Peak-over-threshold flood frequency analysis of streamflow series for insular Newfoundland

Taylor, Kenneth Gordon (2001) Peak-over-threshold flood frequency analysis of streamflow series for insular Newfoundland. Masters thesis, Memorial University of Newfoundland.

[English] PDF (Migrated (PDF/A Conversion) from original format: (application/pdf)) - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (6MB) [English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. (Original Version)

Abstract

In this thesis, regional models for the prediction of flood quantiles for streams on the island of Newfoundland are developed using historical streamflow data which has been subject to peak-over-threshold analysis. The Peak-Over-Threshold method of flood frequency analysis allows extraction of more relevant data from a historical flow series than would be available using the conventional annual maximum flow method. As a result, the peak-over-threshold method is of particular interest in regions where data on streamflows is limited. This is the case in Newfoundland. -- Streamflow series from 63 rivers on the island of Newfoundland are considered. This data is modelled using a Poisson arrival process and the Exponential and Pareto magnitude distributions. Results from single-station peak-over-threshold analysis are compared to those obtained from the annual maxima series modelled using the 3-Parameter Lognormal and Generalized Extreme Value distributions. The island is divided into hydrologically homogeneous regions. Hydrologically homogeneous regions are defined as geographic areas in which flood flows are identically distributed except for scale. Regional index flood estimators are developed using the data generated from the peak-over-threshold approach. -- For the quantile estimates generated for the 63 data series analysed, there is no statistically significant difference between the central position of the results of the 3-Parameter Lognormal, Generalized Extreme Value, Poisson-Exponential, and Poisson-Pareto models. Model error for the single station analysis is tested using a bootstrap approach. For the standard error of quantile estimates generated by resampling, the Poisson-Exponential Distribution model exhibited comparable standard error for lower quantiles and lower standard error for higher quantiles. Because of this, the Poisson-Exponential model was determined to be the most robust for a variety of quantiles. Although the Poisson-Pareto distribution is more flexible, it appears to be inferior to the Poisson-Exponential model in this case. -- Regional models were developed using an index flood approach. The index flood was taken as the two-year return period flood, Q(2), and regional estimators for index floods for each region were developed by non-linear regression on physical basin descriptors. Regional models developed using nonlinear regression exhibited better fit to the underlying data than did the models produced using the traditional log-linear method. The nonlinear models exhibited lower bias, and also less estimation error. The ratios of Q(T)/Q(2) were easily calculated, and allowed estimation of flood quantiles for stations in the regions with a reasonably good fit to the expected values. For most regions RMSE was less than 10% of the mean of the expected values. The estimated values from application of the index flood technique tended to overestimate the quantile slightly and results were somewhat positively skewed from expected values. This will tend to produce more conservative (higher) estimates of flood quantiles. -- In the Southwest Region the equation which performed best (generated estimates with the lowest error) relied on three basin descriptors. The number of gauge records available in this region was only six. The coefficients developed for this equation are also somewhat suspect as they suggest a significant scaling of the result. In this region, the use of the whole island equation may provide a more reliable result and is recommended. -- Quantile estimates generated using the index flood method produced the poorest results in the Northwest Region. However, results were still reasonable and at lower quantiles, the RMSE was less than 10% of the mean expected value. When the estimators derived for the whole island were applied to this region they produced slightly better results. Thus with the exception of the Northwest Region, the use of regional index floods produced improved quantile estimates when compared to the estimates produced by equations developed for the whole island.

Actions (login required)

View Item