Advanced Search  |  

3 Second SRTM Derived Digital Elevation Model (DEM) version 1.0

Note: This metadata describes the dataset in accordance with the ANZLIC (Australia New Zealand Land Information Council) Core Metadata Guidelines Version 2.

Dataset citation

ANZLIC unique identifier: ANZCW0703014182

Title: 3 Second SRTM Derived Digital Elevation Model (DEM) version 1.0


Custodian: Geoscience Australia

Jurisdiction: Australia



The 3 second Digital Elevation Model (DEM)Version 1.0 was derived from resampling the 1 second SRTM derived DEM-S (gridded smoothed digital elevation model). The DEM represents ground surface topography, excluding vegetation features, and has been smoothed to reduce noise and improve the representation of surface shape. The DEM-S was derived from the 1 second Digital Elevation Model Version 1.0 (DEM; ANZCW0703013355) by an adaptive smoothing process that applies more smoothing in flatter areas than hilly areas, and more smoothing in noisier areas than in less noisy areas. This DEM-S supports calculation of local terrain shape attributes such as slope, aspect and curvatures that could not be reliably derived from the unsmoothed 1 second DEM because of noise. A full description of the methods is in progress (Gallant et al., in prep) and in the 1 second User Guide. The three second DEM was produced for use by government and the public under Creative Commons attribution.

ANZLIC search words:

  • CLIMATE AND WEATHER Climate change
  • ECOLOGY Landscape
  • HAZARDS Flood
  • LAND Topography

Spatial domain:

locality map

Geographic extent name: AUSTRALIA EXCLUDING EXTERNAL TERRITORIES - AUS - Australia - Australia

Geographic extent polygon: 113 -10, 154 -10, 154 -44, 113 -44, 113 -10,

Note: The format for each Geographic extent name is: Name - Identifier - Category - Jurisdiction (as appropriate) See GEN Register

Geographic bounding box:
North bounding latitude: -10 °
South bounding latitude: -44 °
East bounding longitude: 154 °
West bounding longitude: 113 °

Data currency

Beginning date: 2000-02-11

Ending date: 2000-02-22

Dataset status

Progress: In Progress

Maintenance and update frequency: Not Known


Stored data format:
DIGITAL - ArcGIS-grid ArcInfo grid Geographic WGS84
Available format type:
DIGITAL - ArcGIS-grid ArcInfo grid

Access constraints:

This material is released under the Creative Commons Attribution 3.0 Australia Licence.

Further information on Creative Commons can be found on the website

Free Data Download

Data quality


Source data

1. SRTM 1 second Level 2 data (Slater et al., 2006), supplied by Defence Imagery and Geospatial Organisation (DIGO) as 813 1 x 1 degree tiles. Data was produced by NASA from radar data collected by the Shuttle Radar Topographic Mission in February 2000.

2. GEODATA 9 second DEM Version 3 (Geoscience Australia, 2008) used to fill voids.

3. SRTM Water Body Data (SWBD) shapefile accompanying the SRTM data (Slater et al., 2006). This defines the coastline and larger inland waterbodies for the SRTM DEM and DSM.

4. Vegetation masks and water masks applied to the DEM to remove vegetation.

5. SRTM 1 second derived Digital Elevation Model (DEM) Version 1.0 ? used to resample the data.

6. Adaptive smoothing applied to DEM to produce 1 second DEM-S.

In order to understand the 3 second DEM the processing of the parent dataset, the 1 second DEM-S is described below.

1 second DEM processing (vegetation offset removal)

Vegetation offsets were identified using Landsat-based mapping of woody vegetation. The height offsets were estimated around the edges of vegetation patches then interpolated to a continuous surface of vegetation height offset that was subtracted from the DSM to produce a bare-earth DEM. Further details are provided in the DSM metadata (ANZCW0703013355).

Adaptive smoothing

The adaptive smoothing process was designed to smooth flat areas to a greater degree than steep areas, and to respond to the degree of noise so that very noisy flat areas are smoothed more than less noisy flat areas. The process operated over multiple resolutions, allowing smoothing over quite large distances in areas of very low relief. The smoothing was performed on overlapping tiles, with sufficient overlap that cells used in the final product were not impacted by edge effects.

The smoothing process was based on the amount of noise in the 1 second DEM. The noise was estimated from the local variation in the difference between elevation and the mean of nearby elevations.

In essence, the smoothing process operated by comparing the variance of elevations in a 3x3 group of cells with the mean noise variance in the group. If the elevation variance was larger than the mean noise it was considered to be due to real topographic variation and the elevations were left unchanged, while if it was smaller it was considered to be due to noise and the elevations were replaced by the mean elevation in the group. This was applied at successively coarser resolutions, producing smoothing over large areas where the topographic variation was small compared to the noise levels. The algorithm actually used statistical tests to make the decisions, and combined the multiple estimates of elevation at different resolutions using variance weighting.

Water bodies

Water bodies defined from the SRTM Water Body Data as part of the DSM processing were set to the same elevations as in the DSM after the smoothing.

The water bodies are also removed from the DEM (set to null) before the smoothing operation to prevent them affecting the land elevations unduly. One cell of water adjacent to land is retained to prevent shoreline elevations from being raised to match the higher elevations further from the shore.

