7.C. Additional Accuracy Assessments by Peer Reviewers
University College London: J.-P. Muller of UCL performed accuracy assessments of GTOPO30 and the JPL/MISR DEMs, as well as the AUSLIG/NGDC DEM (Muller and others, 1998a, 1998b, 1998c). This was done by compiling a large number of point elevation values, such as from the commercial Jeppesen airport data base, and various geophysical data bases from NGDC, such as the elevations of gravity observation stations. He found that the JPL/MISR DEM was slightly more accurate than GTOPO30, with Australia and Greenland particularly better. However, the AUSLIG/NGDC DEM was preferred over the other two available compilations in Australia.
One outcome of this study was the concern over the general quality control of the point elevation data used in the comparisons. One airport elevation appeared to be several thousand meters off and other values seemed unreasonable. Muller noted that "what constitutes the control" and "what constitutes the data set being tested against the control" was unclear. Muller used statistical techniques to reject point data that disagreed with the DEMs by more than 300m, based on the estimated "worst case" RMSE shown in Table 2 (Section 7.B.ii). This served as an attempt to remove point elevation values with significant unknown errors that might have crept into point data bases without quality control.
J.-P. Muller (1997; Muller and others, 1998b) used a global compilation of ERS-1 (Earth Resources Satellite) radar altimetry data at 5' (Bamber and others, 1997; Bamber and Muller, 1999) to compare with other DEMs. The MISR DEM was slightly higher quality than GTOPO30, and the AUSLIG/NGDC DEM was best for Australia by a factor of 2 compared to the other options.
National Oceanic and Atmospheric Administration: Dru Smith of NOAA’s National Ocean Service performed comparisons between full-resolution DTED Level 1 data, other DEMs and geodetic models, and DTED Level 0 means, maxima, minima, and discrete (spot) values. He found that the DTED Level 0 mean values tended to be considerably lower than corresponding derivations made directly from DTED Level 1 data. They also were lower than other DEMs and geodetic models available to him. He was concerned that DTED Level 0 might have had a production bug. However, he found the DTED Level 0 discrete values to be reasonable.
De Montfort University: P.A.M. Berry of De Montfort University (United Kingdom) compared ERS-1 and ERS-2 altimetry with DTED Level 0 means and discrete values. The satellite altimetry should have a longer spatial wavelength view of topography than DTED Level 1, though perhaps shorter wavelength than DTED Level 0. Initial comparisons with DTED Level 0 means also noted a significant discrepancy with ERS-1 and ERS-2 altimetry-derived elevations. DTED Level 0 means seemed significantly lower than ERS-derived elevations produced by Berry. Later comparisons between ERS-derived elevations and DTED Level 0 discrete values in the central United States suggested much better overall agreement. Berry plans further assessments, including the development of a map showing differences in the geographic distribution of GLOBE and her ERS-derived elevations.
Geographical Survey Institute: Hiroshi Murakami of the Geographical Survey Institute of Japan compared DEMs of Japan produced by GSI with others produced by USGS from DCW. This study suggested that DCW for Japan did not use Mean Sea Level as vertical datum, but rather had used the Japanese national datum. In short, it appeared that DCW had been mislabeled as to vertical datum for Japan. This mislabeling could have been caused by inaccurate or inaccurately interpreted source materials documnetation for the Operational Navigation Charts upon which DCW was based.
Use of These Studies in GLOBE: The studies by Muller helped in the selection of candidates for GLOBE. The independent studies by Smith and by Berry helped NGDC select the DTED Level 0 discrete values as preferable for GLOBE, where available between 50o North and South latitude. These values might also have been selected where available poleward of 50o latitude. However, their lower horizontal resolution made the GTOPO30 versions preferable at such higher latitudes, as noted in Section 5.B.i.h. The study by Murakami suggests modest caution when using elevations derived from DCW.