Coastal regions are at risk of flooding because of their natural layout. Evidence of a changing climate, like sea levels rise and more extreme weather events, along with growing populations and cities, can make the impact of floods on society even greater. Additionally, estuary regions are threatened by compound floods, i.e., flood events generated when multiple physical drivers, e.g., the water level and river discharge, interact, even if each driver on its own might not seem threatening.

Along the Chinese coastline, particularly in the south, cities like Shanghai and Haikou are prone to flood, especially during the typhoon season. When typhoons hit the coast, high storm surges and heavy rainfall can interact leading to severe impacts. Characterizing compound flooding in coastal regions poses different challenges, including identifying the physical drivers that potentially generate a flood event, selecting an appropriate numerical model to describe the interaction between these drivers in terms of frequency and magnitude, and ensuring the quality and representativity of the available observations.

This thesis aims to tackle these challenges using cities along the Chinese coast as case studies. It seeks to (i) provide a probabilistic characterization of the physical drivers of compound floods, considering the effect of sea level rise, and (ii) integrate this quantification with a hydrodynamic model to assess the extent and depth of compound flood impacts in inundated areas. This approach can lay a solid foundation for developing flood-resilient strategies and mitigating potential impacts.

Chapter 2 introduces a design approach via conditional probability for quantifying compound flood hazards in coastal regions and its implication for infrastructure design considering the seasonal variation in surge peak occurrence. We found that along the southern coast of China, the severity of the expected rainfall events in case of a storm surge peak is larger compared to the expected severity inferred from the probability distribution of annual maxima of precipitation. Consequently, from a design perspective, implementing rainwater storage systems and facilities to mitigate hydrograph peaks is crucial for these regions.

Chapter 3 focuses on Shanghai and investigates how relative sea level rise (RSLR) affects design values for flood protection systems. We employed the D-Flow FM ocean storm surge model to reconstruct 210 historical typhoon storm surge events in Shanghai to overcome the constraint of unavailable water level records. We then applied a copula-based approach to calculate the joint probability and design value of peak water level and accumulated rainfall with the impact of RSLR. This research improves our understanding of how storm surges, rainfall, and RSLR interact, revealing how they collectively contribute to the risk of flood in coastal areas. Thus, it is crucial to monitor and predict the interplay of these factors for developing future design standards for better flood preparedness.

Chapter 4 reveals distinct patterns in the relationship between flooded areas and volume for both single-driven and multi-driven flood scenarios at the coastal city of Haikou by implementing an ocean storm surge generator and urban overland hydrodynamic model. The results highlighted storm tide (a combination of surge and astronomical tide) as the predominant factor contributing to compound flooding in Haikou. Only examining single-driven factors would underestimate flood hazard.

Chapter 5 investigates the sensitivity of inundated areas to the relative timing between the occurrence of the rainfall peak and the storm surge peak in Shanghai and provides a characterization of the consequent inundated areas based on the main flood driver(s). This is achieved by inferring from the probabilistic model the severity of the expected pairs of storm surges and rainfall events. They are then used as forcing of a hydrodynamic model to generate flood extent. We showed that the relative time between the peak of flood drivers affects the extent and depth of the flood and the flood zone classification. This can better suggest potential strategies for dealing with different types of compound flooding for coastal cities.

Coastal cities and communities are threatened by Sea Level Rise (SLR). Designing adaptations to protect against the rising sea requires a novel approach. With changing conditions, a broader approach considering multiple climate scenarios is required. A city facing an increasing threat from sea levels is Venice, one of the UNESCO world heritage sites. To protect the historic Italian city against floods, the MOSE barrier was constructed by the Italian government.
In 2020, the MOSE barrier was used for the first time. The mobile barrier closes when high water levels are foreseen in the city, preventing floods. Due to SLR, it is anticipated that the barrier will have to close more in the future, leading to questions about the functionality of the barrier when water levels rise. This might require adaptations or alternatives for the MOSE barrier.

The main objective of this thesis is to explore the potential of the framework proposed by Rosner et al. (2014) to evaluate the economic feasibility of an adaptation strategy against sea level rise in the city of Venice.
This framework assesses the potential regrets (monetary losses) on the decision to invest in an adaptation strategy accounting for errors in the evaluation of the level of sea level in the future. The framework incorporates trends based on historical sea level observations, including the uncertainties around these trends.

