Mapping the Ferris State University Katke Golf Course.

Two of the freely available national Digital Elevation Model (DEM) data sets--the National Elevation Dataset (NED) and the Shuttle Radar Topographic Mission (SRTM)--are important for many engineering and mapping applications. Compared to surfaces derived from Real-Time Kinematic (RTK) GPS surveying, these DEMs have various qualities and different limitations. At Michigan's Ferris State University, a team of students and professors performed a study to demonstrate the use of these freely available data for orthophotography creation.

Accuracy comparison of the NED and SRTM data, with respect to the GPS measured points.

Accuracy Requirements for Elevations

Digital elevation models are created by different organizations and agencies for a myriad of diverse applications. Some applications can meet needs with low resolution DEMs while others require denser elevation sampling. Two common standards for describing the accuracy of spatial data are the National Map Accuracy Standards (NMAS) and the ASPRS (American Society for Photogrammetry and Remote Sensing) Standard for Large-scale Mapping. Both standards define the vertical accuracy in terms of the contour interval of the map. The NMAS defines the following two criteria to test the vertical accuracy of a topographic map:

1. Vertical accuracy, as applied to contour maps on all publication scales, shall be such that not more than 10 percent of the elevations tested shall be in error by more than one-half the contour interval.

2. The accuracy of any map may be tested by comparing the positions of points whose locations or elevations are shown upon it with corresponding positions as determined by surveys of a higher accuracy.

A comparison of the slope parameter of the different surfaces.

The NMAS standard limits the number of points with large error to 10 percent. The ASPRS standard uses the Root Mean Square Error (RMSE) statistic to evaluate the accuracy of the spatial data. This RMSE expresses both random and systematic errors of the data. Using this definition, a Class 1 map should have a vertical RMSE of 1/3 the contour interval for well-defined points and 1/6 the contour interval for spot elevations. According to the ASPRS standard, maps compiled within limiting RMSE errors of two or three times those allowed for Class 1 maps are designated as Class 2 or Class 3, respectively.

example 1

Quality Control Evaluation Using GPS

To test the two free data sets, our team at Ferris State University set out with a Leica Geosystems (Norcross, Ga.) GX1230 GPS receiver at the Katke Golf Course in Big Rapids, Mich. We chose the site for its mildly variable topography in an open area with no buildings or trees.

Even though USGS standards for DEM require only 20 check points with at least eight scattered around the edge, this investigation compared the DEM surface and a reference surface of higher accuracy created from ground observations. The observations were collected in RTK mode using the Big Rapids Continuously Operating Reference Station (CORS). This CORS is located one mile away from the test site, which minimized the error due to ionospheric delay and geoid model inaccuracy. The specified accuracy of the GPS instruments we used is 10 mm + 1 ppm in horizontal measurements and 20 mm + 1 ppm in vertical measurements. With the addition of centering error and errors from the geoid interpolation, we assumed an error of no more than 4 cm.

The interpolated surface...

More than 500 points were measured in the test area and the RMSE was calculated with respect to the NED and SRTM data. These GPS observations were interpolated using linear interpolation known as a TIN (Triangular Irregular Network).

We also evaluated the accuracy of the slope in different surface cells to obtain an estimate of the reliability of the DEM in representing the terrain. The slope at a given position is defined as the inclination of the surface at this position. It measures the maximum rate of change in surface values over distance. These results show that the NED data represents the surface better than the SRTM. Despite the fact that the SRTM elevations have a better accuracy in this specific open-area test site, here we found it was not as precise in representing the terrain slope. This is probably attributed to smoothing of the USGS elevation data. The absolute accuracy of the SRTM data is expected to be worse in areas where the first return signal is not the ground, as is the case in forested and developed areas.

...and the set of GPS points.

Orthophotographic Creation

An orthophotograph is a geometrically corrected photograph created from either aerial or satellite imagery. The most expensive part of producing an orthophoto is generally the creation of the DEM. Therefore, many organizations investigate the use of an existing DEM to reduce their costs. This practice is an acceptable procedure provided that the density of the DEM is sufficient to meet the accuracy needs of the ortho product and that no significant changes in the topographic relief have occurred between the time the DEM was created and when the ortho imagery was acquired.

This one (DEM) to many (images) relationship speeds up the orthophotograph creation process by maximizing the use of available data. In the orthorectification process, every DEM point (Z) is linked with a pixel (x,y) in the digital image using the exterior orientation parameters (i.e., exposure station coordinates and rotational parameters of the sensor) through the colinearity condition as follows (see example 1 above):

RMSE of the orthophotograph as a function of the radial distance, circles are drawn at 200 meters interval.

Where x,y are the image coordinates, X, Y, Z are the ground coordinates of the DEM postings, XL, YL, ZLare the exposure station coordinates, f is the focal length of the sensor, and m11,"¦, m33are the elements of the rotation matrix constructed from the rotation angles (ϖ, φ, κ) of the sensor. The error in calculating the ground coordinates in planimetry is directly related to the error in the elevation (Z).

The image pixels may also be transformed to a coordinate framework parallel to the ground coordinate system. This transformation can be performed using a nearest neighbor selection, bilinear interpolation or a cubic convolution process. A bilinear interpolation was used in this study for the rectification to reduce the stair-step effect caused by the raster cells, thereby creating a smoother image.

The accuracy of the NED and SRTM created orthophotographs evaluated against the GPS points.

We created two orthophotographs using Leica Photogrammetry Suite from 1:10,000-scale photography taken at a flight height of 1,582 meters above the average terrain and scanned at ground resolution of 0.15 meters. One orthophoto was created with the NED and the other with the SRTM data. The initial NED and SRTM DEMs (downloaded from the USGS website) were projected from their native geographic coordinates to the Michigan State Plane Coordinate system to create a 35 x 35 meter resolution DEM.

