Making a SPLASH: How digital tools are revolutionising wave overtopping prediction

The SPLASH project   has created an environmental digital twin able to forecast and visualise wave overtopping to support coastal communities in high-risk flooding areas. It provides an excellent example of RSE teams combining user centred design, interactive plotting, machine learning and containerisation to take a scientist’s proof of concept Jupyter notebook to an easy-to-use web dashboard.  

 The SPLASH digital twin uses a random forest machine learning model trained on wind, water level and wave data combined with wave overtopping data from a WireWall sensor. A test scenario is based around the coastal railway in Dawlish, Devon, and a coastal frontage with high social value (recreation and tourism) in Penzance, Cornwall. Overtopping data and forecast conditions are displayed on an interactive dashboard able to display wave overtopping graphs from today up to five days ahead. This dashboard also allows users to adjust the wave, water level and atmospheric features to predict new overtopping scenarios and understand how these variables influence the overtopping hazard. The tool is built using a Plotly Dash frontend and models in scikit-learn.  Up-to-date wave and wind forecast data is obtained from the Met Office through a set cron scripts and the tidal levels are obtained from local predictions. A set of Docker containers are used to run the frontend, backend and downloader making the whole system easy to deploy on a dedicated server, virtual machine or cloud provider. This research shows how integrating machine learning models with digital technologies enables prediction of wave overtopping very efficiently. This digital tool serves as a decision-support system for mitigating coastal hazard, e.g. impact by overtopping waves, benefiting both community policy makers and railway operators with infrastructure in affected regions.

This research was funded by the Natural Environment Research Council (NERC) and the Met Office through their TWINE programme of which the SPLASH project (NE/Z503423/1 and NE/Z503435/1)was part. The digital twin was trained using data collected by the NERC funded CreamT project (NE/V002538/1).