First, different methods of calculating design values under varying conditions are compared. This comparison is focused on different non-stationary Extreme Value Analysis (EVA) distributions. Due to SLR, a stationary and thus constant situation is not valid for Venice. This leads to the choice of the Rosner framework since it evaluates the feasibility of adaptations under varying conditions.

The regret of adapting or not adapting to a trend or SLR scenario is calculated from the damages associated with that scenario and the costs of the adaptation. The damages are computed by calculating the number of MOSE barrier closures for the different scenarios and multiplying this number by the costs that are related to a closure event. The adaptation that was chosen for evaluation is raising the entire city by 30 cm. This is accomplished by injecting seawater into a deep soil layer underneath the city.

The analyses resulted in smaller expected regrets when the choice is made to adapt. This was the case for all SLR scenarios, the difference between the adapt and not adapt regrets is larger for higher levels of SLR. This includes a higher SLR scenario due to a trend with large uncertainty. This is under the assumption that the adaptation, lifting the city, will work, the technical feasibility of this method was not investigated in this thesis.
A more precise calculation of the number of closures is advised. However, it is evident that adaptations are required to keep the MOSE barrier functional and the city of Venice safe from high waters. Raising the city will allow more time to evaluate which alternative to the MOSE barrier is best suited for the future.

The Netherlands, Germany, and Belgium were hit by heavy and prolonged precipitation in July 2021. As time passed, weather warnings escalated, leading to evacuations due to predicted floods, including in the Rur catchment. It was difficult to forecast the flooding of the Rur, raising the question of which elements are crucial in a flood predictionmodel for the Rur river. This question is addressed by addressing both a hindcast of the 2021 flood event and creating forecasts based on the weather forecast of July 13, 2021.

The Rur river basin is characterised by topographic and geological variations, with the steep Eifel responding differently than the flat lowlands, and human intervention in the form of reservoirs and lignite mines. A hydrological Wflow_SBM model has been derived for the Rur river basin, encompassing these characteristics, along with a hydrodynamic ProMaIDes model for the downstream reach of the Rur. These models were compared to investigate various aspects: river routing, floodplain flow, tirbutary interactions, the influence of reservoirs, and the impact of reduced groundwater levels.

The results of the 2021 floods indicate that modelling flows in floodplains is crucial to shaping the flood wave, both in tributaries and the Rur itself. Additionally, the reservoir played a significant role in attenuating the flood wave, with the increase in the outflow of the reservoir primarily affecting the tail of the wave. The reduced groundwater level was simulated by adding a leakage termto the saturated subsurface zone, whose indirect effect is significantly greater than the leakage termitself. Moreover, the tributaries Worm and Inde, particularly, are influential in the Rur’s discharge. These characteristics are also evident in the simulated forecasts, although the spatial and temporal resolution is significantly lower for these meteorological predictions.

Finally, the characteristic response of the Rur demonstrates that not everymodel type is equally practical for flood forecasting. The dominant flow from the reservoirs is highly regulated and is unlikely to induce inundations downstream. Complex flow patterns in floodplains only become relevant in the Dutch Rur, which makes two-dimensional modelling particularly valuable here. Therefore, it is recommended to use a one-dimensional discharge model, incorporating delay effects from winter bed flows. When predicted discharges at the Stah station are exceeded, two-dimensional simulations may provide a solution, the model area reduced to the Dutch Rur, focussing on predictions where a critical value related to floodplain capacity (Qlimit = 300 m^3/s) is exceeded.

