessary changes due to differences in brightness temperature across them. Both the NASA Team and Bootstrap algorithms employ varying tie-points to account for changes in sensors and spacecraft. These tie-point adjustments are derived from regressions of brightness temperatures during overlap periods. The adjustments are made at the product level by adjusting the algorithm coefficients so that the derived sea ice fields are as consistent as possible.

The NASA Team approach uses sensor-specific hemispheric tie-points for each transition (Cavalieri et al.1999; Cavalieri et al. 2011). Tie-points were originally derived for the SMMR sensor and subsequent transitions to the different SSM/I instruments adjusted the tie-points to be consistent with the original SMMR record. The Bootstrap algorithm uses daily varying hemispheric tie-points, derived via linear regression analysis on clusters of brightness temperature values of the relevant channels (Comiso 2009, Comiso and Nishio 2008). Also, in contrast to the NASA Team, Bootstrap tie-points for SMMR and SSM/I are derived from matching fields from the AMSR-E sensor, which is newer and more accurate.

Automated Quality Control

Automated quality control measures are implemented independently on the intermediate NASA Team and Bootstrap outputs. Two weather filters, based on ratios of channels sensitive to enhanced emission over open water, are used to filter weather effects. Separate land-spillover corrections are used for each of the algorithms to filter out much of the error due to mixed land/ocean grid cells. Finally, to screen out errant retrievals of ice in regions where sea ice never occurs, valid ice masks are applied to the Northern Hemisphere and climatological ocean masks are applied to the Southern Hemisphere.

For a complete description of the automated filters and masks, see the C-ATBD (Meier, Savoie, and Mallory 2011).

CDR Algorithm

Different algorithms exist for computing sea ice concentration from brightness temperature data. The two widely used GSFC-developed NASA Team (Cavalieri et al. 1984) and Bootstrap (Comiso 1986) algorithms are describe below. Both of these algorithms have their own inherent advantages and limitations. For this CDR data set, the NASA Team-derived sea ice concentrations are merged with the Bootstrap-derived sea ice concentrations using the CDR algorithm into a single ice concentration estimate. The CDR algorithm steps are as follows:

• First, the sea ice edge is defined using only the Bootstrap-derived data with a 10 percent concentration threshold cutoff. In other words, any grid cell near the ice edge showing a concentration of less than 10 percent in the Bootstrap data is set to open water in the CDR; otherwise it is set to the Bootstrap-derived concentration. Bootstrap is used for the edge because of the ambiguity and potential inconsistencies between how the edge is detected by the NASA Team and Bootstrap algorithms (Meier et al. 2014).
• Second, at each sea ice grid cell within the ice cover, the concentration value given by the NASA Team algorithm and that given by the Bootstrap algorithm are compared; whichever value is greater is selected as the CDR value. This is done because both algorithms tend to underestimate ice concentration, however the source of this bias differs between algorithms (Meier et al. 2014).

NSIDC processes the input brightness temperatures into the two intermediate NASA Team and Bootstrap sea ice concentrations. The processing for the intermediate NASA Team concentrations for the sea ice CDR is nearly the same as the GSFC-processed Sea Ice Concentrations from Nimbus-7 SMMR and DMSP SSM/I-SSMIS Passive Microwave Data with two known differences. First, NSIDC uses a new brightness temperature version for the F8 period from what GSFC processed. Second, NSIDC uses a corrected version of brightness temperatures for F11 and F13, while GSFC used the uncorrected version. Spatial and temporal interpolation is done by GSFC, and they also perform an additional manual QC step. In comparisons between the two, there are occasional unexplained small differences.

The NASA Team algorithm, because it uses a ratio of brightness temperatures, tends to cancel out any physical temperature effects. The Bootstrap algorithm uses relationships between two brightness temperatures that are dependent on physical temperature. Thus, physical temperature changes can affect Bootstrap estimates. Errors occur primarily in regimes with very low temperatures: winter in the high Arctic and near the Antarctic coast (Comiso et al. 1997), where the Bootstrap algorithm can underestimate concentration and give a lower value than the NASA Team algorithm. During winter conditions with more moderate temperatures, NASA Team concentrations also tend to have more of a low bias (Kwok 2002; Meier 2005). During melt conditions, both algorithms tend to underestimate concentration; but the effect is more pronounced in the NASA Team algorithm (Comiso et al. 1997; Meier 2005; Andersen et al. 2007).

