GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 2.1 GNGTS 2024 The initial operation of our workflow consists of the employment of regional hazard disaggregation as weighting information for an “importance” sampling procedure. This enables sampling in areas of significance, where the likelihood of selecting scenarios that have the most impact on the hazard is higher. By choosing scenarios based on their "importance", this sampling strategy significantly enhances efficiency. Source Refinement The combination of importance sampling and hazard disaggregation provides us N IS scenarios that comprehensively capture the total hazard at the most representative point for the target location. Given that the regional hazard is generated with a relatively coarse resolution for both the source and target points, it may be reasonable, at this juncture, to enhance the representation of both the source and target with a more localised perspective (Williamson et al., 2022). This refinement allows for an expanded source discretization, introducing greater variability for the local hazard. If specific local information is accessible, it becomes feasible at this stage to reassign probabilities based on such information, achieving a balanced consideration alongside the regional contribution. This involves perturbing each scenario by sampling alternative values of the source parameters that more accurately characterise it. Following the refinement of the source, we obtain N SR scenarios (greater than N IS ). To maintain the total contribution to the hazard, these new scenarios must undergo reweighting. In the absence of additional information, each new scenario can be assigned to the original scenarios, with equal weight assigned to all scenarios associated with the same original scenario. However, if new local information is accessible (e.g., insights into local fault positions), the weights can vary, with the constraint that they collectively sum to 1 for each original scenario, thus preserving their original balance. Clustering The results of offshore simulations serve as input for an additional filtering step before conducting coastal flood simulations on a fine grid. This step involves a cluster analysis of water column height profiles at specific points along a near-shore bathymetric depth, such as at 10 m. These points are strategically distributed in proximity to the site of interest, ensuring an adequate number for the final hazard calculation. It's important to note that without the application of the refinement step, these points might not be adequately resolved from the regional hazard results. The water height profile, coupled with the vertical coseismic deformation at the corresponding profile location, serves as the proxy variable for the clustering filter (Volpe et al., 2019). For clustering, we employ a standard implementation of K-means clustering (Lloyd, 1982) provided by the Scipy sklearn package (Pedregosa et al., 2011). The selected metric is the Euclidean distance, computed as the square root of the sum of squares of differences in each of the two