DIGITAL ELEVATION DATA AND GIS PROJECTS

ABSTRACT

The U.S. Geological Survey has been a producer and distributor of Digital Elevation Model (DEM) data since the 1970s. These data have been produced by photogrammetric methods and by scanning, tagging, vectorizing, and gridding of hypsographic map separates. Over several years a variety of gridding algorithms have been developed to produce elevation data from x,y,z point data. These algorithms influence the production of the elevation data. It is helpful, for successful GIS projects, if the user has some knowledge about the terrain and elevation data to insure that the ground resolution, horizontal accuracy, and vertical accuracy fit the project criteria. The future production of digital elevation data offers new methods of collection and terrain characterization. This poster session will demonstrate and discuss some of the methodologies and algorithms used in elevation data production, uses of elevation data, and future outlook of digital elevation data production.

METHODOLOGY OF DIGITAL ELEVATION PRODUCTION

Photogrammetric Methods

Digital ground elevation data is predominantly collected photogrammetrically, from stereo aerial photography, stereo satellite imagery, global positioning systems, ground surveying, and from scanned contour data vectorized as x,y,z point data.

One photogrammetric instrument, the Gestalt Photo Mapper 2 (GPM 2), was used by the U.S. Geological Survey (USGS) in the late 1970s and early 1980s to produce orthophotos and a magnetic tape recording of a Digital Elevation Model (DEM). These elevation data are a by-product of the stereo-correlation process. The data was recorded on magnetic tape, patch by patch, in a nominal 182 micron x 182 micron grid. The grid spacing within the patch is a hardware function and is the same regardless of the model scale. It is this patch quilt effect that may be detectable in 1970-1980 DEM data when displayed as shaded relief elevation data on a monitor. Most of the artifacts have been removed during editing prior to entry into the National Digital Cartographic Data Base (NDCDB) for archiving. Figure 1 depicts the USGeoData, Data Users Guides that provide information about Digital Line Graph and DEM data stored in the NDCDB. Quilt artifacts that do remain, generally are less then the required 15 meter maximum NDCDB elevation Root Mean Square Error (RMSE) required for GPM 2 data. However, depending on the application using these data the usual, 1 meter to 10 meter artifacts might cause a problem in the final results. Some of these data were able to meet the minimum 7 meter RMSE when regridded for 30-minute DEM products. For many projects these are not a problem but should be evaluated during project planning.

Another photogrammetric method of elevation production is the collection of elevation data by profiling directly from stereo aerial photography. Figure 2 and Figure 3 both show how this process is accomplished. At the USGS several of the DEMs in the NDCDB were produced using this method. Figure 4 shows what the photogrammetric profiling tracks look like in relation to the 1:24,000-scale topographic map, (1mm=78.74 feet). The aerial photography used was around 1:80,000-scale with a flying height of 40,000 feet. The physical stepover or distance between profiles on the imagery depended on the type of terrain encountered and typically ranged from 2 millimeters to 8 millimeters, but might have gone as high as 16 millimeters in very flat terrain. Figure 5 illustrates the potential for artifacts if the operator is fatigued, inexperienced or the collection rate along the profile is set to a rate that does not allow the operator to keep the floating dot on the model surface as viewed in 3-D stereo. Some of the differences are well within the 7 meter elevation requirements, but when viewed in stereo or if the contour interval is very small a linear pattern is easily seen. This may be important to be aware of when using these elevation data in a hydrologic GIS project.

Collection of specific terrain characteristics, such as ridge line, drain line and breakline data from photogrammetric means is a time consuming, but accurate way to input data into a Triangulated Irregular Network (TIN) method of terrain elevation production. The breakline collection insures that shore lines, cliffs, and man-made surface disturbances will be accounted for in the elevation generation. Figure 6 demonstrates how a portion of a stereomodel was collected using this methodology.

Scanned Vector Data Method

These data, sometimes referred to as, contour-to-grid, is a way to use the contour map information to produce gridded elevation data. The USGS, as well as the U.S. Forest Service, Natural Resources Conservation Service (formerly Soil Conservation Service) and some private companies, such as Infotec Inc., currently are using this method to produce digital elevation data. A film product made from the topographic map contour plate is scanned, usually at 500 or 1000 dots per inch, and the raster data entered into a software system that registers the data to a coordinate system, vectorizes the raster lines and interactively tag the contours, thereby assigning an elevation attribute to the linework. These lines are composed of x,y,z triplets, which will be input to a gridding program that will produce the digital elevation data. Depending on the algorithms used in the gridding process and smoothing ability of certain software products, artifacts can occur during the gridding. An example of a distance weighted algorithm artifact can be seen in the middle left portion of Figure 9. This usually occurs when there is a low density of point data, such as in flat areas where the contours are widely separated.

