Small Island Developing States, including many low-lying atoll islands, are among the most vulnerable countries to natural hazards and climate change disproportionately amplifies this vulnerability. Hence, there is a strong need for disaster risk reduction and risk management. Further development and implementation of methodologies for flood hazard assessment of atoll islands contributes to this. The methodology proposed in this thesis was applied to Majuro, an atoll island and capital of the Republic of the Marshall Islands. More specifically, the flood hazard related to different flood drivers, and including compound events (i.e. the combination of coastal flooding and precipitation) was assessed for the densely populated Delap, Uliga, and Djarrit region, in the east of Majuro Atoll. Main flood drivers are waves during typhoon events and distantly generated (swell) waves, but precipitation and high water levels (mainly tide) are important as well. To include all possible combinations of these flood drivers, and to include the spatial variation in events, 1000 years of synthetic events was generated based on data for historical events. Accurate simulation of inundation depths was computationally unfeasible for all synthetic events. Hence, a method was developed to reduce the number of model simulations, without losing information on the probability of occurrence of each event. The main steps of this method seem applicable to many other study areas where many scenarios are needed to include all (combinations of) drivers. Main steps are: (1) Selection of representative events – by Maximum Dissimilarity Algorithm, based on parameters that characterize the events. Hereby, the most extreme events are included as well. (2) Simulation of inundation depths for the representative events – by use of Delft3D, SWAN, and XBeach models. The XBeach model included a module for rainfall (first application). (3) Weighted interpolation to obtain the inundation depths for the synthetic events – based on the same parameters as in step 1. Based on the inundation depths for 1000 years of synthetic events, flood maps for different return periods and flood drivers were derived. These provide insight in the flood hazard for the DUD region due to the different flood drivers and for different return periods. The importance of different flood drivers varies significantly per area. Generally speaking, flooding related to (swell) waves in combination with high water levels is more frequent, while infrequent typhoons lead to the most severe flooding. Precipitation is an important flood driver as well, and exclusion would lead to underestimation of the flood hazard. Analysis of inundation depths suggests that for many areas these are limited to a maximum, where after excess water drains to the ocean – mainly into the lagoon. The derived flood maps could be used as a base for assessment of flood risk, climate change impacts, and closely related freshwater availability. In relation to the latter, more in-depth understanding of the contributions of precipitation and coastal flooding to the total flood hazard is needed, as on the long term infiltration of precipitation seems favourable, while that of oceanic water is not.

Today's coastal zones are densely inhabited as the majority of the world's population lives in these attractive areas. The shorelines in coastal zones are shaped by complex spatial and temporal variable interactions between natural forcings like changes in mean sea-level, tides, wave and wind conditions, and storm surges. Besides, natural hazards such as coastal erosion, tropical cyclones, hurricanes, typhoons, floods, salt intrusion and tidal surges threaten a major part of the world's population. Furthermore, climate change is likely to increase the risk of natural hazards. As a response, humans changed the world's shorelines and the forcing-driven processes that work on them, to increase protection against hazards and keep supporting their activities. These human interventions are deployed discontinued in time, disperse spatially and might result in negative consequences leading to human-induced hazards. This research focuses on one particular hazard to coastal communities, coastal erosion, which is extended to a term referred to as shoreline evolution by incorporating coastal accretion as well. Up till now, detailed local-scale studies are able to expose human and natural drivers of shoreline evolution and provide a possibility to make a step towards intentional rather than accidental coastal engineering. Digital imaging and, more recently introduced, satellite imagery proved to be a promising new technology to measure and monitor shoreline evolution at bigger temporal and spatial scales. Nevertheless, the opportunity to develop a model that exposes the drivers of shoreline evolution on a planetary scale remains unexplored. This is due to the required computational effort as well as the large variability in coastal systems around the world. With an ever-increasing data availability, data-driven models incorporating Machine Learning (ML) proved to be an efficient alternative approach to heavy computing classical process-driven models in civil engineering practice. Next to this, a coastal classification can be used as a means to inventory the aforementioned variability. Therefore, the research objective in this study is to explore the possibility of exposing and classifying the drivers of shoreline evolution on a planetary scale, by employing ML on satellite imagery. Approximately 390.000 km of shoreline is analyzed for the past 33 years. This resulted in a statistically derived classification of natural and human-induced sandy shoreline evolution. By elaborating on this classification, it is found that natural and human-induced shoreline evolution accounts for approximately 16 and 25% of the total of globally exposed and classified shoreline evolution signals respectively. All outcomes in this research can support detailed local scale investigations and therefore provide an enhanced opportunity to make a step towards intentional rather than accidental coastal engineering. Hence, it is concluded that the developed and applied methods that employ ML on satellite imagery can be used to expose and classify (in)direct human and natural influences on sandy shoreline evolution using spatial and temporal characteristics on a planetary scale. 57% of the shoreline evolution signals is present in a regime with complex and combined (compound) influences, which still requires (rather than supports) local scale investigations to determine the correct influence or driver. More research is required to elaborate on the opportunities that can enhance insight in the compound regime, to improve the obtained results and advance the applicability of this study.

Many tropical, coral reef-lined coasts, are low-lying with elevations less than five meters above mean sea level. Climate-change-driven sea level rise, coral reef decay and changes in (storm) wave climate will lead to greater chance and impacts of wave-driven flooding, posing a heavy threat to these coastal communities. Early warning systems (EWS) are effective for risk management and disaster reduction, however, the vast majority of the world's inhabitants of coral reef-lined coasts have no such system in place. Unfortunately, the complex hydrodynamics and bathymetry of reef-lined coasts make it difficult to establish a global flood prediction model for these areas. This thesis aims to develop a set of 'cluster profiles' that can be used to accurately represent coral reef-lined coasts around the globe. By representing an expansive variety of reef morphology, the cluster profiles are capable of predicting the wave runup over thousands of different coral reef profiles with a fraction of the number. The cluster profiles could be input into a tool such as a Bayesian probabilistic network which can be trained to provide real-time wave runup and flooding predictions given local bathymetry and offshore wave conditions, thus establishing a simplified global flooding EWS. The methodology includes two stages of data reduction. First, cluster analysis techniques are used to group thousands of coral reef profiles into 500 clusters based on morphology alone. Second, agglomerative hierarchical clustering is used to further group the profiles with similar morphology and wave runup response, resulting in a final set of 311 to 45 cluster profiles. Here we show that the cluster profiles are capable of predicting the wave runup for a set of 1000 reef profiles with a mean relative difference of approximately 10\%. The comparison was done using the numerical wave model XBeach with four different wave conditions. The methodology has been developed such that it could be expanded to other coastal environments. A summary of the methodology used in the study is illustrated on the following page.