Data-driven Modeling of Global Storm Surges

The continuous growth of population and critical infrastructure development in low-elevation coastal zones compounded with rising sea levels exacerbates the risk of flooding in these areas. In order to mitigate this risk, drivers of coastal flooding need to be modeled with high accuracy and their long-term variations needs to be studied. In many areas, storm surges caused by tropical and extratropical cyclones are the main contributors to critical extreme sea level events.

Storm surges can be simulated using numerical models that are based on the underlying physical processes, or by using statistical/machine learning techniques that explain the relationship between the target variable (storm surge) and relevant predictors. This study explores the potential of statistical/machine learning based models to predict storm surges globally. A multitude of predictors obtained from remote sensing and climate reanalysis are utilized to model daily maximum surge for the global coastlines, at first for 840 tide gauge locations to train and validate the models based on in-situ observations.

Model results show that daily maximum surge is very well captured, especially in extratropical and sub-tropical regions, with weaker model performance in the tropical regions around the equator. Similar spatial characteristics of model performance are found for extreme events. Models forced with remotely sensed predictors showed a slightly better performance than models forced with predictors obtained from reanalysis products. Results also highlight a significant improvement when compared with the Global Tide and Surge Reanalysis (GTSR), which is based on a hydrodynamic numerical model. The study also shows that for the majority of tide gauges, the most important predictor to model daily maximum surge is time-lagged sea-level pressure. Finally, when adding tides to the analysis, the resulting total still water levels are reproduced with high accuracy everywhere along the global coastlines.