Comprehensive metrics tracking what matters for thriving, sustainable communities, Part of https://walkthru.earth/hormones-cities
AI-powered weather forecasts from NOAA AI-NWP (GraphCast) downscaled to H3 hexagonal cells with topographic corrections. Updated automatically every 12 hours, 5-day forecasts at 6-hour intervals, ~2M cells per timestep at H3 resolution 5. Coordinates are derivable from h3_index via the DuckDB h3 extension.
Six H3 resolutions (0–5) per forecast run. Each is a single data.parquet file, sorted by h3_index. New forecasts are uploaded every 12 hours. Compression: ZSTD level 3, 1M-row row groups for efficient range pushdown.
Partition path components (model, date, hour, h3_res) are encoded in the Hive directory path only and are not stored as columns inside the Parquet file.
All weather values are rounded to meteorologically appropriate precision before ZSTD compression. The source data is 0.25° (~28 km) NOAA AI-NWP — sub-decimal digits in the raw interpolated output are numerical noise, not real atmospheric signal. Rounding improves ZSTD compression by ~63% while preserving all scientifically meaningful information.
h3_index to apply physically-based corrections: variable lapse rates for temperature, slope channelling for wind, and orographic enhancement for precipitationdata.parquet per partition, sorted by h3_index, with 1M-row row groups. h3_index is written as BIGINT (int64); coordinates are derivable via the DuckDB h3 extensionForecasts produced from March 2026 onward use the lean schema described above: h3_index is BIGINT (int64) and geometry/lat/lon/area_km2 columns are omitted. Older forecasts may still contain the previous schema with VARCHAR hex h3_index, geometry, lat, lon, and area_km2 columns.
Weather: NOAA AI-NWP — AI Neural Weather Prediction models hosted on AWS Open Data. GraphCast_GFS produces 0.25° global forecasts on 13 pressure levels + surface variables, 6-hourly timesteps out to 10 days, initialized from GFS analysis. NetCDF format, updated twice daily.
Lam, R., Sanchez-Gonzalez, A., Willson, M., et al. (2023). Learning skillful medium-range global weather forecasting. Science, 382(6677), 1416–1421. doi:10.1126/science.adi2336
Terrain: GEDTM-30m via walkthru-earth/dem-terrain (code) — Global Ensemble Digital Terrain Model at 30m resolution, converted to H3-indexed Parquet with elevation, slope, aspect, TRI, and TPI derivatives. Used for all topographic corrections in this dataset.
Ho, Y., Grohmann, C. H., Lindsay, J., Reuter, H. I., Parente, L., Witjes, M., & Hengl, T. (2025). GEDTM30: global ensemble digital terrain model at 30 m and derived multiscale terrain variables. PeerJ, 13, e19673. doi:10.7717/peerj.19673
This dataset is licensed under CC BY 4.0 by walkthru.earth. The source NOAA AI-NWP data is public domain (US Government work).
| m/s |
| 0.1 |
| None (free-atmosphere) |
wind_v_850hPa_ms | m/s | 0.1 | None (free-atmosphere) |
vertical_velocity_500hPa_Pas | Pa/s | 0.001 | None (negative = upward motion) |
geopotential_500hPa_m | m | 0.1 | None (geopotential height) |
| g/kg·m/s |
| 0.0001 |
| q × u10 |
moisture_flux_v | g/kg·m/s | 0.0001 | q × v10 |
moisture_flux_magnitude | g/kg·m/s | 0.0001 | sqrt((qu)² + (qv)²) |
geopotential_anomaly_500hPa_m | m | 0.1 | H500 − 5574 (ICAO standard reference) |