The Rapid Refresh Forecast System (RRFS) is a computer weather forecasting model developed by the U.S. National Weather Service (NWS) to produce short-term, highly detailed forecasts across North America. It is built to update itself every hour with fresh weather observations and to simulate individual thunderstorms directly, making it especially useful for predicting fast-changing weather such as severe storms, heavy rainfall, and wildfire smoke.
Overview
RRFS is developed by the National Centers for Environmental Prediction (NCEP), the forecasting division within the NWS, which itself is part of the National Oceanic and Atmospheric Administration (NOAA) (NOAA Global Systems Laboratory). It is described by NOAA as the agency’s “next-generation, rapidly updating, high-resolution weather model,” meant to give forecasters a detailed, frequently refreshed picture of the atmosphere for the next one to three days (NOAA Global Systems Laboratory).
How it works
RRFS divides the atmosphere over North America into a grid of roughly 3 kilometers (about 1.9 miles) between grid points, fine enough that the model can simulate individual thunderstorms forming and moving, rather than only estimating storm activity in a general way, as coarser models must do (National Weather Service, 2026a). Over Hawaii and Puerto Rico it runs at a slightly finer 2.5-kilometer spacing, and it also produces a coarser 13-kilometer view of all of North America. A special, relocatable version of the model can run at an even finer 1.5-kilometer spacing over a chosen area to support wildfire-related forecasting (National Weather Service, 2026a). This same fine grid spacing shapes how well RRFS can represent terrain itself. Grid spacing this fine is enough to resolve smaller landscape features, such as ridgelines, valleys, and coastlines, rather than smoothing them into a coarser approximation the way older, wider-grid models had to. This is the same rationale NOAA has documented for HRRR, the similarly fine-grid model that is also folded into REFS: finer resolution was chosen specifically to better capture convective storms, coastlines, and terrain together, rather than any one of them alone (Dowell et al., 2022). The model’s vertical layers work the same way, built around a coordinate that tracks the actual land surface near the ground, then gradually transitioning to simpler, flat pressure levels higher up, away from the influence of terrain (NOAA Geophysical Fluid Dynamics Laboratory).
The model runs on the “FV3” dynamical core, short for Finite-Volume Cubed-Sphere, which is the mathematical engine that calculates how air, moisture, and heat move through the atmosphere over time. This is the same core NOAA uses in its global weather models, including the Global Forecast System (GFS), which switched to FV3 in 2019 (NOAA Global Systems Laboratory; National Weather Service, “Aware,” May 2019). Every hour, RRFS is fed fresh weather observations, such as satellite data, radar, weather balloon soundings, and surface station reports, in a process called data assimilation, so that each new forecast starts from the most current possible picture of the atmosphere. RRFS produces a new short-range forecast (18 hours ahead) every single hour, and four times a day it produces a longer 84-hour forecast (National Weather Service, 2026a).
A companion system, the RRFS Ensemble Forecast System (REFS), runs the model as an “ensemble,” meaning it produces several versions of the forecast (a control run plus six additional members, one of which is the High-Resolution Rapid Refresh model, or HRRR, along with additional time-lagged members) starting from slightly different conditions, meaning each version begins from a slightly different best estimate of the current state of the atmosphere, since no single starting snapshot of the atmosphere can ever be perfectly precise. Comparing these versions helps forecasters gauge how confident to be in a given forecast, since if all the versions agree, confidence is higher than if they diverge. REFS produces 60-hour forecasts four times a day (NOAA Global Systems Laboratory; National Weather Service, 2026a).
RRFS also includes a built-in smoke and dust component, called RRFS-Smoke Dust (RRFS-SD), which NOAA describes as its first operational model of this kind, tracking how smoke from wildfires and windblown dust move through the atmosphere (NOAA Global Systems Laboratory). This is directly relevant to the Sky Islands region, where blowing dust is common enough to have its own local severity scale; see the Phoenix Dust Storm Scale (PHX-DUST) entry.
What it is used for
Because it updates so frequently and resolves individual storms, RRFS is aimed primarily at short-range forecasting of fast-changing, high-impact weather: severe thunderstorms, heavy rain and flash flooding, and, through its fire-weather configuration and smoke and dust component, wildfire behavior and smoke impacts on air quality. Forecasters use it alongside other models to build the short-term outlooks issued by local NWS offices.