The projected increase in sea level is expected to increase the intensity of coastal flooding threatening communities living along the coast. This, in combination with population growth and urban expansion, calls for a better understanding of Extreme Water Levels (EWLs), the mechanisms generating them, and their components, i.e., astronomical tide and storm surge, since they drive the maintenance and design of flood protection systems. Netherlands' flood defense is crucial in facing the risk of flooding given its particular geographical configuration, its large number of inhabitants, and its high value of assets. For this, a better understanding of EWLs and their components is essential to assessing the quality of current structures and developing new adaptation strategies since they drive design and risk assessment procedures. Hence, in this paper, we investigate EWLs in Hook of Holland which represents a strategic location due to the inlet of the port of Rotterdam and the Maeslant storm surge barrier. Here, we present a stepwise procedure that starts by defining EWLs, assessing drivers of storm surges on observed sea levels via spectral analysis and coherence, and ends in estimating the statistics of EWLs based on multiple approaches, i.e., univariate extreme value analysis, copula functions, and Joint Probability Method (JPM). The results show that storms in the Southwest Delta have a duration of about 4 days and that EWLs components, i.e., surge and astronomical tide, present negative dependence (the Kendall's tau $\tau = -0.50$). From the comparison between statistical approaches to model EWLs and infer design values, results show that copulas and JPM lead to an overestimation of EWL. However, EWLs modeled via copulas fit better low quantiles.

Floods are the most frequent natural disaster and due to climate change the frequency and intensity of these events are increasing. Therefore, it is becoming increasingly important to obtain accurate estimations of extreme discharges. Statistical modelling is widely used to estimate extreme discharges by fitting observed extreme discharges to an extreme value distribution. However, limited historical data makes it difficult to confidently model the tail behavior of extremes. Additionally, several modelling assumptions impact extreme discharge estimates including selection of the nonstationary method, extreme value distribution, parameter estimation method, and the impact of seasonality. In an effort to reduce uncertainties, a new method has been developed to derive design discharges for the Meuse in the Netherlands. This method, GRADE (Generator of Rainfall and Discharge Extremes) consists of three components: a stochastic weather generator, a hydrological model, and an extreme value analysis (EVA). However, the stochastic weather generator is not capable of producing daily rainfall that exceeds the range of historical data. Therefore, a physically based climate model, RACMO, is now being studied. RACMO is capable of generating 1,040 years of synthetic meteorological data that can be routed in a hydrological model to obtain 1,040 years of synthetic discharges. The physically based climate model makes it possible to capture the underlying physical processes of extreme events and the hydrological model can provide discharge information at locations where there are no observations. This thesis evaluates the impact various modelling assumptions have on estimated discharges using synthetic data generated by the RACMO through application of a case study in the Meuse.

Understanding the factors that drive extreme water levels is key to an accurate assessment of flood hazard. The city of Venice has always been affected by flooding due to extreme water levels. In this study, we examine the factors driving and influencing extreme water levels in the Venice lagoon, aiming at deriving accurate extreme water level estimates in the Venice lagoon.
Due to the shallowness of the Venice lagoon, extreme water levels are influenced by both atmospheric forcing (surge) and water level of the lagoon (tide and bottom level) and interactions between these two. Furthermore, these extreme water levels have been changing over time due to variations in the bottom level. These variations are reportedly due to local (anthropogenic and natural) subsidence and sea level rise. In this study we resort to the available long-term water level observations of the Punta della Salute tide-gauge. Given the effects of subsidence and sea level rise in these data, we start by homogenizing the data by removing these trends and jumps from the time-series. Using the homogenized time-series, we study the influence of the dependence between tide and surge components on the extreme water level estimates. Finally, we quantify the effect in the estimates of modelling this dependence in the extreme value models.
To homogenize the data and better understand the underlying trends, a time-series analysis was performed on the time-series of water level observations. Mann-Kendall tests for monotonic trend were performed, followed by an analysis using changepoint detection methods. Changepoint detection was performed using the RHtest and BEAST methods on the Punta della Salute time-series as well as time-series from neighbouring tide-gauge stations. Ultimately trend decomposition using the BEAST method was used to detrend and homogenize the Punta della Salute time-series.
After detrending, the tide and surge components were separated using tidal harmonic analysis and reconstruction. The relationship of these now separated components was quantified during extreme water levels using the Pearson r correlation and the Kendall τ rank correlation. This relationship between tide and surge was described using copulas to estimate extreme water levels. Different copula variants were evaluated and extreme water level estimates derived using copulas that describe dependence were compared to extreme water level estimates using a copula that describes tide and surge as independent components. Lastly, these were compared to those derived from univariate extreme value analysis to assess the influence of separation of tide and surge components combined with copulas as opposed to a more traditional univariate extreme value analysis.
The main conclusions of this study are as follows.
• The water level observations of the Punta della Salute tide-gauge are indeed affected by jumps and trends due to subsidence and sea level rise. These can be successfully removed using the applied techniques.
• There is a clear dependence between tide and surge in the Venice lagoon, with lower tide levels leading to higher surge levels. The non-inclusion of this dependence (by assuming independence) in the combined analysis of tide and surge signals to drive total extreme water levels leads to an overestimation of the total water level extremes.
• Extreme water level estimates from the combined analysis of the tidal and surge signal are higher, but compatible with those from the analysis of the total water level signal (without separation of tidal and surge signal). This gives confidence in the combined analysis accounting for the dependence between the signals and allowing for a further application of the models to account for projected climate changes.

