Kristine L. Verdin and Susan K. Jenson
Continental scale digital elevation models (DEMs) are required for climate and global change studies spanning a variety of disciplines: atmospheric science, hydrology, biogeochemistry, wildlife biology, forestry, range science, and others. To date the only available data set with global coverage has been ETOPO5 (National Geophysical Data Center, 1988) which, with its 5 arc- minute (approximately 10 kilometer resolution, has proven inadequate for many applications.
To better address the requirements of the scientific community, the U.S. Geological Survey (USGS) has undertaken to provide DEM data sets for all the earth's land masses at 30 arc-second (approximately 1 km) resolution. USGS scientists are working with the Defense Mapping Agency (DMA) to develop procedures for generalizing 3 arc-second (approximately 90 meter) digital terrain elevation data (DTED 1) for large land masses. For areas where DTED 1 is not available, DMA's Digital Chart of the World (DCW) is being used as a source to generate elevation data, using a gridding algorithm that captures contours, spot heights, and hydrology with drainage enforcement. Techniques were developed to blend the DCW-derived DEM data with the generalized DTED. To date (January 1996), Africa, North America, Europe and Asia are complete and are being made available, free of charge, to the public via internet transfer. A 30 arc-second DEM of Antarctica is being finalized with distribution anticipated for early 1996. Work is progressing on the continents of South America and Australia, with an anticipated completion date of October 1996. In addition to providing the raw elevation data, the DEMs are being processed to extract hydrographic features. The North American DEM was used as a prototype to demonstrate the feasibility of extracting basins and flowlines from the USGS 30 arc-second DEMs. Derived basins and flowlines were verified with independently mapped source material.
The USGS global 30 arc-second DEM project is being carried out at the USGS' EROS Data Center in Sioux Falls, SD. The broad goal of this project is the completion of global one kilometer elevation data for the land surface and the systematic extraction of derivative information to assemble a global database of topographic elevation, slope, aspect, hydrologic flow paths, and watersheds. The primary collaborators and data contributors to date are the USGS, Defense Mapping Agency, NOAA, NASA, UNEP/GRID, USAID, the Instituto Nacional De Estadistica Geografia E Informatica (INEGI) of Mexico, and the Geographical Survey Institute of Japan.
The first data set completed for the project was of the African continent, developed before permission was obtained to incorporate the generalized DTED1 data. This data set is entirely derived from gridded DCW vector data, but work is progressing on incorporating the generalized DTED1 data. The first data set that was developed using the two different data sources was of the North American continent. This data set incorporates generalized 3 arc-second data from various sources, along with gridded DCW data. Techniques were developed with the North American data set to blend the various sources of data to minimize discontinuities between the data types. The European and Asian data sets were developed using the same techniques, as will be the South American DEM. Antarctica is being developed entirely from gridded vector product since DTED1 data do not exist for Antarctica.
The DEMs are being distributed as compressed images, 16-bit binary raster image files in a latitude/longitude coordinate system. The elevation values are given in meters above MSL with ocean areas masked as -9999 (representing no data). Inland water bodies carry representative surface elevations. Vertical accuracies of the elevation values vary with the data source. They range from root mean square error (RMSE) of 18 m in the best case to about 90 m RMSE in the worst case (Gesch, 1994). The DEMs are currently available by ftp transfer from the EROS Data Center anonymous ftp site or tape request. Plans for CD-ROM distribution of the DEMs and ancillary data sets are proceeding. The data sets consist of files of z values, corresponding to geographic subsets of the continental coverage areas, plus 4 ancillary files created with ARC/INFO GIS software (header, world, tic, and statistics files) and the Land Analysis System image processing software (DDR file)along with README documentation files.
The 30 arc-second elevation values were assembled from various data sources. While the African data set is currently a product of the gridded DCW vector product only, the other continental data sets currently available are derived from different sources. Figures 1, 2 and 3 depict the sources used for the North American (Figure 1), European (Figure 2) and Asian (Figure 3) data sets. The sources can be characterized as essentially two types:
The continental data sets are being distributed as they are completed. Coverage of the entire world at 30 arc-second resolution is targeted for completion by the end of 1996. The DEM data sets which have been completed and are currently available to the public are North America (Figure 4) , Africa (Figure 5) , Europe (Figure 6) and Asia (Figure 7) . The data currently are staged and available for free downloading from the EROS Data Center anonymous ftp site.
