Skip to content

HRRR weather forecast — 3 km hourly model output

Find NOAA's High-Resolution Rapid Refresh (HRRR) numerical weather model and locate the GRIB2 files for a specific date.

What this teaches

  • Reading the variable list of a major NWP model
  • Navigating the GCS folder layout for HRRR (hrrr.YYYYMMDD/conus/...)
  • Choosing between CONUS and Alaska domains

Source used: HRRR

Prompt

What variables does NOAA's HRRR model produce, and where can I download
GRIB2 files for May 1, 2026 over CONUS?

Transcript

1. List HRRR domains

search_datasets(collections = "HRRR", limit = 5)

Two items — one per domain:

ID Domain Bbox Resolution
hrrr_conus CONUS + S. Canada + N. Mexico -134.1, 21.1 → -60.9, 52.6 3 km
hrrr_alaska Alaska + Bering / GoA / Arctic -176.0, 40.0 → -126.0, 75.0 3 km

Variables (excerpt): 2-m temperature (T2M), dewpoint (TD2M), 10-m wind (U10M, V10M), wind gust (GUST), precipitation (APCP, PRATE), categorical rain/snow/freezing rain/sleet (CRAIN/CSNOW/CFRZR/CICEP), CAPE, CIN, helicity (HLCY), MSLP, total cloud cover (TCDC), composite reflectivity (REFC), echo top, snow depth (SNOD), soil temperature/moisture, downward/upward radiation, near-surface smoke concentration.

2. Get the asset paths

get_item(collection = "HRRR", item_id = "hrrr_conus")

Assets: gcs_browse (console URL) and gcs_data (bucket path gs://high-resolution-rapid-refresh/).

The HRRR archive on GCS is laid out as hrrr.YYYYMMDD/conus/. For May 1, 2026 that's gs://high-resolution-rapid-refresh/hrrr.20260501/conus/. File types per cycle:

  • wrfnatf{HH}.grib2 — native level fields (~640 MB each)
  • bufrsnd.tar.gz — BUFR soundings
  • class1.bufr.tm00 — surface observations used as input

Result the agent gave

"HRRR produces 24 forecast cycles per day, each running out 18 hours (48 hours for the 00z, 06z, 12z, 18z synoptic cycles). It covers CONUS and Alaska at 3 km horizontal resolution.

Variables include the standard NWP suite — temperature, wind, dewpoint, MSLP, cloud cover — plus storm-relevant fields like CAPE, helicity, composite reflectivity, and operationally important specials like near-surface smoke (for wildfire forecasting).

For May 1, 2026 CONUS GRIB2 files:

Browse: https://console.cloud.google.com/storage/browser/high-resolution-rapid-refresh/hrrr.20260501/conus Bucket path: gs://high-resolution-rapid-refresh/hrrr.20260501/conus/

Open with cfgrib (xarray), pygrib, or wgrib2."

Variations

  • Alaska: switch to hrrr_alaska for the Alaska domain.
  • Specific cycle hour: the GCS folder lists files like hrrr.t{HH}z.wrfnatf{FH}.grib2 — HH is the run hour, FH is the forecast hour.
  • Aviation use case: filter the variable list for ceiling height (HGT_ceiling), visibility (VIS), and turbulence (TKE).
  • Smoke / wildfire: HRRR-Smoke fields are in the same archive — near-surface smoke concentration is the key variable.