Abstract
Debris flows are among the most widespread natural hazards in mountain areas. They play an important role in sediment transfer from low order channel networks to downstream areas and, thus, they can pose a serious threat to infrastructure. For debris flows to be released, in addition to a considerable amount of rainfall, sediment availability and a peculiar morphological setting (e.g. inclined terrain) are required. However, areas prone to trigger debris flows are not always structurally connected to the main channel, and thus do not necessarily provide sediments to the draining channel network. Despite the relatively high number of published research in the fields of debris flow susceptibility mapping and sediment connectivity analyses, to our knowledge, no attempts have yet been made to combine these two aspects, explicitly by means of data-driven procedures.
This work builds upon a novel data-driven approach developed by Steger et al. (under review), which allows to categorize areas based on both aspects, debris flow release susceptibility and structural connectivity. The original method was tested within three mountain catchments in the South Tyrolean Alps (max area 140 km²). This study aims at upscaling this approach at the regional scale and to this end, the whole South Tyrolean Dolomites (1,120 km²) has been selected as study area. In summary, the methodical framework comprised (i) modelling of debris flow release susceptibility using an interpretable machine learning algorithm (generalized additive model) and derivation of a quantitative threshold (susceptible vs. not-susceptible areas), (ii) training a logistic regression model to calculate the probability of an area being structurally connected in terms of debris flow release and subsequent thresholding (connected vs. disconnected areas), and (iii) combination of both previously generated and validated binary maps to create a joint susceptibility-connectivity map.
The results are represented via a variety of maps and associated statistics. It is shown that only 1.7% of the Dolomites were determined to be both susceptible to debris flow initiation and structurally connected to a target channel system, whereas areas that are susceptible but disconnected cover 8.2% of the whole study area. 16.5% of the area represents hillslopes that are potentially connected to the river network but result not susceptible in terms of debris flow release.
The straightforward implementation of this approach allows it to be an effective tool to assist hazard mapping and management. This study shows that the presented method provides spatial information useful to preliminary identify areas of interest for connected debris flows at the regional scale, but at the same time, it also highlights the need for further ground truth data to validate the results quantitatively beyond single catchment analyses.