Weather Geo ERA5 Global Reanalysis

## Overview This dataset contains 1.065 billion hourly/daily weather records derived from the ECMWF ERA5 reanalysis, covering 85+ years (1940–2025) of global weather at 0.25° × 0.25° (~28 km) spatial resolution. Data is stored as parquet and partitioned geographically (48 regional partitions) so consumers can pull a single region without scanning the global archive. Modalities are tabular numeric weather variables plus geo/time keys. ## Schema - latitude — float — grid cell latitude (0.25° step) - longitude — float — grid cell longitude (0.25° step) - time — timestamp — observation time (UTC) - temperature_2m — float — air temperature at 2m (K or °C, see source) - dewpoint_2m — float — dewpoint at 2m - u_component_wind_10m — float — zonal wind at 10m (m/s) - v_component_wind_10m — float — meridional wind at 10m (m/s) - surface_pressure — float — surface pressure (Pa) - total_precipitation — float — accumulated precipitation (m) - region_id — string — geographic partition key - +additional ERA5-derived variables; see source card for full list ## Sources - HuggingFace: https://huggingface.co/datasets/NaaVrug/weather-geo-era5 — license: CC-BY-4.0 - Upstream: ECMWF ERA5 reanalysis via Copernicus Climate Data Store (CDS) — licensed under Copernicus license terms (compatible with attribution) ## Methodology The upstream ERA5 product is produced by ECMWF by assimilating historical observations (surface stations, radiosondes, satellites, etc.) into a fixed numerical weather prediction model to produce a physically consistent global reanalysis on a 0.25° grid. The publisher (NaaVrug) has downloaded ERA5 fields from Copernicus, normalized variable names, partitioned the data into 48 geographic regions, and re-encoded as parquet for efficient columnar/regional access. No additional modeling or quality re-processing beyond ERA5's own is documented. ## Known gaps & limitations - ERA5 is a reanalysis, not direct observation — values are model-interpolated and can diverge from local station measurements, especially in data-sparse regions and over complex terrain. - Pre-satellite era (roughly pre-1979) has substantially weaker observational constraint; uncertainties are higher for early decades. - 0.25° resolution is coarse for sub-mesoscale or urban-scale phenomena. - The source card does not fully enumerate every variable's units or QC flags; buyers should cross-check against the official ERA5 documentation before quantitative use. - Coverage through 2025 implies recent months may use ERA5T (preliminary) rather than final ERA5 — values can be revised upstream. ## Intended use & out-of-scope - Intended: training/eval of weather and climate ML models, RAG over historical weather context, time-series forecasting benchmarks, geospatial analytics, agent tools answering historical weather questions. - Out of scope: real-time/operational forecasting, hyperlocal (sub-km) prediction, or any application requiring directly observed station data — this is reanalysis output, not ground truth. _Federated dataset: 48 parquet shards, 16.63 GB total. Queries and downloads stream through the DataBazaar API._ _PII signals: credit_card_candidate×1 (0.3≤score<0.7) present in the sample. Common in public datasets (papers, logs) but worth knowing before joining with private data._ Original supplier listing: Weather Geo ERA5 — 1.065B Global Weather Records (1940-2025) 1.065 billion ERA5 reanalysis weather records covering 85+ years (1940-2025) of global weather at 0.25° resolution, geographically partitioned parquet for efficient regional queries.