Al-Wardy Malik, Zarei Erfan, Nikoo Mohammad Reza
Center for Environmental Studies and Research, Sultan Qaboos University, Muscat, Oman.
Center for Environmental Studies and Research, Sultan Qaboos University, Muscat, Oman.
Sci Total Environ. 2025 May 10;976:179311. doi: 10.1016/j.scitotenv.2025.179311. Epub 2025 Apr 9.
Coastal vulnerability assessments are crucial for evaluating the potential impacts of environmental hazards. Traditional methods that typically rely on index-based approaches are often limited by their inability to account for the relative importance of individual parameters. This study integrated machine learning models (Random Forest and XGBoost), which were optimized through Particle Swarm Optimization, with an index-based method to determine the weights of vulnerability parameters using feature importance analysis. The results were then compared with those derived from the Analytical Hierarchy Process (AHP) and Shannon's Entropy, producing comprehensive Coastal Vulnerability Index (CVI) maps for Oman's coastline. The analysis considered eight parameters, categorizing them into five vulnerability levels before calculating the CVI. Regarding the influential parameters, results showed that most areas exhibit moderate vulnerability (43.05 %) for the geomorphology, followed by very high vulnerability (31.79 %). In contrast, most areas showed very low vulnerability for elevation and slope parameters, covering 55.82 % and 81.10 % of the coastline, respectively. Other parameters showed a varied distribution of vulnerability, particularly in major urban areas, including Muscat. The CVI results showed significant differences among models, with the index-based method identifying 14 % of the coastline as highly vulnerable, AHP <1 %, and Shannon's Entropy 20 %. This study highlighted how different methods varied in prioritizing coastal factors, with AHP and Shannon's Entropy assigning higher weights to NSM and relative sea level rise, while machine learning models uncovered nonlinear relationships and provided a more flexible approach. This research underscores the integration of machine learning models with index-based methods for CVI calculation.
海岸脆弱性评估对于评估环境灾害的潜在影响至关重要。传统方法通常依赖基于指标的方法，往往因无法考虑各个参数的相对重要性而受到限制。本研究将通过粒子群优化算法优化的机器学习模型（随机森林和XGBoost）与基于指标的方法相结合，利用特征重要性分析来确定脆弱性参数的权重。然后将结果与层次分析法（AHP）和香农熵得出的结果进行比较，生成阿曼海岸线的综合海岸脆弱性指数（CVI）地图。分析考虑了八个参数，在计算CVI之前将它们分为五个脆弱性等级。关于影响参数，结果表明，就地貌而言，大多数地区表现出中等脆弱性（43.05%），其次是非常高的脆弱性（31.79%）。相比之下，大多数地区在海拔和坡度参数方面表现出非常低的脆弱性，分别覆盖海岸线的55.82%和81.10%。其他参数显示出脆弱性的不同分布，特别是在包括马斯喀特在内的主要城市地区。CVI结果显示各模型之间存在显著差异，基于指标的方法将14%的海岸线确定为高度脆弱，AHP方法确定的比例小于1%，香农熵方法确定的比例为20%。本研究强调了不同方法在确定海岸因素优先级方面的差异，AHP和香农熵赋予非结构措施（NSM）和相对海平面上升更高的权重，而机器学习模型揭示了非线性关系并提供了更灵活的方法。这项研究强调了将机器学习模型与基于指标的方法相结合用于CVI计算。
