Search

Abstract: This paper details an automated object-oriented approach to mapping landforms from digital elevation models (DEMs), using the example of drumlins in the Chautauqua drumlin field in NW Pennsylvania and upstate New York. Object-oriented classification is highly desirable as it can identify specific shapes in datasets based on both the pixel values in a raster dataset and the contextual information between pixels and extracted objects. The methodology is built specifically for application to the USGS 30m resolution DEM data, which are freely available to the public and of sufficient resolution to map medium scale landforms. Using the raw DEM data, as well as derived aspect and slope, Definiens Developer (v.7) was used to perform multiresolution segmentation, followed by rule-based classification in order to extract individual polygons that represent drumlins. Drumlins obtained by automated extraction were visually and statistically compared to those identified via manual digitization. Detailed morphometric descriptive statistics such as means, ranges, and standard deviations were inspected and compared for length, width, elongation ratio, area, and perimeter. Although the manual and automated results were not always statistically identical, a more detailed comparison of just the drumlins identified by both procedures showed that the automated methods easily matched the manual digitization. Differences in the two methods related to mapping compound drumlins, and smaller and larger drumlins. The automated method generally identified more features in these categories than the manual method and thus outperformed the manual method. [Copyright &y& Elsevier]