U.S. Numerical Weather Prediction: Time for a Restructuring?

1
U.S. Numerical Weather Prediction Time for a
Restructuring?

• Clifford Mass
• University of Washington

First Symposium on the Weather and Climate
Enterprise Austin, Texas, January 7, 2013
2
Mediocrity is self-inflicted. Walter Russell

3
U.S. NWP starts with huge advantages

• Largest meteorological research community
• Largest governmental weather prediction
  organization.
• NWP began in the U.S. and most NWP breakthroughs
  began here. And still do.
• Largest collection of governmental (e.g., NOAA
  OAR, GFDL) and quasi-governmental research labs
  (e.g., NCAR) in the area.
• Vigorous private sector with substantial
  expertise in NWP and interest in NWP products.
• The NWS has an excellent office system for
  interfacing with local communities.

4
With all those advantages, the U.S. has lost
leadership in NWP

• Our global model (GFS) is clearly inferior to the
  ECMWF model-both for deterministic and
  ensemble-based forecasts.
• Other nations have higher resolution and better
  post-processed mesoscale ensemble systems.
• We have stagnated in the area of model
  post-processing.

5
A consistent story ECMWF is best and the U.S.
GFS is tied for second place with UKMET
Graphics courtesy of NCEP
6
ECMWF model is usually the most skillful for
major storms like Sandyand the media and
others have noticed
7
The most disturbing part of this is story is not
that we are behind the Europeans and others, but
that we are well behind what this nation is
capable of (which is far beyond ECMWF)And this
secondary status is completely unnecessary and
self-inflicted.
8
In this talk I will discuss some of the reasons
for U.S. NWP falling behind

• Poor leadership and lack of vision.
• Inadequate computer resources.
• Lack of cooperation between research and
  governmental NWP.
• Lack of enterprise-wide coordination,
  prioritization and planning.
• Structural organizational problems in NOAA.
• Lack of extramural research support by NOAA/NWS.
• A willingness to accept mediocrity.

9
Lack of computer resources

• NCEPs Environmental Modeling Center (EMC) is
  responsible for a very wide range of numerical
  forecasts (global high-res and ensembles,
  national/regional high res and ensembles, wave
  and storm surge models, climate models, etc.),
  but has FAR less computation resources than ECMWF
  which does only global prediction.

10
Lack of Computer Resources

• ECMWF has two machines, each with 24546 cores and
  a computational ability of .75 petaflops. These
  machines are 37 and 38 on the worldwide list of
  top 500 computers. 
• The National Weather Service has two computers
  that are not even on the top 500 list.   Each has
  4992 processors and an ability to do .07
  petaflops.  
• The NWS has less than 10 of the computer power
  as ECMWF and has many more responsibilities.

11
Implications of inferior computational resources

• ECMWF has the computer power to run their
  high-resolution global forecast at 16 km
  resolution, while ours is 25 km.
• Their global ensemble forecasting system has
  twice the resolution as ours.
• They can use an advanced data assimilation
  approaches (e.g., 4DVAR).
• The U.S. runs a coarse, inferior regional
  ensemble system (SREF).

12
But climate simulations get the big machines!

• NOAA's Fairmont, .38 petaflop
• DOE's Sequoia-20 petaflops
• DOEs Titan-20, 27 petaflops
• and many more!

Fairmont Computer
Jane Lubchenco It will help us tackle the
great big huge problem of generating better
information about climate change at the regional
scale.
13
Poor organization in NOAA

• The research supporting NWS NWP is NOT in the
  NWS, but in NOAA.
• Thus, the heads of NCEP and EMC do not control
  the folks that provide the critical research for
  their mission (e.g., NOAA OAR, GFDL, etc.)
• The result is inefficiency, working at cross
  purposes, and slowed Research to Operations (R to
  O).

14
Research and operations needs to be in one entity
Operational NWP
NWP Research
15
There is very little coordination of the research
and operational communities
U.S. NWP The Uncoordinated Giant
16
A prime example Weather Research and Forecasting
(WRF) model

• After years of separation between the research
  and operational communities (e.g., Eta model
  versus MM5), many hoped that the NWS and NCAR
  could develop one mesoscale model that the entire
  community could use.
• Never happenedthe NWS went its own way with the
  Eta and NMM models, research community went with
  WRF ARW.
• An unnecessary tragedy for our profession.

17
But WRF became a brand name, with little meaning!
18
Lack of Coordination

• There is no group that coordinates the research
  and development of the U.S. NWP community to deal
  with pressing problems
• Deficiencies in boundary layer schemes
• Optional approaches for creating mesoscale
  ensembles
• High-resolution data assimilation.
• And many more.
• The result? Progress is slow and money is
  wasted.

19
Lack of Coordination

• One National Academy Committee after another has
  recommended that NCEP/EMC needs to have an
  advisory committee never happens.
• Might the proposed Weather Commission be a move
  in the right direction?

20
Lack of rational observing system planning

• Recently, the media has been abuzz about the
  expected gap in polar-orbiter weather satellite
  coverage, suggesting weather forecasting skill
  would decline.  
• The NOAA/NWS polar orbiter acquisition program
  has been characterized by mismanagement for
  years, not only delaying the next generation
  satellites, but costing the nation billions of
  dollars.

21
Rational observing system design

• NOAA/NWS needs to assess the optimal collection
  of observations, including dropping old data
  sources that are no longer needed.  There is a
  rational way to do so through Observing System
  Simulation Experiments (OSSEs) and Observing
  System Experiments (OSEs). 
• NOAA/NWS has done very little of this and
  billions of dollars of hardware is being
  purchased without a clear understanding of the
  impacts of new sensors and satellites. 

22
And there is much more I have not had time to
talk about.

• Lack of extramural support by the National
  Weather Service for applicable NWP research.
• Glacial progress towards developing
  high-resolution ensembles and probabilistic
  forecasts, with only one individual at NCEP
  working full time on this critical issue.
• NWS leadership from the military that have not
  had the background and experience to lead a
  highly technological entity.

23
No time

• Fifty year-old NWP post-processing approaches
  (MOS) still dominates.
• Inadequate model verification effort.
• No public strategic plan or vision document for
  U.S. NWP.

24
So what needs to be done?

• Secure a much bigger computer (10-500x) for NCEP
  or allow the NWS to take over Fairmont or another
  government machine.
• Establish a framework for community cooperation
  and coordination.
• Move NCEPs EMC into NOAA and combine with
  relevant groups in OAR.
• Begin a rational approach to observational system
  design.

25
But most of all U.S. operational NWP need more
priority, more leadership, and vision The
problem is much larger than the National Weather
Service.
26
The End