Automated Landslide Identification from Time-Series InSAR Using Improved Hot Spot Analysis

Area of Science:

• Geosciences
• Remote Sensing
• Geohazards

Background:

• Traditional automated landslide detection methods struggle with large training data requirements, low accuracy, and high false positive rates.
• Interferometric Synthetic Aperture Radar (InSAR) offers potential for surface deformation monitoring but requires refined detection techniques.

Purpose of the Study:

• To develop and validate an efficient, training-free automated landslide detection method using InSAR data.
• To improve the accuracy and reliability of landslide detection and boundary delineation.

Main Methods:

• An Improved Hot Spot Analysis (IHSA) method was developed, integrating multi-weight factor coupling with InSAR-derived surface deformation data.
• A spatial weighting matrix incorporating multi-feature fusion was used to optimize the hotspot detection model.
• Morphological processing was applied for refining landslide boundaries.

Main Results:

• The IHSA method achieved a precision of 90.20% and a recall rate of 92.00%, significantly outperforming conventional hotspot analysis.
• The method demonstrated a substantial improvement of 53.61 percentage points in precision compared to traditional approaches.
• Extracted landslide boundaries showed high consistency with manual interpretations, overcoming issues like fragmentation and internal voids.

Conclusions:

• The proposed IHSA method offers an efficient, training-free solution for large-scale, early landslide identification.
• This approach provides crucial methodological support and data for regional landslide detection and hazard mitigation efforts.
• The study effectively addresses key limitations of existing automated landslide detection techniques.