Home | Business News | Browse by Publication | C | Cartography and Geographic Information Science

Shuttle Radar Topography Mission elevation data error and its relationship to land cover.

Article Excerpt
Introduction

The Shuttle Radar Topography Mission (SRTM), flown in February of 2000, collected interferometric synthetic aperture radar (IFSAR) data for over 80 percent of the Earth's land area (NASA 2005). These data have been processed to create the first near-global high resolution (1-and three-arc second interval) digital elevation model (DEM) available to the public and the scientific community. Research-grade DEMs have been available since 2002, while finished DEMs for many parts of the world became available in the summer of 2005 and are available online (USGS 2005; GLCF 2005). Much excitement about the SRTM elevation dataset is due to the impact its availability will have on projects involving vast data-poor regions of world. Prior to SRTM release, the best available global elevation dataset provided roughly one kilometer spatial resolution. The benefits of this product for scientific and planning applications in other parts of the world are clearly tremendous (see Hutchinson and Gallant 1999 and Moore et al. 1991 for overviews of DEM applications). One significant issue regarding the utility of the new DEMs concerns their fidelity to the real-world terrain they depict. In particular, what sorts of errors might one expect, and how might they affect the results of applications employing the data?

Relatively little work on SRTM accuracy has been published to date. Smith and Sandwell (2003) obtained high-resolution laser-collected data for a study site in California's Mojave Desert. They employed spectral analysis of one-dimensional data transects to identify terrain scales at which reliable information could be determined from research-grade one-arc-second SRTM data. Poorer coherence at short terrain wavelengths was identified, as compared to the National Elevation Dataset (NED), which they attribute to filtering steps during SRTM data processing. They also conclude that the three-arc-second dataset will contain about as much information as the nominally higher-resolution one-arc-second dataset. Other papers have used root mean square (RMS) differences with other elevation datasets. Dense vegetation is generally associated with larger error for other DEM products (e.g., Bolstad and Stowe 1994). The known physical properties of IFSAR suggest that height estimates will be too large in areas with dense vegetation or structures (Jensen 2000). Previous research has identified some bias (with respect to bare-earth DEMs) in high slope, forested, and urban areas (e.g., Reinartz et al. 2004; Gamba et al. 2002). Despite this, these papers and others (e.g., Koch and Heipke 2001) indicate that the overall vertical accuracy is better than the 10-16 m objective of the SRTM project (NASA 2002). All of these studies employed preliminary data released in advance of the finished product; error characteristics may be expected to vary somewhat in the final data set.

The SRTM mission employed IFSAR technology to map the Earth's topography. Substantial prior research has been conducted on the accuracy of airborne IFSAR and other high-resolution, remotely sensed elevation production technologies such as lidar; a number of these have identified relationships between vegetation and elevation quality. Norheim et al. (2002) compared spatially coincident IFSAR and lidar data for a study region in Washington State. Survey grade elevations in a variety of different land-cover settings were employed to consider vegetation effects. The IFSAR DEM generalized topography more than the lidar DEM, resulting in a smoother surface, while the lidar DEM consistently overestimated actual heights, particularly in more densely vegetated environments.

Cobby et al. (2001) developed a vegetation height modeling approach for lidar-derived data that segmented the study region into different vegetation cover classes. Their topography model consistently overestimated the actual surface; dense canopy cover resulting in an inadequate number of ground returns was identified as a primary factor in this overestimation. In addition, elevation estimates for locations covered with taller vegetation were more uncertain due to the correspondingly greater return variability.

Relationships between both IFSAR and lidar elevation accuracy and land cover were also investigated by Hodgson et al. (2003) for a study region in North Carolina. IFSAR elevations were afflicted with the greatest amount of error, with substantial overestimations of actual height. Furthermore, average elevation error was significantly higher in forested areas; Hodgson et al. (2003) determined that land cover, as opposed to topographic variability, was the primary factor explaining IFSAR elevation error. Lidar elevation accuracy was much greater, but average error also differed by land-cover type. Interestingly, pine cover had much lower errors than deciduous covers; similar results for pine were reported by Hodgson and Bresnahan (2004) using a different lidar data set and study area.

In contrast to the other studies discussed, Hodgson and Bresnahan (2004) found underestimations of the ground surface for vegetated land covers. Particular post-collection processing methods employed to generate elevations from sensor information in each study--methods which are often sensor-dependent or proprietary--may have substantially affected the results. This makes it difficult to generalize from previous findings and speculate on the specific error characteristics of SRTM elevations, even though similar sensor technologies were used.

This paper investigates the quality of SRTM data in an extremely data-rich environment in southeastern Michigan, USA. The study region possesses several useful characteristics. While its elevation range (83 meters) is unremarkable, the area's glacial past has resulted in locally complex topography and is dotted with small ponds and wetlands, which may be challenging for the sensor (NASA 2002). Second, the study area has highly varied land-cover types due both to the topography and to suburban development. Finally, a wealth of data is available for the area. In addition to SRTM DEM products, high (sub-meter) accuracy elevation postings have been obtained from the Oakland County (Michigan) Information Technology Department. These postings serve as ground truth for the study. Also, a recently collected land-cover dataset is employed to consider its effect on the accuracy of SRTM data. The combination of varied terrain and land cover, along with abundant data on these factors, enables a detailed investigation into the nature of error in the SRTM product.

A standard elevation accuracy assessment strategy, and that employed here, is to employ a comparative approach in which the data set of interest is matched against another data set of higher quality (see for examples Bolstad and Stowe 1994; Kyriakidis et al. 1999;...

View this article FREE - Now for a Limited Time, try Goliath Business News
Free for 3 Days!