MAPPING GLACIAL LANDFORMS ON THE GLACIATED ALLEGHENY PLATEAU USING OBJECT-ORIENTED CLASSIFICATION OF DIGITAL ELEVATION MODELS (DEMS)

Digital Elevation Models (DEMs) provide detailed information about the surface of the landscape on which we now live. In their simplest form, they consist of a matrix of numbers that contain an x and y coordinate and a z-elevation value. Because of the elevation information in DEMs, they are used extensively for slope analysis, viewshed analysis, and watershed studies. However, for landform mapping they have been used most commonly for visualization only through shaded elevation maps. Delineation is then usually based on viewer interpretation which, while important, can lead to considerable mapping error.

This research attempts to eliminate viewer error by objectifying, quantifying, and automating the identification process through the use of the software package eCognition. This remote sensing package classifies spatial datasets through spectral analysis of individual pixels (not unlike regular remote sensing software), but also follows the concept that information necessary to interpret an image is not represented in single pixels but in image objects and their mutual relations. These objects can have form, texture, neighbor relations and context. They are identified through a segmentation process that can be performed at a number of scales depending on the size or complexity of the features to be identified. Objects are also identified based on the border smoothness and shape.

Classification of a DEM is significantly easier than classification of a multi-layered satellite dataset. A single layer is used and it is the inter-relationships of z-values that drive the classification process. On essence, the software package treats the z-values as brightness values. The test site consists of portions of the Glaciated Allegheny Plateau, between Cleveland, Ohio and the border of New York and Pennsylvania. The 30m horizontal resolution National Elevation dataset was used. Using three hierarchies of classification, large plateau tops were easily mapped, different degrees of streamlined drumlins were identified, and valley bottoms (often meltwater channels) were identified. Each landform type was efficiently extracted from the dataset as a shape file for further analysis in GIS. Also, after training the classification for one site, it was easily applied to another to automatically extract the same features.