For detailed information on the various satellites and sensors used to produce this data set, refer to the DigitalGlobe, Inc. Web site. Satellite and sensor details are also provided in Shean et al. (2016).

All DEMs were generated using the NASA Ames Stereo Pipeline (ASP) (v2.5.3 Build ID: 0be5052 to v2.6.0 Build ID: 4521300) and methodology described in Shean et al. (2016).. The data were processed using the NASA High-Performance Computing (HPC) resources, primarily the NASA Pleiades Supercomputer at the NASA Ames Research Center.

The DEMs in these data sets were created from two sets of images. One set was generated from along-track (also referred to as in-track) stereoscopic imagery in which both images were collected minutes apart along the same orbital pass. The other set was generated from cross-track imagery collected at different times on different orbits, and potentially from different spacecraft.

All subscenes for each 11-bit panchromatic image were mosaicked, and new rational polynomial coefficients (RPCs) were derived for each image strip. The mosaicked images were orthorectified using the 1-arcsec (30 m) SRTM-GL1, the void-filled Global Shuttle Radar Topography Mission (SRTM) DEM. Orthorectification using a low-resolution DEM removes most of the feature offsets due to terrain and viewing geometry, providing a significant performance improvement during stereo correlation.

The ASP stereo utility was run for the input orthoimages with the following parameters:

• Integer correlation kernel size of 21x21 pixels
• Parabolic sub-pixel refinement (using the ASP sub-pixel-mode 1 utility) with kernel size 11x11 pixels
• Conservative filtering to remove isolated "islands" up to 1024 pixels
• The ASP point2dem utility was used to produce gridded DEMs from the triangulated point cloud.

Along-track Stereo Strips

These are generated from in-track/along-track stereo pairs, involving images acquired on the same orbit, typically 60-90 seconds apart.

Cross-track Stereo Strips

Cross-track (also referred to as coincident mono) DEMs were generated from pairs of independent monoscopic images with geometry suitable for stereo reconstruction. Candidate pairs in the DigitalGlobe image archive were identified based on the criteria in Table 4. Some of these cross-track pairs were acquired on the same orbit, while others were acquired on different orbits or by different spacecraft (e.g., WorldView-1 and WorldView-2).

The cross-track DEMs potentially have increased error due to residual horizontal displacement errors (i.e., errors due to ice flow between image acquisitions), non-ideal stereo geometry (e.g., smaller convergence angles) and the fact that some errors in ephemeris data for the two images are independent (as opposed to highly correlated errors for along-track pairs). In practice, these issues can result in increased DEM vertical/horizontal bias and increased relative error (e.g., DEM “tilt”). Despite potentially increased error, we include these cross-track DEMs to fill critical gaps in along-track stereo coverage.

Processing Steps

In summary, two images are collected for the same ground extent from different angles. Cross-correlation is used to measure offsets between features in the images, which are related to the different viewing geometry and topography. These offsets are combined with spacecraft attitude/ephemeris data to identify precise sensor locations, and stereo triangulation is used to determine the three dimensional coordinates of matched features. The resulting point cloud is interpolated to create the high-resolution DEM.

See Shean et al. (2016) for details of the full processing workflow. For information regarding ASP, refer to the official documentation (PDF, 21 MB).