Coastal and offshore infrastructure must be designed to withstand extreme wave-induced loading conditions. Extreme Value Analysis (EVA) is often employed to infer probabilistic distributions that provide information about extreme design conditions. In traditional practices, EVA is performed under the assumption of stationarity. This means that the probability of extreme events is constant in time. However, hydraulic loading conditions are expected to exhibit temporal variability in severity and frequency as a result of climate change. Therefore, the assumption of stationarity becomes questionable. Nonstationary extreme value analysis (NEVA) for inferring extreme hydraulic loads have become more attractive in recent years. However, the applicability of NEVA models is debatable ansd differs on a case-by-case basis. Large scale oceanic bodies can be characterized by spatially and temporally varying extreme wave characteristics. Clustering analyses have proven to be successful to identify regions exhibiting similar extreme wave characteristics. Creating clusters based on similar extreme wave characteristics can potentially improve extreme value modelling because intra-cluster information can be pooled to derive more accurate extreme value models.

This research presents a practical assessment of the applicability of clustering analysis and non-stationary extreme value modeling of extreme wave statistics at cluster level in the North Sea. The primary objectives of this research are: (1) Study the temporal variability extreme significant wave height (Hm0) and extreme wind speeds (U10) in the North Sea domain, (2) Investigate how hierarchical clustering analysis (HAC) can be employed to cluster grid points that exhibit similar extreme wave characteristics, (3) How the obtained clusters and temporal variability can be employed to derive extreme value models describing extreme Hm0 statistics at cluster level and (4) assess whether NEVA models at cluster level form a practical alternative compared to conventional stationary analysis in the design and risk assessment of hydraulic infrastructure in light of climate change.

Temporal trend analysis of Hm0 in the North Sea showed that the period between 1990 and 2020 can be characterized by a decreasing trend. Between 1950 and 2020, a decrease in Hm0 intensity is observed in the Western regions and an increase is observed in the East. This is reason to believe that the variability in extreme wave climate is cyclical rather than monotonic. There is reason to believe that temporal variations of extreme U10 are responsible for the temporal variability of extreme Hm0. Initial clustering results partition the North Sea domain into 50 clusters based on characteristic values for the significant wave height (Hm0), peak period (Tp), and dominant wave directions (θ1 and θ2). After splitting clusters based on geo-location and merging clusters based on the intra-cluster statistical properties of the wave parameters, 63 clusters are obtained. The identified clusters and temporal variability are used to define NEVA models describing extreme Hm0 statistics at cluster level. Intra-cluster Hm0 observations are detrended before fitting the GEV parameters by means of Bayesian Inference. Informative priors are constructed by pooling the GEV parameter information from the intra-cluster grid points. Potential non-stationarity is accounted for by adding the Theil-Sen parameters (b and b0) to the location parameter (μ∗), making the location parameter a linear function of time. The model parameters subsequently read: Hm0 ∼ GEV (μ∗ + (b · t + b0) , σ∗, ξ∗). Using the extreme Hm0 data from the clustering centroid yields the most promising results for describing extreme Hm0 statistics at cluster level under the condition that the intra-cluster exhibits homogeneous values for b and b0.

