Bayesian multilevel model calibration of a simplified dam breach model

Abstract

This master thesis aims at giving an overview of how uncertainty quantification techniques can be applied to the specific problem of dam breaching. A simplified physically-based dam breach model is chosen which calculates the peak discharge depending on the geometry of the dam and the breach, the shape of the reservoir and the initial conditions. All material properties are represented by one calibration parameter which is located within the model equations. It allows one to calibrate the model according to measurement data. To quantify the uncertainties, a comprehensive database is built and the relevant quantities are analysed with respect to general differences between dam types, dependences between the input parameters, and their marginal distributions. The calibration parameter is supposed to be lognormally distributed. A sampling of the posterior distributions of the corresponding hyperparameters is created with Bayesian multilevel model calibration. The algorithm is based on Markov chain Monte Carlo simulation and kernel density estimation. The posterior samples of the hyperparameters showed that the mean value converges to one value whereas the behaviour of the standard deviation is erratic with a concentationaround zero. For better convergence, it is suggested to gather more data, to try other prior distributions and to test additional parameteric and nonparametric distributions for the calibration parameter. Considering the complex uncertainty set-up, the algorithms performed satisfactory.

Keywords

Bayesian updating techniques, calibration, dam breach, hierarchical model, simplified physically-based model, stochastic inverse problems, uncertainty quantification

BibTeX cite

MSCTHESIS{AEicherThesis,
author = {Eicher, A.V.},
title = {BAYESIAN MULTILEVEL MODEL CALIBRATION OF A SIMPLIFIED DAM BREACH MODEL},
school = {ETH Zurich, Zurich, Switzerland},
year = {2014}
}