Consolidated S2ID hydro-meteorological disaster records for Brazil (2013–2025) and replication package

This dataset provides a consolidated and validated extraction of Brazil’s official disaster records, together with the complete replication package for the analyses reported in the accompanying article. Source and coverage. The primary source is the managerial report (“Relatório Gerencial — Danos Informados”) of the Integrated Disaster Information System (S2ID), maintained by Brazil’s National Secretariat for Protection and Civil Defence (Sedec). Yearly exports covering 2013–2025 were retrieved in July 2026 with all disaster types, all federative units and all processing statuses. Each record corresponds to one municipality affected by one event and contains a protocol identifier, a COBRADE hazard code, declared human and material damages, and the processing status of the federal recognition request (registered, recognised, or not recognised). Processing. Occurrence dates were extracted from the protocol identifier, because the report’s date column records entry into the system rather than the event itself; the two diverged in year for 1,450 records. After de-duplication by protocol, the consolidated base contains 77,398 unique municipal records, of which 23,152 fall within a combined hydro-meteorological scope defined by COBRADE codes: floods, flash floods and inundations (group 1.2), rainfall-induced mass movements (subgroup 1.1.3), and storms (subgroup 1.3.2, plus cyclones 1.3.1.1). Validation. Record counts were cross-checked against two independently produced sources: the S2ID official historical series of federal recognitions for 2013–2016 (1,832 versus 1,800 hydro-meteorological recognitions; ratio 1.02) and the federal gazette monitoring database maintained by Política Por Inteiro for 2019–2021 (1,801 versus 1,719; ratio 1.05). Contents. • supplementary_analysis_s2id_v3.py — single replication script (Python 3.12): parsing, scope definition, state-level episode construction, descriptive regime classification, intensive- and extensive-margin tests, municipal Kaplan–Meier estimation, recognition models with state-clustered inference, GEE, bootstrap and leave-one-state-out checks, and the full robustness programme. • s2id_master_2013_2025.csv — consolidated analytical base (77,398 records). • tabela_regimes.csv — descriptive regime classification by state, window and universe. • resultados_v6.json — machine-readable output of the replication script. • Table_S1_robustness.md and robustez_janelas.json — robustness of the main results under alternative episode-construction rules (10, 15 and 20-day clustering windows; 30 and 45-day maximum episode durations). • README_supplementary.md — file descriptions, usage instructions and analytical parameters. Reproducibility. Every quantitative result reported in the article is reproduced by a single execution of the replication script over the public S2ID files, with all analytical parameters declared as constants in the script header. Dependencies: xlrd, NumPy, pandas, SciPy, statsmodels. Note on the underlying data. The raw inputs are public administrative records maintained by Sedec and freely retrievable from https://s2id.mi.gov.br. This deposit contains a derived, consolidated and validated version of those records. The dataset contains no individual-level or personally identifiable information: the unit of record is the municipality-event. Note on AI assistance. Data processing, analysis code and documentation were developed with the assistance of an AI tool (Anthropic Claude). All analytical decisions, verifications and final content were reviewed and validated by the authors.