Conceptual models in hydrology are widely used, allow for easy interpretation and require little data. Machine learning models in hydrology often outperform conceptual models but lack the ease of interpretability, require large amounts of data and and do not obey physical laws. Hybrid approaches aiming to combine the advantages of both approaches are becoming more popular. A Neural Ordinary Differential Equations approach is introduced to combine a differential equation-based conceptual model with a neural network. Additionally, conceptual models and LSTM models are used as benchmarks. The models are tested using the LamaH-CE dataset as well as the E-OBS dataset. In many cases the hybrid models outperform the conceptual model. However, to further improve the performance of hybrid models more research is needed to make the models more computationally efficient and optimized training strategies are required to explore the full potential of the approach.

Climate change-induced changes in weather patterns call for the development of hydrological models that perform well under increasingly extreme and varied conditions. Multiple research studies have demonstrated that hydrological models perform poorly when applied to climate conditions that differ from those during the calibration period of the model (Duethmann, Bloschl, & Parajka, 2020). The need for robust hydrological models was emphasised after the Geul catchment, and large parts of Belgium and Germany flooded in July 2021. During this flood, the hydrologists from Waterboard Limburg (WL) indicated that the current model would not have been able to correctly forecast the flood under such an extreme rain event anyway (Expertise Netwerk Waterveiligheid, 2021).

The underlying assumption in this research is that hydrological models cannot perform to the same standard during changing weather conditions because they are overfitted during the calibration period. Meaning that the parameter values are the mathematical best fit for streamflow predictions, but do not represent the internal hydrological processes between precipitation and discharge. One way of trying to improve the internal processes of a hydrological model is to create more relations between these processes and external measurements (Kirchner, 2006). In this research, a secondary calibration data set is used, namely evaporation data.

There are two goals in this thesis, first building a hydrological model of the Geul which represents the streamflow response of the catchment in a meaningful way. Second, comparing if a difference in calibration methodology, specifically comparing calibration on discharge with calibration on both discharge and evaporation, would improve the predictive power of the model. These goals are set up to be able to answer the main research question of this thesis:

Is the predictive power of a discharge calibrated hydrological model of the Geul catchment in the Netherlands, greater than the predictive power of an evaporation calibrated model, in addition to discharge?

The methodology used in this thesis to calibrate and evaluate is Generalised Likelihood Uncertainty Estimation (GLUE), with NSE efficiency as objective function.

The first goal, building a hydrological model of the Geul which represents the stream flow response of the catchment satisfactory, is successfully reached. The model calibrated on streamflow (Model Q) and the model calibrated on both streamflow and evaporation (Model QE) were able to predict streamflow satisfactorily, with DeQ scores between 0.632 and 0.649 across all runs.

The second goal was to compare the performance of Model Q and Model QE. Model QE outperformed Model Q on monthly runoff coefficient, monthly average evaporation and cumulative evaporation, increasing the NSE score of Model Q of the monthly runoff coefficient from 0.774 to 0.801, the NSE log score of the monthly average evaporation from 0.884 to 0.891 and the cumulative evaporation from 0.760 to 0.774. However, Model QE did not outperform Model Q on streamflow.

Therefore, it cannot be concluded that additional calibration on evaporation data increases the predictive power for the streamflow of the model. However, the model now more accurately represents the observed evaporation data, without trading this in for less predictive power of streamflow. Therefore, Model QE can be assumed to be a more accurate description of the hydrological processes in the Geul catchment than Model Q.

The Falsterbo Peninsula is a low-lying area that provides a home to 7,000 residents, as well as various bird and vegetation species. To protect this densely populated area, the municipality was granted a permit to build flood protections. As part of the strategy, the dune system in the study area will be used as natural barriers against storm surges. Despite being part of the protection strategy, the strength of the dune system in safeguarding the hinterland has not been assessed. This had led to the objective of this thesis, which aims to evaluate the strength of the current dune system. To achieve this objective, the following research question were formulated:

To what extent does the dune system on the Falsterbo Peninsula contribute to safeguarding the hinterland against the impact of historical storm conditions?

