This study explores the statistical dependence between wind speed and surge height along the Dutch coast using a large synthetic dataset. Storms were clustered based on wind direction, tidal offset, wind rotation, tidal peak, surge and wind exceedance duration, resulting in 16 clusters per wind direction and per location. Apart from wind direction, comparing clusters revealed a limited impact of clustering based on these storm characteristics on the choice of the best-fitting copula model, suggesting sub-clustering may not be necessary for accurately representing the statistical dependence between extreme wind speeds and surge heights. The BB8 copula generally provided the best fit to the data. However, the observed upper tail dependence did not decrease to zero, particularly for western to northern wind directions, indicating non-negligible dependence in joint extremes of wind speed and surge height. Therefore, applying the BB8 copula (or any other copula model without upper tail dependence) may lead to underestimation of the flood risk, when applied in probabilistic analyses. The findings from this study provide valuable insights for refining hydraulic load models for reliability assessments and design of flood defenses.

Extreme storms over the North Sea drive coastal flood risk in the Netherlands, causing high waves and extreme sea levels. Designing flood defenses requires accurate statistical extrapolation of hydraulic load conditions with return periods of 1,000 years or more. This is a challenging task given limited observational data. This study uses a large, simulated dataset (~9,000 years) to explore the statistical dependence between extreme wind speed u and surge height s. Storms were clustered using several techniques. Self-organizing maps (SOM) effectively captured physical relationships, such as the influence of wind direction and tidal offset on storm dynamics, however variability in statistical dependence between u and s for different clusters was better represented using manual clusters. Copula models were fitted to the cluster data, with the BB8 copula outperforming others. This study illustrates the potential of machine learning to identify patterns in large datasets while emphasizing the relevance of manual clustering approaches for revealing nuanced statistical dependencies critical to flood risk assessment.

Structural reliability analysis often considers failure mechanisms as correlated but non-interacting processes. Interacting failure mechanisms affect each others performance, and thereby the system reliability. We describe such interactions in the context of flood defenses, and analyze under which conditions such interactions have a large impact on reliability using a Monte Carlo-based quantification method. We provide simple examples and an application to levee failure due to landward slope instability and backward erosion piping (BEP). The examples show that the largest interaction effects are expected when the trigger mechanism is relatively likely to occur and the affected mechanism has a relatively large contribution to the system reliability. For the studied levee example, interactions between slope instability and BEP increased the failure probability up to a factor 4. Implications for the assessment and design of flood defenses are discussed.

Like other cities in the Kathmandu Valley, Bhaktapur faces rapid urbanisation and population growth. Rivers are negatively impacted by uncontrolled settlements in flood-prone areas, lowering permeability, decreasing channels widths, and waste blockage. All these issues, along with more extreme rain events during the monsoon due to climate change, have led to increased flooding in Bhaktapur, especially by the Hanumante River. For a better understanding of flood risk, the first step is a return level analysis. For this, historical data are essential. Unfortunately, historical records of water levels are non-existent for the Hanumante River. We measured water levels and discharge on a regular basis starting from the 2019 monsoon (i.e., June). To reconstruct the missing historical data needed for a return level analysis, this research introduces the Classical Model for Structured Expert Judgment (SEJ). By employing SEJ, we were able to reconstruct historical water level data. Expert assessments were validated using the limited data available. Based on the reconstructed data, it was possible to estimate the return periods of extreme water levels of the Hanumante River by fitting a Generalized Extreme Value (GEV) distribution. Using this distribution, we estimated that a water level of about 3.5 m has a return period of ten years. This research showed that, despite considerable uncertainty in the results, the SEJ method has potential for return level analyses.