Annual mosaics were created using data collected during each Greenland winter. Spatial gaps exist either because no data were available or because interferometric correlation was insufficient to produce an estimate in those regions. The latter is most often the case in regions with high snow accumulation. In regions with sufficient data, the averages are based on up to three measurements. Typically, more data are available in regions of swath overlap, especially at higher latitudes.
The data are gridded at 0.5 km resolution, but the true resolution varies between 0.5 and 1 km. Many small glaciers are resolved outside the main ice sheet; however, for glaciers that are narrower than 1 km, the velocity represents an average of both moving ice and stationary rock. So, while a narrow glacier may be visible on the map, its speed is likely underestimated. Furthermore, interpolation produces artifacts where the interpolated value is derived from nearby rock, causing apparent stationary regions in the middle of otherwise active flow. In such instances, the data should be interpreted with care.
Processing Steps
The following sections briefly describe how the mosaics were generated for each winter.
2000/2001
In late 2000 and early 2001, during the RADARSAT-1 Modified Antarctic Mapping Mission, CSA acquired nearly complete coverage of Greenland with multiple passes suitable for InSAR (03 September 2000 to 24 January 2001). All of the available data for Greenland were used to produce the 2000/2001 mosaic. In cases where the data quality was too poor, some products were discarded. All source data were obtained from the Alaska Satellite Facility (ASF).
2005/2006
In 2005 and 2006, RADARSAT-1 imaged most of Greenland on four consecutive missions, producing three InSAR pairs. Once all of the data were processed, poor coherence passes were screened out and the remaining data were used to assemble the 2005/2006 mosaic.
2006/2007
The 2006/2007 mosaic was produced with RADARSAT-1 fine-beam data. Coverage is substantially improved by including ascending JAXA ALOS quad-pol data, including coverage in the southeast of Greenland. The ionospheric errors are often large (>20 m/year) in the ALOS data; therefore, points were manually removed where errors were excessive. This approach was chosen in order to balance maximal coverage with minimal error. Nevertheless, these data should be interpreted with care, particularly in the southeast region.
2007/2008
The 2007/2008 mosaic was produced with RADARSAT-1 fine-beam data in the same manner as the 2006/2007 mosaic, including the use of a substantial volume of ALOS fine-beam data, largely along the northwest coast.
2008/2009
The 2008/2009 mosaic utilizes data from CSA's RADARSAT-1, the DLR's TSX, and JAXA's ALOS satellites.
2009/2010
The 2009/2010 mosaic consists almost entirely of ALOS SAR data collected in Fine-Beam, Single-Polarization (FBS) mode. Because L-band is more subject to ionospheric distortion of speckle-tracked azimuth offsets, streak errors for some areas are large (>10 m/year), often exceeding the magnitude of the accompanying error estimates. In other areas, these errors are barely perceptible. Some of the worst streaks were edited out. However, a number of lesser streaks were left in place to preserve coverage and illustrate the magnitude of these errors with obvious examples. Despite being more susceptible to ionospheric distortion, L-band data correlate well in areas with high accumulation. As a result, this map has better coverage in the southeast than many of the maps from other winters.
Twenty coastal sites in this mosaic utilize 30 km x 50 km TSX scenes. These X-band data greatly improve the results for many of the fast-moving outlet glaciers.
2012/2013
The data for the 2012/2013 mosaic were collected from January 2013 to March 2013, which corresponds to the last months during which RADARSAT-1 was active. These data were combined with TSX winter data from November 2012 to March 2013.
General Information for 2014 to 2018
The 2014 to 2017 mosaics were produced mostly with ESA's Copernicus Sentinel-1A/1B data and supplemented by DLR's TSX/TDX data for coastal outlets. The data for the 2014 to 2017 mosaics were acquired in either 12-day (through 16 September) or 6-day repeat cycles (16 October forward). In cases of missing acquisitions, the repeat periods may be longer (i.e., integer multiples of 6 or 12 days) for some of the image pairs.
Unlike earlier SAR acquisitions, Sentinel-1A/1B provides crossing ascending and descending orbit data over much of the ice sheet. In areas where data from crossing orbits were available, an error-weight range-offset-only solution was included in the velocity product, eliminating azimuth offsets and reducing the error from ionospheric streaking in the azimuth offsets.
To take advantage of the year-round Sentinel coastal coverage, data are collected during Greenland winter periods with little or no melt. This definition might produce small seasonal differences compared to mosaics from other years in which narrower acquisition periods were used. However, such differences are generally small relative to inter-annual variability and to the noise reduction accomplished by averaging a greater volume of data acquired over a longer time period.
Due to the reduction in the resolution of Sentinel-1A/1B SAR data, some systematic differences between the mosaics produced by RADARSAT, ALOS, and TSX/TDX data may exist, especially in regions with sharp gradients or strong curvatures. Smoothing earlier velocity results to approximately 1.5 km resolution (i.e., to roughly the resolution of Sentinel-1) should improve agreement among data sets. In producing the mosaics, higher resolution TSX/TDX data are given more weight, hence the loss in resolution should be smaller in these areas. These mosaics are posted at both 0.2 km and 0.5 km spacing. For work requiring a finer resolution, see Version 1 of the MEaSUREs Greenland Ice Velocity: Selected Glacier Site Velocity Maps from InSAR data set.
As a result of the large volume of data used, the overall quality of the data is good. Compared to earlier products, the coverage in the southeast is generally improved, particularly for 2016/17; however, high accumulation in the southeast reduces image-to-image correlation, resulting in higher noise. Additionally, there may be coherent displacement signals in these regions that are not associated with horizontal ice motion. If such displacement occurs with characteristics other than those assumed in the solution (e.g., predominantly vertical instead of horizontal displacement), then the result will be incorrectly mapped to horizontal motion, contributing to the overall level of noise.
2014/2015
The 2014/2015 mosaic was largely produced from TSX data, with the addition of Sentinel-1A data. As Sentinel-1 data acquisition began during this winter period, there are almost no Sentinel-1 data prior to January 2015, with the exception of the region around the Jakobshavn glacier. As a result of the limited satellite coverage, the 2014/2015 mosaic contains more noise and less spatial coverage than the 2015/2016 and 2016/2017 mosaics.
2015/2016
The 2015/2016 mosaic was largely produced from Sentinel-1A data. The six tracks that covered nearly the entire coast were collected at almost every 12-day interval. In the interior, typically four images (i.e., three pairs) were collected with better coverage and fewer errors than the 2014/2015 mosaic.
2016/2017
The 2016/2017 mosaic was largely produced from Sentinel-1A/1B data. In October 2016, Sentinel-1B started acquiring data over Greenland in an orbit that lags behind Sentinel-1A by six days. As a result, Sentinel-1A/1B pairs are often separated by only six days, providing better correlation and coverage, particularly in the southeast of Greenland. Thus, the mosaic for this winter provides almost complete spatial coverage relative to all prior winter velocity products.
2017/2018
The 2017/2018 mosaic was largely produced from 6-day repeat cycles from Sentinel-1A/1B. In general, more data were collected in the ice sheet interior during this winter mapping campaign. As a result, this mosaic should have a lower level of noise in the data relative to prior winter velocity products. The 6-day sampling also provides better coverage because the image-to-image correlation is improved with shorter time intervals.