Stephanie Contardo: A machine-learning and process-based hybrid workflow for shoreline position prediction
A Machine Learning and Process-based Hybrid Workflow for Shoreline Position Prediction
Stephanie Contardo, Emilio Echevarria, Vanessa Hernaman, Ben Leighton, Bryan Hally, Claire Trenham, Ron Hoeke
We propose a hybrid workflow that combines machine learning techniques with numerical hydrodynamic modelling to predict shoreline position. Beyond generating shoreline predictions, the approach is designed to assess the contribution of infragravity energy to prediction accuracy.
Machine learning methods are integrated at two distinct stages. First, a hybrid model is developed in the hydrodynamic part of the workflow, where a deep neural network model is trained, as a surrogate for the nearshore hydrodynamic process-based model XBeach, significantly reducing computational demand while maintaining predictive capability. Second, a convolutional neural network is used to predict shoreline position, trained with satellite observations of the shoreline (CoastSat) as predictands, and downscaled wave conditions, from both the SCHISM hydrodynamic model, coupled with the WWMIII wave model, and the surrogate XBeach model.
The approach builds on well-established methods and combines process-based modelling with machine learning to capture key shoreline processes, and enable efficient and robust shoreline forecasting while providing insight into the role of infragravity waves.
keywords: nearshore hydrodynamics, morphodynamics, shoreline, waves