Insights into debris-flow dynamics through vertical and horizontal velocity profile measurements at the Gadria creek, Italy

Debris flows are extremely rapid, flow-like landslides composed of fine and coarser-grained components, boulders, woody debris as well as water. They are characterized by large impact forces as well as long runout distances and are one of the most dangerous types of mass movements in mountainous regions. In the past, researchers have mainly measured the velocity of the front or of distinct surges. However, the spatio-temporal distribution throughout a cross-section remains largely unknown. Quantifying the horizontal and vertical velocity profiles is required for hazard assessment, the design of mitigation structures, process understanding and numerical model development.

In the present work, we analyze two debris-flow events that occurred at the Gadria creek (South Tyrol, Italy) in 2023 and 2024. We measure the horizontal as well as vertical velocity profiles for selected phases of the flows and explore how they vary in time. For the horizontal measurements, we use timelapse point clouds from a high-resolution, high-frequency 3D LiDAR scanner (Ouster OS1). We process these 3D point clouds to obtain 2D hillshade images of the moving flow, which we then analyze using Particle Image Velocimetry (PIV). This approach provides a timeseries of dense velocity vector fields of the moving surface, which we then evaluate at a defined channel cross-section to obtain horizontal velocity profiles. In order to derive the vertical velocity profiles, we use a fin-shaped barrier located in the middle of the channel, which is equipped with paired conductivity sensors at different depths along the side-wall. By applying cross-correlation to the paired conductivity signals, we can extract the vertical velocity distribution at the wall. We validated our methods by comparing the velocities to measurements of feature velocities, including boulders or pieces of woody debris, which we tracked manually and/or using a fine-tuned off-the-shelf neural-network-based object detection algorithm (YOLO v8).

For the surface velocity along the barrier, we find good agreement between the different measurement approaches. Over the duration of both events, we observe substantial variations in the shape of the profiles with different degrees of internal deformation: the horizontal profiles vary between parabolic and more plug-flow-like shapes whereas the vertical profiles feature convex to concave shapes.

Our findings highlight the non-uniform and highly variable distribution of debris-flow velocities in a cross-section with important implications for practical applications and process understanding, as for example for discharge and volume estimates. Eventually, the developed methods will be applied to additional events at the Gadria creek, which should allow for further inference into the constitutive flow behavior of debris flows to improve our understanding of these destructive events in the future.