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), accompanying the manuscript “Fractions Skill Score across spatial scales: statistical significance and discrimination” (F. Nerozzi), prepared for Advances in Statistical Climatology, Meteorology and Oceanography. Version 2.1.0 adds the AUC of the FSS at the large-scale limit L = W, computed from the stored domain-mean coverages (script large_scale_limit.py and table large_scale_limit_auc.csv in folder 05; the second Colab notebook runs it). Version 2.0.0 adds the folder 05_revision_rigid_displacement, which supports the revised manuscript. Corresponding forecast systems are now displaced rigidly (outline and core together) by at most half the system diameter (main experiments) or one diameter (position-error experiments). The folder contains: (1) the regenerated alternative ensembles of the four domain/system-count configurations for both displacements, and the morphology and boundary-rule experiments with the new rule; (2) a 200×200 domain with 16 systems, which has the same system density as the 100×100 domain with 4 systems; (3) the false-positive rate of the useful-skill criterion FSS > (1 + f0)/2 under the null, false-positive rates conditional on observed coverage, and intersections of corresponding systems at grid scale; (4) a check of the FSS implementation against the scores and pysteps packages; (5) two Colab notebooks and the corresponding scripts, which regenerate every table and figure of the manuscript and of its Supplement from fixed seeds, together with the stored scores of all new ensembles and the resulting tables and figures. Folders 01–04 are unchanged from version 1.1.2: four domain/system-count configurations (100×100 and 200×200 pixels, 4 and 8 precipitation systems) used to calibrate and validate scale-dependent null critical values at seven neighbourhood widths and two precipitation thresholds; a joint test across widths; a boundary-morphology study with five independent null-only calibration replications per morphology; and independent verification scripts that regenerate the stored scores from their seeds. Their null ensembles are reused by the revised manuscript. Their alternative ensembles follow the displacement rule of versions 1.x and are kept for provenance. No operational forecast–observation data are included, and the stored arrays are scores, not complete gridded precipitation fields. Code is licensed MIT; data and documents are licensed CC BY 4.0. The archive contains 2295 files and is too large for Zenodo’s online ZIP preview: download it and start from README.md. File integrity can be checked with verify_archive_integrity.py against SHA256SUMS.txt.