EQ-INSAR Synthetic Dataset v1: 1000 Physics-Based Synthetic InSAR Earthquake Time Series

A dataset of 1,000 synthetic InSAR (Interferometric Synthetic Aperture Radar) time series generated with the EQ-INSAR Python package using the Davis (1986) elastic half-space point source model. Each sample is an 11-frame time series (5 pre-event + 1 co-seismic + 5 post-event) on a 64×64 pixel grid representing ~100×100 km of surface deformation. Earthquake parameters (Mw 5.0–6.5, depth 5–25 km, all fault mechanisms) are sampled uniformly at random. Three satellite configurations are included: Sentinel-1 ascending, Sentinel-1 descending, and ALOS-2 ascending. All samples are fully reproducible via the included generate_dataset.py script (seed=42). Intended use: training and benchmarking ML models for earthquake detection, segmentation, and anomaly detection in InSAR time series. Array shapes: X (time series, wrapped phase [rad]): (1000, 11, 64, 64) float32 y (binary segmentation labels): (1000, 11, 64, 64) uint8