Scholarlink

<p>Landslide prediction has received considerable attention from both researchers and practitioners due to its extreme societal importance. It can help minimize catastrophic impacts on human lives, economic losses as well as disruptions in critical infrastructures. Existing works mainly rely on relevant predisposing factors and domain knowledge, which makes their shortcomings manifest. For example, they require a case-by-case assessment and context-specific knowledge of landslide mechanisms, which hinders the generalizability. In addition, they only focus on terrain deformations in a period, ignoring the continuous changes and trends of the surface displacements. To address these issues, we present a novel framework for modeling and predicting the landslides using typical InSAR point cloud data. Our method provides a solution for learning the complex structures in InSAR data and is more suitable for 3D surface modeling than generic 2D image mapping approaches. Specifically, we propose a distance- and slope-aware manifold learning to capture the inter-spatial dependencies and maintain the critical terrain characteristics in point clouds. We also propose a GNN model that accommodates the spatial connections and temporal consistency to continually predict landslides. Extensive evaluations conducted on real-world datasets demonstrate the effectiveness of our proposed model. It improves the prediction accuracy and the explained variance score by up to 25.2% and 86.7%, respectively.</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

GNN-Based Spatio-Temporal Manifold Learning: An Application of Landslide Prediction

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Liu Yu,
  • Rongfan Li,
  • Kunpeng Zhang,
  • Siyuan Liu,
  • Goce Trajcevski,
  • Jin Wu,
  • Fan Zhou

摘要

Landslide prediction has received considerable attention from both researchers and practitioners due to its extreme societal importance. It can help minimize catastrophic impacts on human lives, economic losses as well as disruptions in critical infrastructures. Existing works mainly rely on relevant predisposing factors and domain knowledge, which makes their shortcomings manifest. For example, they require a case-by-case assessment and context-specific knowledge of landslide mechanisms, which hinders the generalizability. In addition, they only focus on terrain deformations in a period, ignoring the continuous changes and trends of the surface displacements. To address these issues, we present a novel framework for modeling and predicting the landslides using typical InSAR point cloud data. Our method provides a solution for learning the complex structures in InSAR data and is more suitable for 3D surface modeling than generic 2D image mapping approaches. Specifically, we propose a distance- and slope-aware manifold learning to capture the inter-spatial dependencies and maintain the critical terrain characteristics in point clouds. We also propose a GNN model that accommodates the spatial connections and temporal consistency to continually predict landslides. Extensive evaluations conducted on real-world datasets demonstrate the effectiveness of our proposed model. It improves the prediction accuracy and the explained variance score by up to 25.2% and 86.7%, respectively.