Datasets and Preparation


SSM/I Ice Motion
The Special Sensor Microwave/Imager  SSM/I is flown  on the satellites of the DMSP (Defence Meteorological Satellite Program) since 1987. The sun synchronous orbit, the period of about 100 minutes and the sensors swath wide of 1394 km allows a regular cover of the polar regions at least once a day. (Hollinger et al., 1987). The 19 and 37 GHz channels have a resolution of 25km, the 85 GHz channel has the double resolution of 12,5 km, but is more sensible to atmospheric influence. SSM/I was proceeded by the Scanning Multichannel Microwave Radiometer SMMR on the NIMBUS 7 satellite from 1978 to 1987, which could provide information on 5 frequencies from 6 to 37 GHz. (Cavalieri et al., 1984)
Methods for calculation of sea ice motion out of 85 GHz data are described in detail by Agnew and Le (1997). They give a theoretical error of these estimations of about 4,6 km/d when looking at the geolocation error, tracking error, swath smearing effect and timing error. Velocity errors are divided by the time interval between the images and so can be improved with higher time intervals between the compared images. But this is related to worse temporal resolution and greater tracking errors. SSM/I drift data is available as optimal interpolated and not interpolated  37 GHz and 85GHz values, with errors in 37 GHz being smaller than expected in comparison to the double resolution 85 GHz Channel (Maslanik et al., 1998) .

Drift estimates are  calculated every second day and are available from 1979 to 1997, while one day calculations exist only in the time after 1992 (since the 85 GHz channel was only available more recently on SSM/I).
The size of the search window of the feature tracking algorithm  has an effect of the results, too, cause a large window allows reducing noise but also leads to more spatial averaging and so reduces the number of independent velocity vectors (SEALION- Lemke et al., 2001). The size of the search window leads to the spatial resolution of the acquired drift vectors. The ground resolution of the pixels and the temporal intervals between two images for drift determination lead to the minimum velocity which is detectable. 

The Sea Ice Drift Atlas uses SSM/I Optimal Interpolated data (further cited as OI-data) as an advanced product, combining results from the 37GHz and the 85GHz passive microwave radiometer channel with additional drift information from buoys when geographically and temporally available.
The data is calculated with a weighting function :


out of the two channels with different spatial resolution on a final rectangular grid with 100km cell spacing.
The weighting coefficients alpha, beta and gamma are determined after Colony and Thorndike (1984). Solutions for each point are obtained based on the uncertainties, the expected variance of the motion and the distance to available observations. (Kwok et al., 2000 + personal communications).
SSM/I_OI drift data is only available for month from March to November because satellite ice drift products are not generated for austral summer. The ice surface decorrelates and prevents tracking during the melt season. Data is available at one-day and two-day resolution, with the two-day resolution covering the longer timespan from 1979 to 1997 and therefore is being used for the Atlas products. The ice motion data were produced by the Polar Remote Sensing Group at the Jet Propulsion Laboratory, California Institute of Technology, and are provided by courtesy R. Kwok.


Drifting Buoy Ice Motion
Data from buoys, drifting on sea ice floes and measuring meteorological parameters provide better temporal and spatial higher resolution than satellite based data.  Sea ice drift velocities from buoys were included in the Atlas for a more complete presentation and  for verification of satellite drift fields . Additionally, some information of the buoy data is already included by calculating the SSM/I OI products. Data were collected and provided within the International Programme for Antarctic Buoys (IPAB) since 1995. Former data were collected at the  Alfred-Wegener-Institute (AWI). The figure below shows the absolute number of buoys per month, and how they are divided in the different  ocean regions around Antarctica. Icedrift is calculated from the change of buoy position reports at different times. The accuracy of the ice drift and its derivatives is estimated from the rms error of Argos positioning at approx. 350 m by Doppler shift of the transmission frequency at different satellite positions. It is assumed that position errors are Gaussian distributed and temporally uncorrelated, which is confirmed by the Argos positioning of an automatic weather station on the Filchner-Ronne Ice Shelf. Position reports of different temporal intervals are interpolated to 3h resolution, with gaps being filled by linear interpolation between two points to get regularly spaced time series, see
(Kottmeier and Sellmann, 1996). The errors of the ice drift components decrease with increasing averaging period. Typically they are 4,6 cm/s for three-hourly data and around 1 cm/s for daily data. From 1994 onwards several buoys were equipped with Global Positioning System (GPS) receivers, which provide an improved location accuracy. Using the GPS transmission of the Standard Positioning Service (SPS) and averaging times of 10 min, position accuracies of better than 50 m are obtained. The errors of the drift velocity for daily data goes down to 0.3 cm/s.


