Flood hazard mapping : uncertainty and its value in the decision-making process; Dissertation, UNESCO-IHE Institute for Water Education, Delft and Delft University of Technology.

Floods are natural events that can disrupt vulnerable societies and cause significant damages. Floodplain mapping, i.e. the assessment of the areas that can potentially be flooded, can help reduce the negative impact of flood events by supporting the process of landuse planning in areas exposed to flood risk. Flood inundation modelling is one of the most common approaches to develop floodplain maps. The recent literature has shown that hydraulic modelling of floods is affected by numerous sources of uncertainty that can be reduced (but not eliminated) via calibration and validation. For instance, many studies have shown that models may fail to simulate flood events of magnitude different from that of calibration and validation events. This can be caused by the fact that river flow mechanisms are nonlinear and are characterised by thresholds that demarcate flow regimes. One of the challenges in using uncertain outcomes is that decision makers (e.g. spatial planners) often have to take decisive binary actions, for instance, either to change the landuse (e.g. urbanize) or not. From the perspective of a modeller, one can provide precise (but potentially wrong) results based on both expert knowledge and the results of calibrated-and-validated models. However, this is neither prudent nor pragmatic, given that expert knowledge is variable and unavoidably subjective. As a matter of fact, different modellers using the same input data and models often attain different results. Thus, it is more scientifically sound to provide the results of flood inundation models in probabilistic terms. The objective of this thesis is to contribute to the scientific work on assessing uncertainty of flood inundation models and develops methods to better support the use of probabilistic flood maps in spatial planning. Thus the impacts of diverse dominant sources of uncertainty (such as input flood hydrograph, model parameters and structure) are assessed by focusing on reduced-complexity models of flood inundation dynamics. Subsequently, novel methods to incorporate uncertain model output in decision making, with respect to spatial planning in floodplain areas, are tested. More specifically, the thesis consists of two main (complementary) parts. The first part deals with the analysis of the major sources of errors in flood inundation modelling, which culminates in the production of probabilistic floodplain maps. The second part shows applications of utility based approaches to aid the decision making processes, when binary decisions are to be made on the basis of uncertain information. This thesis provides a contribution to the use of probabilistic floodplain maps in decision making, such as spatial planning under flood hazard uncertainty. Using historical hydrological data, 1D, 1D-2D and 2D flood inundation models are used to simulate flooding scenarios. These models are built for two case studies: (i) a mountainous river reach (River Ubaye, France) and (ii) an alluvial river reach (River Po, Italy). Topographic data are derived from frequently used sources of information of different precision and accuracy, namely SRTM (Shuttle Radar Topography Mission), EUDEM (Digital Elevation Model over Europe) and LiDAR (Light Detection and Ranging). In particular, four major components of uncertainty that affect flood modelling outputs are analysed. They include inflow uncertainty (flood discharge derived from a rating curve), parameter uncertainty, model structure and topographic data uncertainty. Input uncertainty was defined in two ways: (i) single segment rating curve parameter uncertainty and (ii) aggregated peak discharge uncertainty components. The boundary condition (inflow hydrograph) uncertainty was found to be considerably more significant than parametric uncertainty. Probabilistic flood hazard maps are generated using a Monte Carlo approach to capture the impact of these sources of uncertainty. Lastly, a new methodology for assessing the benefits of flood hazard mitigation measures (i.e. the KULTURisk framework as a result of an EU FP7 project) was used. The utility of probabilistic model output is then assessed using two approaches: (i) Value of Information, and (ii) Prospect theory. Implementation of these two approaches is based on the premise of a welfare trajectory, whereby the value of (and generated from) assets and investments in the floodplain accrue over time. Thus, the occurrence of a flood event results in damages that lower the welfare trajectory. Landuse in the floodplain can be altered based on the needs of the community as well as on potential flood risk. In this case, a higher investment yields higher returns, hence, implying a steeper welfare trajectory (and vice versa). A combination of gains of landuse change with a corresponding threat of flood damage (based on a probabilistic floodplain map) exemplifies the spatial planning dilemma that many decision makers have to deal with. In this context, this thesis has demonstrated that probabilistic model outputs can be successfully used to develop flood hazard mitigation strategies and support spatial planning in floodplain areas. Results also point to actual challenges in spatial planning where floodplain locations with higher consequences and uncertainty are identified as requiring additional monitoring