The applied HAC analysis presented in this research is not the optimal strategy. The identified clusters exhibit heterogeneous values for b and b0 Because non-stationarity ofHm0 was not accounted for during the HAC analysis. This hinders the performance of the NEVA models at cluster level. Also, whether the derived methodology can be applied for the long-term projection of future extreme wave events in the North Sea is debatable. The non-stationary of extreme Hm0 is best described by a cyclic pattern. Without a thorough understanding of the underlying causes of the non-stationary in Hm0 and without future projections of the extreme wave climate, the applicability of NEVA for deriving extreme Hm0 design conditions in light of climate change cannot be guaranteed.

Fluvial flooding poses a major threat to mankind and annually leads to major economic losses with many casualties worldwide. The consequences of this can be mitigated when accurate and rapid predictions are available when the water will arrive at which location. Current numerical simulations take a significant amount of time due to their computational demand, which is not affordable during a calamity where each second can make the difference between life and death. Literature has shown promising results in fluvial flood forecasting by using a deep learning model, which generally takes only a fraction of the computation time compared to conventional models. However, these models have thus far only been trained on static landscapes with changing boundary conditions, which prevents the model’s application elsewhere.

This research explores therefore the possibilities of creating and training a generic non-location bound deep learning model which can predict the spatial distribution of fluvial flood arrival times per grid cell. The architecture of the created deep learning model consists of five parallel encoder-decoders, which takes the elevation, slopes, elevated elements, land roughness, and initial water levels into consideration, depending on the dataset. The model is trained, validated and tested on four unique datasets, which consists of 30,000 flooding samples. The degree of complexity within a sample increases with each dataset number.

The average error for the four consecutive test datasets were 0.91, 1.41, 1.25, and 1.76 hours per cell. The differences in the predicted and the groundtruth are relatively small although the deviation tends to become larger at the end of the simulation, in regions with a strong gradient in arrival time, and in hilly and complex landscapes.

In addition, the model has been tested on various benchmark landscapes to examine specific flow phenomena, as well as a realistic test scenario in the Dutch dike ring 48. The model shows satisfactory performances for landscapes other than those present in the dataset, untill it encounters a feature on which the model was not trained for, such as undershots or irregularly shaped waterways.

This research has shown the potential of deep learning in predicting fluvial flood arrival times on unseen before landscapes. Recommendations for further studies include the use of an active dike breach and a variable flood location.

Floods and droughts, also known as hydro-hazards, are phenomena that generally involve detrimental consequences to society and environment. Traditional practices for risk assessment consider flood and drought independently. However, they are two opposite extremes of the same hydrological cycle. Omitting their interaction might lead to an under- or overestimation of the current and future risks associated with such natural hazards. In history, a number of drought-flood interactions have been observed in various parts of the world. Research to these drought to flood interactions is still in its infancy. Therefore, this research explores the concept of consecutive dry and wet (CDW) events in the Netherlands. The aim of this research is two sided. First, the Consecutive Events Graph (CEG) is introduced. This is a radar chart type of graph used to quantify spatial and temporal changes in dry and wet indicators in consecutive seasons. These can be used to identify hot-spots prone to opposite extremes. Second, a fully probabilistic framework based on Non-parametric Bayesian Network is developed to model the dependence between dry and wet indicators. Such model can be used to infer expected wet conditions in a given region when dry conditions are known.
For the CEG and probabilistic model a number of settings were introduced to quantify meteorological dry and wet extremes and to couple them spatial-temporally. First, a number of indicators were selected to quantify both type of extremes. Second, the dry period is defined in summer (June, July and August) and consecutive wet period in fall (September, October and November). Third, the Netherlands was subdivided in five homogeneous regions such that both wet and dry indicators were characterized on a regional scale. Maximum values of the indicators in its corresponding period were calculated for each single region and for every single year between 1965 to 2020.
This resulted in a dataset consisting of quantities for dry extremes in summer and wet extremes in fall for 5 unique regions over 56 years. Application to the CEG shows potential to identify and quantify CDW extremes. Region-to-region and year-to-year comparison is possible to quantify changes between years or regions. Application to the NPBNs disclosed limited interdependencies across the dry and wet indicators. Using the NPBNs for precise forecasting of expected wet conditions is deemed unsuitable as of low precision. Making inference of wet indicators based on hypothetical mild to extremely dry indicators revealed multiple trends of those wet indicators. These trends are increasing for short term precipitation indicators (R1D, R3D, and R5D) and simple precipitation intensity index (SDII) and are mildly decreasing for the total precipitation (Ptot).
Extreme dry events, extreme wet events and consecutive occurrences of these events are inevitable. It is expected that these phenomena will occur more frequently and become more severe due to a changing climate. A number of recommendations for future research is proposed. Findings from this thesis will help to smooth the path towards better understanding of the identification, quantification and interaction of CDW events or multi-hazard events in general.

