08:30
A mortar finite element method with embedded ensemble propagation for marine ice-sheet flow
Maarten Arnst | Université de Liège | Belgium
Show details
Authors:
Maarten Arnst | Université de Liège | Belgium
Kim Liegeois | Université de Liège | Belgium
Kevin Bulthuis | Université de Liège | Belgium
Eric Phipps | Sandia National Laboratories | United States
This work is concerned with the numerical simulation and uncertainty quantification of marine ice-sheet flow in the coupling zone between grounded and floating ice. Whereas the evolution of marine ice sheets is usually modeled by using a multi-domain formulation, we explore the use of a contact formulation, in which Signorini-type contact conditions govern the motion of the grounding line that separates the grounded portion from the floating portion of the ice. This contact formulation allows us to draw numerical methods from computational contact mechanics to enable the efficient solution of marine ice-sheet models on high-performance computers. At the conference, we will present the contact formulation, its discretization by using a mortar finite element method, the implementation of the discretized formulation by using software components from the Trilinos library, the use of Trilinos’s automated embedded ensemble propagation method for efficient uncertainty quantification, and performance evaluations on a benchmark problem from glaciology.
08:50
Estimation of the basal sliding coefficient in ice sheet flow problems with uncertain rheological parameters
Ruanui Nicholson | University of Auckland | New Zealand
Show details
Authors:
Olalekan Babaniyi | Rochester Institute of Technology | United States
Ruanui Nicholson | University of Auckland | New Zealand
Umberto Villa | Washington University in St. Louis | United States
Noemi Petra | University of California, Merced | United States
Modelling of the dynamics of polar ice sheets is critical to enable relevant projections of future sea levels. Key parameters which influence predictions are the basal sliding coefficient and constitutive relations employed to describe rheological parameters of the ice, both of which contain significant uncertainty. Here, we consider inferring the distributed basal sliding coefficient field from surface flow observations under random distributed rheological parameter fields. The standard approach to deal with multiple uncertain parameters is to infer all parameters simultaneously, or each parameter individually. However, joint estimation problems can be highly ill-posed and computationally infeasible. To avoid these issues, we approximately premarginalize over rheological parameters, and infer the basal sliding coefficient only. This leads us to "ignore" rheological parameters at the estimation stage, which, if unaccounted for, leads to overly confident and biased estimates. To mitigate these effects, and to carry out the premarginalization we use the Bayesian approximation error (BAE) approach. We present linearized uncertainty analyses, which indicate that fixing rheological parameters at an incorrect, but otherwise well justified, value can result in infeasible and misleading posterior estimates. Conversely, the BAE approach leads to feasible results, and is computationally less expensive (measured in the number of online PDE solves) than the conventional error approach.
09:10
Kriging metamodeling of functional-inputs computer code for coastal flooding hazard assessment
José Betancourt | Institut de Mathématiques de Toulouse - Ecole Nationale de l'Aviation Civile | France
Show details
Authors:
José Betancourt | Institut de Mathématiques de Toulouse - Ecole Nationale de l'Aviation Civile | France
François Bachoc | Institut de Mathématiques de Toulouse, Université Paul Sabatier | France
Thierry Klein | Institut de Mathématiques de Toulouse - Ecole Nationale de l'Aviation Civile | France
Deborah Idier | Bureau de Recherches Géologiques et Minières | France
Rodrigo Pedreros | Bureau de Recherches Géologiques et Minières | France
Jeremy Rohmer | Bureau de Recherches Géologiques et Minières | France
This study is part of the ANR RISCOPE research project, which deals with the construction of surrogate models for coastal flooding early warning [1]. In this work we discuss the development of a functional-input metamodel where the inputs represent time varying maritime conditions. We concentrate on Kriging metamodels to emulate the behavior of the hydrodynamic code. To model the inputs, we use a functional decomposition. The main challenge addressed here is to simultaneously select: (i) the projection dimension (i.e., number of basis functions and coefficients to keep); (ii) the distance used to measure similarity among observations; and (iii) the inputs that will remain active. We call these, structural parameters of the model. We propose a staged approach, where dominated levels of the structural parameters are progressively discarded, allowing an efficient identification of attractive configurations. Our approach uses the prediction capability of the metamodel as criterion to discriminate among configurations. In contrast, the classic approach in the literature is to select the projection dimension based on the error of the projection itself. We compared both methods through a case study using real data gathered at Gâvres coast in France. The classic approach lead to unnecessarily large projection dimensions while the proposed method was able to efficiently identify a fast and accurate metamodel.
[1] RISCOPE site: https://perso.math.univ-toulouse.fr/riscope/.
