Workflow-induced uncertainty in probabilistic landslide hazard maps

Abstract

Increasing complexity and capacity of computational physics-based landslide run-out modelling over the last few years yielded highly efficient model-based decision support tools, e.g. landslide susceptibility or run-out maps, or geohazard risk assessments. Prior to applying such computational models for decision making, however, one has to carefully decide for appropriate and compliant building blocks of the underlying simulation tool chain. For example, one has to decide on resolution and quality of the data products representing topography or vegetation, on the complexity of the underlying process model and on the numerical solution scheme used to solve the process model. Probabilistic hazard mapping based on applied uncertainty quantification (UQ) furthermore requires to decide for model parameter's prior probability distribution, e.g. of friction parameters, as well as how to deal with the UQ-related high-throughput challenge. Surrogate modelling techniques based on Gaussian process emulation reduce the computational costs needed to generate hazard maps while accounting for the error introduced. A comparative study is presented herein by considering a collection of design criteria for a hazard mapping workflow, such as rheological parameters and intensity. Results of multiple model runs for a synthetic case on a real-world topography are illustrated to demonstrate how uncertainty is reflected in hazard maps. Implications regarding the design of model-based decision support tools in practice are discussed