Site specific emulators for tsunami run-up simulations

Local Probabilistic Tsunami Hazard Analysis (PTHA) aims to quantify the probability distribution of inundation intensity parameters, such as maximum flow-depth, at a given location over a specified period of time. In a Monte Carlo framework such an analysis is dependent on the simulation of a large number of scenarios. A particularly expensive step, from a computational point of view, is the solving of the nonlinear shallow water equations associated with the tsunami run-up. This problem is even more pronounced in the context of Tsunami Early Warning and Probabilistic Tsunami Forecasting (PTF). A site specific (local) tsunami run-up emulator, trained on precalculated simulation results, enables rapid estimation of inundation maps allowing an assessment of a large number of scenarios with limited computational resources. While high dimensional input and output, dependence on topography and nonlinear dynamics has made the problem intractable for traditional statistical methods, the problem has recently been approached using new techniques developed within the field of Machine Learning. In this work we consider the problem of predicting onshore maximal flow-depth based on timeseries associated with simulated offshore gauge measurements. The site of study is the town of Catania in eastern Sicily. The dataset comprises more than 32,000 tsunami simulations for different earthquake sources in the Mediterranean Sea. Promising results have been obtained using only a small fraction of the total number of simulations as training data. The ML-based inundation predictions for locations close to the water's edge, which are flooded in many of the scenarios, show excellent correspondence with the numerical simulation results. Predicting inundation at locations further inland, which are flooded in only a small number of the simulations, is more challenging.