Uncertainty Quantification of Tsunami Models

Event Start
Event End
Location
B1 East Side MPR
Serge Guillas, Professor of Statistics, University College London (UCL)

Abstract

In this talk, we first show various strategies for the efficient emulation of simulators having uncertain inputs, with applications to tsunami wave modelling. A fast surrogate of the simulator's time series of outputs is provided by the outer product emulator. Dimension reduction of the outputs can be achieved by employing functional principal components, to account for smoothness, on registered curves. It can enable real-time warnings according to the uncertain characteristics of the mechanism (landslide or earthquake) that generate the tsunami. It can also help provide a risk assessment for a specific region as it the emulator covers well the possible ranges of likely events. We show some new realistic simulations for earthquake-generated tsunamis in Cascadia (Western Canada and USA), using the tsunami model VOLNA. VOLNA is a solver of nonlinear shallow water equations on unstructured meshes that is now accelerated on the GPU system Emerald. We propagate uncertainties from the uplift to the time series of wave elevations. We also present an analysis of the influence on the tsunami wave of the uncertainties in the bathymetry, due to gaps in the depth soundings.  Finally, enhancements in terms of sequential design of computer experiments are presented and applied to the tsunami model VOLNA. Points in the design are chosen according to a new adaptive strategy and compared to current approaches in terms of statistical and computational efficiencies. 
Joint Authors:  Joakim Beck (UCL), Simon Day (UCL), Xiaoyu Liu (UCL), Andria Sarri (UCL).

Brief Biography

PhD (Paris 6, France) 2001, Research Associate (University of Chicago, USA) 2002-2004, Assistant Professor (Georgia Institute of Technology, USA) 2004-2008, lecturer (UCL) 2007-2009, Reader (UCL) 2009.

Contact Person