B. Romstad1, B. Etzelmüller2
1 Department of Geosciences, University of Oslo, PO Box 1047 Blindern, 0316 OSLO, Norway
and International Centre for Geohazards (ICG), PO Box 3930 Ullevaal Stadion, 0806 OSLO, Norway and CICERO Center for Climate and Environmental Research, PO Box 1129 Blindern, 0318 OSLO, Norway
Telephone: +47 2285 8530
Fax: +47 2285 8751
2 Department of Geosciences, University of Oslo, PO Box 1047 Blindern, 0316 OSLO, Norway
Telephone: +47 2285 7229
Fax: +47 2285 4215
The spatial prediction of landforms and surface processes is an important component in the understanding and modelling of an environmental system. Therefore a fundamental research topic within geomorphometry is to extract and classify landform elements and landform types. The general topic is thoroughly introduced and reviewed in the recent chapter by MacMillan and Shary (2008) and the paper by Minár and Evans (2008), but in this paper the focus is on the matter of automatic extraction of landform elements.
Landform elements are segments characterized by simple geometry and can be viewed upon as the basic building blocks for landforms, landform types and landform systems. While many applications (implicitly or explicitly) define the grid cell itself as the basic landform element, several studies have pointed out the weaknesses of this approach (e.g. Rowbotham and Dudycha 1998, Blaschke and Strobl 2001, Romstad 2001, Dragut and Blaschke 2006). If a landform element instead can be defined by a group of connected cells, we have effectively made the transition from a field based to an object based representation of the terrain. This is a powerful approach as it allows for the calculation of contextual information such as the shape and size of regions. Important contributions on how to construct geomorphologically significant landform elements in this way include those of Friedrich (1996), MacMillan et al. (2004), Dragut and Blaschke (2006) and Strobl (2008).
The by far most common method for this type of terrain segmentation is to delineate local catchments by use of flow modelling. This type of segmentation has the favourable property that the method for delineation of regions is based on an explicitly defined physical process (hydrological flow). Thus the resulting regions represent meaningful real world objects by definition. A weakness is that this method does not ensure the geometric simplicity of the resulting elements and significant changes in slope gradient may have to be treated separately. Dragut and Blaschke (2006) delineated homogenous landform objects by applying the image segmentation algorithm described by Baatz and Schäpe (2000) to a set of topographic attributes. This algorithm convincingly created landform elements that were both geometrically simple and geomorphologically meaningful, but the algorithm is complex and relies on a number of parameters. Thus it may be difficult to predict how the algorithm will behave when applied to the same topographic attributes in different areas.
In this paper we explore whether a simple watershed segmentation of curvature maps will produce meaningful landform units. We explain how this segmentation procedure creates elements that are geometrically simple and we evaluate the method by comparing the resulting regions to a geomorphological map.