This section gathers methods that flag abnormal displacement patterns in the InSAR catalogue. The goal is to compare an interpretable baseline against deep models that exploit temporal and spatial structure.

Baseline

Strong classical reference model.

• RANSAC Baseline - consensus-based robust fitting used for outlier detection and regularized resampling.

Deep Learning Models

Reconstruction-based methods with increasing relational capacity.

• Autoencoder - point-wise reconstruction on reservoir embeddings.
• Graph Autoencoder - reconstruction with explicit neighborhood context.