To seek answers to this question, the research was didvided into three parts. The first part of the methodology involves collecting the environmental data such as the wind, water level and wave data. Additionally, the data on the dune's morphology was collected during the field work. The second step of the methodology involves identification of extreme conditions within the time series spanning from 1959 to 2022. Considering the complex interaction between the waves and water levels, the extreme conditions were identified based on the combined effect of the two variables, which was represented in the total water level (TWL). The sampling method was based on the peak over threshold method applied to the time series of TWL. The choice of the threshold value was based on the scenarios of potential coastal flooding in the study area. The largest storm surge, the 1872 storm, was included in the analysis to evaluate its impact on the present dune system.

The dune erosion due to the selected extreme conditions was determined in the last part of the methodology. Two morphological models, the XBeach model and the storm impact model were employed to estimate the dune erosion in four transects withing the dune system.

The obtained dune erosion was expressed as a fraction of the available dune volume. The maximum dune erosion was found in the transect situated at the far-right end of the dune system when facing north. The maximum dune erosion under extreme conditions in the period 1959 to 2022, estimated by the XBeach and the storm impact model are 7.67% and 32.89%, respectively. Based on these results, it can be concluded that the present dune system is strong enough to provide protection to the hinterland against the impact of extreme condtions.

For the 1872 storm, the XBeach model estimated erosion percentage of 67.89%, whereas the storm impact model estimated more than 100%. This indicates that in the event of recurrence of the 1872 storm, a dune breach could be expected. While the 1872 storm may not be the design storm condition for the dune system, the storm impact model result highlights the need of reevaluation of the formulation of potential plans to reinforce the dune system.

For future studies, it is strongly recommended to establish a long-term monitoring program for the dune system on the Falsterbo Peninsula. The obtained dune erosion data can be used to calibrate the morphological models to enhance its accuracy in the predictions. Additionally, dune recovery data can aid in the understanding of the dune system as a whole.

The authorities of the Port of Genoa have requested the construction of a new vertical breakwater for the Sampierdarena canal, as well as the subsequent demolition of part of the existing breakwater. The reason for this project is the need to expand the size of the canal, to allow larger vessels to safely use it. A choice is made for a vertical-type breakwater, i.e., a concrete caisson placed on top of a rubble mound foundation. The main goal of this thesis is to provide a preliminary probabilistic design of a cross-section of the new breakwater at the port of Genoa by applying a vine-copula model for the wave climate in a Monte-Carlo simulation.

All possible regular vines were obtained by permuting the six equivalence classes for 5 nodes. 13 different copula types and all 480 possible 5-node regular vines were fitted on extreme wave data. The best vine-copula was selected based on the lowest AIC value. The performance of the vine-copula model was compared to that of an independent model.

Offshore waves were transformed to onshore waves by applying SWAN. A MATLAB loop-function was made to automate this process. Prof. Goda's method was used for determining the wave-induced loads on the vertical breakwater. 10 failure modes were considered in total. A Monte-Carlo simulation was built in MATLAB. Using the Monte-Carlo simulation, several dozen designs were tested in an iterative matter to find an optimized design. The proposed preliminary cross-sectional design fulfilled all design criteria.

Since 2019, Rijkswaterstaat has been applying flexible water level management in the IJsselmeer and Markermeer to make the Dutch water system more robust and future proof. The fixed target water level of -0.20 m NAP during summer is replaced by a flexible water level in which the water level can fluctuate between -0.10 and -0.30 m NAP to enlarge the fresh water supply and to anticipate on weather conditions. In the IJsselmeer region, 15,600 hectares of land is located outside primary levees. These areas are mainly pasturelands, recreation areas, nature areas and some buildings and infrastructure areas. By applying flexible water level management, the probability of high water levels increases and therefore flood risk in areas outside primary levees increases. However, the relation between applying flexible water level management and flood risk has not been investigated in detail yet. This research explores this relation further, on smaller scale as well as for the whole IJsselmeer region.