While these characteristics of the algorithm are true in an overall general sense, ice conditions and algorithm performance can vary from grid cell to grid cell; and in some cases, this approach of choosing the larger value will result in an overestimation of concentration (Meier 2005). However, using the higher concentration between the two algorithms will tend to reduce the overall underestimation of the CDR estimate (Meier et al. 2014). For a more in-depth discussion on the reasoning behind the algorithm, see the C-ATBD (Meier, Savoie, and Mallory 2011).

NASA Team Algorithm

The NASA Team algorithm uses brightness temperatures from the 19 GHz V, 19 GHz H, and 37 GHz V channels. The methodology is based on two brightness temperature ratios, the polarization ratio (PR) of the 19 GHz V and H channels (Equation 1) and the spectral gradient ratio (GR) of the 19 GHz V and 37 GHz V channels (Equation 2).

Where:

When PR and GR are plotted against each other, brightness temperature values tend to cluster in two locations, an open water (zero percent ice) point and a line representing 100 percent ice concentration, roughly forming a triangle. The concentration of a grid cell with a given GR and PR value is calculated by a linear interpolation between the open water point and the 100 percent line segment. See Figure 2.

For a detailed description of the NASA Team algorithm, please see the NASA Team Sea Ice Algorithm Web page, the NASA Technical Memorandum 104647 (Cavalieri et al. 1997) that includes information about differences (for example, tie points) between

the original algorithm and the revised NASA Team algorithm, and the NASA Team Algorithm section of the C-ATBD (Meier, Savoie, and Mallory 2011) for a table of tie-point values.

Bootstrap Algorithm

Like the NASA Team algorithm, the Bootstrap algorithm is empirically derived based on relationships of brightness temperatures at different channels. The Bootstrap method uses the fact that scatter plots of different sets of channels show distinct clusters that correspond to two pure surface types: 100 percent sea ice or open water.

Figure 3 shows a schematic of the general relationship between two channels. Points that fall along line segment AD represent 100 percent ice cover. Points that cluster around point O represent open water (zero percent ice). Concentration for a point B is determined by a linear interpolation along the distance from O to I where I is the intersection of segment OB and segment AD. This is described by Equation 3.
 

Where:

The Bootstrap algorithm uses two such combinations, 37 GHz H versus 37 GHz V and 19 GHz V versus 37 GHz V, denoted as HV37 and V1937, respectively. Points that fall within 5 K of the AD segment in a HV37 plot, corresponding roughly to concentrations greater than 90 percent, use this approach. Points that fall below the AD-5 line, use the V1937 relationship to derive the concentration. Slope and offset values for line segment AD were originally derived for each hemisphere for different seasonal conditions (Table 2 in Comiso et al. 1997). However, a newer formulation, employed in this CDR, was developed where slope and offsets are derived for each daily field based on the clustering within the daily brightness temperatures (Comiso and Nishio, 2008). For a detailed description of the Bootstrap algorithm, please see the Bootstrap Algorithm Web page.

For the NOAA/NSIDC CDR data (seaice_conc_cdr), NSIDC uses brightness temperatures from SSM/I sensors on the DMSP-F8, -F11, and -F13 platforms and from the SSMIS sensor on DMSP-F17. See Table 17 for a description of orbital parameters of the different platforms. The rationale for using only these satellites was made to keep the equatorial crossing times as consistent as possible to minimize potential diurnal effects of data from sun-synchronous orbits of the DMSP satellites. Note that the SMMR sensor on the Nimbus-7 platform is used for the merged GSFC NASA Team/Bootstrap data (goddard_merged_seaice_conc), the GSFC NASA Team data (goddard_nt_seaice_conc), and the GSFC Bootstrap data (goddard_nt_seaice_conc) from 1978 to 1987.

*Indicates sensor and spacecraft orbital characteristics of the three sensors used in generating the sea ice concentrations.

Table 18 lists the footprint size of the SSM/I instrument.

For a complete description of the SSM/I instrument, see the NSIDC Special Sensor Microwave Imager (SSM/I) Web page; and for the SSMIS instrument, see the NSIDC Special Sensor Microwave Imager/Sounder (SSMIS) Web page. For a complete description of the different DMSP platforms, see the NSIDC DMSP Satellite F8, F11, F13, and F17 Web pages.