Comparison of Geomorphometric Methods for Semi-Automated Extraction of Landform Objects
Abstract Code: 10641

Abstract

We present results from a project testing the feasibility and challenges associated while using existing terrain analysis algorithms for semi-automated mapping areal extents of three broad categories of landforms: non-linear eminences (e.g., peak, mount, pillar, mountain, hill, mesa, butte), linear eminences (e.g., ridge and spur) and linear depressions (e.g., channel, valley, and hollow). Three popular geomorphometric methods based on geomorphometric features (Wood, 1996), geomorphons (Jasiewicz and Stepinski, 2013), and the topographic position index (Weiss, 2001) were evaluated for three study areas in the Great Smoky Mountains (NC-TN), White Mountains (NH), and Colorado Plateau (NM). Experimental results were compared in 2D and 3D map views in GIS software, followed by quantitative comparative analysis to answer questions about the impact of input algorithmic parameters and the terrain type on quality of extracted feature’s areal extents. Additional comparative analysis was conducted to evaluate the semantic similarity between landform classes from different methods (e.g., summit vs. peak, valley vs. channel). The major finding from this project was that only smaller neighborhood scales between 300 to 400 meters are the optimum scales for extracting landform objects that correspond well to intuitive expectations of landforms shapes and extents. The type of terrain is not as critical as initially assumed, but more careful analysis is warranted for low relief areas which make it harder to detect and delineate landform boundaries. The comparison of landform classes raised fundamental questions about landform class semantics that can be answered only in the next phase of the project.