09:30
A polynomial chaos framework for probabilistic predictions of storm surge events
Pierre Sochala | BRGM | France
Show details
Authors:
Pierre Sochala | BRGM | France
Chen Chen | UT Austin | United States
Clint Dawson | UT Austin | United States
Mohamed Iskandarani | University of Miami | United States
We present a polynomial chaos-based framework to quantify the uncertainties in predicting hurricane-induced storm surges. Perturbation strategies are proposed to characterize poorly known time-dependent input parameters, such as tropical cyclone track and wind as well as space-dependent bottom stresses, using a handful of stochastic variables. The input uncertainties are then propagated through an ensemble calculation and a model surrogate is constructed to represent the changes in model output caused by changes in the model input. The statistical analysis is then performed using the model surrogate once its reliability has been established. The procedure is illustrated by simulating the flooding caused by Hurricane Gustav 2008 using the ADvanced CIRCulation model. The hurricane's track and intensity are perturbed along with the bottom friction coefficients. A sensitivity analysis suggests that the track of the tropical cyclone is the dominant contributor to the peak water level forecast, while uncertainties in wind speed and in the bottom friction coefficient show minor contributions. Exceedance probability maps with different levels are also estimated to identify the most vulnerable areas.
09:50
- CANCELED - Uncertainty quantification of subsurface properties and the forecasting of aquifer contamination
Arunasalam Rahunanthan | Central State University | United States
Show details
Authors:
Abdullah Al-Mamun | University of Texas at Dallas | United States
Felipe Pereira | University of Texas at Dallas | United States
Arunasalam Rahunanthan | Central State University | United States
In monitoring subsurface aquifer contamination, we want to predict quantities--fractional flow curves of pollutant concentration--using subsurface fluid flow models with limited data. We consider a Bayesian approach for the forecasting aquifer contamination, a research topic where the complexity associated with the simulation study in the forecasting presents an ongoing practical challenge. To reduce the computational burden in the approach, we use a Karhunen-Loeve expansion for the permeability field of the aquifer within a two-stage Markov Chain Monte Carlo (MCMC) method. Further reduction in computing cost is addressed by running several MCMCs. Running parallel MCMCs requires a careful study of the convergence of the chains. We first propose a fitting procedure for the Multivariate Potential Scale Reduction Factor (MPRSF) data that allows us to estimate the number of iterations for the convergence of the parallel chains. Then we present an analysis of ensembles of fractional curves suggesting that, for the problem at hand, the number of iterations required for the convergence through the MPRSF analysis is excessive. Thus, for practical applications, we provide a criterion to terminate MCMC simulations for a reliable forecasting.
10:10
Uncertainty Quantification for the inverse atmospheric dispersion problem with optical measurements
Robert Lung | University of Edinburgh | United Kingdom
Show details
Authors:
Robert Lung | University of Edinburgh | United Kingdom
Nick Polydorides | University of Edinburgh | United Kingdom
We consider the inverse problem of fitting atmospheric dispersion parameters of plume models from time-resolved back-scattered Lidar data. The obvious advantage of light-based remote sensing modalities is their extended spatial range which makes them less sensitive to strictly local perturbations/modelling errors or the distance to the plume source. However, the complex behaviour of photons in heterogeneous scattering media makes this inverse problem computationally a lot more challenging than those related to point measurements of gas concentration. Motivated by environmental emergency response applications and the need to solve the problem in nearly real-time, we address this issue by proposing a method that utilises, instead of the conventional voxelized representation, a non-linear global image parameterisation that avoids a high-dimensional ill-posed inverse problem that is usually encountered in 3D image reconstruction allowing us to consider arbitrary orders of scattering. The obtained parameters are directly related to the dispersion model which means that any point estimate or UQ can be associated with meaningful physical units. The latter is an important aspect of the inverse problem as measurements of scattered light will usually have a relatively low signal-to-noise ratio, hence any reconstruction is significantly affected by the Poisson noise that inevitably corrupts optical measurements.
10:30
- NEW - The optimization and UQ tool box Nodeworks
Aytekin Gel | Arizona State University | United States
Show details
Author:
Aytekin Gel | Arizona State University | United States
The open source tool box Nodeworks (\url{https://mfix.netl.doe.gov/nodeworks/nodeworks-applications/} ) is being developed and initial release was made in 2019. This toolbox enables the user to build their custom workflows for UQ analysis through the use of an intuitive graphical user interface and with Python backend using both available libraries and also recently integrated with PSUADE UQ toolkit from Lawrence Livermore National Laboratory.
The features of Nodeworks are going to be demonstrated for an industrial multiphase flow problem.