The majority of the test points (60%) were located in the northeast quadrant. Results of the comparison show that the orthophotograph created with the SRTM data was more accurate than the NED with respect to the test points.

The distribution of this error is critical for orthophoto users. As was expected, the errors were larger on the edges of the orthophotograph and very small near the center of the image (nadir point). The plot of cumulative RMSE versus the radial distance from the ground principal point (GPP) shows that it is possible to achieve a submeter RMSE within a 1,000-meter distance from the perspective center.

The RMSE as a function of the distance from the center of the photograph. This table was derived from 54 GPS measurements of points that were identified on the orthophotograph created with SRTM DTM.


From the results of this experiment, it is clear that these government data sets can be used to create orthophotos at a scale of 1:10,000 and meet acceptable industry standards such as those developed by ASPRS. Since the test site was of limited size and consisted of only one type of land cover, further experiments need to be performed under different situations.

This study found that the SRTM data had slightly better accuracy than the NED data but it may not represent the terrain properly and may have larger errors in computing slope and aspect parameters. It is also important to note that the SRTM data is a Digital Surface Model (DSM) while NED data is a DEM measuring ground topography. SRTM data is current, which is an important advantage providing a proper model that can be used for many applications, even for updating the NED. Our future research will include evaluating NED and SRTM data in other surveying applications such as flow accumulation and basin determination. Moreover, we will study the use of these data sets as soft data to be fused with sparse data with higher accuracy (hard data) to obtain an improved surface.

Acknowledgment: The support for this research was granted by the National Geospatial-Intelligence Agency under contract No. HM1582-04-1-2026. For a bibliography of this article, please see this article online at

Technical information for the downloaded data (NED & SRTM) and the collected data (GPS).


A Digital Elevation Model (DEM) is any digital representation of a topographic surface. It normally consists of a file of elevation points, measured either systematically at equally spaced intervals (raster DEMs) or in an irregular pattern. DEM collection methods include conventional surveying techniques using a total station or a GPS receiver, photogrammetric method using aerial photographs or satellite stereo images, and active sensors using laser scanners like LiDAR (Light Detection And Ranging) or radar systems like IFSAR (Interferometric Synthetic Aperture Radar). A DEM may also be derived from an existing contour map through digitizing or scanning of the map detail.

The SRTM and NED data sets with a horizontal spacing of one arc-second interval (~90 feet at the equator) can be obtained from the United States Geological Survey's (USGS) Seamless Data Distribution (SDD) website at Before downloading these data from the Internet site it is important to turn off spyware and pop-up blocker options within the browser program.

Data for the NED have been developed by merging existing USGS elevation data available across the United States and its territories into a seamless raster format. These DEM quadrangles were created using high altitude aerial photography and are often referred to as Digital Terrain Models (DTMs) because they represent the ground surface.

The development of NED began in the early 1990s and was completely assembled in 1999 by merging and processing the individual 7.5 minute DEM (with 10 and 30 meter resolution). The horizontal datum for NED is the North American Datum of 1983 (NAD83) and the vertical data is the North American Vertical Datum of 1988 (NAVD88), except for Alaska where NAD27 and NAVD29 are still employed. Adjacent tiles were edge-matched and feathered while remaining artifacts and other anomalies were removed with directional filters, ensuring the most accurate seamless DEM possible.

The SRTM data were acquired by the National Geospatial-Intelligence Agency (NGA) and the National Aeronautics and Space Administration (NASA) using a radar system that flew onboard the Space Shuttle Endeavour during an 11-day mission in February 2000. The current products available have a resolution of one arc second for the United States and its territories, and three arc seconds for all the areas between 60º North and 56º South latitudes. Interferometry was used during the SRTM to measure interference patterns of radar signals and to determine 3D positions of elevation points. The radar data underwent extensive processing and noise filtering before they were released to the public. SRTM DEM uses WGS84 datum and EGM96 geoid model. NED data is smoother than the SRTM data.

Unlike a DEM, which represents ground topography, a Digital Surface Model (DSM) describes a surface that includes buildings, structures and vegetation (tree canopy). SRTM data are DSMs because the radar measurements were derived from the first return signal with the exception of forested areas where the radar signal may penetrate the tree coverage and measure the ground. A few tools are available to filter the SRTM data into a DEM. SRTM data may have gaps in very steep areas due to radar layover. Furthermore, radar return off water creates noisy data with many spikes due to waves. To solve this problem, large water bodies were delineated and the surface was flattened to a determined elevation.

The internal consistency and accuracy of these data sets have been described by a few authors. Smith and Sandwell (2003) performed spectral analysis of the 1 arc second SRTM and NED data, and found that the Root Mean Squared (RMS) differences between the SRTM data and an accurate test surface is 2.7 m, while the RMS differences between the NED data and an accurate surface is higher (3.5 m). They noted that the SRTM data is worse in short wavelength (smaller than 350 m), likely due to the application of filtering algorithms. Holmes et al (2000) investigated the impact of error in the NED data on hydrological modeling using geostatistical simulation, and Reinartz et al (2005) studied the accuracy of SRTM data and SPOT 5-derived DSM in open areas, low built-up areas and forest areas. They concluded that in forest areas the elevation accuracy decreases drastically since it neither represents the tree canopy or the ground.

Chad M. Schaeding is a recent graduate of the Ferris State University Surveying Engineering program. He plans to attend Purdue University to pursue graduate education. Yaron A. Felus, PhD, PS, and Robert R. Burtch, PS, are professors of the Surveying Engineering Department at Ferris State University.