Automated delineation and morphometry of unclassified subglacial bedforms.

In the context of climate change, ice sheets are strongly influenced by the reorganization of the subglacial hydrological system and the dynamics of ice flow. Interactions between meltwater, ice flow and subglacial sediments give rise to a unique assemblage of periodic subglacial landforms composed of sediments known as bedforms. These subglacial bedforms therefore provide a large-scale observational window into the subglacial environment, which is difficult to analyze beneath current ice masses.
Mapping subglacial bedforms is traditionally performed using digital elevation models (DEMs) and/or aerial or satellite imagery through manual digitization in GIS software. This method is time-consuming and introduces operator subjectivity, heavily dependent on the expertise level of the operator. This manual approach is also a significant barrier to the use of new datasets with increasingly higher resolution (e.g. ArcticDEM, RGE ALTI®, HiRISE) and coverage of ever larger areas. Addressing these limitations is essential to efficiently analyze the distribution and morphometry of subglacial bedforms over large territories.
To overcome these challenges, we designed an automated tool to delineate and analyze the shape of subglacial bedforms using a recently defined land surface parameter, the Volumetric Obscurance. This parameter highlights convex and concave surfaces while minimizing the impact of noise from the topographic signal, making it particularly suited for detecting and mapping subglacial morphologies. The automated tool is based on the assumption that the diversity of subglacial bedform shapes reflects a continuum: therefore, unlike traditional methods, no pre- or post-mapping classification of bedforms is performed.
Our method uses DEMs and optical satellite images, including ArcticDEM and Sentinel-2 data, to generate regional morphological maps (bedform outlines and crestlines) and regional morphometric maps (spatialized statistical analysis of bedform morphometrics). It employs a multi-threshold segmentation approach to extract bedform features and calculate both dimensional morphometric parameters (e.g., volumes, areas) and dimensionless parameters (e.g., sinuosity, circularity, elongation). These provide synthetic and spatialized information on the distribution of morphological parameters across entire bedform fields.
We tested the tool on ArcticDEM data over a portion of the former Laurentide Ice Sheet bed, specifically the Keewatin Dome region in northern Canada, which displays a wide diversity of bedform shapes. The produced morphological maps demonstrated strong consistency, with approximately 75% correspondence between individual bedform outlines generated automatically and reference maps manually digitized by two distinct glacial geomorphologists. Despite a 25% difference in individual bedform outlines, the derived morphometric maps were highly comparable and provide reliable insights into subglacial deformation and hydrology.
By reducing subjectivity and significantly accelerating the mapping process, this tool enables the analysis of larger areas with greater precision compared to manual methods. The derived datasets allow for reconsideration and refinement of ice-sheet scale reconstructions of ice flow and meltwater dynamics. The tool is developed in Python and is freely accessible to the research community.