Data Source

The data originate from various sources. The developers and data contributors compiled a comprehensive collection of digital topographic source data. The data used can be grouped into the following three categories:

• cartographic data
• remotely sensed data
• survey data

Cartographic data include contours, spot height points, and surface structure lines digitized from paper topographic map sheets. Remotely sensed data consist of ERS-1 satellite radar altimetry data and airborne radar echo-sounding data. Survey data include ground-based survey data and satellite-based GPS measurements.

The following investigators contributed data:

• Jay Zwally of the NASA Goddard Space Flight Center, USA
• Anita Brenner and John DiMarzio of Raytheon Corporation, USA
• Jonathan Bamber of the Center of Remote Sensing, University of Bristol, UK
• Paul Cooper, David Vaughan, and Phil Homes of the British Antarctic Survey, UK
• Ted Scambos of the National Snow and Ice Data Center, USA
• Craig Lingle of the University of Alaska, USA
• Lee Belbin, Ursula Ryan, and Mike Craven of the Australian Antarctic Division, Australia
• Cheryl Hallam and Jerry Mullins of the USGS, USA
• Johannes Ihde of the Institut fur Angewandte Geodasie, Germany
• Ian Whillans, Paul Berkman, and Terry Wilson of the Ohio State University, USA

Data Format

All DEM data are provided in ARC/INFO and binary grid formats, and the 1 km and 400 m DEMs are also available in ASCII format. Following is a chart that summarizes characteristics of the binary grid.

ASCII grids contain fields for latitude, longitude, elevation relative to the WGS84 ellipsoid, and elevation relative to the OSU91A geoid. Data are represented in decimal degrees, from -180 degrees (west) to 180 degrees (east) longitude; and -90 to -60 (south) degrees latitude.

ARC/INFO coverages of RAMP DEMs are organized into individual directories for each resolution (1 km, 400 m, and 200 m) and geoid/ellipsoid.

Unit of Measurement

Elevations for points in this data set are measured in meters [m] above both the WGS84 ellipsoid and the OSU91A geoid. (See the section titled Data Granularity for a list of files referencing these two models.)

While the WGS84 ellipsoid is based on an approximation of the Earth's shape using only an equatorial radius and a polar radius (or a radius and an eccentricity), the OSU91A geoid is a more complex surface representing mean sea level. The OSU91A geoid is reported as a height above or below the WGS84 ellipsoid. The relationship between the two for the RAMP DEM can be described algebraically as follows:

S = W - G

Where:

W are the WGS84 elevations
G are the OSU91A elevations
S are mean sea level elevations of the OSU91A geoid, relative to the WGS84 ellipsoid

Data Range

Values are in meters for each grid.

Note: Maximum values decrease with increasing grid spacing because a larger region is averaged for each grid cell. Maximum elevation values are found in the Ellsworth Mountains near Vinson Massif. Zero values are at the coast; there are no points in the interior of Antarctica that are at or below sea level (i.e. with a geoid elevation of zero or less).

Data Granularity

A granule of RAMP DEM data (i.e. the smallest aggregation of data that is independently retrievable) includes coverage of the entire continent at a given resolution and geoid/ellipsoid model. Compressed and uncompressed file sizes are summarized below.

Data Manipulaton

Derivation Techniques and Algorithms

Interpolation of Satellite Radar Altimeter Data

The radar altimetry data sets for the RAMP Antarctic DEM had already been corrected for tracking and slope errors and preprocessed into evenly distributed points with a spacing of about 5 km, prior to being combined with the many additional data sets. Refer to NSIDC's Radar Altimeter document for more information, (Davis and Zwally 1993, Zwally et al. 1983, and Brenner et al. 1983.) The RAMP DEM development team used the Triangulated Irregular Network (TIN) Quintic interpolation method to further interpolate the satellite radar altimeter data (Liu, Jezek, and Li 1999).

Interpolation of Traverse Airborne Radar Data

Airborne radar data are densely sampled along flight lines but widely separated between flight transects. Most interpolation algorithms have difficulty resolving such a pattern. The development team for the RAMP Antarctic DEM used a procedure that combines the quadrant neighborhood-based Inverse Distance Weight (IDW) method to stabilize the interpolation result, with the TIN method to retain the topographic details present in the source data (Liu, Jezek, and Li 1999).

Interpolation of Contour-based Cartographic Data

Contour data are characterized by oversampling of information along contour lines and undersampling between contour lines, especially in low relief areas with widely spaced contours. It is the most difficult data type to interpolate with general-purpose interpolation techniques. The development team chose to use the TOPOGRID-based method (Hutchinson 1988; Hutchinson 1989; ESRI 1991; Gesch and Larson 1996) to interpolate the cartographic data in the RAMP Antarctic DEM. The team modified the TOPOGRID method slightly (Liu, Jezek, and Li 1999) to compensate for spurious sinks that occur in contour sparse areas corresponding to low slope areas like glacial valley floors (Bliss and Olsen 1996).

