Natech accidents in France: weather signal versus coding artifacts (ARIA × Météo-France, 1995–2026) – code, derived data and results
Abstract
> Code, derived data and results accompanying the article "Rising Natech Accidents or Better Bookkeeping? Separating Weather Signal from Coding Artifacts in Three Decades of French Industrial Accident Records" (submitted to the International Journal of Disaster Risk Science). > > The study links 51,821 technological accidents recorded in mainland France in the ARIA database (BARPI, export of 21 October 2026; 1 January 1995 – 31 May 2026) to daily observations from 6,684 Météo-France stations. It estimates exposure–response relationships for maximum temperature, precipitation, wind gusts and thunderstorms with a time-stratified case-crossover design and conditional logistic regression. It then separates the coding-independent weather contribution from changes in cause coding over time, and validates narrative-based detection of weather triggers that are not coded. > > Contents: > - code/: complete Python pipeline (00_download.py – 10_figures.py) that downloads the public source data and reproduces all results exactly; > - data_derived/case_crossover_dataset.csv.gz: 228,103 case and referent days (51,821 strata) with station exposures, distances and coding flags; > - data_derived/aria_records_metadata.csv.gz: record-level metadata (date, department, activity, coding flags) without narratives; > - results/: all model estimates reported in the article and the manual audit sample; > - figures/: Figures 1–4 (TIFF, 300 dpi). > > Original ARIA narratives are not redistributed. ARIA data © BARPI, reused for non-commercial research with attribution. Météo-France data are under Licence Ouverte / Etalab 2.0.
