cosmicraysandearthquakes/claude.md
root e5a812fa14 Initial commit: full analysis pipeline source code
Scripts 01-08 implement the complete cosmic-ray/earthquake correlation
analysis from data ingestion through out-of-sample validation and
combined timeseries sinusoid fitting.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-22 02:45:10 +02:00

1.7 KiB
Raw Blame History

Cosmic Ray / Earthquake Correlation Study

Scientific context

Testing Homola et al. 2023 (JASTP 247, 106068, DOI 10.1016/j.jastp.2023.106068) which claims a 6σ correlation between cosmic ray rate variations and global seismic activity at a 15-day lag, with ~11-year periodicity. Prior replication attempt using independent NMDB data failed. This project is a rigorous methodological replication + stress test.

Key hypotheses to test

  1. The 15-day lag signal survives phase-randomised surrogate null models
  2. The signal survives solar cycle detrending (both series share an ~11-year component)
  3. The signal is stable across NMDB station subsets
  4. The signal is stable across time windows (e.g., 1980-95, 1995-2010, 2005-19)
  5. The signal appears with seismic moment (M₀ ∝ 10^(1.5Mw)) not raw magnitude sum

Data sources

  • NMDB: http://nest.nmdb.eu (44 stations, pressure-corrected hourly neutron counts)
  • USGS: earthquake catalogue (M≥4.5 for completeness)
  • SIDC/KSO: daily sunspot numbers
  • Reference: Pierre Auger scaler data (optional, secondary)

Hardware

  • CPU: Xeon E5-2680 v4 (14 physical / 28 logical cores)
  • RAM: 60 GB
  • GPU: Nvidia M40, 12 GB VRAM, compute capability 5.2 (CUDA 11.x max)
  • Prefer CuPy over PyTorch for FFT work; M40 does not support bfloat16

Code standards

  • Python 3.11+, type hints required for public functions
  • pytest for all data-transformation functions
  • Keep raw data in data/raw/, processed in data/processed/, results in results/
  • All station metadata (lat, lon, altitude, rigidity cutoff) in config/stations.yaml
  • Seed all random operations; log seed + git SHA in every results file
  • Use polars or pandas with pyarrow backend (not default pandas) for speed