Journal article 199 views 96 downloads
Predicting shoreline changes using deep learning techniques with Bayesian optimisation
,
Doug Pender,
Demetra Cristaudo,
Rob Lamb ,
Harshinie Karunarathna
Coastal Engineering, Volume: 203, Start page: 104856
Swansea University Authors:
Tharindu Manamperi, Alma Rahat , Harshinie Karunarathna
-
PDF | Version of Record
© 2025 The Authors. This is an open access article under the CC BY license.
Download (18.36MB)
DOI (Published version): 10.1016/j.coastaleng.2025.104856
Abstract
Accurate prediction of shoreline change is vital for effective coastal planning and management, especially under increasing climate variabilities. This study explores the applicability of deep learning (DL) techniques, particularly Long Short-Term Memory (LSTM) and Convolutional Neural Network-LSTM...
| Published in: | Coastal Engineering |
|---|---|
| ISSN: | 0378-3839 |
| Published: |
Elsevier BV
2026
|
| Online Access: |
Check full text
|
| URI: | https://cronfa.swan.ac.uk/Record/cronfa70210 |
| Abstract: |
Accurate prediction of shoreline change is vital for effective coastal planning and management, especially under increasing climate variabilities. This study explores the applicability of deep learning (DL) techniques, particularly Long Short-Term Memory (LSTM) and Convolutional Neural Network-LSTM (CNN-LSTM) models, for shoreline forecasting at monthly to inter-annual timescales, under two modelling approaches—direct input (DI) and autoregressive (AR). All models demonstrated the ability to reproduce temporal shoreline variability, while the autoregressive DL models were performing better.Further, a noise impact assessment revealed that seasonal decomposition and noise filtering significantly enhanced the model performance. In particular, the models using 52-week data decomposition and residual noise reduction improved the model performance. The reduction of data noises also resulted in narrower ensemble prediction envelopes, indicating that ensemble candidate models behave with low diversity. The temporal data resolution analysis showed that lower data resolutions reduce the predictive performance of the model and at least fortnightly data are required to satisfactorily capture the trend of variability of the shoreline position at this beach.The use of ensemble predictions, derived from a selected subset of model trials based on their collective performance, proved beneficial by capturing diverse temporal behaviours, thereby offering a quasi-probabilistic forecast with minimal computational cost. Overall, the study underscores the potential of DL models, particularly with autoregressive architectures, for reliable and transferable shoreline change prediction. It also emphasizes the importance of data quality, resolution, and preprocessing in improving model robustness, laying the groundwork for future research into use of DL in multi-scale shoreline predictions. |
|---|---|
| Keywords: |
Shoreline prediction; Deep Learning; LSTM; Bayesian Optimisation |
| College: |
Faculty of Science and Engineering |
| Funders: |
The Authors would like to acknowledge the JBA Trust (project No. W22-1128), UK & Engineering and Physical Sciences Research Council (EPSRC) - Doctoral Training Partnerships (DTP) (EP/W524694/1) for funding the Doctoral Research study of the first author. |
| Start Page: |
104856 |