Scholarlink

<p>This study focuses on the application of advanced data‐driven modeling techniques to assess and predict high‐tide flooding (HTF) risks in Oman. The research analyzes the duration, frequency, and economic impacts of HTF and employs machine learning (ML) and deep learning (DL) models as practical tools for improving local flood prediction and management. At three Omani tide gauge stations—Salalah, Masirah, and Muscat—temporal analysis reveals clear seasonality, with Salalah exhibiting maximum HTF durations during summer months (July–September) and Muscat displaying more irregular month-to-month variability. Major floods account for disproportionately high economic losses (67–74% of total exposure), though they occur less frequently than minor floods. Non-tidal residuals (NTR) and tidal anomalies (TA) are the dominant mechanisms driving HTF, with NTR prevailing in Salalah (median contribution: 126.52%) and TA in Muscat (103.11%). Tree-based ML classifiers (Random Forest, CART, GBM, XGBoost, LightGBM) were tested for deterministic HTF prediction, and GBM achieved the best performance (F1-score: 0.915). For probabilistic forecasts, DL models (LSTM, TCN, GRU) were integrated with a Bayesian framework and Monte Carlo (MC) dropout, with LSTM achieving the highest short-term accuracy (0.925 at a 1-h lead time). The analysis shows a clear decrease in reliability for longer lead times (F1-score: 0.687 at 3 h), consistent with rising model uncertainty. While the modeling tools proved effective in capturing short-term HTF dynamics in Oman, the study acknowledges certain limitations, including the simplified treatment of compound effects and the need for broader validation under different coastal conditions.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Toward enhanced coastal flooding forecasting using deterministic and probabilistic models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月15日

查看全文
  • Malik Al-Wardy,
  • Erfan Zarei,
  • Mohammad Reza Nikoo,
  • Ghazi Al-Rawas

摘要

This study focuses on the application of advanced data‐driven modeling techniques to assess and predict high‐tide flooding (HTF) risks in Oman. The research analyzes the duration, frequency, and economic impacts of HTF and employs machine learning (ML) and deep learning (DL) models as practical tools for improving local flood prediction and management. At three Omani tide gauge stations—Salalah, Masirah, and Muscat—temporal analysis reveals clear seasonality, with Salalah exhibiting maximum HTF durations during summer months (July–September) and Muscat displaying more irregular month-to-month variability. Major floods account for disproportionately high economic losses (67–74% of total exposure), though they occur less frequently than minor floods. Non-tidal residuals (NTR) and tidal anomalies (TA) are the dominant mechanisms driving HTF, with NTR prevailing in Salalah (median contribution: 126.52%) and TA in Muscat (103.11%). Tree-based ML classifiers (Random Forest, CART, GBM, XGBoost, LightGBM) were tested for deterministic HTF prediction, and GBM achieved the best performance (F1-score: 0.915). For probabilistic forecasts, DL models (LSTM, TCN, GRU) were integrated with a Bayesian framework and Monte Carlo (MC) dropout, with LSTM achieving the highest short-term accuracy (0.925 at a 1-h lead time). The analysis shows a clear decrease in reliability for longer lead times (F1-score: 0.687 at 3 h), consistent with rising model uncertainty. While the modeling tools proved effective in capturing short-term HTF dynamics in Oman, the study acknowledges certain limitations, including the simplified treatment of compound effects and the need for broader validation under different coastal conditions.

Graphical abstract