Performance Evaluation of Four DEM-Based Fluvial Terrace Mapping Methods Across Variable Geomorphic Settings: Application to the Sheepscot River Watershed, Maine

Abstract:

Fluvial terraces are utilized in geomorphic studies as recorders of land-use, climate, and tectonic history. Advances in digital topographic data, such as high-resolution digital elevation models (DEMs) derived from airborne lidar surveys, has promoted the development of several methods used to extract terraces from DEMs based on their characteristic morphology. The post-glacial landscape of the Sheepscot River watershed, Maine, where strath and fill terraces are present and record Pleistocene deglaciation, Holocene eustatic forcing, and Anthropocene land-use change, was selected to implement a comparison between terrace mapping methodologies. At four study sites within the watershed, terraces were manually mapped to facilitate the comparison between fully and semi-automated DEM-based mapping procedures, including: (1) edge detection functions in Matlab, (2) feature classification algorithms developed by Wood (1996), (3) spatial relationships between interpreted terraces and surrounding topography (Walter et al., 2007), and (4) the TerEx terrace mapping toolbox developed by Stout and Belmont (2014). Each method was evaluated based on its accuracy and ease of implementation. The four study sites have varying longitudinal slope (0.1% – 5%), channel width (<5 m – 30 m), relief in surrounding landscape (15 m – 75 m), type and density of surrounding land use, and mapped surficial geologic units. In general, all methods overestimate terrace areas (average predicted area 136% of the manually defined area). Surrounding topographic relief appears to exert the greatest control on mapping accuracy, with the most accurate results (92% of terrace area mapped by Walter et al., 2007 method) achieved where the river valley was most confined by adjacent hillslopes. Accuracy decreased for study sites surrounded by a low-relief landscape, with the most accurate results achieved by the TerEx toolbox (Stout and Belmont, 2014; predicted areas were 45% and 89% of manual delineations). Our work informs future studies by highlighting the strengths and drawbacks of each method tested and by making recommendations for the types of geomorphic settings where each is most appropriate. The tested algorithms represent powerful new ways to analyze landscape history over large regions using high-resolution, lidar DEMs.