Uncertainty propagation and sensitivity analysis in hydrology

Abstract

This project presents an application of uncertainty quantification techniques, namely polynomial chaos expansions (PCE) and sensitivity analysis (SA), in a hydrology context. More specifically, a Sobol’ analysis on a subset of the parameters of the physically based and fully distributed rainfall-runoff model TOPKAPI-ETH (TE) is performed.
The analysis uses a PCE-based approach, and for a relatively low amount of model evaluations is shown to perform reasonably well for all the investigated hydrological quantities. These are a total of 30 scalar values obtained from roughly 1000 TE model runs using a 3-year simulation time span, and describe: magnitude and volume of discharge flood events, 7-day low flows, average snow cover and evapotranspiration averages. For each one a PCE was computed using the (hybrid) LAR algorithm, and afterwards the Sobol’ indices were obtained from post-processing of the coefficients of said PCE.
The results obtained can help hydrologists start the calibration process while appropriately considering the importance of all the model parameters.

Keywords

Uncertainty quantification, rainfall-runoff models, polynomial chaos expansions, surrogate modeling, sensitivity analysis

BibTeX cite

MSCTHESIS{RArrigoniThesis,
author = {Arrigoni, Riccardo},
title = {Uncertainty propagation and sensitivity analysis in hydrology},
school = {ETH Zurich, Zurich, Switzerland},
year = {2020}
}