Comprehensive analysis of sustainable flood retention basins
To adapt to climate change which results in increasing flood frequency and intensity, the European Community has proposed Flood Directive 2007/60/EC. It requires member states to conduct risk assessments of all river basins and coastal areas and to establish Flood Risk Management Plans focused on prevention, protection and preparedness by 2015. Sustainable Flood Retention Basins (SFRB) that impound water are a new concept that arose in 2006. They can have a pre-defined or potential role in flood defense and were supposed to facilitate the implementation of the Flood Directive. Early and preliminary studies of SFRB were derived from case studies in Southern Baden, Germany. In Scotland, there are a relatively high number of SFRB which could contribute to flood management control. This research aimed to produce a guidance manual for the rapid survey of SFRB and to propose a series of frameworks for comprehensive analysis and assessment of SFRB. Precisely 372 SFRB in central Scotland and 202 SFRB in Southern Baden were investigated and characterized by 43 holistic variables. Based on this practical experience, a detailed guidance manual was created, guiding users to conduct a SFRB survey in a standardized and straightforward way. To explore the hidden data structure of data arising from the SFRB survey, various widely used machine learning algorithms and geo-statistical techniques were applied. For instance, cluster analysis showed intrinsic groupings of SFRB data, assisting with SFRB categorization. Principal Component Analysis (PCA) was applied to reduce the dimensions of SFRB data from the original 43 to 23, simplifying the SFRB system. Self-organizing Maps (SOM) visualized the relationships among variables and predicted certain variables as well as the types of SFRB by using the highly related variables. Three feature-selection techniques (Information Gain, Mutual Information and Relief) and four benchmark classifiers (Support Vector Machine, K-Nearest Neighbours, C4.5 Decision Tree and Naive Bayes) were used to select and verify the optimal subset of variables, respectively. Findings indicated that only nine important variables were required to accurately classify SFRB. Three popular multi-label classifiers (Multi-Label Support Vector Machine (MLSVM), Multi-Label K-Nearest Neighbour (MLKNN) and Back- Propagation for Multi-Label Learning (BP-MLL)) were applied to classify SFRB with multiple types. Experiments demonstrated that the classification framework achieved promising results and outperformed traditional single-label classifiers. Ordinary Kriging was used to estimate the spatial properties of the flood-related variables across the research area, while Disjunctive kriging was used to assess the probability of these individual variables exceeding specific management thresholds. The results provided decision makers with an effective tool for spatial planning of flood risk management. To assess dam failure hazards and risks of SFRB, a rapid screening tool was proposed based on expert judgement. It demonstrated that the levels of Dam Failure Hazard and Dam Failure Risk varied for different SFRB types and in different regions of central Scotland. In all, this thesis provided a guidance manual for rapid survey of SFRB and presented various effective, efficient and comprehensive frameworks for SFRB analysis and assessment, helping to promote the understanding and management of SFRB and thus to contribute to Flood Risk Management Plans in the context of the Flood Directive.