Future Extreme Weather: a Data and AI driven approach to Understand Future Coastal Flooding

Coastal flooding can be regarded as the most damaging extreme weather event. Careful
planning of mitigation and adaptation strategies requires a deep understanding of the event’s
likelihood and intensity.
This project provides a framework for assessing those changing statistics of coastal floods in
the future. We use historical records of coastal floods on the coasts of the UK historical
weather variables data (sea surface temperature, sea-level pressure, zonal and meridional
wind speeds and daily precipitations) from remote sensing sources, reanalysis data and
global climate models and future predictions of those weather variables from global climate
models. The method consists of using machine learning models to classify days as being
either ‘flooded’ (i.e. containing a coastal flood event) or ‘non-flooded’, at tide gauge
locations in the past 2 decades; both ‘out-of-the-box’ and more complex machine learning
models are trained on historical data. The models are then further used to assess the future
statistics of coastal flooding, by classifying days with or without flooding in the future
decades, from global climate models data. Currently, the method is showing promising
results on predicting the future number of ‘flooding days’, while the models used and trained
still show gradual improvement.
Using the same intensity scale as in the dataset of historical records of floods, it can be
assessed whether those events are becoming stronger or not. As well, the frequency, or the
return period, for the upcoming decades can be inferred from this project. This framework
produces an actionable set of information, that can be used by policy-makers, businesses,
governments and people, to plan accordingly for future floods.