There has been a lot of discussion of how U.S. global numerical weather prediction has fallen behind others, such as the European Center and the UK Met Office. A serious problem that the National Weather Service is finally giving some attention.
But perhaps the greatest weakness of U.S numerical weather prediction doesn't regard global prediction, but rather forecasting in our own nation.
A deficiency that undermines the ability of the National Weather Service (NWS) and the private sector to forecast severe thunderstorms, flash flooding, heavy precipitation in terrain, and other critical weather features. A gap in NWS capabilities that is costing lives and resulting in large and unnecessary economic losses.
The lack of a high-resolution ensemble prediction system over the U.S.
Why does the U.S. need a high resolution ensemble system?
All numerical weather forecasts possess uncertainty, either due to uncertainty in the initial conditions of the forecasts, in the descriptions of physical processes (model physics), or in the approximations used in solving the equations.
This uncertainty must be estimated and communicated to the public. For example, instead of saying the forecast for tomorrow is 63F, we must tell folks the probabilities of various temperature ranges. Instead of saying it will snow or not, we must tell people the probabilities for snow of various amounts. This is called probabilistic prediction.
Fortunately, meteorologists have developed a technology to make probabilistic forecasts: ensemble prediction, in which forecasts models are run many times, each time a little differently. We can slightly vary the initial conditions, or how we describe cloud processes or other physics in the model, or by varying how we solve the equations.
The resolution of an ensemble system is important and is closely related to the spacing of the grid points on which the forecasts equations are solved. When the grid points are close together, resolution is higher, and vice versa.
One of the most critical forecast problems deals with convection and thunderstorms, associated with cumulus and cumulonimbus clouds. These are the systems producing severe thunderstorms, tornadoes, hail and lightning. here is deep experience showing that demonstrates that forecasts models should have grid spacing of 3-4 km or less to have a chance of simulating and predicting these critical weather features.
Similar grid spacings are needed to get mountain weather right, such as flow through gaps (like the Columbia Gorge or the Fraser River Valley) or heavy precipitation associated with fronts and terrain. The critical details of hurricanes (like their eye wall circulations) also require high resolution.
In short, to produce ensemble forecasts of critical weather features, the National Weather Service needs to run an ensemble of at least 3-4 km grid spacing, and 2-3 km would be better. I am not the only one saying this. A long list of National Academy of Sciences reports (see below), national workshops, and advisory committees have recommended exactly this. A large (30-50 members) ensemble at 2-4 km grid spacing, run several times a day is required. (each of the separate ensemble forecasts is called a member)
What is the National Weather Service Actually Doing?
Today, the National Weather Service is running two operational ensemble systems: a 21 member GEFS ensemble (35 km grid spacing) and a 26 member SREF ensemble (26 members, 16 km grid spacing). The resolution of these ensemble systems is not even close to what is needed for local weather prediction. And their deficiencies don't end there. GEFS is too small and is underdispersive, which means the forecasts frequently do not encompass truth. SREF is the combination of two different models that often produce distinctly different solutions, and again often does not encompass the truth.
And now a bit of dirty meteorological laundry. The National Weather Service knows they need a high resolution ensemble system at convection-allowing resolutions (again 2-4 km). But they have not been ready to invest in doing it right. So what did they do? Kludge together a collection of forecasts run by different groups into a small stop-gap ensemble system: the Storm-Scale Ensemble of Opportunity (SSEO) run out of the NOAA/NWS Storm Prediction Center.
SSEO is made up of 7 members with grid spacing of 4-5 km and run by several organizations. It is too small, with a poor design, and not reliable. But it is better that nothing.
Somewhat embarrassing for the National Weather Service, a major U.S. research organization, the National Center for Atmospheric Research, is running a 10-member ensemble at 3-km grid spacing, once a day (see below). It is a hit with both the research and operational communities, with NWS forecasters using and citing it heavily. It is not clear how long NCAR will continue this wonderful service.
Time for the National Weather Service to Act
The National Weather Service has been kicking this can down the road too long and it is time that hey finally build a state-of-science high-resolution ensemble system over the U.S. A major excuse has been lack of computer power, but recently the NWS has acquired new computers that are running half-empty as I write this. There is enough spare computer resources to immediately begin a 5-10 member, 3-km ensemble, 2-4 times a day out to 48 hr.
NOAA/NWS leadership must work with the new administration to secure the additional computer resources to build out our national high-resolution ensemble prediction system. Trump wants to make American great again. Fine. Let's start by making America's numerical weather prediction state-of-the-science. You can call it critical infrastructure.
The computer requirements are large, but quite possible (the cost is equivalent to one fighter jet). Support for this new capability should be bipartisan since everyone will benefit. There are a lot of red states whose citizens are receiving mediocre quality forecasts for very active weather (thunderstorms, hurricanes, flash floods) and the proposed system will provide a radical improvement.
A high-resolution ensemble system would improve snow forecasting in the Northwest, heavy rain forecasting in California, thunderstorm forecasting in Nebraska, and coastal storm prediction over the East Coast. It would lay the groundwork for better use of observations though improved data assimilation. Every NWS forecaster I have talked to acutely wants it. Every relevant advisory committee I have been on recommends it.
The time for delay in building this critically needed weather infrastructure is over. NOAA and NWS management need to make a high-quality convection-allowing operational ensemble prediction system a reality in this country.