Improved Methods for Regional Flood Frequency Analysis

摘要

The overall objective of the research has been to develop flood frequency models and estimation methods for prediction of floods at gaged and ungaged watersheds. Several topics were studied under the project including, estimation of Pearson III model, properties and estimation of non-conventional distributions such as mixture models, and regional maximum likelihood estimation (MLE). In addition, applications to flood data in Colorado and Illinois regions were made. Population moments, moment-ratio diagrams and separation conditions of mixture and two-component product models were studied. New families of mixture of quantile models as well as estimation based on censored MLE for mixture models were developed. Likewise, moment and MLE procedures for estimating the Box-Cox transformation parameter were derived. Results based on simulation experiments indicated that regional MLE may be beneficial in estimating flood quantiles at a given site as compared to at site estimates or estimates derived by alternative least square methods, under the assumption of a reliable regional parametric structure. However, as the parametric structure becomes unreliable, the benefits of regional MLE become less apparent. Likewise, in comparing regional MLE with generalized least squares methods based on observed flood data, results indicate that both methods perform about the same.