Reproducibility Dataset and Code for Fractions Skill Score across Spatial Scales: Statistical Significance and Discrimination

Reproducibility package for controlled synthetic-precipitation experiments on the Fractions Skill Score (FSS). It contains Monte Carlo null distributions and critical values at seven neighbourhood widths, independent validation, a joint test across widths, and power and ROC/AUC discrimination against alternatives with a known fraction of corresponding precipitation systems. Version 2.2.0 adds the folder `06_summer_convective_cells`, which supports the current version of the paper: - synthetic hourly precipitation from isolated summer convective cells on a 2.2-km grid (domains of 100 × 100 and 200 × 200 grid points, 4 or 8 cells, and a Poisson number of cells with mean 4), thresholds 2, 5 and 10 mm/h; - the FSS of every field pair under four treatments of the domain boundary (valid windows, zero padding, reflection, true field around the domain), with the stored numerators and denominators needed for seasonal (aggregated) scores; - Bernoulli random forecasts paired with every observed null field, used to show that the random-forecast reference of Antonio and Aitchison (2025) is the asymptotic limit of the calibrated critical value of the aggregated score; - analyses of seasons of 90 cases, three modes of computation (single case, mean of the daily scores, aggregated score), transfer of critical values between configurations, and the published reference values compared at matched false-positive rate; - two Colab notebooks: (A) regeneration of stored blocks from their seeds and a comparison with the `scores` and `pysteps` implementations; (B) every table and figure of the paper and of its Supplement from the stored scores. Folders 01–05 are unchanged and document the earlier versions of the study. No operational forecast or observation data are included. The map used as the backdrop of Figure 1 uses Natural Earth (public domain) and ISTAT boundaries via openpolis/geojson-italy (CC BY 4.0). Related publication. The archive supports the preprint: Nerozzi, F. and Tomozeiu, R.: Fractions Skill Score across spatial scales: statistical significance and discrimination, EGUsphere [preprint], https://doi.org/[EGUSPHERE-2026-6860], 2026. How to use. The whole archive, including the stored blocks of folder 06 (1,827 files), is the single file `FSS_reproducibility_archive_v2.2.0.zip`. After extracting it: - `python archive/verify_archive_integrity.py` checks all 4,357 files against `SHA256SUMS.txt`; - `archive/06_summer_convective_cells/run_all_analyses.sh` recomputes every table and figure of the paper and its Supplement from the stored blocks; - notebook A regenerates stored blocks from their random seeds and compares the FSS with the `scores` and `pysteps` packages; notebook B reproduces the tables and figures. Both run on Google Colab or locally (Python 3 with numpy, scipy, pandas, matplotlib, pyproj, shapely, cmcrameri, scores and pysteps). Licences: code MIT; data, figures and documents CC BY 4.0. Keywords (add): seasonal aggregation; asymptotic distribution; delta method; convective precipitation; boundary treatment Version notes (field "Version" / "Notes"): Adds folder 06_summer_convective_cells (summer convective-cell generator, four boundary treatments, Bernoulli forecasts, seasonal and asymptotic analyses, two Colab notebooks). Folders 01–05 unchanged. Rodica Tomozeiu, co-author of versions 1.x, is again a co-author. Related identifiers (Zenodo form, section "Related works"): - is supplement to: [EGUSPHERE-2026-6860] (Preprint), once the preprint is posted; - references: 10.1175/MWR-D-24-0120.1 (Antonio and Aitchison, 2025), 10.1175/2007MWR2123.1 (Roberts and Lean, 2008), 10.1175/MWR-D-18-0106.1 (Mittermaier, 2021).