Other Current Elevation Collection Methods

The Global Positioning Systems (GPS) have become more widely used in collecting positional information about natural and man-made features. The vertical component is roughly half as accurate as the horizontal component, but for many projects would be more than adequate. The GPS hardware have become very reliable and portable allowing collection from a variety of means such as backpack, car, truck, dirt bike, train, etc..

Total ground survey stations are a very precise method for elevation data collection for small area projects or linear projects, such as pipe lines, transmission lines, and road construction. Generation of TIN elevation data from this collection process can yield very good results if significant changes in slope are captured.

Another method of collection is from stereo satellite digital imagery. Systeme Probatoire l'Observation de la Terre (SPOT), level 1A stereoscopic imagery is a source for the production of digital elevation data with greater than 10 meter vertical accuracy.

FACTORS AFFECTING DIGITAL ELEVATION DATA

Photogrammetrically profiled terrain digital data, as with other collection methods is dependent on the characteristics of the terrain. An in-house study at the USGS in 1978 reported that for 1:48,000-scale profiled stereo aerial photography resulted in sigma values of 10-15 feet for flat terrain, 10-25 feet for moderate hills and drains, 15-35 feet for mountain tops, and 15- 40 feet for steep sidehills. It has been observed that after many hours of collection some operators will exhibit fatigue by collecting digital elevation data lower than the terrain along a profile, and tend to float above the terrain along the adjacent return profile. This can be demonstrated from a contour plot, such as shown in Figure 5..

Producing a TIN from vector hypsographic data without breaklines can have an effect upon the formation of the triangles that compose the TIN. Drains do not form sharp patterns in the valley and tend to be smoother. Ridges tend to be slightly lower, unless an algorithm attempts to extrapolate beyond the last contour value at the top of the ridge, which might cause problems in discrepencies at the tops when trying to smooth. Shorelines are not clearly defined and blend with the ground and water. Without overedge data at map neatline edges void data tends to restrict the correct formation of the triangle at the edge. Figure 8 is an example of this problem when editing did not correct the problem. Incorporation of breaklines help to enforce the correct formation of the triangles and will produce very good results if the data being used is a good representation of the actual terrain and the significant changes in slope. Figure 7 demonstrates a TIN with only the use of hypsographic vector data.

Figure 10 illustrates with contours how lack of registration of drainage re-entrants with actual drainage, especially in mild- slope areas occur when only the hypsographic data is used in the production of the elevation data. Some algorithms will generate better surfaces with specific data types and all will generate good terrain surfaces when there is sufficient data that represents the significant changes in slope on the original surface.

SOME USES OF DIGITAL ELEVATION DATA

When a good digital representation of the terrain has been produced to meet the requirements of the project for which it was intended, then the use of these data in a GIS analysis will result in valid conclusions. When you know that the digital elevation model does not meet the needed requirements, but is the only data available then that will have to be conveyed to the customer of the GIS analysis in the final data to account for any discrepancy in the results.

An example of how digital elevation data is used in geophysical and exploration of mineral deposits is shown if Figure 11 and Figure 12 respectively. A point to keep in mind when incorporating elevation data with modeling algorithms is to have an understanding of the algorithm with respect to the elevation data available. If the elevation data used in your project is know to have deficiencies, then this information should be conveyed to the user to insure that the results are not being misinterpreted.

Visual portrayal of digital elevation data of the terrain is becoming a popular method in business, recreation, science, and military activities. As such, it is important to choose the best level of content for which the intended visualization is to be used. In the case of business locations, a coarse graphic representation is sufficient for a particular business theme. Other areas of business might want to portray demographic information on the digital elevation data to better convey the location of the information in 3-D representation. In the recreation area some companies of hiking trail maps want to show the terrain in a shaded relief representation derived from elevation data, and even overlay trails or imagery on the digital surface displayed in 3-D. Figure 13 is an example of digital shading created on a Mylar and overlaid on a 7.5-minute topographic quadrangle visually enhances the shape of the terrain. Many earth scientists are interested in the results of modeling the effects of their particular area of study with the digital elevation data. The military are active users of digital elevation data in tactical evaluations and training involving marine, air, and ground forces.

