Comparison of roving-window and search-window techniques for characterising landscape morphometry

Neighbourhood analysis in a Geographical Information System (GIS) calculates the value of a given raster cell from the values of its neighbouring cells. Common operations include filtering (highpass, low-pass, etc.) and smoothing (mean, mode) of data, operations that can be done by means of roving-windows or search-windows. Digital terrain analysis (or geomorphometry) relies on neighbourhood operations to calculate morphometric variables such as slope, aspect, local relief or surface roughness (among many others) at scales ranging from local (i.e., single landforms) to regional (entire mountain chains). The intent of this paper is to compare both techniques in a multi-scale study of geomorphometry, in central-eastern Brazil. The study area is limited by coordinates 0 and 26 S latitude and 34 W and 56 W longitude, with approximately 4:900:000km. The roving-window approach can be considered the standard filter technique in raster GIS operations and in image processing (Demers, 2004; Lillesand et al., 2004). It determines the new value for a given cell in a raster map using a mathematical function (mean, mode, standard deviation, etc.) of the cells values inside a n n neighbourhood (with odd n) centred in the cell of interest U 93