Drainage analysis software (Jenson and Domingue, 1988; ESRI, 1992) is being employed to derive river basin data sets from the 30 arc-second DEMs. These tools are based upon an algorithm which solves for cell to cell flow directions by finding the path of steepest descent between a cell and its eight neighbors. Thereafter, values for flow accumulation (the number of tributary cells) are computed for each cell, permitting the definition of drainage networks of varying density, and, for any cell of interest, the identification of the area tributary to it.
An abundance of data covering the North American continent can be used to verify the basins derived from the 30 arc-second DEM. Foremost among these data sources is the Hydrologic Unit System developed by the U.S. Geological Survey (Seaber, 1987). This system divides the entire United States into 21 major regions, 18 within the conterminous United States. These regions are further subdivided into 222 subregions including areas drained by a river system, a reach of a river and its tributaries in that reach, closed basins or a group of streams forming a coastal drainage area. Beyond these subregions the areas are broken into successively smaller accounting units and cataloging units. The shapes of accounting and cataloging units are often influenced by planning or administrative boundaries, whereas the region and subregion divisions are, in general, topographically delineated. An 8-digit code, commonly referred to as the Hydrologic Unit Code (HUC), is used to allow each of the four levels of classification to be uniquely identified by 2-digit fields within the 8-digit number.
To create a river basin data set for North America, drainage analysis software was applied to produce basin delineations for comparison with existing USGS vector coverages for the 18 regions (2-digit HUC) and 222 subregions (4- digit HUC) of the conterminous United States. These were compiled and digitized at 1:250,000 scale (Seaber, 1987), and are distributed to the public as a Digital Line Graph (DLG) data set. Upstream areas were derived from the 30 arc-second DEM for the cells corresponding to the mouths of these basins. Results were then overlaid with the vector HUC's for visual inspection and tributary areas calculated for comparison.
Development of major basins from a continental scale DEM presented challenges in the application of the existing drainage analysis algorithms. A DEM of the size of the North American 30 arc-second DEM contains many sinks, locations in the DEM from which there does not exist a natural drainage path. The vast majority of these sinks are spurious, resulting from the methodology used to develop the original DEM or introduced in the generalization algorithm. Before proper flowlines and resulting basins can be extracted from the DEM, these spurious sinks must be removed. The drainage analysis tools developed by Jenson and Domingue (1988) and implemented in ARC/INFO provide the means to remove these spurious sinks by using a filling function. However, due to the complexity of the landscape on a continental scale, not all sinks occuring in the DEM are, in fact, spurious. Many sink features in the DEM are representations of natural surface features. Exclusion of these features by blindly filling all sinks will result in a DEM which will not correctly represent actual topography. The Great Basin of the western United States is a case in point. This large basin, one of the 18 major regions of the conterminous US, is entirely closed; that is, there is no natural outlet to the sea. Blindly applying filling algorithms to the North American DEM would result in the Great Basin being filled and drained to the Pacific Northwest.
The North American 30 arc-second DEM was processed using standard GIS software tools to create a DEM from which major drainage basins can be delineated. Application of the existing tools in the GIS, such as filling, flow accumulation and watershed functions, was done on the entire data set with techniques developed to ensure the preservation of actual sink features. The procedures developed for delineating basins for the North American continent can be summarized as:
Several iterations were carried out with the North American DEM before the thresholding criterion for the sinks was properly identified. The unique situation of the Great Lakes region in the data set also required some iterating to ensure that the flow through the Great Lake system was properly modeled.
Figure 8 presents a comparison of the 2-digit HUCs derived from the 30 arc-second DEM, shown as colored polygons, and the boundaries of the 2-digit HUCs as recorded in the 1:250,000 DLG product, shown as black lines.
Disagreements are seen to be minor relative to the extent of the continental area that has been subdivided. Figure 9 highlights the discrepancy areas as small colored polygons along the divides between the 2-digit HUCs.
Most of these coincide with areas of low relief, and are explained by the inadequacy of a 30 arc-second elevation spacing for resolving the subtle features which control drainage in those areas. An exception to this observation is the Riviere Richelieu, which the DLG product includes as part of the Mid-Atlantic Basin for administrative purposes, though, in fact, it drains into the St. Lawrence River, and is therefore part of the Great Lakes Basin, which drainage analysis of the 30 arc-second DEM has correctly delineated.
