Untangling Scale Effects: Resolution, Neighborhood Size, and Analysis Scale.

This presentation explores the differences between resolution, neighborhood size, and analysis scale. Of particular interest is determining which concept is most important for identifying optimal scales of covariates for the use of correlation in digital soil informatics. Many variables that are useful as predictors in digital soil mapping are dependent upon spatial context. For example, the calculation of slope gradient on a digital elevation model requires an analysis neighborhood that considers cells surrounding the cell being calculated. The context or area considered is then dependent upon both the resolution and the number of cells (window size) used to define the neighborhood. Raster resolution (cell size) and analysis neighborhood are different generalization methods for spatial data and thus should not be assumed to affect the data in the same way. Resolution is widely recognized as an important factor in identifying covariates in digital spatial analysis. Despite some studies having demonstrated that neighborhood size is also influential, the mode in digital soil mapping publications has remained to emphasize the importance of resolution. Nonetheless, it is the combination of the two that best corresponds to the geographic concept of analysis scale. Identifying the analysis scale at which different covariates best explain the variation of soil properties reflects the classic geographic strategy of using map generalization (relative size of map delineations) to identify the scale at which phenomena occur. While polygon delineations mostly correspond with the definition of resolution, the raster data model in a geographic information system changes many concepts from traditional mapping. Analysis scale is the area over which the generalization is made, which in raster analysis is the result of the cell size, multiplied by the number of cells. This study confirms earlier results that these variables of spatial analysis indeed affect the amount of correlation found between digital terrain analysis and soil properties. However, regardless of the combination of resolution and analysis neighborhood, when the analysis scale was the same, the correlation was the same.