RRFS Operationalization Update

After many postponements, the Rapid Refresh Forecast System (RRFS) and RRFS Ensemble Forecast System (REFS) are still scheduled to become operational on November 3rd.  

The RRFS and REFS are the latest "high resolution" modeling systems developed by NOAA for the National Weather Service.  I have high resolution in quotes because the reality is that many countries are now far ahead of the US National Weather Service in resolution and ensemble design.

The RRFS and REFS have 3-km horizontal grid spacing and are based on the Finite-Volume Cubed-Sphere FV3 dynamical core developed by NOAAs Geophysical Fluid Dynamics Laboratory.  A dynamical core is basically the guts of a numerical weather prediction model as it represents the mathematics and numerics used to simulate the motion of the atmosphere on Earth.  

In 2016, NOAA decided that the FV3 dynamic core would be the guts of their numerical weather prediction models, starting with the Global Forecast System (GFS).  The FV3-based GFS became operational in 2019 and FV3-based Global Ensemble Forecast System (GEFS) became operational in 2020.  Following those upgrades, NOAA also hoped to replace their High Resolution Rapid Refresh (HRRR) forecast system, which provides higher resolution (3-km grid spacing) for the Continental US with an FV3-based modeling system that covered a larger area and also provided ensemble forecasts.  

However, the development of that FV3-based modeling system (i.e. the RRFS) and ensemble (i.e. the REFS) has taken far longer than planned.  The FV3 dynamical core has proven challenging at high resolution.  So much so that NOAA is going to pivot to a different dynamical core, the Model for Prediction Across Scales (MPAS) for future modeling upgrades.  MPAS was developed by the National Center for Atmospheric Research and is expected to perform better at 3-km grid spacings.  

At the University of Utah, we have been producing snowfall forecasts with the experimental RRFS and REFS for a couple of years.  These are available on weather.utah.edu and hopefully the operational version will automatically be available when it starts running in a few days.  That said, I don't have a good handle on the biases in the new RRFS and REFS, so we will all be learning as we go this winter.

One thing that is clearly different from the HRRR is that the terrain in the RRFS is smoother, despite the similar grid spacings.  Below is a comparison of the RRFS terrain with the HRRR, including a difference plot for the central Wasatch.  Contours of real terrain based on a digital elevation model with ~900 m grid spacing is contoured.  Neither model at this grid spacing is able to resolve the corrugated terrain of the central Wasatch, but the RRFS is considerably smoother than the HRRR.  The HRRR actually has a fairly pronounced ridge where the Alpine Ridge is south of Little Cottonwood.  This feature is as much as about 500 meters lower in the RRFS.  

Whether or not other aspects of the RRFS and REFS design "compensate" for the lower terrain when it comes to precipitation forecasts is unclear.  Sometimes the approaches used to produce precipitation can be tuned to be more productive.  We shall see.  

The REFS consists of the RRFS forecast plus five additional members run every 6 hours.  On weather.utah.edu we are providing plan view plots and plume plots based on these six runs.  Below is the forecast from 0600 UTC last night for Alta Collins as an example.  This latest run is a bit more bullish on the potential for snow later today and tonight at high elevations than prior runs, although none of the members produce more than about 8" of sloppy, high-density snow (not that there's anything wrong with that).  

It's been sad to see the operational modeling in the US stagnate over the past few years.  As I discussed in May (see "I Have Seen the Future"), Europe is far ahead of the US National Weather Service in operational modeling resolution.  MeteoSwiss for example runs an 11-member 1.1 km ensemble eight times a day to 33 hours and a 21-member 2.1 km ensemble four times a day to 120 hours.  The bright light in the US is in the private sector, which appears to be driving a lot of innovation these days in areas such as satellite remote sensing and its use to improve forecast models, not to mention artificial intelligence weather prediction.  

Reflections on the Prior Post

Some readers raised concerns about the use of an AI generated image to lead the prior post "It's Time to Pray for Snow" (reproduced below). Fair enough.  Although I was simply looking for something humorous to start the post, the image should have been identified as an AI-generated conceptual illustration.  It was not my intent to mislead or misinform.  I appreciate readers raising these concerns and I will be more cautious moving forward.