Flood risk is defined as the yearly probability of exceedance of hydraulic loads (lake water level, wind set up and wave run-up) multiplied by the consequences of inundation and is expressed in euros per year. Risk is a set of scenarios, with each a probability and a consequence, and therefore discretization is necessary. Hydraulic loads are calculated in Hydra-NL, a probabilistic model. By uploading hydraulic loads associated to their exceedance probability in the Waterschadeschatter, the consequences are defined and flood risk can be computed.
Flood risk depends on the lake water level, wind set up and wave run-up. The lake water level and wave run-up are in general higher when flexible water level management is applied and wind set up is lower. This observation is valid for all areas outside primary levees in the IJsselmeer region. About 90% of the total damage comes from water damage in buildings and infrastructure areas, and then especially from residential areas. Since less than 3% of areas outside primary levees consist of buildings and infrastructure areas and less than 10% of buildings and infrastructure areas consist of residential areas, the flood risk is reduced. Moreover, most computations show overestimations. Water levels computed in Hydra-NL are higher than occurred water levels and the upper boundary of -0.10 m NAP is used to calculate effects of flexible water level management. Because both overestimations are used and there are just a few residential areas, most areas located outside primary levees in the IJsselmeer region will hardly suffer from applying flexible water level management. This research shows that interests of areas outside primary levees in the IJsselmeer region and applying flexible water level management go quite well together. Therefore, the IJsselmeer region offers many opportunities.

Smallholder farmers cultivate more than 75% of the available agricultural land in Africa and therefore this form of agriculture is crucial to the global food supply. At present, little is known about these smallholder agricultural and related irrigation practices, yet the increasing availability and accessibility of remotely sensed data provides significant opportunities to assess the status quo of these practices. However, these characteristic agricultural landscapes can create complexity in identifying land use with the help of remote sensing. The cultivated plots are small and consist of intercropping systems with dynamic spatio-temporal practices concerning planting, irrigation and harvesting. 
This research aims to provide insight into the usefulness of remotely sensed passive Sentinel-2 Level-1C and active Sentinel-1 SAR data for land use classification of these complex landscapes with a focus on irrigated agriculture, using a case study in Central Mozambique. For this purpose, an open source-code is written that uses open-source satellite data from Google Earth Engine to execute the supervised image classification methodology, using confusion matrices as an assessment method. 
The results of this research appear to show that a nonparametric RF classifier ( = 88.0%) is preferred over a parametric ML classifier ( = 85.0%) for processing the data that is high in variability, in which classifications based on the chlorophyll sensitive Red Edge and SWIR bands provide the highest overall accuracies (>88.0%). However, the classifier overestimates the amount of irrigated areas by a factor of 1.5 in the first and a factor of 3 in the second irrigation season. The opportunistic sampling method appears to cause inflated accuracy outcomes and an optimistic bias towards classification of the main class in training. Spectral analysis of the temporal behavior of various S-2 bandwidths does not provide insight into the underlying mechanisms on which the algorithm performs classification. Although it appears that irrigated agriculture with S-2 data can be identified on the basis of an increase in vegetation biomass and that the classifier benefits from more information through the use of multiple bands. Research into the use of Sentinel-1 SAR data appears to have potential for identifying irrigation. Time series of the VV backscatter signal show a difference between the irrigation class and the classes non irrigated and light seasonal vegetation in irrigation season 2. However, high standard deviations do reflect the high intra variability of the data, and classification accuracies in this period, do not exceed an overall accuracy of 64.1%. The main confusion as identified by the confusion matrices, comes from classes that are often identified as irrigated, whereas they are not, overestimating the amount of irrigated areas as with the use of S-2 data.The results of this research show that the used method and data collections do not provide accurate information for the intended classification goal. This research demonstrates in several ways the complexity of supervised image classification in complex agricultural landscapes: the unbalanced and variable reference data of different land uses, which often consist of only a few satellite pixels, make it difficult to identify characteristics of land classes, from which the classifier can derive information. In which Sentinel-1 as added and used in this research, offers no additional insights. Therefore, in order to improve the identification of farmer-led irrigated agriculture in Manica, other technologies for smart agriculture can be explored in addition to deploying satellite data, such as citizen science. Further research is recommended on the field of using S-1 and S-2 data for classification of complex agricultural landscapes. This may include more advanced methods of performing image classification and accuracy assessement with imbalance datasets, such as: the use of a weighted confusion matrix for accuracy assessment or exploring the use of spatial-spectral instead of pixelwise random forest algorithms. These algorithms seem to be better at handling spatial dependencies and intrinstic heterogeneity which is characteristic of these complex agricultural landscapes. Lastly, it is strongly recommended to assess the use of speckle filters when using SAR data for small target objects. These filters make use of a buffer zone, consisting of a few pixels in size and about the same size of the target object. The main challenge with be to balance the need of speckle reduction and class specific information preservation.

