Analysis and Implementation of Multiple Models and Multi-Models for Shallow-Water Type Models of Large Mass Flow

Dense large scale granular avalanches are a complex class of flows with physics that has often been poorly captured by models that are computationally tractable. Sparsity of actual flow data (usually only a posteriori deposit information is available) and large uncertainty in the mechanisms of initiation and flow propagation make the modeling task challenging and a subject of much continuing interest. Models that appear to represent the physics well in certain flows turn out to be poorly behaved in others due to intrinsic mathematical or numerical issues. Nevertheless, given the large implications on life and property many models with different modeling assumptions have been proposed.;While, inverse problems can shed some light on parameter choices it is difficult to make firm judgements on the validity or appropriateness of any single or set of modeling assumptions for a particular target flow or potential flows that needs to be modeled for predictive use in hazard analysis. We will present here an uncertainty quantification based approach to carefully, analyze the effect of modeling assumptions on quantities of interest in simulations based on three established models (Mohr-Coulomb, Pouliquen-Forterre and Voellmy-Salm) and thereby derive a model (from a set of modeling assumptions) suitable for use in a particular context. We also illustrate that a useful approach is to use a Bayesian modeling averaging based on the limited available observation data to combine the outcomes of alternative models.;We juxtapose observation data to the simulation results, only with the purpose to maintain a link to the real occurrence of a geophysical flow representing a possible outcome of the case studies described. The fundamental focus of this study is to explore the dynamics of the simulated flows, enabling a notion of the contribution of different mechanisms or models elements inside the simulation procedure, in a fully quantitative, predictive-use oriented and statistical framework. Subsequently, using Bayesian model averaging for hazard analysis based on an identified quantity of interest. The combined model strategy will be embedded in a UQ framework that attempts to quantify two types of so-called "epistemic uncertainty" in resulting simulated maps. This consists of the uncertainty on the selection of rheology model and the one affecting the specific parameters space of the chosen rheology.