Work

Morphodynamics-guided spatiotemporal network for modeling the morphological evolution of sand beaches in response to extreme storm events: alphaBeach-X

Year: 2025

Type: article

Abstract: In this study, we propose alpha Beach- X , a morphodynamics-guided spatiotemporal deep neural network designed to model the morphological evolution of a sandy beach and its changes in response to extreme storm events. The model was developed through a knowledge transfer learning strategy, in which morphodynamics were pretrained using numerical simulation data and then fine-tuned using field survey data to improve accuracy. The training and test datasets are curated based on hydrodynamic energy conditions to reduce the data imbalance issues commonly associated with rare extreme events. To further improve prediction accuracy during such extreme events — especially given the non-periodic temporal resolution of the field survey data — a recursive autoregressive process is incorporated into the model. The proposed model was applied to Maengbang Beach, located on the east coast of South Korea, and demonstrated skillful predictability in modeling storm-induced morphological changes through an ensemble prediction. As assessed using the Brier skill score (BSS), alpha Beach- X outperformed previous studies, achieving improved accuracy in prediction pre- and post-storm morphological changes. Based on 49 bathymetric survey datasets collected over six years, and using a 70:30 training-to-testing data split, the proposed alpha Beach- X model demonstrates average morphological change prediction performance in the swash zone (above −5 m relative to mean sea level) with a Root Mean Square Error of 0.63 m, a Bias of 0.10 m, a Structural Similarity Index Measure of 0.45, and a BSS of 0.52. • Development of alpha Beach- X , a morphodynamics-guided spatiotemporal network for modeling the morphological evolution • Knowledge-transfer learning via simulation-data pretraining and field survey-data fine-tuning • Using alpha Beach- X to predict storm-driven morphological changes on sandy beaches • Training/test dataset configuration using storm-energy computation to reduce data imbalance • Skillful predictability of extreme storm-induced morphological beach changes.

Source: Journal of Ocean Engineering and Science

DOI: 10.1016/j.joes.2025.11.003