Flooding is one of the most damaging natural disasters worldwide, and presents a signi_cant risk for a large amount of the global population. For the development of ood disaster management strategies, policy makers make use of ood hazard maps to inform investment strategies to reduce risk. In many current ood hazard mapping methods, the role of embankments is either implicit or completely unaccounted for. However, these defense systems can play a key role in both the _nal extent of the inundation as well as in the development of the inundation. Limitations in available data are often the reason for the lack of explicit implementation of embankments. The goal of this study is the development of a method that can explicitly include the e_ects of ood defenses in a ood hazard map based on limited available data. Next, this has been tested for a low-lying riverine area and the Tisa river basin in Serbia was selected as a case study area for this purpose. First, an idealized approach has been followed to better understand the model behavior using settings resembling this river. Finally, the method has been applied with a more realistic river schematization.
To develop a method for a more explicit inclusion of ood defenses, an understanding of the di_erent current approaches has been generated based on a literature review. On the scale of global ood maps, the failure probabilities of ood defenses are not used. Only with post-processing, certain areas are considered protected by removing inundation from the maps. Many studies of ood hazard mapping use the so-called bathtub approximation. Hereby is assumed that the complete oodplain will ood and that the inundation depth is found by extrapolating the water surface level outside of the embankments to the area inside of the embankments. The inuences of the ood defenses on the inundation are not included. Regional ood hazard maps are still made without ood defenses in many cases. While the improved resolution allows for more detailed maps, the lack of available data still limits the implementation of ood defenses. Flood mapping methods exist that include failure probabilities for ood defenses but these require large amounts of data that is not available everywhere.
An idealized model based on the Tisa river characteristics is used to test a ood mapping method which includes explicitly the presence and potential failure of embankments. Based upon data on the river geometry, hydraulics as well as land-use in the oodplains, a model schematization has been set up with the SOBEK software. The embankments were simpli_ed to a crown height for the purpose of overow and an estimated failure probability. The breaching of the levee is simulated according to the Verheij-vdKnaap breach growth model. The estimation of the failure probabilities was based on historical failure rates of a comparable ood defense system along the Elbe River. The levees were schematized into segments based on the maximum breach width of the breaching model…

The effects of climate change are felt all around the world. An increased sea level goes hand in hand with an increased risk of flooding. To combat this, the coastlines must be reinforced to withstand future sea levels. However, repeatedly reinforcing coastlines to keep up with the sea level rise (SLR) could prove extremely costly. An alternative approach would be to shorten the coastline, as the Netherlands did with the Afsluitdijk to enclose the IJsselmeer. Looking at Europe, Groeskamp and Kjellsson (2020) proposed the construction of the Northern European Enclosure Dam (NEED) — a dam that would disconnect the North and Baltic Seas from the Atlantic Ocean. In this way, it would protect fifteen northern European countries against the accelerated global mean SLR (GMSLR), as it simultaneously shortens the coastline that requires reinforcement. This thesis aims to determine whether the NEED (Adaptation Strategy 2) would be a financially favourable adaptation strategy over raising the coastal defences on a country-by-country basis (Adaptation Strategy 1) around the North Sea to combat future GMSLR, and if so, at which GMSLR.

Through the use of the GLOFRIS model framework it was possible to estimate that by 2080 a total of 15,000 km2 would be inundated, affecting 9.5 million people and resulting in damages up to 1 trillion € for all fifteen countries combined. Reinforcing the regional flood protections (Adaptation Strategy 1) is estimated to have a total cost range of 245 to 335 billion € for a 1-metre GMSLR with an increase in costs between 170 and 235 billion € per metre GMSLR. The construction of the NEED (Adaptation Strategy 2), using an earth-fill dam design with 1:6 slopes on either sides, is estimated to be just under 1.1 trillion €, with an increase in costs of 11 billion € per metre GMSLR.