Determination of DEM Grid Spacing

The horizontal grid spacing of DEMs is an important parameter that needs to be specified during interpolation. In general, a small grid spacing is required to obtain an accurate representation of the surface details for a rugged and mountainous terrain, while a large grid spacing is sufficient for a low-relief terrain. For the satellite radar altimeter data and the airborne radar data, a post spacing of 1 km was used. For the contour data, 200 m grid spacing was used for rugged mountainous areas where the contour density is very high, while 400 m grid spacing was used for the sloped coastal area where the contours are relatively smooth and regularly, broadly spaced (Liu, Jezek, and Li 1999).

Data Integration

For mountainous and sloped coastal margins, the development team integrated the contour data, the spot elevation points, coastlines, grounding lines, and limited GPS data during the interpolation process. To avoid the edge effects, all the source data layers are merged into a number of overlapping blocks, and the interpolation extent at each time is set much smaller than that of input data. Individual DEM data sets are merged by using GIS logical "clipping" and "inserting" operations along coastlines and grounding lines, and by using a cubic Hermite blending function (S-shaped) along irregular buffer zones (Liu, Jezek, and Li 1999).

Accuracy

Horizontal (Spatial) Resolution
The real horizontal resolution of the DEM varies from place to place according to the density and scale of the original source data. The developers of the data set estimate that the horizontal resolution of the DEM is about 200 m in the Transantarctic Mountains and Antarctic Peninsula, and about 400 m in the sloped coastal regions. For ice shelves and the inland ice sheet covered by satellite radar altimeter data, the horizontal resolution is about 5 km, but where the airborne radar sounding data were used, the horizontal resolution is about 1 km. For the plateau inside 81.5 degrees south latitude, horizontal resolution is estimated at about 10 km (Liu, Jezek, and Li 1999).

Geolocation Accuracy
The accuracy of geolocation (i.e., the accuracy of the position of a given feature on the DEM) is governed by the accuracy of topographic data sources, and is generally better than the horizontal resolution of the DEM.

Vertical Accuracy
Vertical accuracy of the RAMP Antarctic DEM is ± 100 m over rugged mountainous areas, ± 15 m for steeply sloped coastal regions, ± 1 m on the ice shelves, ± 7.5 m for the gently sloping interior ice sheet, and ± 17.5 m for the relatively rough and steeply sloped portions of the ice sheet perimeter. For latitudes south of 81.5 degrees south, within the interior East Antarctic ice sheet and away from the mountain ranges, vertical accuracy is estimated to be ± 50 m (Liu, Jezek, and Li 1999).

Error Handling

Potential errors in the RAMP DEM include imperfections in the measuring instrument, faulty readings or recordings, calculation and execution faults, and digitizing errors. Errors were noted in the ARC/INFO contour coverages, with mislabeled contours and intersections of contours. In some cases, digitized contour lines deviated from their original position on the source map and often intersected one other, resulting in some positions having two or more conflicting values. Also, poor ground control and inaccurate navigation techniques used to acquire the original topographic data were noted. Some ground control points were assigned erroneously large values, due to data entry errors. The RAMP DEM development team employed a variety of techniques to detect and correct these errors (Liu 1999).

Global Statistical Analysis
Summary statistics were calculated from the ARC/INFO attribute tables where elevation data reside. These global statistics were used to identify extreme erroneous values, namely the elevation range, from prior knowledge about a specific region or from the frequency distribution of elevation measurements. Data points with elevation values outside the reasonable range were flagged, and erroneously large values and negative values for elevation were removed. Contour lines with irregular elevation values were detected and corrected according to the values of neighboring contours.

Cross-Validation
Cross-validation methods were used for multiple data sets that overlap in the same area. Spot height points were checked against corresponding contour coverages by first predicting the elevation values at the positions of spot points by interpolating between contours, then computing the differences between the interpolated values and spot height values. Points that had an absolute difference greater than one contour interval in the flat area and two times greater than the contour interval in highly variable areas were removed. Cross-checking was similarly conducted between contour data and satellite radar alimeter data.

Visual Inspection
Errors in elevation data were detected with a variety of interactive methods including perspective views, color sequence in contour lines, hill shading, and synthetic stereo display. In areas of question, contour lines were overlain with the source data to reveal errors in elevation values. When the DEM grid was rendered as a hill shaded image or synthetic stereo image, errors would appear as anomalous valleys or scars, especially when vertical exaggeration was increased or the illumination angle was adjusted.

Image Simulation
This method integrated a digital synthesis of a satellite image according to a DEM grid, with information about the satellite illumination angle and image geometry. In an area with homogenous land cover like that of Antarctica, a comparison and correlation analysis between the simulated image and real image can often reveal errors in the DEM.

Spatial Autocorrelation
Subtle errors in elevation values were detected using rigorous statistical methods, and by tracing errors back to the original source data after locating areas of spatial discontinuity. In checking the consistency and continuity of each data point relative to nearby points, erroneous data points are flagged as local outliers if they are inconsistent with neighboring points.

Topographic source data were collected by various instruments, teams, and projects between the 1940s and 1990s.