7.B.i. Absolute Accuracy: Data from Raster Sources

Digital Terrain Elevation Data (DTED): The full resolution 3" DTED have a vertical accuracy of ± 30 meters (two-sigma) linear error at the 90 percent confidence level (Defense Mapping Agency, 1986), where they meet specifications. If the error distribution is assumed to be Gaussian with a mean of zero, the statistical standard deviation of the errors is equivalent to the root mean square error (RMSE). Under these assumptions, vertical accuracy expressed as ± 30 meters linear error at 90 percent can also be described as a RMSE of 18 meters.

The areas of GLOBE derived from DTED might be interpreted to retain (or slightly improve upon) the level of vertical accuracy found in the source 3" data. This is because the representative value computed during compositing to 30" is either a simple value (as in nearest-neighbor or spot sampling) at the same vertical accuracy as the source data, or is a computation (as a mean or median) that should reduce some of the random error in a sample of source 3" values.

A visual inspection of DMA's (1995) map, Available DTED1 Cells by Accuracy Range, suggests that about half of DTED coverage meets specifications, while the other half fails to meet horizontal, vertical, or both horizontal and vertical specifications. About 25% of DTED has 30-50 meter vertical accuracy and 0-50 meter horizontal accuracy; about 15% has 0-50 meter vertical and 50-131 meter horizontal accuracy. Ten percent has either (1) 50-9999 meter vertical and 0-131 meter horizontal accuracy, or (2) 0-9999 meter vertical and 131-9999 meter horizontal accuracy. Specific accuracy information for each 1^o x 1^o DTED cell is contained in DTED Level 0 header information on NIMA's Web site.

Shaded-relief images produced from DTED sources show two fairly common shading patterns suggestive of vertical errors:

1. Striping apparently caused by stereoprofiling techniques or use of satellite scanner imagery. The amplitude of these errors is usually small, in the 0-20 meter range. These appear to be within NIMA’s tolerances for DTED.
2. Blocking around the edges of individual 1^o x 1^o DTED Level 1 tiles (DTED cells). These appear to be edge effects due to subtle de-facto disagreements about (most likely) vertical or (possibly) horizontal datum between DTED tiles. These also appear to be well within vertical accuracy specifications set by NIMA for DTED for their respective DTED tiles.

Note that NIMA allows for vertical errors as great as 9999 meters, for some areas. This appears to be an extremely conservative figure.

Other Raster Data Sources: Data for Italy, Japan, and New Zealand were supplied as raster data. However, their documentation all notes that they were derived from cartographic sources. These are discussed below. Zwally (and others)/NSIDC/JPL data for Greenland were derived from satellite altimetry. Assessment of vertical accuracy of this data set is not yet complete.