In mountainous areas, the use of DEM as environmental parameter in local and global models is widely accepted. Derived information like slope, aspect and others are often used as substitutes for missing data on specific habitat parameters to model the probability of species occurrence or absence. For habitat analysis of species, point observations were combined with underlying terrain information. The use of DEM with a resolution of 25 m is still accepted as sufficient. Different studies have shown the influence of specific topographic elements like ridges, depression, steep and flat areas on species distributions. Data providers often provide an overall accuracy description for the total area. The spatial correlation of error is hardly ever described, although this influences notably the derived data from DEM and could consequently be a major source of uncertainty in models using such derivatives.
We investigated the error of neighbouring cells of two different DEMs with a resolution of 20 m and 25 m, derived from photogrammetry and digitized from contour lines, respectively. 5 sampling areas each with 5*5 control points of the precise lattice coordinate of the DEM were established in the area of the Swiss National Park to investigate error correlations in open and forested as well as in planar and steep areas. Moreover, 246 survey points of a slightly irregular grid in the open area were used. The reference points were measured with surveying techniques.
The results show a spatial correlation of the error on all plots as well as significant differences between the different subareas. Nevertheless, the average difference of error between neighbouring points is 1.01 m. The derived mean slope error from cell to cell (20m) is therefore 2.9°. The maximum error between two neighbouring cells of 25 m is 8.14m which is resulting in a slope mismeasurement of 19.0°.
These results show the importance of the knowledge about spatial correlation of error in DEM, especially its consequences on derived data of DEM used in more complex model calculations such as species habitat models.