The mosaic includes data from:

• CSA's RADARSAT 1 and 2 data from 2000/2001, 2005/2006, 2006/2007, 2007/2008, and 2008/2009, 2012/2013.
• JAXA's ALOS PALSAR data from 2007/2008, 2008/2009, 2009/2010, 2010/2011.
• DLR's TerraSAR-X (TSX)/TanDEM-X (TDX) data from 2008-2015.
• ESA's ERS-1/2 Tandem data from 1995/1996.
• USGS's Landsat 8 optical imagery data from 2014/2015.

The resulting mosaic represents an error weighted average of all the available source data at each point. See Figure 2 for examples of three of the data sources (Joughin, 2002). This averaging process means that at any given point, the resulting velocity estimate could be derived from a single image pair or more, and may represent an average of as many as dozens of pairs. As such, it is not a uniformly averaged velocity for the 20-year period. For slow moving regions in the interior where velocities are slowly changing, the mosaic should provide a reasonable approximation of velocity for the period. By contrast, for fast changing glaciers at the coast, the velocity may have fluctuated wildly over the 20-year period. For time series analysis of a basin, use the full suite of MEaSUREs products from the TerraSAR-X (NSIDC-0481) and Landsat velocity products (NSIDC-0646) for the most rapidly changing areas, the annual mosaics (NSIDC-0478) for the more slowly changing regions farther inland (Joughin et al., 2010), and this mosaic for the relatively constant interior.

Much of the source data can be accessed individually through other MEaSUREs products: RADARSAT and ALOS (NSIDC-0478), LANDSAT (NSIDC-0646), and TerraSAR-X (NSIDC-0481).  Due to limited data and success working with the data (i.e., poor interferometric correlation) in the high-accumulation Southeast region, some 1995/1996 ERS-1 and 2 data were used to obtain full coverage. Used in conjunction with the other data, these products improve the overall result. They are not of sufficient quality, however, to produce stand-alone products.

For much of the interior of the ice sheet, the crossing ascending (ALOS) and descending (RADARSAT-1) image pairs allowed velocity estimates entirely from the interferometric phase (Joughin et al., 1998). Because the errors for phase are much lower than for speckle-tracking, where they exist (Fig. 1), phase estimates are weighted the most heavily in the mosaicking process.

There are gaps in coverage, particularly in the southernmost regions of the ice sheet where coverage was often dropped in favor of other radar modes optimized for mapping sea ice in shipping lanes. As a result, the mosaic includes Landsat 8 data. The errors for Landsat generally are greater than for individual SAR estimates (speckle or phase), so the Landsat data are weighted less heavily than radar results.

Although the data are posted at 250 meters, the true resolution varies, depending on the data source. In most cases, the true resolution is closer to 500 meters. As a result, results near sharp velocity transitions may be biased. For example, velocities along the centerline of a narrow (< 1 km) glacier may read low because some nearby rock could be included when the resolution is 500-m.

Error estimates are provided for all velocity estimates in separate GeoTIFF files appended with _ex.tif and _ey.tif.  In general, the errors maps represent the average behavior of the data, and errors could be much less than reported in some areas and much greater in others and care should be taken when assigning statistical significance based on the errors, especially given that the errors can be correlated over large areas. In the interior of the ice sheet where phase estimates are used, the reported errors are small (< 1 m/year). Comparison with control points suggest the errors are approaching but are not quite at this level, within a factor of 2; although with many cases, the errors agree closely with the estimates when slope effects are included.

The assumption of surface parallel flow means that there can be errors of up to about 3% due to errors in the slope data used to calculate velocity (Joughin, 2002b. These errors are not included in the formal error estimates. As a result, this error should be added in quadrature to the error map (i.e., Ex_total = sqrt(Ex_map)^2 + (0.03Vx^2).

Users should keep in mind that although the data have been screened carefully, errors well outside the formal estimates may exist, particularly in the Southeast where data quality is the poorest.

Interpolated Points

Small holes in the final maps have been filled via interpolation. These points can be identified as those which have valid velocity data but no corresponding error estimate. See Joughin, et al, 2002 for more detail on errors and how they were computed.