Upgrade of the 9 Second Australian Digital Elevation Model

A Joint Project of CRES and AUSLIG

Michael Hutchinson, John Stein and Janet Stein
CRES, ANU

Contents

1. Introduction
2. Sample images
3. Project information
4. Purchasing the DEM
5. Brief description of the upgraded DEM
6. A brief history of the GEODATA 9 second DEM
7. The revised source data
8. The ANUDEM gridding algorithm
9. Accuracy estimates
10. References

1. Introduction

The land surface is a fundamental control on earth surface, and near earth surface atmospheric, processes. So strong is this linkage that an understanding of the nature of terrain can directly confer an understanding of the nature of these processes, in both subjective and analytical terms. Thus digital elevation models (DEMs) have been used widely over the last twenty years to support assessment and analysis of climate, hydrology, agriculture, forestry and biodiversity.

The ANUDEM digital elevation modelling program has also been developed over the last twenty years, with applications to the generation of hydrologically sound DEMs from local to regional and continental scales. ANUDEM was first applied in the late 1980s to a DEM of the Australian continent at the relatively coarse resolution of 1.5 minutes (approximately 2.5 km). This DEM supported much of the natural resource assessment of the Australian continent during the 1990s.

The new 9 second (250 m) national digital elevation model is a very significant upgrade over the first 9 second DEM produced by AUSLIG, AGSO and CRES in 1996. The new DEM is the result of a two year collaboration between CRES and AUSLIG, during which a major upgrade of the ANUDEM program was combined with extensive revision and augmentation of the source data. The new DEM accurately represents, for the first time at this scale, drainage structure and high points of the entire Australian continent. The DEM is distributed by AUSLIG.

2. Sample images

Click on any of these thumbnails to see a larger image.

	Canberra, ACT. Looking soutwest over Black Mountain in the foreground towards the Brindabella Ranges
	Northern Flinders Ranges, SA. From lake Frome looking west
	Prince Frederick Harbour, Kimberley Region, WA, looking southeast
	The Grampians, Victoria, looking southwest from above Stawell
	The Grampians, Victoria, looking west from above Stawell
	The Blue Mountains, New South Wales, looking northeast into the Grose Valley
	Coastal Range, south of Cairns, QLD, with Bellenden Ker in the background
	Tasmania, looking southwest towards Cradle Mountain
	Hobart, Tasmania, looking south towards the mouth of the Derwent River with Mt Wellington in the background
Additional images	Many more images, including elevations maps and hill-shaded relief maps for the entire continent, are available here.

3. Project Information

A full description of the DEM and the proceedures used to produce it are given in the user guide available through the AUSLIG DEM website.

For further information relating to the research that underlies this project,
please contact Dr Michael Hutchinson or visit his personal website

4. Purchasing the 9 second DEM

The 9 second DEM is available commercially for the whole of Australia or for specific 'tiles' through the AUSLIG DEM Website.

5. Brief Description of the Upgraded 9 Second DEM

The 9 Second DEM is a grid of elevation points covering the whole of Australia with a grid spacing of 9 seconds in longitude and latitude (approximately 250 metres) in the GDA94 coordinate system. The data are held in tiles that generally have extents of 6 degrees in longitude by 4 degrees in latitude.

The upgraded DEM incorporates comprehensive revision of all source point and stream line data used for Version 1, additional point, sink point and cliff line data, and a comprehensive upgrade of the elevation gridding procedure, as implemented in ANUDEM Version 5.0.

The revisions and additions to the source data were completed by the Centre for Resource and Environmental Studies (CRES) of the Australian National University in collaboration with the Australian Surveying and Land Information Group (AUSLIG).

The revised version of the ANUDEM elevation gridding procedure was completed by Dr Michael Hutchinson of the Centre for Resource and Environmental Studies (CRES) of the Australian National University.

The model has been improved by including the national trigonometric data points in the source data and by revising the ANUDEM gridding procedure to model high points more closely. Abrupt changes in landform have also been modelled by incorporating cliff line data in selected areas.

