Abstract

Sensitivity analysis (SA) systematically assesses and quantifies a model’s uncertainty by examining the impact of parameter variations on the model’s output. An objective of sensitivity analysis is to ascertain model-input parameters that contribute the most to the uncertainty of a model’s output and could serve as a gauge to determine whether a parameter’s variation is suitable, expected, or a realistic representation of the phenomenon being modeled. Incorporating sensitivity analysis for assessing the importance of a model feature is imperative for decision-making. Therefore, as a statistical methodology, SA’s application in investigating coastal regions that are vulnerable to sea-level rise and extreme weather events—exacerbated by climate change—is crucial to effectively plan for and mitigate the consequences that follow. Quantifying the uncertainty of future forcings of coastal hazards and their impacts on risk estimates would better inform decision and policy making. Using random forest (RF) regression, this research provides a framework for assessing uncertainty in coastal risk predictions by proposing an extension of permutation variable importance for second-order interactions of model parameters. The second-order permutation variable importances assesses the importance of a combination of features by computing the mean-squared-error between the model’s true value and its prediction computed over randomly permuted spaces that include the combination of features itself. The second-order permutation variable importances are tested on a simple mechanistically-motivated emulator, using a semi-empirical model and dataset for global mean sea-level (GMSL) change and a coupled sea level-coastal impacts model for the U.S. Gulf Coast. Our results demonstrate and provide another relationship between permutation variable importances and Sobol sensitivity indices.

Library of Congress Subject Headings

Coast changes--Forecasting; Uncertainty (Information theory); Permutations; Variables (Mathematics)

Publication Date

5-2-2024

Document Type

Thesis

Student Type

Graduate

Degree Name

Applied and Computational Mathematics (MS)

Department, Program, or Center

Mathematics and Statistics, School of

College

College of Science

Advisor

Mihail Barbosu

Advisor/Committee Member

Lucia Carichino

Advisor/Committee Member

Tony Wong

Campus

RIT – Main Campus

Plan Codes

ACMTH-MS

Download

Share

COinS