The Rapid Refresh Revolution

There is a major revolution developing  in weather prediction, something many professionals are becoming aware of and using, but relatively unknown to most others:  The Rapid Refresh Revolution.

How many times do you get the latest forecast over the web, radio, or TV and the short-term forecast is different than what is happening outside?  This is not rare, unfortunately.   We may be able to fix this.

Short-term forecasts are really important; they help us decide whether to take that bike ride, mow the lawn, take a run, and innumerable other tasks.   For folks worrying about wind energy, a good short-term forecast is worth huge sums and for those who maintain our roads, a good short-term forecast can make the difference between gridlock and a free-flowing highway.

Typically, the numerical weather forecasts made by major centers are only done every 6 hours (National Weather Service GFS and NAM models) or 12 hours (like the European Center for Medium Range Forecasts).   Since forecast errors increase in time, these forecasts are often a bit stale when we get them. Furthermore, many of the global and national modeling systems are run a modest resolution and are not started (or initialized) with a full set of local weather data.   The result of these and other issues is that short-term forecasts based on them are frequently unskillful, even in the short-term.

But some imaginative atmospheric scientists at NOAA's Earth System Research Lab in Boulder, Colorado have developed an alternative approach to weather prediction, one designed to optimize short-term forecasts:  the Rapid Refresh (RR)  approach (also called the Rapid Update Cycle).

In the current RR approach, high-resolution (13 or 3 km grid spacing) numerical weather prediction models (WRF model) are initialized and run EVERY HOUR using as many local weather assets as possible, including surface networks, weather radars, radiosondes, aircraft data, and others.  The model closely duplicates the detailed structure of the atmosphere at the starting time and then is run out 15-18 hours, not the days or weeks commonly used for the national or global models.

Begun at the Rapid Update Cycle in 1994 (60 km grid spacing, update every 3 hours), the Rapid Refresh methodology has been perfected over the years, particularly as computer power and high resolution data have increased.  5 years ago the Rapid Refresh methodology did not impress me, but during the past few years I and others interested in forecasting have noted a very large improvement in skill.   To put it succinctly, the skill of NOAA/NWS Rapid Refresh is now sufficiently accurate in the short term to be a game changer for short term forecasting.   But not perfect by any means.

There are two resolutions of the current  NOAA/NWS Rapid Refresh system, one at 13 km (called Rapid Refresh) and 3-km (the High Resolution Rapid Refresh or HRRR).  You can view these yourself here and here.  Every hour they construct a high-resolution analysis as well as short-term (15 or 18 h) forecasts of all major weather parameters.

Let's consider an example.  Here is the radar image at 2 PM on Friday.   The yellows are heavy precipitation.

Friday afternoon a very strong convective line developed over north Puget Sound and adjacent areas. Another line of heavy precipitation in eastern Washington.  

The HRRR two hour forecast had the Puget sound band in the right location and orientation, as well as a heavy band of rain in eastern Washington. Not too bad.

Compare the HRRR 2-h forecast to the 9 h prediction from the UW WRF, which was initialized at 5 AM.   The Puget Sound band is not there and the precipitation in eastern Washington is too weak.

This is not an exceptional case.   Today, the Rapid Refresh (13km) system is operational and available on National Weather Service web servers.  The HRRR is not operational, but only available in test mode.  The reason?  LACK OF COMPUTER POWER available to the National Weather Service operational center (Environmental Modeling Center).  You know what I think of that!

The Rapid Refresh approach will make possible greatly improved NOWCASTING, short term descriptions of what is happening now and in the near term.  Imagine a whole new generation of smartphone apps, showing you exactly what weather to expect in the next few hours.  Greatly enhanced renewable energy guidance for the hours ahead. A whole cottage weather industry could be built around such detailed local weather forecasts, and I suspect will.