Table 1 presents a comparison of the two data sets by listing the respective areas for the 2-digit HUCs as represented by the DLG and DEM-derived data sets. Differences are seen to be typically less than 5% of the area in questions, except for the Lower Mississippi and Mid-Atlantic Basins, already mentioned, and New England.
| Huc # | Region Name | HUC area (sq.mi.) | HUC area (sq.km.) | DEM generated (sq.km.) | Discrepancy (%) | ||||||
| 1 | New England | 64090 | 165990 | 155280 | -6.5% | ||||||
| 2 | Mid-Atlantic | 111360 | 288420 | 255470 | -11.4% | ||||||
| 3 | South Atlantic-Gulf | 278680 | 721780 | 697300 | -3.4% | ||||||
| 4 | Great Lakes | 178300 | 461800 | 482730 | +4.5% | ||||||
| 5 | Ohio | 161250 | 417640 | 411650 | -1.4% | ||||||
| 6 | Tennessee | 40670 | 105340 | 105800 | +0.4% | ||||||
| 7 | Upper Mississippi | 189100 | 489770 | 491940 | +0.4% | ||||||
| 8 | Lower Mississippi | 104030 | 269440 | 255060 | -5.3% | ||||||
| 9 | Souris-Red-Rainy | 60350 | 156310 | 162370 | +3.9% | ||||||
| 10 | Missouri | 509547 | 1319730 | 1308270 | -0.9% | ||||||
| 11 | Arkansas-White-Red | 245500 | 635840 | 650460 | +2.3% | ||||||
| 12 | Texas-Gulf | 183140 | 474330 | 461160 | -2.8% | ||||||
| 13 | Rio Grande | 132510 | 343200 | 344810 | +0.5% | ||||||
| 14 | Upper Colorado | 112110 | 290360 | 293790 | +1.2% | ||||||
| 15 | Lower Colorado | 139130 | 360350 | 372120 | +3.3% | ||||||
| 16 | Great Basin | 140110 | 362880 | 372540 | +2.7% | ||||||
| 17 | Pacific Northwest | 277660 | 719140 | 712930 | -0.9% | ||||||
| 18 | California | 159650 | 413490 | 397400 | -3.9% |
Table 1 Comparison of Basin Areas
Development of prototype basin extraction procedures for 30 arc- second DEMs for the North American case permitted checking results against existing digital depictions of basin boundaries for the U.S., as described, and for Canada. (Most other continents are without these valuable reference data sets.) The experience revealed that it is necessary to work closely with sink areas in the DEMs, in order to discriminate between spurious sinks and natural closed basins. Even so, the results favor the application of the approach to all continents, in order to obtain a consistent and comprehensive global basins data set. Discrepancies with delineations that can be obtained with higher resolution elevation data are not significant at global scales.
When completed, the USGS global basins data set will be a valuable resource for partitioning results obtained by observing and modeling atmospheric and hydrologic processes. This will facilitate the calculation of meaningful continental and regional energy and water balances.
Defense Mapping Agency, 1989, Digital Chart of the World Database (MIL-D-89009), Washington, D.C., U.S. Government Printing Office. Defense Mapping Agency, 1990, Digitizing the future (3d ed.): Defense Mapping Agency, Washington, D.C., 105 p. Defense Mapping Agency, 1992, Development of the Digital Chart of the World: Washington, D.C., U.S. Government Printing Office. ESRI, 1992, "Cell Based Modeling with GRID", ESRI, Inc. Gesch, Dean B., 1994, "Topographic Data Requirements for EOS Global Change Research", U.S. Geological Survey, Open-File Report 94-626, 60 p. Hutchinson, M.F., 1989, A new method for gridding elevation and stream line data with automatic removal of pits: J. Hydrol, 106, 211-232 p. Jenson, S.K., "Applications of Hydrologic Information Automatically Extracted From Digital Elevation Models," Hydrological Processes, 5:1, 31-44 (1991) Jenson, S.K. and Domingue, J.O., 1988, "Extracting topographic structure from digital elevation data for geographic information system analysis: Photogrammetric Engineering and Remote Sensing, v. 54, p. 1,593-1,600. National Geophysical Data Center, 1988, Topography data base - data announcement 88-SE-1102: National Geophysical Data Center, Boulder, Colo. Seaber, Paul R., F. Paul Kapinos, and George L. Knapp, 1987, Hydrologic Unit Maps, United States Geological Survey Water- Supply Paper 2294, 63 p.
Kristine L. Verdin EROS Data Center Sioux Falls, SD 57198 (605) 594-6002
email: kverdin@usgs.gov