Landslide susceptibility assessment for engineered slopes using statistical and deterministic approaches

Abstract

Landslides cause hundreds of deaths and billions euros of damage to infrastructure and the environment each year. In order to predict the locations most susceptible to landslides, the field of landslide hazard assessment has gone through a massive development in the last twenty years by introducing a wealth of statistical and geotechnical landslide susceptibility models.
However, these efforts have been largely restricted to landslides occurring in natural terrain even though landslides occurring on geotechnical assets on transportation networks can result in even greater consequences. Current risk assessment approaches for earthworks on large transportation networks still largely take form of subjective risk matrices with inputs gathered by visual walkover surveys using data stored in an asset database.
This paper shows the application of two distinctive objective landslide susceptibility approaches on a case study of Irish rail. The first is a ‘statistical’, or ‘data-driven’ approach uses logistic regression as a statistical tool to establish the influence of slope-describing variables that have led to landslide occurrence. This approach draws the data from the asset database containing records of slope variables, and the adjoining landslide register. The same asset database is used as a basis for the second, ‘geotechnical’ or ‘deterministic’ approach. In this approach, geometrical and geotechnical properties of each slope are used to carry out probabilistic slope stability analysis, resulting in probability of failure for each slope.
Both approaches result in susceptibility zoning for earthwork assets across the network, effectively ranking them in the criticality terms. This study compares the requirements, applicability and outcomes of each approach, and discuss the methods needed for developing each of them into hazard and risk assessments.