Sea Ice Concentration
Ice concentration data is taken out of the CD of the EU funded Project Pelicon (Project for Estimation of Long-term variability in Ice Concentration, Heygster et al., 1996). For temporal constancy, data calculated with the NASA2 Team  algorithm (Comiso et al., 1997) are chosen, which are available for the whole examination period. The algorithm uses 17 and 37 GHz  channels to derive ice concentrations from brightness temperatures. In the Pelicon algorithm, a correction scheme of Thomas (1998) for areas with ice concentration below 50%  is included, which reduces the overestimation due to the atmospheric influences.

Meteorological Data
Model output data from NCEP Reanalysis  Project (NNRP) datasets, a 40 year record of global analysis of atmospheric fields,  are used to combine ice drift and atmospheric parameters. The datasets in use were  at pressure level partition or as surface data on a 2,5x 2,5 latitude/longitude grid and have a temporal resolution of 6 hours.
The principle of this long time reanalysis dataset is to use a fixed system for assimilation, analysis and forecasting for using the data from 1957 to present.  This allows investigation of interannual variability with being sure that no model changes will have influence. It involves land surface, ship, rawinsonde, pibal, aircraft, satellite and other data from various organisations. For an analysis as accurate as possible for the past 40 years, all available data at given times are used.  Additional information is  available from Kalnay et al., (1996).

Elevation Data
Antarctic land topographic data, which are included in the geoinformation system come from DEM GLOBE . The Global Land One-kilometer Base Elevation (GLOBE) digital elevation model is the most thoroughly designed, reviewed, and documented global digital elevation dataset to date. GLOBE was developed by an international group of specialists, cooperating with the Committee on Earth Observation Satellites (CEOS) Working Group on Information Systems and Services (WGISS), International Geosphere-Biosphere Programme's Data and Information System (IGBP-DIS), and IGBP Working Group IV/6. GLOBE comprises a global 30 arc-second latitude-longitude array, with land areas populated with integer elevation data. Full GLOBE documentation is available on the CD-ROM and GLOBE Website. (Hastings et al. 1998)

Geographic Information System
All data is uniformly georeferenced and combinated in the Geographic Information System PCI. This allows the spatial and temporal overlay of the different data and makes a combined analysis, for example like the ice drift-wind forcing relation, possible.


The following steps were undertaken, to get homogenous, temporal averaged datasets of vector- and raster-data:
  • Georeferencing:
    Data with cartesian x-y-coordinates (polarstereographic) like the SSM/I and Pelicon data are converted to latitude-longitude-coordinates, according to the grid in which they exist. Georeferencing is done using Hughes Ellipsoid with a latitude of true scale at 70S. This leads to the smallest distortions in length at the areas around Antarcticas coast.
  • Temporal averaging:
    Out of daily data, different mean values are calculated. Due to the restricted availability of satellite products of only nine months per year, there are monthly means from March to November, seasonal means for the three seasons March-May, June-August and September-November and annual means including all the nine months. For the same three averaging intervals, interannual means, including the mean of a month or period out of the complete data-range from 1979 to 1997, are available.  Due to  an average ice motion of around 200 km in 30 days (Kottmeier and Sellmann, 1996) , a spatial averaging of ice drift must be taken in account in the mean values.
  • Buoy data preparation:
    Buoy data at different temporal resolution is careful checked and controlled, according to the buoy documentation, and then put together to daily means. These data, as well as daily means of 6hourly NCEP data is further on treated like the satellite data to calculate monthly means and statistics, so that the monthly or seasonal temporal variances always relate to daily input values. Combinated images for comparison of satellite and buoy drift data is only built for monthly means, because averaging intervals longer than one month is not suitable for most of the buoy data. In this images, OI-vectors that are calculated with inclusion of buoy data are highlighted by white dots. So it is to see, that not all available buoy data was included in the OI data. A more detailed description on the effects of this can be found in the Quality Check.
  • Statistical parameters and calculation of covariance ellipses:
    For all kind of mean products, statistical values like the variance of each velocity component, the complete drift variance and covariance is calculated and added to the ASCII-datasets in the Download area.  For better illustration of the coupling variability, covariance ellipses are calculated and plotted. Their size and orientation describe the magnitude of the variance as well as the relation of variance in the longitudinal u-component and meridional v-component.

    Covariance ellipses are developed out of the covariance matrix like described in Nikolaj Nawri (1996).  

