|
Abstract:
|
Hydrologists have long used unmixed distributions to estimate flood frequency distributions, and have recently proposed improvements to their methods, including use of extreme value distributions. This presentation looks at alternative approaches to this type of analysis to account for multiple mechanisms for flood events. We study finite mixture models for 3-parameter log gamma (log Pearson Type III), normal, lognormal and extreme value distributions. These techniques will then be used on flood frequency data from watersheds in South Carolina to estimate, e.g., the 1% chance flood for these estuaries. We will review standard error estimates for these methods and propose new approaches.
|
