Generating realistic storms using a joint return period sampling of intense precipitation

Planning flood-resilient infrastructures requires an accurate estimation of the flood hazard, which is commonly achieved by modelling the flood responses to synthetic extreme precipitation events known as design storms. Current methods for estimating design storms fail to account for observed joint return period dependencies across different durations within events. The common block-maxima approaches, for example, follow the entire intensity-duration-frequency curve throughout the event. To overcome the limitations of the current design storm approaches, we develop a method based on vine copula and a constrained micro-canonical cascade model to generate design storms that reproduce observed return period dependencies. Taking Zurich (Switzerland) as a case study, we analysed the dependencies between precipitation volumes over duration intervals ranging from 10-min to 6-h and found strong pairwise dependencies between return periods over different duration intervals, with a maximum Kendall’s τ rank correlation coefficient of 0.69. With our new approach, we find high variability in possible duration-frequency profiles, leading to an average reduction in total storm volume compared to common block-maxima approaches. For example, events with a 50-year return period over the 10-min duration interval have a total storm volume that is on average 56% lower than that of design storms generated using the block-maxima approach. Additionally, the variability in possible duration-frequency profiles indicates that multiple design storm events should ideally be used to ensure a robust flood risk assessment.