Relationship to other forecast models
RRFS and REFS are intended to replace several older NWS forecast systems: the North American Mesoscale Forecast System (NAM), the High-Resolution Ensemble Forecast system (HREF), the Short-Range Ensemble Forecast system (SREF), and most domains of the High-Resolution Window (HiResW) forecast system (National Weather Service, 2026a). The HRRR model, a separate high-resolution model that has been in operational use since 2014, is not being retired. It continues to run on its own and its output is also folded into the REFS ensemble alongside RRFS (National Weather Service, 2026a).
Development and current status
NOAA had already begun evaluating the FV3-based, convective-scale model configurations that became RRFS by 2018, when NOAA’s Hazardous Weather Testbed began including them in its annual Spring Forecasting Experiments (Developmental Testbed Center, 2022). Its path to full operational use has taken longer than originally planned and was pushed back several times.
NOAA’s implementation timeline for this changeover has shifted more than once, moving from an original August 31, 2026 target to October 6, 2026 and then to its current date, October 14, 2026 (National Weather Service, 2026b). Implementation could still be postponed further if a significant severe-weather event is underway when the date arrives.
Independent testing by NOAA’s forecasting testbeds and outside researchers has found mixed results so far. A peer-reviewed 2024 study found that RRFS tends to depict storms as substantially larger than they turn out to be in reality, by roughly 40 to 60 percent on average and by as much as 140 percent for the most intense storm cores, though it captured the timing of new storm formation somewhat better than the HRRR model does (Grim et al., 2024). Separately, in blind evaluations run by NOAA’s Hazardous Weather Testbed, RRFS has performed competitively against other high-resolution models, for instance ranking third out of five models compared for severe-weather forecasting and outperforming most other models besides HRRR for warm-season rainfall forecasting (NOAA Global Systems Laboratory, 2026). NOAA is already developing a planned successor, sometimes referred to as RRFS Version 2 or RRFS-MPAS, which would replace the FV3 core with a different modeling approach called MPAS (Model for Prediction Across Scales); this next version is targeted for roughly the 2027 to 2028 timeframe (NOAA Global Systems Laboratory).
Sources
National Oceanic and Atmospheric Administration, Global Systems Laboratory. “Rapid Refresh Forecast System (RRFS).” https://gsl.noaa.gov/rrfs/.
National Weather Service. Service Change Notice 26-48, “RRFS and REFS Implementation.” https://www.weather.gov/media/notification/pdf_2026/scn26-048_RRFS_and_REFS_Implementation.aab.pdf. 2026.
Dowell, D.C., et al. “The High-Resolution Rapid Refresh (HRRR): An Hourly Updating Convection-Allowing Forecast Model. Part I: Motivation and System Description.” Weather and Forecasting, 37(8), 1371-1395, 2022.
National Oceanic and Atmospheric Administration, Geophysical Fluid Dynamics Laboratory. “FV3: Key Components.” https://www.gfdl.noaa.gov/fv3/fv3-key-components/.
National Weather Service. “Aware” (newsletter), May 2019. https://www.weather.gov/media/publications/Aware/19may-aware.pdf.
Grim, J.A., J.O. Pinto, and D.C. Dowell. “Assessing RRFS versus HRRR in Predicting Widespread Convective Systems over the Eastern CONUS.” Weather and Forecasting, 39(1), 121-140, 2024. https://repository.library.noaa.gov/view/noaa/61118/noaa_61118_DS1.pdf.
Developmental Testbed Center. “Advances in the Rapid-Refresh Forecast System as Seen in NOAA’s Hazardous Weather Testbed’s Spring Forecasting Experiment.” 2022. https://dtcenter.org/news/2022/03/advances-rapid-refresh-forecast-system-seen-noaa-s-hazardous-weather-testbed-s-spring-forecasting-experiment.
Mahalik, Matthew. “NOAA Testbeds evaluate the Rapid Refresh Forecast System.” NOAA Global Systems Laboratory, March 25, 2026. https://gsl.noaa.gov/news/noaa-testbeds-evaluate-the-rapid-refresh-forecast-system/.
National Weather Service. Service Change Notice 26-48 Updated, “RRFS and REFS Implementation.” https://www.weather.gov/media/notification/pdf_2026/scn26-048_Updated_RRFS_and_REFS_Implementation_aad.pdf. 2026.