Titan U glossary

High-Resolution Rapid RefreshHRRR

The High-Resolution Rapid Refresh (HRRR) is NOAA's 3-km, hourly updated, convection-allowing weather model for the United States. Each run starts from fresh observations, including radar data, which makes it a tool for the next several hours to two days rather than the week ahead.

The High-Resolution Rapid Refresh (HRRR) is NOAA's 3-km, hourly updated, convection-allowing weather model for the United States. Each run starts from fresh observations, including radar data, which makes it a tool for the next several hours to two days rather than the week ahead.

How often it runs and how far ahead

Over the contiguous United States the HRRR runs every hour. Most runs go out 18 hours, and the 00, 06, 12 and 18 UTC runs go out 48 hours.

The 48-hour runs arrived with HRRR version 4, which took effect 12/02/20. A separate Alaska version runs every third hour.

Starting from what radar sees

Radar data is assimilated, meaning blended into the starting state, every 15 minutes during the hour before each run. Version 4 added the HRRR Data Assimilation System, a 36-member, 3-km ensemble that assimilates radar reflectivity directly.

That gives storms already on radar a head start in the first hours of the forecast, instead of leaving the model to build them from nothing.

How forecasters and chasers use it

Because the HRRR is a convection-allowing model, it outputs simulated radar reflectivity and storm-scale fields such as updraft helicity. Forecasters look at when storms first develop in the model, whether they stay discrete or merge into lines, and how that picture shifts from one hourly run to the next.

Status and limits as of October 2026

As of October 2026, HRRR version 4 is still operational. The Rapid Refresh Forecast System (RRFS) is scheduled to become operational on 10/14/26, but NOAA has said the HRRR is not being retired with it, and the HRRR supplies two members to the new RRFS ensemble.

NOAA expects to retire the HRRR with a later RRFS upgrade. Until then, treat each HRRR run as one plausible scenario, since simulated storms can arrive early, late, displaced or not at all.

Keep going

How High-Resolution Rapid Refresh connects

Builds on 2

  • Rapid Refresh

    The HRRR starts from the Rapid Refresh's hourly analysis and adds detail on its finer grid.

  • Radar Reflectivity

    The HRRR blends radar reflectivity into its starting state, which gives storms already on radar a head start in the forecast.

Leads to 4

  • National Blend of Models

    The National Blend includes rapid-update models such as the HRRR among its inputs.

  • National Operational Hydrologic Remote Sensing Center

    The NOHRSC snowfall analysis starts from a background that uses HRRR precipitation forecasts, among other inputs.

  • Nowcasting

    Rapidly updating models such as the HRRR support the human nowcast by ingesting the latest observations each hour.

  • Storm-Chase Targeting

    Targeting weighs the ingredients against high-resolution guidance such as the HRRR, though current observations usually win when they disagree.

Part of 2

Includes 2

  • Near-Surface Smoke

    Near-surface smoke is a field from HRRR-Smoke, the smoke version of the HRRR model, produced alongside vertically integrated smoke.

  • Vertically Integrated Smoke

    Vertically integrated smoke is a field from the HRRR smoke model, totaling smoke through the full column from the surface to high aloft.

Contrasts with 1

  • Rapid Refresh Forecast System

    The RRFS is NOAA's next storm-scale system, but the HRRR is not being retired with it and even supplies members to its ensemble.

Appears here 1

  • Model Bias

    The HRRR offers a documented case of model bias, with its version 1 running warm and dry in the summer daytime boundary layer.