Improved pooled flood frequency analysis using soft computing techniques

Pooled flood frequency analysis can be used to improve flood quantile estimation at catchments with short streamflow records. Commonly used pooled flood frequency analysis methods are the index flood method and the quantile regression method. The index flood method assumes that distributions of flood peaks for different sites in a pooling group are the same aside from a site-specific scaling factor. Applying the index flood method for flood frequency analysis involves two major steps: delineation of homogeneous pooling groups and deriving the pooled growth curve. Regression methods can be used to build simple models to predict flood quantiles, or the index flood, as a function of site characteristics.; A number of soft computing techniques are introduced in this thesis to deal with several issues in pooled flood frequency analysis and to generate improved flood estimates. The methods developed in this thesis are applied to selected catchments in Great Britain. Most flood events in the study region occur in the winter season. Most areas, except in the mountains, have moderate climate as a result of the surrounding temperate ocean, and snow events are rare. According to the criteria used for catchment selection, 424 catchments are selected. The average record length of annual maximum flood series is 24.3 years at the sites in the study region.; The artificial neural network ensemble method is introduced to estimate the index flood and flood quantiles. The method can be used to obtain flood estimates at an ungauged site. A review is given of popular ensemble methods. Six approaches for creating artificial neural network ensembles are applied in pooled flood frequency analysis for estimating the index flood and the 10-year flood quantile. The results show that artificial neural network ensembles generate improved flood estimates and are less sensitive to the choice of initial parameters when compared with a single artificial neural network. Factors that may affect the generalization of an artificial neural network ensemble are analyzed. In terms of the methods for creating ensemble members, the model diversity introduced by varying the initial conditions of the base artificial neural network to reduce the prediction error is comparable with more sophisticated methods, such as bagging and boosting. When the same method for creating ensemble members is used, combining member networks using stacking is generally better than using simple averaging. An ensemble size of at least 10 artificial neural networks is suggested to achieve sufficient generalization ability. In comparison with parametric regression methods, properly designed artificial neural network ensembles can significantly reduce the prediction error. (Abstract shortened by UMI.)...