National Multi-Network River Water Quality Benchmark with River-Flow Topology for South Korea (2015–2025)

A frozen, machine-learning-ready dataset integrating three national river water-quality monitoring networks of South Korea (automatic, algae-alert, and water-quality networks) into a masked daily grid of five target parameters (chlorophyll-a, turbidity, total nitrogen, total phosphorus, dissolved oxygen), 2015–2025, together with a directed river-flow topology graph derived from HydroRIVERS (2,208 stations, 2,366 directed edges, 612 confluences, with dam/weir/free edge types). The package is organized in two tiers. Tier 1 is the benchmarked core (node table, flow-edge list, masked daily grid) accompanied by reproducible reference baselines for held-out station imputation (spatial generalization); on this benchmark, flow-topology graph neural networks do not exceed strong geostatistical and pooled-covariate baselines, establishing a calibrated reference. Tier 2 provides additional license-cleared layers as-is for broader reuse: daily weather (Open-Meteo), daily water level (WAMIS), livestock facility locations, and the original per-network raw measurement tables. All processed/derived data are released under CC BY 4.0. See ATTRIBUTION.md for required upstream citations (HydroRIVERS/HydroSHEDS, Open-Meteo, Korea NIER / data.go.kr, WAMIS).