Scholarlink

<p>Coastal zones are highly dynamic environments, where climate variability exerts strong influence on shoreline evolution and beach morphology. This study presents a data-driven framework for predicting short-term morphological adjustments using ensemble machine learning techniques, emphasizing the Random Forest (RF) algorithm. The objective was to test the predictive capability of the models for estimating short-term changes in beach slope and width under varying wave climate and synthetic storm scenarios representative of near-future conditions (2022–2025). Two ensemble configurations (M1 and M2) were calibrated and validated using cross-validation methods, and their performance was evaluated through R<sup>2</sup>, MAE, RMSE, and maximum error metrics. Both models demonstrated high predictive accuracy (R<sup>2</sup> &gt; 0.98), with MAE and RMSE ranging between 0.12–1.12&#xa0;m and 0.21–1.40&#xa0;m, respectively. Spatial variations in predictive skill were linked to site-specific geomorphological characteristics. The results reveal that beaches predominantly exhibit slope flattening accompanied by width expansion or contraction, driven by local exposure and sediment supply. Relative width prediction errors ranged from 0 to 77%, with larger discrepancies on steep and narrow profiles. The ensemble model was tested under short-term wave climate and synthetic storm scenarios (H<sub>m</sub>0 = 4&#xa0;m and 8&#xa0;m) to evaluate its ability to predict morphological responses under high-energy conditions. The study demonstrates the suitability of RF-based ensemble models for achieving accurate prediction of short-term morphological changes under projected wave climate variability and storm-induced forcing. The methodological framework applied herein is also applicable to inform erosion mitigation strategies, habitat preservation, and infrastructure-related planning in different coastal settings. The results underline the inclusion of machine-learning tools in coastal monitoring programs, especially in regions experiencing accelerated environmental shifts.</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

Modeling Future Coastal Morphodynamics Under Climate Forcing Using Ensemble-Based Machine Learning Approaches

  • 期刊论文

  • 发表日期：

    2026年1月12日

查看全文
  • Ghadah Aldehim,
  • Randa Allafi,
  • Abdulwhab Alkharashi,
  • A. Sagai Francis Britto,
  • J. Vijayalakshmi,
  • A. Sumaiya Begum,
  • M. C. Sashikkumar,
  • V. Priya

摘要

Coastal zones are highly dynamic environments, where climate variability exerts strong influence on shoreline evolution and beach morphology. This study presents a data-driven framework for predicting short-term morphological adjustments using ensemble machine learning techniques, emphasizing the Random Forest (RF) algorithm. The objective was to test the predictive capability of the models for estimating short-term changes in beach slope and width under varying wave climate and synthetic storm scenarios representative of near-future conditions (2022–2025). Two ensemble configurations (M1 and M2) were calibrated and validated using cross-validation methods, and their performance was evaluated through R2, MAE, RMSE, and maximum error metrics. Both models demonstrated high predictive accuracy (R2 > 0.98), with MAE and RMSE ranging between 0.12–1.12 m and 0.21–1.40 m, respectively. Spatial variations in predictive skill were linked to site-specific geomorphological characteristics. The results reveal that beaches predominantly exhibit slope flattening accompanied by width expansion or contraction, driven by local exposure and sediment supply. Relative width prediction errors ranged from 0 to 77%, with larger discrepancies on steep and narrow profiles. The ensemble model was tested under short-term wave climate and synthetic storm scenarios (Hm0 = 4 m and 8 m) to evaluate its ability to predict morphological responses under high-energy conditions. The study demonstrates the suitability of RF-based ensemble models for achieving accurate prediction of short-term morphological changes under projected wave climate variability and storm-induced forcing. The methodological framework applied herein is also applicable to inform erosion mitigation strategies, habitat preservation, and infrastructure-related planning in different coastal settings. The results underline the inclusion of machine-learning tools in coastal monitoring programs, especially in regions experiencing accelerated environmental shifts.