Analysis of Mixed Distribution Statistical Flood Frequency Models and Implications for Dam Safety Assessments

Restricted (Penn State Only)
Author:
Roop-Eckart, Kenneth Joel
Graduate Program:
Geosciences
Degree:
Master of Education
Document Type:
Master Thesis
Date of Defense:
May 31, 2018
Committee Members:
  • Klaus Keller, Thesis Advisor
  • Elizabeth Ann Hajek, Committee Member
  • Tess Alethea Russo, Committee Member
  • Christopher Duffy, Committee Member
Keywords:
  • mixed distributions
  • dam
  • safety
  • risk analysis
  • flood frequency
  • two population
Abstract:
Water is a vital resource, but also a source of hazards. Flooding poses considerable hazards to human lives and property. Dams and levee systems are key components of modern flood defenses. However, these flood defenses can fail catastrophically. This thesis addresses mixed distributions in statistical flood frequency analysis and implications for dam safety assessments. Previous studies in dam safety assessment have established a variety of statistical and physical modeling methods (Swain et al., 2006). Statistical flood frequency analysis represents a popular, low cost method (Swain et al., 2006). However, current flood frequency methods can neglect mixed distributions and under predict true flood risk. Here, I improve on the standard flood frequency methods (England et al., 2018) by: A. implementing single and mixed distribution models to assess flood frequency analysis sensitivity to model choice and model structural uncertainty, B. statistically test for the presence of mixed distributions in peak flow data, and C. demonstrating the implications of accounting for mixed distribution peak flows in dam safety assessments. I find that current methods in flood frequency analysis can lead analysts to disregard mixed distributions of peak flows. Goodness-of-fit metrics can be used to identify mixed distributions of peak flows at a location. Additionally, implementing mixed distribution statistical flood frequency analysis at mixed distribution peak flow sites can produce better fits (as judged by statistical tests) and can greatly increase predicted flood risk. These findings have potential safety implications for flood-frequency analysis based dam safety assessments.