Scholarlink

<p>Landslides are among the most frequent natural hazards that damage infrastructure and human lives. Therefore, analyzing the regions likely to be hit by natural disasters is critical in mitigating losses. Therefore, this study is focused on evaluating an area's potential for landslides based on the system of a recurrent neural network (RNN). The geographical dimension of the given analysis includes the provinces of Chabahar and Konarak in Iran. By implementing the two-layer RNN architecture in the given research, the accuracy of prediction and the error are optimized. Instead of the conventional approach of feeding the happenings of the first layer into the setup to enhance the performance, the latest development is about feeding the previous output into the system. The model's robustness depends not only on the learning algorithm but also on specially considered variables; therefore, the algorithm should pass through a rigorous optimization. This research is a significant achievement in the study of landslide susceptibility mapping, providing valuable data for future researchers both within the region and globally. The study dataset consists of 400 historical landslide points. Among these, 80% are allocated as a training subset, and the remaining 20% as a validation subset. The top 16 factors, covering geological, topographical, and environmental parameters, represent the multiple attributes considered in a geospatial database. The landslide locations were validated through detailed fieldwork and, subsequently, confirmed by the use of high-resolution imagery on Google Earth. Because RNN technology utilizes dynamic data patterns, they can link recurrent nodes together in the hidden layer, thus differentiating them from traditional neural network ones. Establishing a two-layer RNN provides the best accuracy of predictions by repeatedly remodeling finalized outcomes. The validation findings confirm a remarkable accuracy rate of 87%, which underpins or corroborates the proposed methodology. The results of this study, supplemented by the RNN model approach, show the horizon to improve landslide susceptibility evaluations and develop strategies to eliminate uncertainties in landslide risk assessment.</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

Landslide Susceptibility Assessment Using Recurrent Neural Network (RNN)—A Case of Chabahar and Konarak in Iran

  • 期刊论文

  • 发表日期：

    2025年4月3日

查看全文
  • Vahid Isazade,
  • Abdul baser Qasimi,
  • Mahdi Safari Namivandi,
  • Mehdi Sheybani Amin,
  • Guerzou Ahlem

摘要

Landslides are among the most frequent natural hazards that damage infrastructure and human lives. Therefore, analyzing the regions likely to be hit by natural disasters is critical in mitigating losses. Therefore, this study is focused on evaluating an area's potential for landslides based on the system of a recurrent neural network (RNN). The geographical dimension of the given analysis includes the provinces of Chabahar and Konarak in Iran. By implementing the two-layer RNN architecture in the given research, the accuracy of prediction and the error are optimized. Instead of the conventional approach of feeding the happenings of the first layer into the setup to enhance the performance, the latest development is about feeding the previous output into the system. The model's robustness depends not only on the learning algorithm but also on specially considered variables; therefore, the algorithm should pass through a rigorous optimization. This research is a significant achievement in the study of landslide susceptibility mapping, providing valuable data for future researchers both within the region and globally. The study dataset consists of 400 historical landslide points. Among these, 80% are allocated as a training subset, and the remaining 20% as a validation subset. The top 16 factors, covering geological, topographical, and environmental parameters, represent the multiple attributes considered in a geospatial database. The landslide locations were validated through detailed fieldwork and, subsequently, confirmed by the use of high-resolution imagery on Google Earth. Because RNN technology utilizes dynamic data patterns, they can link recurrent nodes together in the hidden layer, thus differentiating them from traditional neural network ones. Establishing a two-layer RNN provides the best accuracy of predictions by repeatedly remodeling finalized outcomes. The validation findings confirm a remarkable accuracy rate of 87%, which underpins or corroborates the proposed methodology. The results of this study, supplemented by the RNN model approach, show the horizon to improve landslide susceptibility evaluations and develop strategies to eliminate uncertainties in landslide risk assessment.