Operational InSAR at Scale – Emilia–Romagna HPC Benchmark

This repository accompanies the manuscript:

Mehdi S.S. Joud, "Operational InSAR at Scale: Reproducible HPC Pipeline Benchmarking for Sentinel-1 and SWIN-Based Groundwater Inference over Emilia–Romagna", submitted to IEEE Transactions on Geoscience and Remote Sensing (TGRS).

The goal is to provide a reproducible, end-to-end MT-InSAR processing chain for Sentinel-1 IW data over the Emilia–Romagna region and to enable hardware-to-hardware performance comparisons between CPU-only and CUDA-enabled backends.

Contents

• environments/ – Conda environment definitions for:
  • ISCE3 CPU build (conda_insar_cpu.yml)
  • ISCE3 CUDA build (conda_insar_cuda.yml)
  • MintPy / SBAS post-processing (conda_mintpy.yml)
• configs/ – AOI geometry, SBAS settings, and example path configuration for the AAU Strato cluster.
• pipeline/ – Python drivers for:
  • Valid pair selection and baseline graph construction
  • Running the ISCE3 CPU and CUDA chains
  • Running SBAS / MintPy time-series inversion
  • Timing and logging utilities
• slurm/ – Slurm job scripts for single-node and multi-node (strong/weak scaling) experiments, and SBAS runs.
• benchmark_slice/ – Metadata and example lists for the fixed benchmark subset used in the paper.
• docs/ – Instructions to reproduce the main tables and figures.

Quick start (conceptual)

1. Create environment (example for CUDA build):

   conda env create -f environments/conda_insar_cuda.yml
   conda activate insar_cuda