Assessing erosion and flood risk in the coastal zone through the application of multilevel Monte Carlo methods

Coastal zones are vulnerable to both erosion and flood risk, which can be assessed using coupled hydro- morphodynamic models. However, the use of such models as decision support tools suffers from a high degree of uncertainty, due to both incomplete knowledge and natural variability in the system. In this work, we show for the first time how the multilevel Monte Carlo method (MLMC) can be applied in hydro-morphodynamic coastal ocean modelling, here using the popular model XBeach, to quantify uncertainty by computing statistics of key output variables given uncertain input parameters. MLMC accelerates the Monte Carlo approach through the use of a hierarchy of models with different levels of resolution. Several theoretical and real-world coastal zone case studies are considered here, for which output variables that are key to the assessment of flood and erosion risk, such as wave run-up height and total eroded volume, are estimated. We show that MLMC can significantly reduce computational cost, resulting in speed up factors of 40 or greater compared to a standard Monte Carlo approach, whilst keeping the same level of accuracy. Furthermore, a sophisticated ensemble generating technique is used to estimate the cumulative distribution of output variables from the MLMC output. This allows for the probability of a variable exceeding a certain value to be estimated, such as the probability of a wave run-up height exceeding the height of a seawall. This is a valuable capability that can be used to inform decision-making under uncertaint