This master thesis is about the positive effects of river system behaviour along the river IJssel. Due to failure of the flood defences along the IJssel, the protected areas will flood and water will flow out of the river. The discharge in the river will decrease and thus the water level at the downstream ends of the river will also decrease. This will have an impact on the required dimensions of the flood defences downstream as the loads will be lowered due to the decrease in water level. The design of the dimensions will change and the costs of the flood defences downstream are lower. In this research, flood simulations are done in the dike-rings along the river IJssel due to overtopping with and without breach formation. The simulations are done in a developed HEC-RAS model and eventually the decrease in water level is generated within the model computations. The tool OKADER is used for the calculations of the required dimensions of the flood defences in 2050 and the associated costs when changes in water level is incorporated. At the end the interest is to quantify the positive effects of system behaviour in terms of differences in required dimensions of the flood defences along the river IJssel and the amount of saved costs.

The Kampereiland is an island located at the south eastern part of the IJsselmeer in the middle of the Netherlands. This island serves a purpose to lower water levels in the IJssel- and Vechtdelta during high water, by being inundated. This inundation is meant to result in lower water levels in the IJssel near Kampen. However, the flooding of the Kampereiland is a complex case. There are numerous stochastic variables that are used to determine the water levels, as well as breaching possibilities and a compartment structure which is present on the Kampereiland. All these aspects have influence on inundation of the island. In this thesis research, the flooding of the island is analysed, during multiple different types of high water scenarios, as well as with a variation of breaches or no breaching. This is all done while taking the compartment structure on
the Kampereiland into account.

To analyse how the Kampereiland floods, three main types of highwater scenarios have been analysed. These types are stormdominated scenarios, which are defined by high wind speeds and low discharges. The second type of high water is the discharge dominant scenario. The high water levels in these scenarios are reached by high discharges from the Vecht and the IJssel rivers, in combination with low wind speeds. Lastly, a number of extreme events have been analysed, to give a worst case perspective of the flooding of the Kampereiland. In these scenarios, high wind speeds coincide with high discharges from the two rivers, resulting in extreme water levels in the Ketelmeer, ZwarteMeer and in the Ganzendiep.

Each of the analysed scenarios have outside water levels exceeding the crest height of the flood defenses surrounding the Kampereiland. The scenarios have been analysed for multiple breaching cases. Three different breach locations have been determined. The locations have varying crest heights, type of flood defense (overflow resistant or non overflow resistant), type of high water (storm dominated or discharge dominated) and different hinterland compartments. Each scenarios has then been used to run the model to determine water levels in each compartment. This is done for situations without a breach, a single breach or two breaches at one location. The determination of the water levels is calculated by the use of spillway formulas.

The results show that the Kampereiland mainly floods from west to east. Storm dominated scenarios lead to high water levels in the Ketelmeer, leading to overflow from here. In scenarios up to 32 m/s wind speeds without breaches, the flooding is contained to the compartments to the west of the N50. The entire island floods once water levels increase due to wind speeds of 37 m/s and higher. Adding breaching into the storm dominated scenarios, the results are shifted. Now the evaluated scenarios with 32 m/s also lead to flooding
of the entire island. This case still leads to less high water levels on the Kampereiland than the scenarios of 37m/s without breaching, but every compartment is inundated.

The discharge dominated events are contained in the Willem Meyer- and Stikkenpolder. This is due to the presence of the old flood defense, which contains the flooding in cases with and without breaches.

Finally, the extreme events flood the majority of the Kampereiland, similarly to the stormdominated scenarios. The water levels that occur in the Zwarte Meer and Ganzendiep are higher during the extreme events. They also last longer than in the storm dominated scenarios. This results in a small inflow from the north
and southeast, while the western side of the Kampereiland rapidly fills up. Breaching during these extreme events leads to higher water levels when the breach occurs in the overflow resistant part of flood defense 225. The rising velocities in the west are also the highest in this case. Eastern compartments have higher rising velocities in case of breaching in flood defense 101 (along the northern boundary).

