Influence of model configuration for coastal flooding across Europe

Coastal flooding estimation at large scale, e.g. pan-European is usually performed using static method while dynamic method, in which numerical flood models are used to solve hydrodynamic equations, have proven to perform better. However, a numerical flood model can rapidly become computationally demanding. Thus, to respect the balance between efficiency and quality, models need to be properly configured. Usually, the model configuration is supported by calibration and validation. In the cases where it is not possible to appropriately implement calibration and validation through comparison against observed and measured data, sensitivity analyses can be applied in order to identify the key parameters that could influence the model capability to properly represent the modelled process. The present work aimed to identify influential model parameters across Europe and their relative importance in flood model configuration. Seventeen test cases were selected for which a LISFLOOD-FP model was developed, covering several sites across Europe and considering different storm events. A panoply of local morphologies and boundary conditions derived from the sites and storm event characteristics were used. For each test case, 72 simulations with different configurations were performed by varying the grid resolution, the numerical solver, the bottom friction and the wave set -up formulation used to estimate the total water level as a boundary condition. Two sensitivity analyses were performed on the modelled maximum flooded areas and water volumes using One-Driver-At-a-Time and variance-based methods. By using a k-means clustering method, the results of these sensitivity analyses allowed us to identify patterns through the test cases related to the geographical region, providing important information for the configuration of flood models across Europe. Both sensitivity analysis methods led to similar results highlighting dominant relative influences from the floodplain solver on the Atlantic coasts and from the boundary conditions on the Mediterranean ones. In addition the grid resolution was found to have great effect on the North and Baltic seas, while globally the friction was shown to impact the model's results less. The test cases were clustered using a k-means method using as input both the sensitivity analysis results and morphological factors. Depending upon the inputs, two different sets of clusters were generated revealing a complex relationship between the influence of the model's parameters and the selected morphological indicators.