Natural Hazards Research Summit 2022: Reinforcing coastal resiliency: how advancements in machine learning can guide ongoing efforts

Among the most hazard-prone regions today are coastlines, where the population density and the lack of preventive actions against climate change has created multi-faceted challenges. In order to offer holistic solutions that are tailored to each region, one cannot rely simply on historical data, which might be sparse or insufficient. Breakthroughs in simulating natural phenomena the past decades have allowed the more in-depth investigation of various hypothetical scenarios using high fidelity models. Although such tools are of extreme detail, they tend to be computationally intensive, preventing the repetition of multiple analyses in a risk assessment setting. Surrogate models are data driven tools, that are able to develop mathematical approximations between the input and the output of complex and expensive numerical models. They are able to maintain similar accuracy to the high fidelity model they try to emulate, but at a significantly lower computational cost. Such tools have been proven particularly useful in coastal hazard assessment applications, especially for predicting storm surge for large regions of interest, taking into account sea level rise, complex local geomorphology and even the spatio-temporal surge evolution. Future endeavors involving such tools, are aiming to incorporate the tidal component/precipitation in those predictions, investigating additionally the intensification of storms and how available databases can be enriched towards this direction.