Resampling to 3 seconds

The 1 second SRTM derived smoothed Digital Elevation Model (DEM-S) was resampled to 3 seconds of arc (90m) in ArcGIS software using bilinear interpolation.

Further information on the processing is provided in the 1 second and three second User Guides (Geoscience Australia and CSIRO Land & Water, 2010).

Positional accuracy:

The horizontal positional error is estimated to be three times that of the 1 second product which is around 21m. See Rodriguez et al. (2006) for more information on SRTM accuracy.

Attribute accuracy:

Elevation accuracy is essentially the same as for the raw SRTM 1 second data, with 90% of tested heights within 9.8 m for Australia, with improvements due to the removal of vegetation offsets. Relative elevation accuracy between adjacent cells is improved in DEM-S due to the reduction in noise levels; this has not been quantified but is evident in the comparison of slopes calculated before and after smoothing as shown in the User Guide (Geoscience Australia and CSIRO Land & Water, 2010).

Accuracy of the 1 second DEM (before smoothing to form DEM-S) was tested using 1198 Permanent Survey Marks distributed across the Australian Continent relative to the Australian Height Datum (AHD71). Results of this comparison show the absolute accuracy of the data as tested relative to AHD71 to be 7.582m at the 95th percentile with a RMS error of 3.868 in open, flat terrain. 99 percent of points are within a height difference of less than 9.602m.

The removal of striping artefacts in the DEM improves the representation of the landform shape, particularly in low relief areas, but it is not clear whether this also produces an improvement in overall height accuracy. Some striping remains in the data at a much reduced level (mostly less than 0.3 m amplitude). Additional artefacts including long-wavelength (~10 km) striping have not been corrected.

The removal of vegetation offsets in the DEM provides a significant improvement in the representation of the landform shape, particularly in low relief areas, and areas of remnant vegetation. Elevation accuracy varies in forested areas. Comparisons with several higher resolution datasets suggest that elevation accuracy varies depending on the height and structure of the existing vegetation, quality of vegetation input masks and local relief. Further details of these comparisons are provided in the User Guide (Geoscience Australia and CSIRO Land & Water, 2010).

Height accuracy is likely to be poorer in areas where voids have been filled using the 9 second DEM, particularly in high relief areas.

Logical Consistency:

The 1 second DEM-S represents ground elevation with greatly improved relative elevations between adjacent grid cells in low relief areas due to the smoothing process. Slopes as small as 0.02% (2 m in 10 km) can be resolved in this DEM-S.

The removal of vegetation involves estimation of vegetation height at the edges of vegetation patches, and interpolation of those heights across areas of continuous vegetation cover. Variations in vegetation height within large areas of vegetation are not captured by this method. The vegetation removal process guarantees that no elevations have been increased as part of the process.

All void areas have been filled and there are no discontinuities due to original tile boundaries.

The SRTM editing rules relating to water bodies have been respected in the processing: lakes are flat, rivers decline continuously in a downstream direction and sea surfaces are at 0 m elevation. Flattened water bodies occupy the same areas as in the original SRTM 1 second data. Grid cells adjacent to water bodies are at least 1 cm above the water surface. Void areas within water bodies (small islands not represented in the original SRTM data) are at least 1 cm above the water surface over their entire area.


The DEM covers all of continental Australia and near coastal islands land areas including all islands defined by the available SRTM 1 second elevation and SRTM Water Body Data datasets. Some fragments of mainland or pieces of islands may be missing.

Contact information

Contact organisation: Commonwealth of Australia (Geoscience Australia) (GA)
Contact position: Manager Client Services
Mail address: Cnr Jerrabomberra Ave and Hindmarsh Dr
Mail address: GPO Box 378
Locality: Canberra
State: ACT
Country: Australia
Postcode: 2601
Telephone: 02 6249 9966
Facsimile: 02 6249 9960
Electronic mail address:

Metadata information

Metadata date: 2013-03-08

Additional metadata

Metadata reference XHTML:

Metadata reference XML:

Conversion to floating point format

The smoothing process alters all data values in the 1 second DEM by varying amounts and the result is a floating point data set capturing in some places very small but meaningful differences in elevation between adjacent cells.

Ancillary data layers

- A water mask at 1 second resolution showing the cells that are part of the flattened water bodies.

- JPEG Image of the 3 second DEM.


Gallant, in (in prep) An adaptive smoothing method for improving noisy DEMs.

Geoscience Australia (2008) GEODATA 9 Second DEM Version 3

Geoscience Australia and CSIRO Land & Water (2010) 1 Second SRTM-Derived Digital Elevation Model (DSM, DEM, DEM-S and DEM-H) User Guide. Version 1.0. Geoscience Australia.

Rodriguez, E., Morris, C.S., and Belz, J.E. (2006) A global assessment of the SRTM performance. Photogrammetric Engineering and Remote Sensing 72 (3), 249-260.

Slater, J.A., Garvey, G., Johnston, C., Haase, J., Heady, B., Kroenung, G., and Little, J. (2006) The SRTM data "finishing" process and products. Photogrammetric Engineering and Remote Sensing 72 (3), 237-247.

For technical queries please contact:

c/o Geoscience Australia

or GA Sales on 02 6249 9966

Authors:Wilson, N. Tickle, P. Gallant, J. Dowling, T. Read, A.