Effects of digital elevation model resolution on topographic representation: A case study in the Tama area, western Tokyo

Abstract

The effects of digital elevation model (DEM) resolution on topographic representation need to be evaluated to perform meaningful terrain analysis using a raster DEM. DEMs with a nominal resolution of 50 m, published by the Geographical Survey Institute, have often been used for terrain analysis in Japan. The resolution of the DEMs, however, may be insufficient to characterize complex topography such as finely dissected hilly land. This paper discusses relations between DEM resolutions and topographic parameters derived from DEMs, using data for the Tama area in western Tokyo. The area includes dissected hilly land and adjacent low-altitude mountains. Ten-meter DEMs representing current topography were produced from high-resolution vector digital data including contour lines and information about survey points. Part of the hilly land have undergone extensive artificial landform modification since the mid-1980s to provide flat residential areas. For these areas, analytical aerial photogrammetry was employed to construct 10-m DEMs representing topography in 1984. Coarser DEMs were derived from the 10-m DEMs by sampling elevation values with a certain interval. Slope angle and profile curvature were calculated from the DEMs with different resolutions. The mean slope angle for a 1 km X 1 km unit area consistently decreases with increasing DEM grid interval. The rate of the decrease in slope angle is mostly larger than that reported in previous studies, reflecting highly complex topography or mosaics of short valleys and ridges in the Tama area. Mean profile curvature tends to converge into null with decreasing DEM resolution, which conforms to the decrease in mean slope angle due to flatter representation of topography. These observations indicate that DEMs with a resolution finer than 50 m are needed for effective terrain analysis in the Tama area. Because DEMs finer than 50-m resolution are often unavailable in Japan, we examined whether 10-m DEMs interpolated from the 50-m DEMs are useful for terrain analysis in the Tama area. Twenty-seven methods were employed for interpolation, and their accuracy was evaluated using the mean error and the root mean square error of altitude, slope angle, and profile curvature calculated from the original 10-m DEMs. The results revealed that the three most accurate interpolation methods are the radial basis function of thin plate spline, minimum curvature, and modified Shepard's method without a smoothing parameter. Kriging also yields relatively good results if functions corresponding to actual variograms are used. These results are common to hilly land both with and without extensive landform modification. The mean error of slope angle is always negative, showing that the DEMs produced by interpolation can only represent smoothed topography. The 50m DEMs before interpolation, however, demonstrate further smoothed topography, meaning that interpolation can improve topographic representation. In contrast, profile curvature calculated from the DEMs produced by interpolation seems to be inaccurate, suggesting that real high-resolution DEMs should be used when analysis focuses on curvature.