  • NCEP and additional data preparation:
    NECP reanalysis data is first being interpolated to daily values to  have the same temporal resolution like drift data for getting the statistic parameters of different averaging periods. In the graphical outputs, a vector thinning for points in high latitudes is done, because the 2,5 degrees resolution could not be optical resolved in a polarstereographic projection.
  • Interpolation of gridded fields:
    SSM/I and NCEP data is imported to the geoinformation system as vector data with its geographic coordinates for positioning and the drift components and statistic values as vector attributes. For evaluation of climatological patterns and changes in sea ice motion and atmospheric data over the last 20 years, gridded fields of this values are best to use. Gridded fields for sea ice motion, variance, air  pressure and temperature are created by interpolation of vector data to raster data. Interpolation is done by using a nearest-neighbour-interpolation with 6 nearest neighbours (i.e., the 6 nearest of all sample points to the grid point in question) weighted by their distances from the grid point.  The interpolation equation then becomes:


    Where for each k-nearest neighbour, Zi is its grey level, Di is its distance from the current grid point being interpolated. For each grid point, the k nearest neighbours must be identified from the entire data set before the distance-weighted equation can be employed. This spatial interpolation range for the single grid points is similar to interpolation of  gridded fields out of buoy positions wit an acceptance circle of around 300 km like done in Kottmeier et al.(1997).
    All gridded fields related to ice motion data are circumscribed by the ice margin out of Pelicon data. Because often drift vectors are not available at all points of the ice covered region, it should be noted that interpolated values are only representative in the regions with drift arrows.


  • Calculation of differential kinematic parameters:
    Differential kinematic parameters like vorticity, divergence and shear give a detailed overview of the motion of the sea ice field.

    Based on its mathematical sense of the divergence of a continuous vector field, the divergence of the 2-D vector field of sea ice motion means the relative  change of an area covered all neighbouring ice floes within an enveloping curve, whereas changes of the shape or orientation of this curves does not affect divergence . Divergent ice motion causes opening of leads, causing  enhanced heat losses from the  ocean in direct contact to the atmosphere. , Convergent ice motion decreases the size of leads between the floes and may result in larger internal ice stress.

    Vorticity describes changes of the orientation of the enveloping curve. Ice drift vorticity has been found empirically to be related to the vorticity of the geostrophic wind (Kottmeier and Sellmann, 1996).  This is  mainly documented by investigation of buoy drift for certain regions and can be seen here for all areas and longer periods.

    Shear means the changes in shape,  having two components in  direction  of the mean motion and perpendicular to it. The shear rate of ice drift is found to be  relatively close to the shear rate of the geostrophic wind field, too in case studies of drift.

     The parameters are calculated from the drift vector D with components (u,v) in the following way:

    Divergence                                   div(D) = du/dx + dv/dy

    Drift vorticity                            curl z(D) = dv/dx - du/dy

    Shearing rate of ice drift          shear(D) = 

    consisting of

    Normal Deformation Rate         NDR(D) = du/dx - dv/dy

    and Shear Deformation Rate     SDR (D) = dv/dx + du/dy.

    To suppress effects of noise errors of u and v on derivatives, dx and dy are chosen to be 600km, which comes close to correlation lengths of ice motion detected in many investigations of ice motion in the Southern Ocean. For different seasons during the Winter Weddell Sea Project, Kottmeier et al. (1992) calculated the spatial correlation lengths of the buoy drift speeds and components. The longitudinal correlation lengths of the drift speeds varied between 490 and 680 km; the lateral ones between 270 and 540 km. Vihma et al. (1996) provide correlation lengths of 700 km for the longitudinal and 550 km for the lateral drift speeds for the central and western Weddell Sea. As a result of the large correlation lengths, spatial gradients of ice drift obtained from SSM/I are able to reflect meaningful vorticity and divergence on these scales. Smaller-scale DKPs may be much more important for lead and polynya formation, but cannot be obtained reliably from SSM/I-based ice drift.

  • Quality Check on Ice Motion Data
  • An extensive quality check of satellite OI data compared to buoy data is done out of the whole dataset, see the pdf-file Quality Check. Detailed studies on comparisons of satellite drift data with drifting buoy measurement (Maslanik et al., 1998; Kwok et al., 1998; Geiger et al., 2000) show that it is not sufficient to just compare a speed value at a certain buoys position and time to the nearest gridpoint on SSMI grid. Thus a method is used and tested for best temporal and spatial scales for comparison. In the coastal regions of the western Weddell Sea, more perennial ice is expected and the buoy motion is constrained with the ice being compressed against the coastal barrier. In the central region, there is predominantly seasonal ice with relatively free drift and divergent motion (Kottmeier and Sellman, 1996; Drinkwater, 1998; Drinkwater and Liu, 1999). Results of seasonal and regional rms differences show a quite smaller error of OI data compared to not interpolated drift vectors and furthermore illustrate the big impact of different ice regimes on the drift estimations accuracy.

    Maps of seasonally/spatially varying estimates of rms errors are provided in the Ice Drift Altas Datasets section and as additional quality flag included in the datasets for download.