The major source data for the product were revised national spot height elevation data taken from 1:100 000 scale topographic mapping and revised river information from 1:250 000 scale topographic mapping. These data are components of AUSLIG's GEODATA TOPO-250K digital map product (AUSLIG 1994). All revisions to the source data were made by CRES. These data were augmented by national trigonometric data supplied by AUSLIG from the National Geodetic Data Base (NGDB). Additional spot height, sink point, stream line and cliff line data were digitised by CRES.

The ANUDEM algorithm used to grid the point elevation data utilises a drainage enforcement algorithm that improves the elevation model by removing spurious sinks (Hutchinson 1989, 2000). ANUDEM Version 5.0 incorporates a substantially revised drainage enforcement algorithm and permits the modelling of cliff lines. By digitising additional sink point data, the drainage enforcement algorithm was applied to the whole continent. This has yielded an upgraded DEM with a substantially improved representation of surface drainage structure, especially in the large areas of the continent with low relief.

The source elevation data have a standard deviation of around 7.5 metres. Errors in the DEM are closely related to terrain complexity. Theoretical estimates and tests of the DEM against trigonometric data distributed evenly across the continent indicate that the standard elevation error of the DEM varies between about 7.5 metres and 20 metres for most of the continent. Errors are larger in highland areas with steep and complex terrain where the largest errors can exceed 200 metres.

The density of source data points used to create the grid and the horizontal resolution of the grid warrant that the final grid be considered as having a scale of no coarser than 1:250 000. This makes the DEM useful for national, statewide and regional applications.

6. A brief history of the Geodata 9 Second DEM

Between the mid-1960s and 1988 AUSLIG captured selected points from 20 metre contour data of 1:100 000 scale maps over the whole of Australia (Manning et al. 1989). The ANUDEM elevation gridding program was also initially developed in the mid-1980s, with applications to such data a primary objective (Hutchinson 1989). It was first used to generate a national DEM, at the relatively coarse resolution of 1.5 minutes, from coarser scale national topographic data by Hutchinson and Dowling (1991).

An early version of the ANUDEM program was also applied to the digitised 1:100 000 scale point elevation data for use in telecommunication applications. AUSLIG also applied an early version of the ANUDEM program to the 1:100 000 scale point elevation data to produce an 18 second DEM for about one third of Australia. Production of the 18 second DEM was only done on user demand and was not of adequate resolution for many applications.

The GEODATA 9 SECOND DEM Version 1 project grew out of agreements between AUSLIG, AGSO, the Australian Heritage Commission (AHC) and CRES. AUSLIG and AGSO had agreed to undertake cooperative programs in geosciences, geodesy and geographic information and had developed preliminary specifications for a national DEM in early 1994. The AHC and CRES had also developed similar specifications for a national DEM in early 1994. It was agreed that the AHC and CRES would join with AUSLIG and AGSO in the production of the DEM. It was also agreed that Version 4.4 of the ANUDEM program (Hutchinson 2000) would be used.

This joint effort commenced in June 1994. CRES/AHC worked on Victoria and most of Queensland as well as the Kimberley and south-west Western Australia regions that were a high priority in their Wild Rivers Project. AGSO concentrated on Tasmania, New South Wales and South Australia, with AUSLIG picking up the remainder of the continent. Version 1 of the 9 SECOND DEM was released in July 1996, as described by Carroll and Morse (1996).

Further examination of the DEM and its supporting source data by CRES in their Wild Rivers Project for the AHC (Stein et al. 1998) made it clear that there were significant deficiencies in the 9 SECOND DEM Version 1. These related particularly to remaining errors in the directions of stream line data, as well as disparities in the spatial density of the same data. There were also sporadic large errors remaining in the point elevation data. All of these deficiencies led to significant errors in the representation of terrain shape and surface drainage structure. These included large anomalies in terrain slope later found by Kirby and Featherstone (1999) in making terrain based corrections to gravity data.

It also became clear that Version 4.4 of the ANUDEM elevation gridding program could be upgraded to improve representation of peaks, to overcome problems caused by disparities in density of stream line data and to improve overall accuracy and connectivity of surface drainage. CRES therefore began independent revision of the ANUDEM program, as well as systematic revision of the stream line and point elevation data.

