Employing a data-driven approach to process-based modelling in Delft3D-FM: a hybrid model for simulating shoreface nourishment evolution

Abstract

In recent decades, a growing number of shoreface nourishments have been carried out along the Dutch coast to mitigate coastal erosion. Designing shoreface nourishments requires a comprehensive understanding of their complex morphological evolution to develop optimal solutions. However, accurately predicting this evolution in Delft3D-FM has proven challenging. Excessive flattening of nourishments in morphological process-based models leads to an inaccurate representation of the cross-shore profile, and the implications of this inaccuracy on longshore predictions remain unclear. To address this issue, a hybrid model has been developed that seamlessly integrates a data-driven component into the process-based numerical model Delft3D-FM through Basic Model Interface. Contrary to previous study efforts, the focus is not on improving the representation of physical processes, but on incorporating a data-driven component into Delft3D-FM to more accurately represent observed behavior in cross-shore modeling. The presented modelling framework facilitates the prevention of excessive flattening of shoreface nourishments in Delft3D-FM by enabling user-defined manipulation of bed level changes during simulation. The hybrid model is then applied to reproduce a shoreface nourishment at the coast of Ter Heijde for two main objectives: (1) to demonstrate the proof-of-concept of a hybrid model approach and (2) to examine how inaccuracies in cross-shore modelling affect longshore predictions, thereby highlighting the added value of a hybrid model approach.
The results demonstrate the proof-of-concept of the hybrid model. Unlike the standard model, which showed excessive flattening of the nourishment, the hybrid model maintained its nourishment shape and followed observed cross-shore evolution over three months of morphological modelling. We found that a hybrid model approach has the potential to more accurately represent the nourishment’s lee effect, which plays a crucial role in the morphological response to a shoreface nourishment. By preventing excessive nourishment flattening, larger waves break earlier and/or more frequently in the hybrid model, causing a calmer wave climate in the lee of the nourishment compared to the standard model. As a result of increased wave sheltering, the flow velocity in the lee reduces which causes sediment supplied supplied by longshore currents to settle at a higher rate compared to the standard model, leading to increased sedimentation in the nourishment’s lee.
As a result of the enhanced representation of the lee effect, the alongshore redistribution of sediment differs between the hybrid model and the standard model. After three months of morphological modeling, differences of 10-25% between the models attributed to the nourishment sediment are observed in the lee of the nourishment. Additionally, the hybrid model shows a trend of increasing divergence from the standard model over time. This indicates a sustained added value of the hybrid model over time.
This thesis represents a contribution towards a data-integrated approach in process-based modelling of the complex evolution of shoreface nourishments, highlighting the potential added value of such an approach. Therefore, we anticipate this thesis to be a starting point for more sophisticated ways to incorporate accurate cross-shore evolution in numerical models in the context of shoreface nourishments.