Related Experiment Video
Updated: Aug 6, 2026
11:41
Evaluation of an Exclusive Spur Dike U-Turn Design with Radar-Collected Data and Simulation
Published on: February 1, 2020
Deep learning-based early warning of tailings storage facility instability using Sentinel-1 and Radarsat-2 InSAR
Maral Bayaraa1, Brian Sheil2, Cristian Rossi3
1Engineering Science department, University of Oxford, Oxford, UK. contact@maral.space.
Scientific Reports
|July 20, 2026
Summary
This study integrates satellite radar, geomechanical models, and AI to monitor tailings dams, improving early detection of ground movement and reducing failure risks. Horizontal deformation data proved most effective for predicting potential failures.
Area of Science:
- Geotechnical Engineering
- Remote Sensing
- Artificial Intelligence
Background:
- Tailings Storage Facilities (TSFs) require robust monitoring to prevent catastrophic failures.
- Traditional monitoring methods have limitations in detecting subtle ground movement precursors.
- Interferometric Synthetic Aperture Radar (InSAR) offers potential but faces challenges like phase unwrapping errors.
Purpose of the Study:
- To advance InSAR applications for TSF monitoring by integrating multi-source data and advanced modeling.
- To develop a novel calibration strategy for InSAR measurements using geomechanical finite element (FE) models.
- To create a deep learning framework for early warning systems using multi-source InSAR data.
Main Methods:
- Utilized multi-scale and multi-source satellite data (Radarsat-2, Sentinel-1).
- Employed different InSAR processing algorithms (PS-InSAR, ISBAS).
- Integrated geomechanical finite element (FE) modeling with InSAR data for calibration and uncertainty analysis.
- Developed a deep learning framework for early warning detection.
Main Results:
- Demonstrated a novel calibration strategy to overcome InSAR phase unwrapping limitations using FE model predictions.
- Showcased ISBAS processing's ability to maintain high spatial density while reducing phase ambiguity expression.
- Identified horizontal deformation components as the clearest precursors to TSF failure.
- Validated the effectiveness of an integrated framework combining InSAR, FE modeling, and deep learning.
Conclusions:
- The integrated approach significantly reduces uncertainties inherent in individual monitoring methods.
- This framework enhances the ability to predict TSF failures, mitigating environmental and economic risks.
- Combining multi-sensor InSAR, FE modeling, and deep learning provides a powerful tool for TSF safety assurance.