Scholarlink

<p>With an increase in the number of landslides throughout the globe, landslide susceptibility mapping is the initial step towards landslide mitigation. Accurate mapping not only helps in identifying vulnerable areas but also plays an important role in the planning and implementation of preventive measures. Conventional landslide susceptibility mapping utilizes a range of statistical and machine learning methodologies, which requires many landslide pre-conditioning factors, such as slope, land use, and geological features to demarcate the areas susceptible to landslides. But these models face big challenges in multidimensional feature extraction, model generalization capabilities and the prediction quality of these non-linear data types. In recent years, Deep Learning methods like convolutional neural networks (CNN), recurrent neural networks (RNN) and long short-term memory (LSTM) have been applied for landslide susceptibility. This paper proposes a novel Deep Learning approach called DenseNet (169) Neural Network, applied to predict landslide susceptibility in Chamoli District of Uttarakhand, India. The geospatial dataset is prepared on 18 Pre conditioning factors based on geomorphological, geological, environmental, and anthropological data. The DenseNet model is validated with popular benchmark machine learning models comprising MLPNN and XGBoost. The accuracy assessment of model and ground validation has been calculated using the AUC-ROC curve, Precision-Recall and F1-score, where the model accuracy of DenseNet (87%) is higher than MLPNN (85%) and XGBoost (84%). Thus, these models can effectively use the spatial information about landslide occurrences and can predict landslide susceptibility in complex mountain systems.</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

A Novel Deep Learning Approach for GIS Based Landslide Susceptibility Prediction in Himalayas: A Study of Chamoli District, Uttarakhand

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Sayantan Mandal,
  • Ashis Kumar Saha

摘要

With an increase in the number of landslides throughout the globe, landslide susceptibility mapping is the initial step towards landslide mitigation. Accurate mapping not only helps in identifying vulnerable areas but also plays an important role in the planning and implementation of preventive measures. Conventional landslide susceptibility mapping utilizes a range of statistical and machine learning methodologies, which requires many landslide pre-conditioning factors, such as slope, land use, and geological features to demarcate the areas susceptible to landslides. But these models face big challenges in multidimensional feature extraction, model generalization capabilities and the prediction quality of these non-linear data types. In recent years, Deep Learning methods like convolutional neural networks (CNN), recurrent neural networks (RNN) and long short-term memory (LSTM) have been applied for landslide susceptibility. This paper proposes a novel Deep Learning approach called DenseNet (169) Neural Network, applied to predict landslide susceptibility in Chamoli District of Uttarakhand, India. The geospatial dataset is prepared on 18 Pre conditioning factors based on geomorphological, geological, environmental, and anthropological data. The DenseNet model is validated with popular benchmark machine learning models comprising MLPNN and XGBoost. The accuracy assessment of model and ground validation has been calculated using the AUC-ROC curve, Precision-Recall and F1-score, where the model accuracy of DenseNet (87%) is higher than MLPNN (85%) and XGBoost (84%). Thus, these models can effectively use the spatial information about landslide occurrences and can predict landslide susceptibility in complex mountain systems.