Articles | Volume 5, issue 4
https://doi.org/10.5194/esurf-5-821-2017
https://doi.org/10.5194/esurf-5-821-2017
Research article
 | 
07 Dec 2017
Research article |  | 07 Dec 2017

Bumps in river profiles: uncertainty assessment and smoothing using quantile regression techniques

Wolfgang Schwanghart and Dirk Scherler

Abstract. The analysis of longitudinal river profiles is an important tool for studying landscape evolution. However, characterizing river profiles based on digital elevation models (DEMs) suffers from errors and artifacts that particularly prevail along valley bottoms. The aim of this study is to characterize uncertainties that arise from the analysis of river profiles derived from different, near-globally available DEMs. We devised new algorithms – quantile carving and the CRS algorithm – that rely on quantile regression to enable hydrological correction and the uncertainty quantification of river profiles. We find that globally available DEMs commonly overestimate river elevations in steep topography. The distributions of elevation errors become increasingly wider and right skewed if adjacent hillslope gradients are steep. Our analysis indicates that the AW3D DEM has the highest precision and lowest bias for the analysis of river profiles in mountainous topography. The new 12 m resolution TanDEM-X DEM has a very low precision, most likely due to the combined effect of steep valley walls and the presence of water surfaces in valley bottoms. Compared to the conventional approaches of carving and filling, we find that our new approach is able to reduce the elevation bias and errors in longitudinal river profiles.

Download
Short summary
River profiles derived from digital elevation models are affected by errors. Here we present two new algorithms – quantile carving and the CRS algorithm – to hydrologically correct river profiles. Both algorithms preserve the downstream decreasing shape of river profiles, while CRS additionally smooths profiles to avoid artificial steps. Our algorithms are able to cope with the problems of overestimation and asymmetric error distributions.