Index flood model development using hydroinformatics

A reliable model for flood prediction is crucial for many aspects of engineering design in reducing flood risk impact to human lives, both economic and social. Despite an abundance of research on improving flood regionalisation modelling, there are still potential difficulties that require attention by the hydrological community. This thesis attempts to solve some problems concerning the index flood model development encompassing catchment information constraint, model selection process, calibration data issues, model structure and linearisation of regression models. These interrelated factors have largely affected the prediction accuracy of index flood models. First, catchment information data are problematic if acquisition of digital maps is a problem. In a country like UK, not all the digital maps and catchment information are available for free. As a result, freely downloadable digital maps from the internet are useful in deriving catchment information. Besides, by employing these freely available data in conjunction with the automated catchment delineation algorithm and a powerful tool of GIS for data derivation, a large number of alternative catchment characteristics can be produced. As such, the derivation method and catchment characteristics presented in this study are more applicable to other parts of the world. Second, in practice, there are a large number of catchment characteristics (or input variables in mathematical terms) that could be used to develop a model. The problem is how to choose a set of significant input variables as not all of them are important and some may be irrelevant and redundant. Furthermore, the process of input variable selection is time demanding since each validation will need a fully calibrated formula. The Gamma Test tool in conjunction with the cross validation method is proposed which has reduced model development workload besides gaining a reliable prediction model. Third, calibration data are one of the crucial aspects that generally affect the reliability of developed models. Uncertainty of model occurs regardless how many data points are used. This study provides a platform to those in a region where data availability is limited, to choose good quality calibration data so as to enhance prediction estimation. Fourth, model structure is another factor that affects prediction accuracy of the index flood model. This has been explored in this study to demonstrate that there is a room for further improvement over the existing power form models. Finally, a multiple regression approach is a popular method used to relate flood estimate to the catchment characteristics. Linearisation of nonlinear multiple regression model into logarithmic form is the most commonly adopted technique that eventually leads to the bias. To avoid this problem, model parameters are directly solved by nonlinear optimisation techniques.EThOS - Electronic Theses Online ServiceGBUnited Kingdo