In view of the wide potential applicability of an accurate continental DEM, CRES approached AUSLIG in 1997 with a proposal to carry out a comprehensive revision of all source data and to produce an upgrade of the DEM for general distribution. The proposal was accepted, with AUSLIG agreeing to fund the source data revision and DEM upgrade work by CRES, but with CRES continuing to support the revision of the ANUDEM elevation gridding program. Final checking on data quality was to be completed by AUSLIG in consultation with CRES.

During the course of the project, the improved version of the ANUDEM program revealed a new generation of significant source data errors that had hitherto been undetected. The errors were in both stream line data and point elevation data. All detected source data errors have been corrected in this project. CRES corrected 25 000 errors in point elevation data and 9000 errors in stream line data. CRES also digitised an additional 87 000 spot heights and an additional 11 000 stream lines.

The improvements to the drainage enforcement algorithm within ANUDEM also made it feasible for the first time to apply the drainage enforcement algorithm to the low relief areas with ill-defined drainage structure that cover nearly half of the continent. This eventually resulted in CRES digitising 21 000 sink data points to assist the definition of drainage structure in these areas.

These unscheduled additions to the work significantly extended the duration of the project, but have led to a DEM of much improved quality over Version 1 of the 9 SECOND DEM. Other unscheduled additions included the incorporation of trigonometric data points from the National Geodetic Data Base to improve the representation of high points in the DEM. These data also required significant editing and revision before they could be used. Radar altimeter data for Lake Eyre were also added to the source data to replace existing GEODATA-250K point data for Lake Eyre.

Late in the project it was recognised that cliffs inland from the Nullabor coast could only be modelled satisfactorily by implementing a new cliff data type in ANUDEM. This was successfully designed and implemented in July 2000. The success with which ANUDEM then modelled the Nullabor cliffs led to the incorporation of additional cliff line data for the Kimberley coast and for the Blue Mountains in New South Wales. Time precluded inclusion of all significant cliff lines around the continent, although such a task would be desirable in a future upgrade.

Despite the best efforts of CRES and AUSLIG it is apparent that some source data errors are likely to remain and that additional data could be digitised to better define terrain shape and surface drainage structure in some areas. Nevertheless, Version 2 of the DEM embodies a very significant improvement over Version 1. It has already been used successfully by separate projects for the National Land and Water Resources Audit to model sediment transport across the continent and to define a nested series of sub-catchments catchments across the continent for systematic reporting of land and water resources (Hutchinson et al. 2000).

7. The revised source data

The source data sets used to create the AUSLIG GEODATA 9 SECOND DEM Version 2 are listed below. A full description of the revised source data is given in the User Guide for the 9 Second DEM published by AUSLIG.

The first three sets of source data listed below came from the GEODATA TOPO-250K digital product from AUSLIG. They have been comprehensively revised by CRES. More information on the GEODATA TOPO-250K data and its origins can be found in the GEODATA TOPO-250K User Guide for this product (AUSLIG, 1994).

It should be noted that lakes and reservoirs from the waterbody layer of GEODATA TOPO-250K Hydrography theme were not used for Version 2 of the DEM. The aim was to provide a representation of the ground surface that maximised hydrological connectivity. For applications that need representations of waterbody surfaces, these may be added from the GEODATA TOPO-250K Hydrography theme using standard GIS techniques.

1. Revised spot heights from GEODATA TOPO-250K Relief theme. A total of 6000 data points were corrected and 19 000 erroneous points were deleted.

2. Revised linear watercourse features from the Drainage layer of GEODATA TOPO-250K. A total of 6000 stream lines were corrected and 3000 streamlines were deleted.

3. Revised coastline of Australia from GEODATA COAST-100K data and coastal inlets from the GEODATA TOPO-250K Framework layer;

4. Trigonometric data points supplied by AUSLIG from the National Geodetic Data Base and revised and converted to the GDA94 coordinate system. A total of 19 0000 corrected points were used from this data base.

