Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples

Accurate estimation of extreme sea levels caused by storm surges is critical for coastal engineering, particularly during typhoon seasons. Data-driven approaches have emerged as efficient tools for storm surge prediction. This study presents the development of deep learning neural network models to...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhuo Zhang, Lu Zhang, Songshan Yue, Dong Zhang, Zhaoyuan Yu, Di Hu, Peng Chen
Format: Article
Language:English
Published: Taylor & Francis Group 2025-08-01
Series:International Journal of Digital Earth
Subjects:
Online Access:https://www.tandfonline.com/doi/10.1080/17538947.2025.2536074
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849224324155703296
author Zhuo Zhang
Lu Zhang
Songshan Yue
Dong Zhang
Zhaoyuan Yu
Di Hu
Peng Chen
author_facet Zhuo Zhang
Lu Zhang
Songshan Yue
Dong Zhang
Zhaoyuan Yu
Di Hu
Peng Chen
author_sort Zhuo Zhang
collection DOAJ
description Accurate estimation of extreme sea levels caused by storm surges is critical for coastal engineering, particularly during typhoon seasons. Data-driven approaches have emerged as efficient tools for storm surge prediction. This study presents the development of deep learning neural network models to predict storm surge levels in the estuarine and coastal waters of the Pearl River Estuary, China, using typhoon characteristics and previous surge data. Unlike traditional single-point neural network models that focus on individual tide stations, this study introduces multipoint models and incorporates a convolutional layer to extract spatial features from tide levels at neighboring stations. Comparative results reveal that multipoint models significantly increase prediction accuracy by integrating spatiotemporal information from surrounding stations. This advantage is particularly evident when forecasting lead times exceed six hours, where multipoint models demonstrate superior accuracy and stability compared with single-point models. Furthermore, this study evaluates the impact of limited training sample sizes on prediction accuracy, offering valuable insights into the data requirements for robust model training. The findings highlight the potential of using multipoint neural network models as effective tools for storm surge prediction, offering increased accuracy and contributing to enhanced coastal risk management and decision-making in the face of extreme weather events.
format Article
id doaj-art-cdbb6f1cb39d463fbdeb50132b6006a4
institution Kabale University
issn 1753-8947
1753-8955
language English
publishDate 2025-08-01
publisher Taylor & Francis Group
record_format Article
series International Journal of Digital Earth
spelling doaj-art-cdbb6f1cb39d463fbdeb50132b6006a42025-08-25T11:28:21ZengTaylor & Francis GroupInternational Journal of Digital Earth1753-89471753-89552025-08-0118110.1080/17538947.2025.2536074Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samplesZhuo Zhang0Lu Zhang1Songshan Yue2Dong Zhang3Zhaoyuan Yu4Di Hu5Peng Chen6State Key Laboratory of Climate System Prediction and Risk Management, Nanjing Normal University, Nanjing, People's Republic of ChinaKey Laboratory of Virtual Geographic Environment, (Nanjing Normal University), Ministry of Education, Nanjing, People’s Republic of ChinaState Key Laboratory of Climate System Prediction and Risk Management, Nanjing Normal University, Nanjing, People's Republic of ChinaSchool of Marine Science and Engineering, Nanjing Normal University, Nanjing, People’s Republic of ChinaState Key Laboratory of Climate System Prediction and Risk Management, Nanjing Normal University, Nanjing, People's Republic of ChinaState Key Laboratory of Climate System Prediction and Risk Management, Nanjing Normal University, Nanjing, People's Republic of ChinaPearl River Hydrology and Water Resources Survey Center, Guangzhou, People’s Republic of ChinaAccurate estimation of extreme sea levels caused by storm surges is critical for coastal engineering, particularly during typhoon seasons. Data-driven approaches have emerged as efficient tools for storm surge prediction. This study presents the development of deep learning neural network models to predict storm surge levels in the estuarine and coastal waters of the Pearl River Estuary, China, using typhoon characteristics and previous surge data. Unlike traditional single-point neural network models that focus on individual tide stations, this study introduces multipoint models and incorporates a convolutional layer to extract spatial features from tide levels at neighboring stations. Comparative results reveal that multipoint models significantly increase prediction accuracy by integrating spatiotemporal information from surrounding stations. This advantage is particularly evident when forecasting lead times exceed six hours, where multipoint models demonstrate superior accuracy and stability compared with single-point models. Furthermore, this study evaluates the impact of limited training sample sizes on prediction accuracy, offering valuable insights into the data requirements for robust model training. The findings highlight the potential of using multipoint neural network models as effective tools for storm surge prediction, offering increased accuracy and contributing to enhanced coastal risk management and decision-making in the face of extreme weather events.https://www.tandfonline.com/doi/10.1080/17538947.2025.2536074Deep learningneural networksurge predictionestuarine and coastal watersPearl River Estuary
spellingShingle Zhuo Zhang
Lu Zhang
Songshan Yue
Dong Zhang
Zhaoyuan Yu
Di Hu
Peng Chen
Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples
International Journal of Digital Earth
Deep learning
neural network
surge prediction
estuarine and coastal waters
Pearl River Estuary
title Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples
title_full Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples
title_fullStr Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples
title_full_unstemmed Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples
title_short Short-term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples
title_sort short term prediction of storm surges in estuarine and coastal waters via a multipoint deep learning neural network with limited training samples
topic Deep learning
neural network
surge prediction
estuarine and coastal waters
Pearl River Estuary
url https://www.tandfonline.com/doi/10.1080/17538947.2025.2536074
work_keys_str_mv AT zhuozhang shorttermpredictionofstormsurgesinestuarineandcoastalwatersviaamultipointdeeplearningneuralnetworkwithlimitedtrainingsamples
AT luzhang shorttermpredictionofstormsurgesinestuarineandcoastalwatersviaamultipointdeeplearningneuralnetworkwithlimitedtrainingsamples
AT songshanyue shorttermpredictionofstormsurgesinestuarineandcoastalwatersviaamultipointdeeplearningneuralnetworkwithlimitedtrainingsamples
AT dongzhang shorttermpredictionofstormsurgesinestuarineandcoastalwatersviaamultipointdeeplearningneuralnetworkwithlimitedtrainingsamples
AT zhaoyuanyu shorttermpredictionofstormsurgesinestuarineandcoastalwatersviaamultipointdeeplearningneuralnetworkwithlimitedtrainingsamples
AT dihu shorttermpredictionofstormsurgesinestuarineandcoastalwatersviaamultipointdeeplearningneuralnetworkwithlimitedtrainingsamples
AT pengchen shorttermpredictionofstormsurgesinestuarineandcoastalwatersviaamultipointdeeplearningneuralnetworkwithlimitedtrainingsamples