UNIPLU-BR Processed Rainfall Dataset: Quality-Controlled Brazilian Precipitation Observations (1855–2025)

Quality-controlled Brazilian rainfall observations derived from the UNIPLU-BR dataset (Lemos et al., 2026). This processed version applies per-network quality control, removes structurally invalid records, aggregates sub-hourly data to hourly resolution, and provides standardised output with documented QC flags. Source data: 911 million raw observations from 21,395 stations across six Brazilian monitoring networks (1855–2025), with no quality control applied at source. Processed output: 396,646,593 observations from 21,277 stations, split across six network files plus a station catalogue. Each observation carries a QC flag (PASS or SUSPECT_*) enabling users to filter by quality level. Networks included: Hidroweb diário — 12,051 stations, 133M daily observations (1885–2025) Telemetria — 2,818 stations, 144M hourly observations (2014–2025) INMET subdiário — 629 stations, 68M hourly observations (2000–2025) ICEA — 176 stations, 30M hourly observations (1951–2025) CEMADEN — 5,014 stations, 14M hourly observations (2014–2025) INMET diário — 589 stations, 8M daily observations (1889–2025) QC methodology: Per-station IQR-based outlier detection, absolute upper bounds at 2× WMO world records, duplicate removal, dry spell analysis, flat station detection, and constant rain checks. Full criteria documented in the processing pipeline repository. Files: One ZIP per network containing observations CSV, QC summary, and processing log. Separate ZIP for the station catalogue (21,395 stations with coordinates, elevation, network, operating agency, and metadata conflict flags). Processing pipeline: https://github.com/AIRCentre/UNIPLU-BR Live dataset (API): This dataset is also served live via the AIR Centre Atlantic Cloud API at https://services.aircentre.org/access/, where it can be queried alongside Portuguese meteorological observations. The API supports station-based and state-based queries with QC flag filtering. The Zenodo archive represents a frozen snapshot (v1.0) of the processed data for citation and reproducibility purposes.