JPDH: a national-scale hydro-meteorological dataset for Japan

Companion dataset to Leveraging Japan's streamflow records for end-to-end learning of hydrological simulations at national scale. A river-network dataset covering Japan's four main islands (Hokkaido, Honshu, Shikoku, Kyushu) — approximately 367,000 km² of drainage area — discretised into 8,893 catchments at a 40 km² mean mesh resolution. It contains: the river network graph: catchment connectivity, local catchment polygons, full upstream drainage basins, and vectorized channel lines; catchment-aggregated dynamic meteorological forcing for five precipitation products (JMA gauge-adjusted radar, JMA radar, University of Tokyo gauge analysis, JMA MSM analysis, ERA5-Land) plus temperature, wind, surface pressure and cloud cover, daily 2008–2020 with sub-daily binning variants; static catchment descriptors (382) and channel descriptors (17); observed discharge at 1,089 MLIT gauges, daily and hourly; hourly reservoir inflow, outflow and storage records for 116 monitored dams, 2000–2024; the canonical 10-fold Prediction in Ungauged Regions (PUR) train/validation/test split definitions. Forcing arrays are stored as float32 with blosc/zstd compression; geometries are GeoParquet. SHA256SUMS and manifest.json describe and checksum each archive. A full data dictionary (variable naming, units, CRS, split definitions) will be added in a subsequent version of this record; in the meantime the layout is documented in the code repository. Code to download and use the dataset: https://github.com/TristHas/jpdh Data provenance. River gauge observations, dam discharge records and dam infrastructure information were provided by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) and the Japanese Dam Association. Derived products from Hatono et al. (2022), Nagata (2011), Saito et al. (2006), Muñoz-Sabater et al. (2021), Hirayama et al. (2022), Yamazaki et al. (2020) and Linke et al. (2019).