Assessment of flood hazard over large geographical areas using data-driven approaches

The mapping of flood hazard can be time and resource consuming, but it is essential for assessing flood risk and for designing strategies to avoid consequences and recover faster in the event of flooding. This generally involves setting up complex numerical hydrologic/hydrodynamic models to simulate the flow of water in river channels and over the floodplains. Although such approach is considered standard, it is not always feasible. For example, it is challenging to simulate floods over large areas, produce a number of scenarios, represent flood mechanisms in a detailed way, and make use of all the data that is increasingly available in the field of water resources. Clearly, flood managers could use more options. Machine learning (i.e., algorithms that learn from data, in contrast to physically-based equations) has been seldomly used until now, but is a good candidate because of simplicity, typically faster runtimes and ability to handle large amounts of data. In combination with geographic information systems attractive tools can potentially be created. The combination of nearly instantaneous results with a web-GIS provides the possibility of near-real time analysis using any modern web browser. This thesis seek for additional clues that can help in the answering of the following questions: can data-driven models live to their expectations in flood hazard assessment? to what extent they offer viable alternatives to standard approaches and what are the concrete advantages and limitations? Several aspects of flood hazard assessment are addressed by developing and employing different state-of-the-art data-driven approaches, namely for the estimation and mapping of areas that may be subject to flooding across geographic scales, their downscaling, their extrapolation and regionalisation, or the transfer between catchments based on physical similarity. In each part of the thesis, the viability of selected methods are demonstrated and possible ways to overcome limitations are highlight.