FUTURE TRENDS IN DIGITAL ELEVATION PRODUCTION

There are methods of digital elevation data collection that show promise for the future. The first is the autocorrelation and generation of elevation data from stereo digital imagery of any surface with identifiable control points. These data are produced from the softcopy photogrammetric workstation, which is currently becoming the next generation of photogrammetric workstation tool. Photogrammetry began with the analog plotter, then moved to analytical workstations, and now replacing the hardcopy stereo pair prints of imagery with the digital softcopy representation of a stereo pair of imagery. The digital elevation data produced using this technology represents a surface of natural and man-made features. For example, the elevation data will be autocorrelated on the tops of buildings, trees, towers, bridges, ground vegetation, etc.. Users of these data may find that by not having only the ground represented is a benefit in modelling activities such as: surface roughness in air flow over terrain, pollution tracking, commercial and military flying, tree heights for various deciduous and evergreen stands, line of sight for transmission signals, etc.. The need to have the elevation represent the ground level only involves editing of these data. Algorithms and methods to do this as an option is currently being worked on by vendors and academia.

Another automated method of elevation data collection for natural and man-made surfaces is from Interferometric Synthetic Aperture Radar (IFSAR). The positive aspect of this collection is the cloud penetration and limited surface penetration characteristics. The sensor can be located in high altitude aircraft and satellite systems allowing collection above storms and other flying aircraft. Choice of systems with specific band frequencies permit a sensor to collect information about surface characteristics such as vegetation, forestry, soil, water, and ice. These digital data when combined with the GPS result in a georeferenced image and elevation data derived from electromagnetic waves generated from the IFSAR active system. The products from this process is a digital radar orthoimage graphic and digital elevation radar data with a wide range of vertical accuracies from centimeters to tens of meters depending on sensors, reflights, processing techniques, and surface movement.

Airborne Laser Mapping is seeking a niche in the elevation market. Linear collection of elevation data using an airborne laser assembly combined with GPS and aircraft inertial reference system produce a three-dimensional positional XYZ digital file with horizontal and vertical components in latitude, longitude, and elevation which can be converted into a specific projection and rectangular coordinate system.

COMMONLY USED ALGORITHMS FOR ELEVATION DATA

1. BI-HARMONIC FILTERING - USED FOR INITIAL ESTIMATE OF SURFACE FROM RAW DATA DERIVED BY SOME INTERPOLATION PROCEDURE. TENDANCY TO HONOR DATA POINTS IN MOST CASES.

A. RESULTS IN SURFACES OF MINIMUM CURVATURE AND TENSION. RESULTS IN VERY SMOOTH APPEARANCE ACROSS DATA AND EXHIBITS A CONTINUOUS TREND AWAY FROM THE DATA.

B. CONTOURS GENERALLY TEND TO HAVE GOOD CONNECTIVITY AND PARALLELISM.

2. LAPLACEAN FILTERS - USED FOR INITIAL ESTIMATE OF SURFACE FROM RAW DATA DERIVED BY SOME INTERPOLATION PROCEDURE.

A. RESULTS IN SURFACES WITH CONSIDERABLE CURVATURE PEAKING AT THE DATA POINTS.

B. CONTOURS GENERALLY TEND TO HAVE MORE CURVATURE.

3. POLYGON OR PROXIMAL - GENERATES GRID VALUES EQUAL TO THE NEAREST POINT VALUE.

A. RESULTS IN FLAT POLYGONAL AREAS SURROUNDING THE DATA POINTS.

B. A GOOD METHOD OF SURFACE DETERMINATION FOR MINING ACTIVITIES.

4. DISTANCE WEIGHTING - ASSIGNS MORE WEIGHT TO NEARBY POINTS THAN TO DISTANT POINTS.

A. DEPENDING ON THE SPATIAL LOCATION OF RAW DATA AMBIGUITY CAN RESULT DUE TO CHOICE OF WEIGHTING FUNCTION.

B. ARTIFACTS CAN OCCUR WHEN USED WITH AN UNEVEN DISTRIBUTION OF DATA POINTS. THIS CAN BE SEEN WHEN APPLIED IN AREAS OF FLAT TERRAIN WITH A SPARSE DISTRIBUTION OF DATA POINTS.

5. SPLINES - SOME POLYNOMIAL OF DEGREE M; LINEAR FOR M=1, QUDRATIC FOR M=2, CUBIC WHEN M=3 AND CONSIDERED BICUBIC WHEN USED IN THE THREE-DIMENSIONAL CASE.

A. SPLINE FUNCTIONS IN SPATIAL INTERPOLATION ARE FORMED BY PIECING TOGETHER SUCCESSIVE CURVE SEGMENTS INVOLVING RELATIVELY FEW POINTS AT A TIME AND SHOULD BE CLOSELY RELATED TO THE VALUE BEING INTERPOLATED; THEY ARE ANALYTIC; AND ARE FLEXIBLE. WHEN APPLIED TO A SPLINE SURFACE, SUCCESSIVE RECTANGULAR SURFACE PATCHES FORM A COMPOSITE SURFACE SIMILAR TO A PATCHWORK QUILT.

