Shuttle Radar Topography Mission


1. Resolution, Projection and Coverage
The SRTM data from JPL comes in geographical degree tiles. All elevations are in meters referenced to the WGS84 EGM96 geoid and the horizontally geo-referenced to the WGS84 ellipsoid using a geographic projection. The naming scheme is tied with the geographic coordinates of the data content. For example, the coordinates of the lower-left corner of tile N40W118.tif are 40 degrees north latitude and 118 degrees west longitude. The degree tiles for the 3 arc-second product contain 1201*1201 pixels, and the tiles for the 1 arc-second product contain 3601*3601 pixels. All GeoTIFF files are 16-bit GeoTIFFs. Any two neighboring tiles share one column/row of common data. In all three versions of data we serve, some tiles have data voids more or less. Those void pixels are of value -32767/-32768 depending on the data version. Therefore for the users, it is important that your work platform supports GeoTIFF files with negative values. The Global Land Cover Facility provides SRTM data at three resolutions:
A. 3 arc second SRTM data of the world from
B. 1 arc second "Unfinished" SRTM data of continental US
C. 30 arc-second "Unfinished" SRTM-GTOPO30 of the world Coverage of Datasets:

30 meter series coverage
90 meter series coverage
1 km series coverage

2. Conversion from degrees tiles to WRS-2 tiles
This WRS-2 conversion for SRTM data is meant to enable the users to use SRTM data seamlessly with the Landsat GeoCover dataset also available from the GLCF website. The GeoCover dataset is also in the WRS-2 convention, however, due to satellite mechanics, the Landsat imagery actually taken may have a shift up to 5 kilometers. Therefore, GLCF SRTM-UTM dataset was generated with a 7.5-km data-buffer around the WRS2 tile.
The GLCF performed the following processing to create the GLCF SRTM layers:
A. Make WRS-2 tiles of 3 arc-second SRTM data that matches the Landsat GeoCover dataset, and re-project the data into UTM coordinates of 90-m resolution using nearest-neighbor resampling, for all three versions.
B. Make WRS-2 tiles of 1 arc-second "Unfinished" SRTM data that matches the Landsat GeoCover dataset and re-project the data into UTM coordinates of 30-m resolution using nearest-neighbor resampling.
C. Mosaic the 30 arc-second SRTM-GTOPO30 data into one global file.

DEM with SRTM Tile Boundaries
DEM with WRS-2 Boundary

3. Filling data voids
The voids in the unfinished SRTM dataset are mostly in three types of topographies: mountain valley, water body, and sand dune. In the finished versions, mountain valleys and sand dunes are greatly improved. The amount of voids is reduced to about 1/20 of before. The water body has been filled and cleaned to the same height. Therefore, in the finished versions, the remaining void is now a minor issue. However, most commercial software tends to load terrain data as it is and treat -32767/-32768 as real data, which causes a visual display problem. And more importantly, some voids are so small that it may escape the examination of human eyes and will cause problem if directly applied into engineering applications. Therefore we applied the triangulation algorithm of ENVI to fill the finished versions. As the triangulation algorithm matches fine with the mountain terrain but not with water bodies, the unfinished versions do not have a void-filled version for fear of introducing errors.

4. Editions of released SRTM data
The first edition was released in early 2004, which carried the original code name 'Unfinished version'. This edition has notable amount of missing data in steep regions, and it also record the exact height of water body surface. The second edition was released in early 2005, with the code name 'Finished version'. This edition handles missing data in steep regions much better, and smoothes out the wavy look on water body surfaces. In late 2005, a third edition of SRTM data was released by JPL for the special needs of researchers in some fields. This edition has no code name and it was processed similar to the 'Finished version' with just one difference. The difference lies in the resampling scheme from 30-m data to 90-m data as we mentioned that each 90-m pixel comes from 9 classified 30-m pixels. In the following illustration, A1-A9 are 9 cells each 30m by 30m and they together form a cell 90m by 90m. In the second version, the data value is the same as the value of A5, but in the third version, the data value is the average of A1 to A9.

A1 A2 A3
A4 A5 A6
A7 A8 A9

At the Global Land Cover Facility we serve all three versions of data because we realize that different research fields might prefer different processing levels. We also realize that the original data from JPL is not very user friendly. It was a combination of HGT and DEM formats in UNIX byte-swapping order. The global data was segmented into each degree block which may be too small for applications. We converted all of them to GeoTIFF format and serve them online in both degree tiles and Landsat tiles (WRS-2 tiles). Moreover, all three editions have holes more or less. We also provide the hole-filled versions ready for download for our users without proper software for that purpose. The following charts listed what JPL released and what we serve now:

JPL Version Void Property 90m Data Property
Unfinished A lot of voids Not Available
Finished_A Few voids Average of 3x3 30m resolution grid

GLCF Version Void Property Void Filling Method Availability of 30m US data 90m Data Property
Unfinished A lot of voids None Yes Not Available
Finished_A Few voids JPL filling No Center pixel of a 3x3 30m resolution grid
Finished_B Few Voids JPL filling No Average of 3x3 30m resolution grid
Finished_B_Filled No Void Mainly JPL filling+ Minor triangulation filling No Average of 3x3 30m resolution grid

Copyright © 1997-2016 University of Maryland. All rights reserved
University of Maryland National Aeronautics and Space Administration GOFC_GOLD