中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
首页 机构 成果 学者
  
  1. CAS IR Grid
  2. 烟台海岸带研究所
  3. 中国科学院烟台海岸带研究所
  4. 中科院海岸带环境过程与生态修复重点实验室
Probabilistic storm surge forecasting in the Bohai Sea: A deep learning framework with adaptive uncertainty quantification

文献类型：期刊论文

作者Su, Changyu1,3; Mao, Miaohua1,2,3
刊名ESTUARINE COASTAL AND SHELF SCIENCE
出版日期2026-08-01
卷号336页码:16
关键词Storm surge forecast Deep learning Probability forecasting Uncertainty quantification
ISSN号0272-7714
DOI10.1016/j.ecss.2026.109877
通讯作者Mao, Miaohua(mhmao@yic.ac.cn)
英文摘要Storm surges pose a significant threat to coastal communities, yet operational forecasting relies primarily on deterministic predictions, leaving associated uncertainties largely unexplored. This study developed a probabilistic deep learning framework to quantify the storm surge prediction uncertainty in the Bohai Sea. We integrate Bidirectional Long Short-Term Memory (BiLSTM) networks with Adaptive Bandwidth Kernel Density Estimation (ABKDE). Unlike standard KDE, ABKDE dynamically adjusts bandwidths to effectively model the heteroscedastic and non-Gaussian characteristics of surge residuals, thereby capturing complex tail behaviors. Furthermore, Sequential Forward Selection (SFS) is employed to disentangle the contribution of different typhoon predictors to forecast uncertainty. Comparative analysis demonstrates that the BiLSTM-ABKDE model reduces the CWC to 0.27, showing higher efficiency than traditional KDE, Bootstrap, and Quantile Regression methods in generating reliable 90% prediction intervals. Spatiotemporal analysis reveals that forecast uncertainty is amplified in coastal bays compared to offshore zones and escalates with lead time. Interpretability analysis indicates that incorporating atmospheric pressure and maximum wind speed reduces Absolute Uncertainty (AU) by 27% compared to a wind-speed-only baseline. This result highlights the synergistic interaction between the inverse barometer effect and surface wind stress in constraining surge generation. Adding latitude further reduces it by 32.8%, accounting for track-dependent wind fetch variations. These probabilistic outputs provide critical, lead-time-dependent risk information, supporting more informed decision-making for early warnings and emergency resource allocation.
WOS关键词DENSITY
WOS研究方向Marine & Freshwater Biology ; Oceanography
语种英语
WOS记录号WOS:001741172100001
资助机构National Natural Science Foundation of China ; Chinese Academy of Sciences BRJH Program ; Key R&D Program of Shandong Province, China
源URL[http://ir.yic.ac.cn/handle/133337/42390]  
专题烟台海岸带研究所_中科院海岸带环境过程与生态修复重点实验室
通讯作者Mao, Miaohua
作者单位1.Univ Chinese Acad Sci, Beijing 100049, Peoples R China
2.Shandong Key Lab Coastal Zone Environm Proc & Ecol, Yantai 264003, Peoples R China
3.Chinese Acad Sci, Yantai Inst Coastal Zone Res, Yantai 264003, Peoples R China
推荐引用方式
GB/T 7714		 Su, Changyu,Mao, Miaohua. Probabilistic storm surge forecasting in the Bohai Sea: A deep learning framework with adaptive uncertainty quantification[J]. ESTUARINE COASTAL AND SHELF SCIENCE,2026,336:16.
APA Su, Changyu,&Mao, Miaohua.(2026).Probabilistic storm surge forecasting in the Bohai Sea: A deep learning framework with adaptive uncertainty quantification.ESTUARINE COASTAL AND SHELF SCIENCE,336,16.
MLA Su, Changyu,et al."Probabilistic storm surge forecasting in the Bohai Sea: A deep learning framework with adaptive uncertainty quantification".ESTUARINE COASTAL AND SHELF SCIENCE 336(2026):16.

入库方式： OAI收割

来源：烟台海岸带研究所

浏览0
下载0
收藏0
其他版本

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。