Delineating valley segments of variable length across a catchment using coarse resolution DEMs

Abstract:

We present a quick and easy-to-use, automated technique fully within ArcGIS for delineating valley segments using DEM-derived network scale metrics of valley bottom width and slope using only publicly available, coarse-resolution DEM input. This technique harnesses the power of unsupervised machine learning via a k-means clustering algorithm to solve a conundrum in GIS-based geomorphic analysis of rivers: the delineation of valley bottom segments of variable length. The delineation of valley bottom segments provides an entry point into automated geomorphic analysis (such as measuring sinuosity and confinement) and characterization (or classification) of river systems.

Also, we present an approach for delineating valley extent (or polygon) across a large catchment using only coarse resolution DEMs by extracting landscape morphometries that are most likely to characterize the morphology of a valley bottom (i.e. flat, pit, valley, footslope and shoulder). In comparison to other models that rely on break in slope thresholds to identify valley extent, our methodology deals with floodplain geomorphic unit complexity. We assess the sensitivity of our results to variable DEM resolution and find that coarse-resolution datasets (90m resolution) provide superior results. We also find that LiDAR-derived DEMs produce more realistic results than satellite-derived DEMs across the full range of topographic settings tested. Satellite-derived DEMs perform more effectively in moderate topographic settings, but fail to capture the subtleties of valley bottom extent in mild gradient, low-lying topography and in narrow headwater reaches.