Many quay walls in Amsterdam have surpassed their structural lifetime and have started showing signs of damage. The city of Amsterdam is currently tackling the problem and have published a plan of action. This plan includes the renovation of hundreds of kilometres of quay walls. Given this enormous amount, it is necessary to prioritize certain quay walls over others based on the severity of their damage. Some quay walls have reached total collapse, of which the most recent case involves the "Grimburgwal" quay. The municipality has no accurate view of the current condition of quay walls in Amsterdam. On top of that, the vast majority of quay walls have not been assessed on their safety. It is known that the most vulnerable quay walls types consist of masonry walls, supported by wooden foundation structures. Given that the quay wall renovation project requires prioritisation, it is necessary to gain more information on how the most vulnerable walls are recognised. Preferably, a method should be developed in which only visual cues given by the masonry wall are required, as it is quick and relatively cheap. To gain information on what these visual cues might be, a three-dimensional finite element model is made to run simulations on possible behaviours of quay walls. In this thesis, it is attempted to model a quay wall as realistically as possible. Several different deterioration conditions will be applied to see how the masonry responds. The 3D model is built using a parametric model coded in Python. This code can be used to run simulations in the finite element software DIANA FEA. Many behavioural aspects have been incorporated into the model, with the purpose to make the model more realistic. The model consists of a masonry wall, planks on which the wall rests, and supporting piles. The behaviour of each component has been applied in the code and have been obtained through other literature and European norms. The model is loaded by simulating the weight of the soil and its effect on the quay wall structure. The masonry is simulated using a smeared cracking model (macro-model). Long-term deterioration of quay walls is simulated by changing the material properties of each respective component. This thesis focuses on three deterioration conditions: 1. Non-uniform pile degradation: application of broken piles, simulated by removal of those piles from the model. This is subdivided into two categories: removal of entire rows (a row consisting of a front, middle and end pile) and removal of front piles only. 2. Non-uniform soil removal: formation of soil pits at the foundation level, which result in decreased bedding around the foundation piles. 3. Uniform degradation: application of uniform deterioration along a stretch of quay walls. The simulations yield fairly consistent cracking patterns, in which the same crack fields appear in each simulation depending on the chosen case mentioned before. Displacement patterns are also documented and presented in all cases. The quay wall model is able to display in-plane and out-of-plane movement simultaneously. The effect of each parameter on the crack/displacement patterns are analysed as well. This includes masonry and wood quality. The results show that the largest in-plane settlements are reached by damaged piles, while the largest out-of-plane displacements are caused by a loss of soil bedding around the piles. The results can be used to provide better insight on how quay walls with poor quality present themselves in real life and what their cause might be. This research contributes to the possibility of improving recognition of quay walls which find themselves in critical condition, which can then be prioritized for renovations. For future research, it is recommended to see whether time-dependent simulations can be run, to see if it makes a difference in the outcome of displacement/cracking patterns. Another important recommendation is to look into deterioration rates of materials, which could be used as another indicator for critically damaged quay walls.

The design of offshore and coastal infrastructures, sand nourishment and other ’soft’ coastal interventions require the analysis of environmental variables (e.g. wind, waves, rainfall) that can potentially cause the failure of such structures. Processes such as overtopping, beach erosion, and coastal flooding can result from a combined action of two or more physical processes. Traditional infrastructure design practices assume the highest load previously experienced as the design load, regardless of possible interactions between variables (or processes). This may lead to a misrepresentation of critical design loads. This thesis presents a methodology for defining infrastructure design loads accounting for their interdependence. The methodology is general and is based on regular vines. Vines are graphical tools for defining high dimensional distribution functions through pair-copula construction. With this premise in mind, the main effort was concentrated in formulating a series of steps to integrate several stages of the design: from the processing of raw data up to the choice of design loads for any specific design purpose. The vine-based methodology was applied to the design of a breakwater in Galveston Bay, Texas. This application showed that accounting for the interdependence between design variables provides a more comprehensive description of the physical system acting on the infrastructure. However, the vine-based method is computationally demanding. Hence, the applicability of this methodology should be evaluated on a case by case basis. In parallel, the possibility to define goodness of fit test for vine-copula based on the concept of tree-equivalent classes is explored. The focus is on model selection strategies based on graphical and statistical properties of the vines. The main motivation to investigate model selection strategies for vines is the considerably large computational time needed to fit all regular vines in more than 6 nodes to the data. In this thesis, a novel algorithm is developed to facilitate the implementation of vines in higher dimensions (vines with more than 6 nodes). This algorithm significantly reduces the computational effort to select a regular vine by allowing the user to test only a subgroup of vines in n-nodes constructed on specific characteristics of the vines in (n-1)-nodes.