Testing space-scale methodologies for automatic geomorphic feature extraction from LiDAR in a complex mountainous landscape

The next generation of digital elevation data (≤3 m resolution) calls for the development of new algorithms for the objective extraction of geomorphic features, such as channel networks, channel heads, bank geometry, landslide scars, and service roads. In this work, we test the performance of two newly developed algorithms for the extraction of geomorphic features: the wavelet-based extraction methodology developed by Lashermes et al. (2007) and the GeoNet nonlinear diffusion and geodesic paths methodology proposed by Passalacqua et al. (2010). The study area is part of the Rio Cordon basin, a headwater alpine catchment located in the Dolomites, a mountainous region in the eastern Italian Alps. The aim of this work is to compare the capability of the two new algorithms in extracting the channel network and capturing channel heads, relevant channel disruptions corresponding to landslides, and representative channel cross sections. The extracted channel networks are also compared to the ones obtained using classical methodologies on the basis of an area threshold and an area-slope threshold. A high-resolution digital terrain model of 1 m served as the basis for such analysis. The results suggest that, although the wavelet-based methodology performs well in the channel network extraction and is able to detect channel heads and channel disruptions, the local nonlinear filter together with the global geodesic optimization used in GeoNet is more robust and computationally efficient while achieving better localization and extraction of features, especially in areas where gentle slopes prevail. We conclude that these new methodologies should be considered as valid alternatives to classical methodologies for channel network extraction from lidar, in addition to offering the potential for calibration-free channel source identification and also extraction of additional features of interest.