Scholarlink

Accurate monitoring of ground deformation is crucial for hazard mitigation, infrastructure management, and environmental protection. Interferometric Synthetic Aperture Radar (InSAR) and Global Navigation Satellite System (GNSS) are two complementary geospatial technologies whose integration relies predominantly on physical modeling and geometric transformations for fusion. This paper introduces a novel deep learning model that predicts GNSS-like three-dimensional ground displacements at InSAR measurement locations, using weak supervision from spatially sparse GNSS data. Our approach leverages a Dynamic Graph Convolutional Neural Network (DGCNN) backbone to model spatial dependencies among localized InSAR-derived features, effectively calibrating InSAR measurements to correct for viewing geometry limitations. The proposed method is evaluated in an area in the Netherlands affected by induced seismicity and ground subsidence across different experimental scenarios, with a particular focus on predicting ground deformations in time windows not experienced at training time.

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 Deep Learning Model to Predict GNSS from InSAR Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Michelangelo Caretto,
  • Antonio Alliegro,
  • Rosario Milazzo,
  • Osmari Aponte,
  • Andrea Gatti,
  • Eugenio Realini,
  • Lia Morra,
  • Tatiana Tommasi

摘要

Accurate monitoring of ground deformation is crucial for hazard mitigation, infrastructure management, and environmental protection. Interferometric Synthetic Aperture Radar (InSAR) and Global Navigation Satellite System (GNSS) are two complementary geospatial technologies whose integration relies predominantly on physical modeling and geometric transformations for fusion. This paper introduces a novel deep learning model that predicts GNSS-like three-dimensional ground displacements at InSAR measurement locations, using weak supervision from spatially sparse GNSS data. Our approach leverages a Dynamic Graph Convolutional Neural Network (DGCNN) backbone to model spatial dependencies among localized InSAR-derived features, effectively calibrating InSAR measurements to correct for viewing geometry limitations. The proposed method is evaluated in an area in the Netherlands affected by induced seismicity and ground subsidence across different experimental scenarios, with a particular focus on predicting ground deformations in time windows not experienced at training time.