A new tool for studying shallow landslides at the basin scale: BEMSL

Shallow landslides are characterized by a superficial sliding surface whose depth is at least five meters below the ground. Their occurrence has increased in recent decades due to climate change, especially in Northern Italy where extreme meteorological events (the main triggering factors) have been reported to increase in intensity. Since shallow landslides are a very common geohazard in mountain and hilly areas, whose consequences can be catastrophic both for people and the natural environment, new methodologies that aim to better estimate landslide susceptibility have been explored in the literature. Here a new tool called BEMSL (“Basin Ensemble Models for Shallow Landslides”) has been developed to forecast effectively shallow landslides at the basin scale.

The BEMSL is a model that considers an ensemble approach for susceptibility mapping, and it is conceptually divided into three parts. Primarily, it includes different limit equilibrium and infinite slope formulations that describe the stability of a slope in terms of safety factor (FS), which is defined as the ratio between stabilizing and destabilizing actions. Even if, theoretically, the actions acting on a slope should be always the same, many authors in this field have proposed different FS equations, trying to choose the most relevant acting actions depending on the local geology, soil composition and other predisposing factors. Consequently, it is difficult to choose the most suitable FS formulation that fits best to the considered situation. To provide a unique answer, the second part of BEMSL includes the Random Forest (RF) approach that creates a model ensemble able to merge the outputs from the implemented FS formulations. Since RF is a machine-learning algorithm that works autonomously on FS data provided, countermeasures to avoid overfitting have been considered. In the last part, the output validation was assessed using the ROC (“Receiver Operation Characteristics”) approach, which essentially consists of the quantification of how many true positives (TP), false positives (FP), true negatives (TN) and false negatives (FN) compared to the available landslide census.

BEMSL was applied to retrieve dynamic landslide susceptibility maps, giving site-specific insight into the probability of shallow terrain failures. The reliability of this BEMSL tool was tested considering the event that happened in July 1987 in Tartano Valley (Sondrio province, located in Northern Italy). In the late afternoon of 18 July 1987, an extreme storm triggered several shallow landslides across Tartano Valley, which evolved into a catastrophic debris flow, resulting in 21 casualties and extensive infrastructure damages. In this case study, the risk of failure of punctual and linear electrical powerlines was investigated using the BEMSL. A dependence on the risk of failure due to the rainfall intensity temporal evolution has shown the vulnerabilities of the Tartano Valley electrical infrastructures developed during the extreme geo-hydrological event.