It was found that the NEED flood protection adaptation strategy will eventually become more financially favourable over the regional strategy. Estimated to be more cost-effective beyond 5.15 metres GMSLR, which is associates with construction costs of roughly 1.15 trillion €. This GMSLR for scenario SSP5-RCP8.5 is expected to occur between 2280 and 2660, approximately. However, as the total costs are greatly contingent upon the core material, modifying slope angles of the NEED design will lead to a significant reduction in volume and, hence, costs. For the alternative designs with a 1:4 and 1:5 slope, the total costs are reduced by 17% and 34%, respectively. For these designs, the NEED will already become favourable at 3.35 and 4.25 metres GMSLR, respectively.

Several cost distributions have been created based on the four aspects that have been investigated, namely (i) coastline reinforcement length, (ii) size of inundated area, (iii) population exposed and (iv) economic damages caused by flooding. Together with the extrapolated regional costs per country, it is possible to determine which distribution is the most and least financially favourable for each country and whether contributing to the NEED is even favourable at all from the perspective of each country.

This research scrutinised the costs, effects and consequences of the most extreme scenario that generated the greatest exposure indicator values (i.e. SSP5-RCP8.5 combined with a return period of 1000 years). Through this
assessment, it was possible to estimate the costs associated with both adaptation strategies, and furthermore, to determine at which GMSLR one strategy surpasses the other in financial attractiveness and when this GMSLR can be expected. However, it should be noted that recent studies have shown that, in reality, the most extreme scenario might, unfortunately, turn out to be even more extreme than the most extreme scenario assumed in this thesis. And as the consequences strongly dependent on how climate change will unfold in the future, the costs to combat and the timing of such GMSLR occuring will differ.

The results retrieved from this research provide insight into when the NEED flood protection adaptation strategy will become a better alternative to regional flood protection reinforcement. However, it should be borne in mind that it is not a matter of ‘either-or’, but rather ‘both-and’, as regional dike reinforcement cannot entirely be omitted when deciding to construct the NEED. Instead, a balance must be found in the extent to which regional dike reinforcement is required to protect the countries while the NEED is under construction. So there are plenty of uncertainties and questions that require additional research to fully comprehend all the effects of this massive operation and making it feasible.

Extreme value analyses (EVA) are often used to determine the frequency of extreme events. The length of the available observations is an important aspect when performing EVA. It is generally known that more available data results in better estimates with less uncertainties. The main objective of this research report was to assess what the influence of the length of the observations is when inferring rare events. This was done by first analyzing the sensitivity of inferred return levels from synthetic data from three known distributions. Also, three case studies were analyzed to observe the sensitivity of inferred return levels and return periods from the observations. The results of the analyses were that a larger sample size generally leads to a higher confidence in the estimates of inferred return levels from synthetic data. However, there will always remain some uncertainty associated with the estimates. The confidence in the inferred return levels from observations also generally increase for an increasing sample size. However, this can not always be observed and other aspects can be more dominant than an increasing sample size. No correlation could be observed between the sample size and the inferred return periods. The conclusion of the research was that, while there is a positive correlation between the sample size and the confidence in the estimates, there will always remain some uncertainties. It is therefore important to always communicate the uncertainties associated with estimates.

