The damaging effects of natural hazards on communities and the environment can be avoided and reduced through accurate forecasting, issuing warnings to populations at risk and preparedness. Tsunami events are responsible for greatest of losses compared to other more frequent hazards. Probabilistic tsunami forecasting (Selva et al. 2021) has been proposed to assimilate observational data from multi-channel information sources such as GNSS, seismic stations, ocean bottom sensors, and tidal gauges to reduce uncertainties in a Bayesian probabilistic framework. This project aims to take this one step further and to move towards impact-based type of forecasting for impact metrics such as number of people and critical infrastructure affected. It will convolve tsunami simulations with time-dependent human exposure scenarios to generate impact scenarios. The main objective of this project is to develop fast and accurate machine learning-based on-the-fly tsunami impact simulations accompanied by transparent uncertainty characterization/quantification.
The student will use a comprehensive portfolio of available tsunami scenarios and simulations for training, validation and testing using Machine Learning (ML) techniques as a basis to emulate tsunami inundation simulation (e.g., Salmanidou et al. 2021). The student will also use ML/AI methods for mapping elements at risk from satellite images and for processing traffic data (along with Origin-Destination analysis, e.g., Kim et al. 2018). This information will be used to extract time-dependent human exposure patterns. The student will be trained in tsunami modelling, probabilistic risk analysis and forecasting, Artificial Intelligence, advanced machine learning techniques, and uncertainty characterization and quantification.