Integrity Auditing and Monte Carlo Limits in Probabilistic InSAR Forecasting: A Willcox Basin Pilot Study

Abstract

Code and derived results for a development-stage probabilistic forecasting workflow that applies a conditional diffusion model to NASA OPERA DISP-S1 InSAR line-of-sight displacement in the Willcox Basin, Arizona. The workflow implements executable checks of artifact consistency, evaluation eligibility, numerical scope, and Monte Carlo stability, and the archive includes the single-fault injection suite used to evaluate those checks. Contents: pipeline source code, configurations, existing tests, redacted historical gate records, six training-metric records, the analysis entry points, the fault-injection experiments, and the result and figure JSON files from which the result tables and data figures are replayed on CPU. Released source omits comments and docstrings. Native observational arrays, model checkpoints, and full predictive draws are not included; third-party inputs (OPERA DISP-S1, NWIS, PRISM, ADWR, AZGS, USGS, OpenStreetMap) are available from their providers under their own terms. The archive supports a development-stage study and does not contain a held-out forecast evaluation. Version 3 corrects the georeference of the basin raster used for the basin statistics and the calibration-window development-target domain. The raster had been rasterized with the spatial-partition corner as its origin, displacing the basin 1.57 km north. Version 3 adds analysis/make_basin_mask_final.py, which regenerates the raster on the frame grid, and replaces the four result files that depend on it; all other result files are unchanged from version 2. It also removes workstation-specific drive checks, lists all dependencies directly in requirements.txt, and updates README.md, CITATION.cff, and LICENSE. Licenses: code is released under the MIT License; documentation and derived artifacts are released under CC BY 4.0. See README.md and CHANGES.md for details.