Matching Attribute Resolution to Scale: The Effects of Filtering on DEM Resolution

Raster datasets are important for spatial analysis and modeling as well as for cartographic display. As raster data becomes more readily available at finer spatial resolutions, generalization is required to meet project needs. The modification of detail in Digital Terrain Models through the generalization process of smoothing using a Gaussian Filter will be examined. While it can be assumed that the smoothing process will simplify the terrain through reduction of attribute complexity, the rate of generalization, variations for rough and smooth terrain, and differences in specific landscape conditions, as well as any intermediate spatial resolutions are unknown after iterative filtering. The theoretical contribution of this thesis is to systematically explore the concept of “implicit” resolution (attribute resolution) change, defined in the thesis as a modification to resolution which is the consequence of data processing or modeling without knowing in advance of the operation what is the precise resolution of the output. Several concerns arise after smoothing regarding data resolution and vertical integration with other datasets. This thesis will compare filtered Digital Elevation Models (DEM) to other National Elevation Datasets and National Hydrography Datasets of various spatial resolutions. A standard deviation and semivariogram analysis will relate the attribute resolution of a filtered DEM to a known spatial resolution, which can then be linked to a target mapping scale. A conflation analysis will determine the success rate of vertical data integration between a filtered DEM and an external, vector dataset. Results of this analysis have identified that aggressive smoothing can have a large, global impact on the spatial dependencies within the DEM. This suggests that smoothing is only useful for small changes in scale or resolution. Additionally, smoothing can have strong impacts on the rate of vertical integration on flat landscapes, where features (i.e. valleys) are only defined by a low relief