The past decades, the increasing availability of data has paved the way for a new, data-driven generation of models. This research proposes a non-parametric Bayesian network (NPBN) to model hydrologic processes. The Bayesian network (BN) is a directed, acyclic graph in which the variables are represented by the nodes, and the conditional probability distribution between variable pairs is represented by the arcs. NPBNs are computationally less expensive than many conceptual hydrologic models and are sufficiently flexible to be able to handle different continuous data sources. The goal of this thesis is to make an NPBN for a lowland catchment and test its performance. The case study concerns the catchment of the Vledder, Wapserveense and Steenwijker Aa. This catchment makes this research the first one in which an NPBN is comprehensively implemented for (1.) a single catchment in which the catchment processes are modelled, (2.) a Dutch catchment, and (3.) a lowland, partially managed, catchment. For the BN model, seven hydro-meteorological variables have been selected for the model, complemented by the target variable, which is the monthly maximum daily average discharge (MMDAD). The aim of the BN is to be able to accurately predict the MMDAD, and the Kling-Gupta efficiency (KGE) acts as a performance indicator by which to optimize the BN’s parameters. For this thesis, the Gaussian copula was selected to be implemented for all variable combinations in the BN, because this type allows for the use of the multivariate normal distribution to calculate a conditioned network. The fit of the Gaussian copula to the data is tested in this thesis. This method is far more convenient than the alternative called the vine-copula method and most likely gives a better fit than the other alternatives. Three distributions are compared to model the marginal distributions, of which the Gaussian mixture model has been selected. This function extrapolated too little, so a novel alteration function has been proposed to shift the predictions. Several other parameters in the BN have been analysed as well. A sensitivity analysis has been performed to understand what influences of artificial errors would be. In general, random errors have a low influence on the prediction of the model, whereas new systematic errors have a larger influence. Criteria for a practical, well-performing BN have been presented and a strategy to create such a model that satisfies these criteria has been assembled. This strategy left the selection of some connection implementations up for interpretation. The chosen implementation has been decided based on which implementation produced the best predictions of the relevant variable within the network. The final model gave a median, k-fold tested KGE of 0.73 when predicting the MMDAD. It is also analysed how well the MMDAD is predicted if not all other variables are fixed. Another novelty is that a BN model is benchmarked against a SOBEK model, a neural network, and a multiple linear regression model. Compared to these models, the BN performs well. Moreover, all these other models lack some advantages that the unsaturated BN has.

The objective of this study is find out whether maximum daily discharge of the Geul and Rur catchments can be forecast using machine learning (ML) methods, and if so, to what extent. In addition, these ML models are compared to a conceptual model to see which performs better. A second objective is to test whether soil moisture content (SMC) and NDVI increase performance of the two ML models. The Geul and Rur catchments are both partly situated in the administrative area of Waterschap Limburg, a water authority in the Netherlands. They use discharge forecasts in order to prepare flood defenses and to monitor high water levels more closely. Currently, discharge is forecast using the conceptual HBV model for the Geul. Forecasting is done based on experience for the Rur and only in case of high water levels. Conceptual and physical models are based on physical laws, i.e. conservation of mass and energy. However, some relations are not yet fully understood, or are hard to translate to equations, and assumptions have to be made. This is why a data-driven approach is used, as no explicit relationship between variables has to be specified. In this study a gradient boosted decision tree framework (XGBoost) and a recurrent deep learning model (LSTM) are used to map input to output. XGBoost is a relatively new framework that has shown promising results in other water resources related studies. Long Short-Term Memory is a type of recurrent neural network and chosen for its ability to handle long-term dependencies and for its ability to model non-linearities. In order to see whether the machine learning methods outperform conceptual models, they are compared to the GR4J model. GR4J is a simple yet effective soil moisture accounting model. Beside SMC and NDVI, meteorological variables are used as input. Results show that the deep learning model performs best for simulating today’s discharge and when forecasting up to three days ahead. The GBDT model has a slightly higher Nash-Sutcliffe Efficiency (NSE) for the daily simulation of the Geul, but also a higher mean absolute error (MAE) compared to the deep learning model. The same holds for the three-day-ahead forecast for the Geul and the Rur. Peak timing is accurate for most models but peak discharge is often underestimated. When comparing the ML models to the conceptual model for the daily simulation, deep learning performs best in terms of MAE, but GBDT is better in terms of NSE. When looking at the one-day-ahead forecast, deep learning outperforms the GBDT and conceptual model in both NSE and MAE. In any case, when looking at the metric they outperform the conceptual model. However, the conceptual model has only a couple of parameters to calibrate, is transparent and has only two input variables. The ML models, on the other hand, have my parameters to train, are difficult to physically interpret and have four to five input variables. Besides comparing the two types of models, it is tested whether adding soil moisture content and NDVI as input improve performance of the machine learning models. The former undoubtedly improves performance, whereas NDVI at best improves performance as much as some other meteorological variables. Overall, this study finds that a conceptual model still outperforms the two ML models from a holistic point of view. However, machine learning is not yet fully exploited in water resources management. It already gives promising results and is likely to perform even better in the future.