5. A total of 300 radar altimeter point elevation data supplied by AUSLIG for Lake Eyre.

6. An additional 87 000 spot heights were digitised from digital 1:100 000 scale mapping.

7. An additional 11 000 stream line data were digitised from digital 1:100 000 scale mapping;

8. An additional 21 000 sink point data were digitised from digital 1:100 000 scale mapping;

9. Cliff line data and associated contour line data for selected areas digitised from digital 1:100 000 scale mapping.

8. The ANUDEM gridding algorithm

The ANUDEM program has been designed to produce accurate digital elevation models with sensible drainage properties from point elevations, stream lines, contour lines and cliff lines (Hutchinson 2000). It was first applied to the generation of a national DEM, at the relatively coarse grid resolution of 1.5 minutes of latitude and longitude, by Hutchinson and Dowling (1991). The algorithm implemented by the program interpolates the elevation data onto a regular grid by minimising a suitably weak roughness penalty on the fitted grid values and by simultaneously imposing constraints that:

1. Ensure connected drainage structure by imposing a global drainage condition on the fitted grid values that automatically removes spurious sinks or pits and by calculating drainage constraints directly from input stream line data (Hutchinson 1989). These actions make up one of the principal innovations of the program. They eliminate one of the main weaknesses of elevation grids produced by general purpose interpolation techniques that has limited their usefulness in hydrologic applications, particularly those that rely on the automatic calculation of catchment areas.

2. Ensure proper representation of ridges and streams as deduced automatically from input contour line data. This is achieved by inserting ridge and stream lines deduced from corners of contour lines that indicate where these lines cross the elevation contours, as described in Hutchinson (1988).

3. Smooth point elevation data according to the natural discretisation error associated with the incorporation of point data onto a regular grid (Hutchinson 1996).

The imposed global drainage condition has been found in practice to be a powerful condition that can significantly increase the accuracy, especially in terms of their drainage properties, of digital elevation models interpolated from sparse sets of surface specific data (Hutchinson 1989). The size of such data sets can be at least an order of magnitude smaller than the number of points normally required to adequately describe elevation using digitised contours. This can minimise the expense of obtaining reliable digital elevation models in terms of the capture, correction and storage of primary elevation data. The global drainage condition also virtually eliminates the need for detailed manual editing of interpolated elevation grids to remove spurious drainage features.

The revised version of ANUDEM included a number of significant enhancements for this project as follows:

• The drainage enforcement algorithm was revised to remove deficiencies associated with dense stream line networks.

• The drainage enforcement algorithm was revised to improve performance in low relief areas with ill-defined surface drainage structure.

• The algorithm was revised to provide additional diagnostics to detect errors in point elevation data and stream line data.

• Data smoothing was reduced to allow the DEM to more closely match elevation peaks.

• A new cliff line facility was added to permit breaking of surface continuity across digitised cliff lines.

9. Accuracy estimates

The elevation error at a single point in a DEM, the discretisation error, depends on the resolution (cell size) of the DEM and the roughness of the surface that is being modelled (Hutchinson 1996). The most important aspect of this roughness is the slope. Hence the discretisation error is greater in mountainous terrain.

Tests using various cell sizes indicated that the cell size of 9 seconds is the optimum across Australia for the available source data. Theoretical error estimates based on the root mean square slope and the maximum slope of the DEM for selected 1:250 000 map sheets are given below. These estimates ignore non-linear variation of actual elevations within each grid cell and are therefore somewhat optimistic in steep and complex terrain.

The elevation source data for the DEM has a standard elevation error of around 7.5 metres (AUSLIG, 1994). The shape-based standard elevation error is a measure of the discrepancy between the final grid and the source elevation points. Both are combined to calculate the overall standard elevation error estimates in the table.

The root mean squared (RMS) slopes hide variation within the actual sheet so the maximum slope within each sheet is also quoted. For Tallangatta, which includes Mt. Kosciuszko, the largest slopes can be over 100% giving local shape-based standard elevation error of about 70 metres and a maximum elevation error of about 200 metres. These estimates have been verified by comparison with trigonometric data from the National Geodetic Data Base.

