A Spatial Filter for the Removal of Striping Artifacts in Digital Elevation Models

Elongated topographic artifacts, such as the striping production artifacts described for USGS 7.5-minute DEMs, can result in globally biased estimates of slope and aspect. As such, developing methods to reduce these artifacts and their resulting biases is important. This study presents an algorithm for the mitigation of these artifacts, using Terrain Resource Information Management (TRIM) digital elevation models (DEMs) of the Fort St. John Forest District, in British Columbia, Canada, as the test bed. The algorithm uses a theoretical error model, where elevation measurement errors are assumed to be autocorrelated along the collection lines of the photogrammetric model, and takes advantage of the entry order of DEM points to apply a sequence of spatial filters to the elevation. A probability function is used to constrain the elevation changes to an acceptable range. The algorithm is effective in mitigating the artifacts’ effects on slope and aspect while preserving the original topographic detail. Introduction Digital elevation models (DEMs), a gridded numerical representation of terrain elevation, are used in an increasingly wide range of analytical applications (Weibel and Heller, 1991; Pike, 2000). As DEM applications become more widespread, so does concern about the quality of the available elevation data and the propagation of DEM errors through the analyses. It is now well documented that the results of most DEM-based quantitative operations are influenced not only by the magnitude but also by the spatial distribution of elevation errors (Fisher, 1991; Lee et al., 1992; Bolstad and Stowe, 1994; Hunter and Goodchild, 1997; Fisher, 1998; Heuvelink, 1998). However, most DEM producers, like the United States Geological Survey or the British Ordnance Survey, report only the average magnitude of DEM errors as the root-mean-square error (RMSE), which does not provide information on systematic bias (Li, 1988) nor on the spatial patterns of the DEM errors (Guth, 1992). Artifacts, i.e., spatially structured errors of a systematic nature, are often associated with the production of DEMs. The interpolation of DEMs from contour line maps, for example, can produce several kinds of artifacts (Carrara et al., 1997), an example of which is that of contour line “ghosts” in USGS Level 2 DEMs (Guth, 1999). Different kinds of artifacts are associated with DEMs produced directly from photogrammetric models (Kok and Rangayyan, 1987). In particular, striping artifacts have been described for level 1 USGS DEMs (Brunson and Olsen, 1978; Hassan, 1988; Klinkenberg and Goodchild, 1992; Brown and Bara, 1994; Garbrecht and Starks, 1995; Oimoen, 2000). Striping artifacts are also found in the Terrain Resource Information Management (TRIM) digital map product of the Province of British Columbia (BC), Canada. In this paper we describe an algorithm that mitigates these artifacts by taking advantage of the specific structure of the TRIM product. While the methodology described below was applied to BC TRIM data, it could be applied to any digital elevation data collected using a profiling technique (Petrie, 1990). Production Artifacts as Spatially Structured Elevation Error In December of 1996, the Ministry of Environment, Land, and Parks (MoELP) of the Province of British Columbia, Canada, announced the completion of the digital production of 7027 1:20,000-scale Terrain Resource Information Management (TRIM) maps that cover the entire province (GDBC, 1999). The TRIM product contains digital elevation data that are designed to produce triangulated irregular networks (TINs). TRIM elevation data is organized into two vector files: a file of elevation (mass) points, hereafter referred to as the DEM file, and a file of break lines (GDBC, 1992). Elevation points, typically collected using a semi-automated profiling technique, are distributed on a semi-regular grid spaced between 75 and 100 m apart, and are collected in sequence along profiles generally oriented north-south (GDBC, 1992). As a result of their collection methodology, groups of sequentially measured elevation points, hereafter referred to as “collection lines,” are generally recognizable. As one might expect, elevation values within collection lines have a higher degree of spatial autocorrelation than do elevation values between collection lines, especially in areas with limited topographic variability. This produces topographic artifacts manifested by “stripes” or “ridges” aligned with the collection lines. These collection line artifacts are particularly evident in analytical hill-shading images or in aspect maps generated from the TRIM data source. Oimoen (2000) provides several examples of similar problems encountered in USGS DEMs. The authors encountered collection line artifacts while working with TRIM digital elevation models on a landscape ecology investigation in the area of the Halfway and Peace River watersheds in northeastern British Columbia. Sixtythree 1:20,000-scale map sheets were made available for digital terrain analysis by the Fort St. John Forest District, British Columbia Ministry of Forests. Several of these, when converted to 30-m-resolution raster DEMs, showed evidence of “stripes” aligned with the collection lines of the DEM points. Their perfect alignment with the collection lines, M. Albani was with the Department of Forest Sciences, University of British Columbia, 2424 Main Mall, Vancouver, BC V6T 1Z4, Canada; he is presently with the Department of Organismic & Evolutionary Biology, Harvard University, HUH 22 Divinity Avenue, Cambridge, MA 02138 ([email protected]). B. Klinkenberg is with the Department of Geography, University of British Columbia, 1984 West Mall, Vancouver, BC V6T 1Z2, Canada. IPC_Grams_02-046.qxd 6/6/03 4:49 PM Page 755