B. ADVANTAGEOUS FOR DENSE OR RECTANGULAR DATA, BUT CAN INTRODUCE ANOMALIES NOT IN THE ORIGINAL SURFACE DEPENDING ON THE INTERPOLATION AND BLENDING METHODS USED.

6. UNIVERSAL KRIGING - A STATISTICAL APPROACH TO A SURFACE OF IRREGULAR POINTS BY DETERMINING A SEMIVARIOGRAM, WHICH IS A FUNCTION RELATING THE COVARIANCE OF THE DIFFERENCE BETWEEN POINTS TO THE DISTANCES BETWEEN THE POINTS.

A. EFFECTIVE USE DEPENDS UPON THE PROPER SELECTION OF THE SLOPE OF THE SEMIVARIOGRAM, DEGREE OF POLYNOMIAL DRIFT, AND THE VARIANCE.

B. INTENSIVE COMPUTATION, BUT CAN PRODUCE AN ACCURATE GRID. A DIFFERENT SET OF EQUATIONS ARE USED FOR EACH POINT ESTIMATE IN DIFFERENT NEIGHBORHOODS.

7. TRIANGULATED IRREGULAR NETWORK (TIN) - THIS METHOD, IS NOT, AN INTERPOLATED REGULAR GRIDDED SURFACE, BUT RATHER A METHOD OF GENERATING A SURFACE FROM A SERIES OF X,Y,Z POINTS RESULTING IN A SERIES OF TOPOLOGICALLY STRUCTURED TRIANGLES WHICH CONNECT THOSE EXACT DATA POINTS TO THEIR NEIGHBORS.

A. A GOOD METHOD TO HONOR DATA COLLECTED ALONG RIDGES, DRAINS, AND SIGNIFICANT BREAKLINES OF CHANGING SLOPE ON THE TERRAIN SURFACE.

B. THE INITIAL SURFACE GENERATED WILL APPEAR PEAKED AT THE DATA POINTS AND HAVE ANGULAR SHARP CORNERS.

C. LOCATION OF THE DATA POINTS ON THE TERRAIN SHOULD BE ASSESSED BEFORE ANY APPLICATION OF SMOOTHING IS USED.

D. TOPOLOGY OF TRIANGLES ALLOWS FOR A VARIETY OF APPLICATIONS, SUCH AS SLOPE, SHADING, ASPECT, AND CONTOURS.

8. CONVERGENT GRIDDING - A PROCESS WHEREBY GRID NODE VALUES ARE CONVERGED-UPON THROUGH ONE OR MORE ITERATIONS OF SNAPPING CONTROL POINTS TO NEARBY NODES. THIS IS DONE USING A DISTANCE-WEIGHTING TECHNIQUE, SUCH THAT CONTROL POINTS CLOSER TO THE NODE HAVE A LARGER AFFECT ON THE OUTCOME OF THE NODE Z-VALUE. A BLENDING FUNCTION INSURES NON-COLLISIONS BETWEEN WEIGHTED Z- VALUES AND AVERAGES THE CONTROL POINT Z-VALUE USED.

A. A VERY GOOD ALGORITHM FOR HONORING HYPSOGRAPHIC DATA SUPPLEMENTED WITH HYDROGRAPHIC OR OTHER VECTOR DATA FOR FEATURE DEFINITION ENFORCEMENT, SUCH AS WATERLINE, DRAINS, OR GEOLOGIC FAULTS.

B. PROCESS IS VERY COMPUTATIONAL INTENSIVE DEPENDING ON CHOICE OF SIZE OF MOVING WINDOW AREA.

9. ITERATIVE FINITE DIFFERENCE INTERPOLATION - A DISCRETISED VERSION OF THE THIN PLATE SPLINE TECHNIQUE FOR WHICH THE ROUGHNESS PENALTY (DEFINED IN TERMS OF FIRST AND SECOND ORDER DERIVATIVES OF THE FITTED GRID) IS USUALLY A LINEAR CURVATURE OF THE FITTED SURFACE. THE ITERATION TECHNIQUE EMPLOYS A SIMPLE MULTI-GRID STRATEGY WHICH CALCULATES GRIDS AT SUCCESSIVELY FINER RESOLUTIONS, STARTING WITH A COARSE INITIAL GRID AND SUCCESSIVELY HALVING THE GRID SPACING UNTIL THE FINAL USER SPECIFIED GRID RESOLUTION IS OBTAINED.