It is important to note that for many applications, accurate representation of terrain shape is more important than absolute elevation accuracy. Measures of accuracy of overall shape and drainage structure, as principally determined by slope and aspect, are more difficult to quantify than standard elevation errors. This is because measures of slope and aspect are scale or resolution dependent and also because independent measures of these quantities are not generally available.

A simple independent assessment of shape accuracy is the maximum slope of the DEM across each 1:250 000 scale map sheet. Slopes in real landscapes rarely exceed 50 degrees (equivalent to 120%) except in the steepest terrain associated with mountains and cliffs. Slopes of Version 2 of the DEM exceed 50 degrees in only four 1:250 000 map sheets, two with mountainous terrain and two where cliff lines have been incorporated in the source data. This is a significant improvement on Version 1 of the DEM, where slopes exceeded 50 degrees in coastal ranges with dense streamline data.

An independent assessment of the accuracy of the drainage structure represented by Version 2 of the DEM is the close agreement between the boundaries of the national river basins (AUSLIG 1997) and the catchment boundaries calculated by Hutchinson et al. (2000) from the flow direction grid associated with the DEM. The DEM and these derived catchments have been used to successfully support continent-wide sediment transport modelling and systematic reporting of analyses for the National Land and Water Resources Audit.

10. References

AUSLIG (1994). GEODATA TOPO-250K Data User Guide, Version 1 Data, Ed 2. Australian Surveying & Land Information Group, Commonwealth Department of Administrative Services, Canberra, Australia.

AUSLIG (1997). Australia's River Basins, Version 1.0. September 1997. Australian Surveying & Land Information Group, Commonwealth Department of Administrative Services, Canberra, Australia.

Carroll, D. and Morse, M.P. (1996). A national digital elevation model for resource and environmental management. Cartography 25: 395-405.

Hutchinson, M.F. (1988). Calculation of hydrologically sound digital elevation models. Proceedings of the Third International Symposium on Spatial Data Handling, August 17-19, Sydney. International Geographical Union, Columbus, Ohio, 117-133.

Hutchinson, M.F. (1989). A new method for gridding elevation and stream line data with automatic removal of pits. Journal of Hydrology 106: 211-232.

Hutchinson, M.F. (1996). A locally adaptive approach to the interpolation of digital elevation models. In: NCGIA (ed.), Proceedings of the Third International Conference Integrating GIS and Environmental Modeling, Santa Fe, New Mexico, 21-25 January, 1996. University of California, Santa Barbara, National Center for Geographic Information and Analysis: CD-ROM and http://www.ncgia.ucsb.edu/conf/SANTA_FE_CD-ROM/main.html

Hutchinson, M.F. (2000). ANUDEM Software. Centre for Resource and Environmental Studies, Australian National University, Canberra. http://cres.anu.edu.au/outputs/software.php

Hutchinson, M.F. and Dowling, T.I. (1991). A continental hydrological assessment of a new grid-based digital elevation model of Australia. Hydrological Processes 5: 45-58.

Hutchinson, M.F. and Gallant, J.C. (2000). Digital elevation models and representation of terrain shape. In: J.P. Wilson and J.C. Gallant (eds), Terrain Analysis. John Wiley & Sons, New York, 29-50.

Hutchinson, M.F., Stein, J.L. and Stein, J.A. (2000). Derivation of nested catchments and sub-catchments for the Australian continent. Centre for Resource and Environmental Studies, Australian National University, Canberra.

Kirby, J.F. and Featherstone, W.E. (1999). Terrain correcting Australian gravity observations using the national digital elevation model and the fast Fourier transform. Australian Journal of Earth Sciences 46: 555-562.

Manning, J. and Menzies, R.W., (1988). Vertical Control for Australian Topographic Mapping. Australian Surveying Conference Proceedings, Sydney, 1988.

Stein, J.L., Stein, J.A. and Nix, H.A. (1998). The identification of Wild Rivers. Methodology and Database Development. Environment Australia, Canberra, Australia, 73 pp.