Size does matter: relationships between image pixel size and landscape process scales

This paper briefly reviews the application of digital elevation models (DEMs) to the study of landscapes. Such studies can involve both the enhancement of DEM images to highlight particular patterns, and the use of DEMs to model attribute values of landscapes. Recognition of palaesurfaces is an example of the first use, while modelling hydrological properties based on slope attributes derived from a DEM is another. Following the review, the paper presents work on the character and scale of slopes and the processes that form them in a study area near Picton, NSW. These slope and process scales are then considered in the context of digital elevation models as a source of data about slopes. Slope angles are clustered around modal values that may be referred to as characteristic and threshold slopes. Characteristic slopes are those most commonly occurring in an area, and their inclination is controlled by the material on which they are formed and the processes that control their formation. They are closely related to threshold slope angles, which are those where sudden changes of slopes processes take place. Most DEMs have generalisations of the land surface built into them. If these generalisations are within the spatial range of the processes that are operating in the landscape of interest, there is no problem. However, if the generalisations are greater than the resolution of landscape processes, any results or indices derived from DEMs must be treated with caution. In the Picton study area only an original ground survey and to a lesser degree a 25m "DEM" give any indication of the shape of the ground surface. 50m and 100m DEMs barely resemble the ground surface, and values derived from these latter DEMs in no way reflect the original slope form. Moreover, they give no indication of the characteristic slope values, so they do not reflect the nature of the processes operating in this landscape. A SRTM 90m DEM over the same profile line similarly provides no real ground surface information. This study shows that, although DEMs are frequently used to derive values for slope angles, the accuracy of these derived values depends on the pixel resolution of the DEM from which they are derived. That accuracy of slope angle and shape depends on DEM resolution is obvious. What is not so obvious, and in many cases seems to be ignored, is that DEM resolution must be better than landsurface process scale if DEMs are to be used to predict spatial patterns of, say, soil attributes. Slope angles derived from most available DEMs are therefore limited as descriptions of real landscapes and processes unless the data are at a resolution that equals or is better than the scale of slope and regolith processes. The appropriate scale for a particular landscape can only be determined by geomorphic analysis of landform shape and processes; in most cases this will mean ground survey. In the Picton area a pixel resolution of 5m is adequate to capture the scale of surface processes and therefore likely variation of, say, soil attributes. In other areas the resolution required may be as small as 1m or as large of 100m. In other words, landscape process scale will dictate useful pixel resolution scale. And although this paper does not consider other raster image data, the results imply that the same conclusions apply to them as well.