A. A VERY GOOD ALGORITHM FOR HYDROLOGIC MODELING FROM CONTOUR DATA.

B. RIDGE AND DRAIN INFORMATION IS AUTOMATICALLY CALCULATED FROM THE CONTOUR DATA.

C. STREAM LINE DATA CAN BE USED WITH THE DRAINAGE ENFORCEMENT ALGORITHM TO INSURE A MORE ACCURATE PLACEMENT OF STREAMS.

Selected Bibliography

Abdel, S.M. 1982) Study of Factors Affecting Accuracy of Digital Terrain Models. M.Sc. Thesis, Cairo University, Egypt

Avecedo, W. (1991) First Assessment of U.S. Geological Survey 30- Minute DEMs: A Great Improvement over Existing 1-Degree Data. Proceedings of the 1991 ACSM/ASPRS Annual Conference, Baltimore, Maryland, 2:1-12.

Balce, A. (1986) Determination of Optimum Sampling Interval in Grid Digital Elevation Models Data Acquisition. Proceedings ISPRS Commission III Symposium, Finland, Int. Archieves of Photogrammetry and Remote Sensing, Vol. 26, Part 3.1, pp. 40- 55.

Braile, L. (1978) Comparison of four random to grid methods. Computer & Geoscience, Vol.4:341-349.

Douglas, D.H., and Peucker, T.K. (1973) Algorithms for the reduction of the number of points required to represent a digitised line or its caricature. Canadian Cartographer, 10(2):112-122.

Garbrecht, J., and Starks, P. (1995) Note on the Use of USGS Level 1 7.5-minute DEM Coverages for Landscape Drainage Analyses. Photogrammetric Engineering & Remote Sensing, 61(5):519-522.

Gruen, A.W., and Baltsavias, E.P. (1987) High Precision Image Matching for Digital Terrain Model Generation. Photogrammetria, 42:97-112.

Hayes, J.G., and Halliday, J. (1974) The least squares fitting of cubic spline surfaces to general data sets. Journal of the Institute of Mathematics and its Applications, 14(1):89- 103.

Hunter, G.J, and Goodchild, M.F. (1995) Dealing with Error in Spatial Databases: A Simple Case Study. Photogrammetric Engineering & Remote Sensing, 61(5):529-537.

Hutchinson, M.F. (1991) The application of thin plate smoothing splines to continent-wide data assimilation. In: J.D. Jasper (ed), Data Assimilation Systems, BMRC Research Report No. 27, Melbourne: Bureau of Meteorology, pp.104-113.

Jenson, S.K. (1991) Applications of Hydrologic Information Automatically Extracted from Digital Elevation Models. Hydrological Processes, 5:31-44.

Jenson, S.K., and Domingue, J.O. (1988) Extracting Topographic Structure from Digital Elevation Data for Geographic Information System Analysis. Photogrammetric Engineering & Remote Sensing, 54(11):1593-1600.

Lam, N.S. (1983) Spatial interpolation methods: a review. American Cartographer, 10(2):129-149.

Lue, Y., and Novak, K. (1991) Simultaneous Extraction of Digital Elevation Models and Orthophotos. Technical Papers, ACSM- ASPRS Annual Convention, Vol.5:412-418.

Mirante, A., and Weingarten, N. (1982) The radial sweep algorithm for constructing triangulated irregular networks. IEEE Computer Graphics and Applications, 2(3):11-21.

Norvelle, F.R. (1992) Window shaping and DEM corrections. Photogrammetric Engineering and Remote Sensing, 58(1):111- 115.

O'Callaghn, J.F., and Mark, D.M. (1984) The Extraction of Drainage Networks form Digital Elevation Data. Computer Vision, Graphics, and Image Processing, 28:323-344.

Schurt, G.M. (1976) Review of Interpolation Methods for Digital Terrain Models. The Canadian Surveyor, 30(5):389- 411.

Sibson, R. (1981) A brief description of natural neighbour interpolation. Interpreting Multivariate Data, (Editor V. Barnett). John Wiley and Sons, Chichester, pp. 21-36.

U.S. Geological Survey (1990) Digital Elevation Models Data Users Guide, Reston, Virginia, 51 p.

Yoeli, P. (1977) Computer executed interpolation of contours into arrays of randomly distributed height points. Cartographic Journal, 14(2):103-108.

Yoeli, P. (1986) Computer executed production of a regular grid of height points form digital contours. American Cartographer, 13(3):219-229.