The History, Future and Impact of
Regional Numerical Environmental
Prediction at the UW
Cliff Mass, David
Ovens, Rick Steed,
Jeff Baars, Mark
Albright, and Neal
Johnson
University of
Washington
A Major Department Facility
• Has led to dozens of papers
and millions of dollars of
grant support.
• A major resource for
department classes.
• Has led to spin-off business
that employ a large number
of our grads.
• Has served as a national
model that has been
duplicated elsewhere.
A Major Department Facility
• Led to and supported major field programs
and research initiatives.
• Has employed 3-5 staff members that in turn
have been a resource to the department.
• Established a new model of cooperative
funding.
Key Concept: End to End Regional
Environmental Modeling
Regional Environmental Testbed
• To serve as a platform to test the impacts of
resolution and modeling innovations.
• To find and solve issues with high-resolution
mesoscale modeling
• To test the concept
of regional environmental
prediction.
This Talk
• Will describe the history and impacts of this
department facility.
• Will examine a new type of organizational and
funding entity: the Northwest Modeling
Consortium
• Will propose we can extend this idea to create
a parallel departmental/college entity, with
perhaps as much potential: a center for
regional climate simulation and impacts.
History
• The NW modeling
consortium can be traced
back to the early 1990s
when regional air quality
agencies, the UW, and the
NWS started meeting
about another
problem…the lack of
upper air soundings over
the Puget Sound region.
The Solution: Purchasing a 915 Mhz
Profiler
• After some discussion, the group decided that a
radar wind profiler would be the best solution, but
no one entity had enough funds.
• The solution: form a consortium of local agencies
and groups to share the costs and work, with the
device located at NOAA Sand Point.
• A 915 Mhz Radian radar wind profiler was purchased
in 1992 ($200K) and the data has been available
every since.
• The Consortium was born.
915 Mhz Profiler With RASS
Seattle
Profiler
Observations
Consortium Phase II
• During the 1980s and
early 1990s, my group
was doing extensive
experimentation with
first RAMS and then
MM4/5.
• It became clear that with
sufficient resolution,
mesoscale models could
simulate and predict
many local weather
features of importance to
the air quality and other
communities.
Early MM4 simulation (10 km) of a Puget
Sound convergence zone (Jim Steenburgh)
The Ask
• I pitched to the Consortium that they
purchase one of the new UNIX workstations
and support a Jim Steenburgh postdoc to
simulate some cases of interest, particularly
for air quality issues (e.g., heat waves,
onshore push). In 1994 they agreed.
DEC Alphaserver 250: 1 processor
Jim’s and Mark Albright’s
Inspiration
• Jim not only did a number of research runs,
but in 1995 decided to try (with Mark)
creating a real-time system: running MM5
once a day at 27 km over the Northwest. (the
NWS Eta model was at 48 km then, only 80 km
grids available)
• Became apparent that not only could a
university do this, but the forecasts were often
very good, particularly for regional terraininduced features.
A Major Early
Success: The
December
12, 1995
windstorm
The NW Modeling Consortium
• The agencies supporting the project were excited by
the results, and decided to invest in a far larger
computer and to jump to much higher resolution.
The NW modeling consortium was formed.
• In 1996, Consortium funds, plus a major discount
from SUN Microsystems, allowed us to secure at 14
processor SUN server, and try a new domain
structure: 36-12 km….really high resolution for that
day.
• In 1997, more upgrades allowed us to add 4-km (new
machine ES 6500)
• All based on MM5 (Penn. State/NCAR mesoscale
model)
Domains in 1997
4 km
12 km
1996-1997
ES4000: 14 processors
ES 6500: 30 processors
Early Members of the Consortium
• Puget Sound Air Pollution
Control Agency
• UW
• National Weather Service
• EPA
• Washington State Depts. of
Ecology and Transportation
• U.S. Forest Service
• US Navy
• Seattle City Light
Naydene Maykut, PSCAA, First Chair,
NW Modeling Consortium
Regional High Resolution
Forecasting: Fast Forward to 2012
• MM5 replaced by WRF
• Larger domains … plus a new 1.3 km inner
nest—perhaps the highest resolution
operational NWP in the U.S.
• Model run on 136-core SAGE cluster using
commodity (Intel Xeon) chips. Highly
parallelized using 40 Gbit per sec.
Infinitband
• Over 100 TB of disk storage
SAGE
Cluster
One of the
greenest
computer
facilities
on the
planet
36 km
12
km
4 km
1.33 km
Current Real Time Runs
• WRF driven by NWS GFS model
– 36-12 km out to 180h, twice a day
– 4 km, 84 h, twice a day
– 1.33 km, 48h, twice a day
• MM5 driven by NWS NAM model (early look)
– 36-12 km, 72 h, twice a day
Reliability: 99% plus.
How many are looking at UW NWP
Products on the Web?
How many are looking?
A Number of Organizations Are
Getting the Model Grids
• National Weather Service: brought into their
interactive system
• Forest Service: to support wildfire and smoke
operations.
• Washington State Univ: Air quality and
smoke modeling
• Washington State: to force air quality models
• KING-5 TV for on-air graphics
• …and others.
WSU Air Quality Modeling
Managing Field Burning
WSDOT
Ventilation Index for AQ Agencies
U.S. Forest Service
King-5 Futurecasts Driven by UW
WRF
2012 Consortium
• Rob Elleman, Chair (replaced Rob Wilson)
• Members:
–
–
–
–
–
–
–
–
Private Sector (Iberdrola, King)
UW
Puget Sound Clean Air Agency
EPA
State of Washington (Ecology, Transportation, DNR)
NWS
Forest Service
City of Seattle (SPU, City Light)
A Reliable Funding Approach
• Consortium members come and go, but the
total support has been fairly stable.
• Supports 2-4 staff members, lots of computer
acquisition and departmental support services
Back to History
Hydrology and 3-Tier
• During the late 90s an undergrad (Ken Westrick)
was accepted into our grad program.
• In our M.S. program, he took a class given by
Dennis Lettenmaier on hydrology and learned
about Dennis’ distributed hydro model
(DHSVM).
• Ken asked: why not connect DHSVM to MM5 to
create a high-resolution hydrological prediction
system?
• The result was a success and a M.S. degree
The results were good enough that
it was used by NWS and City Light.
3-Tier
• Working for a few years as department staff
member, Ken decided to start a business to
provide hydrological forecasts using his new
technology.
• Called the company 3-Tier and took a second
loan on his home to get the funding.
• My parting gift…his first contract…the City
Light support I had been getting.
• The result: a great success.
Leading wind energy prediction firm in the U.S.
Five overseas offices.
3-Tier
Ken Westrick: Founder and
First CEO. M.S. UW Atmos.
Pascal Storck: First
President Ph.D. UW Civil Eng,
UW Grads Employed at 3-Tier
Atmos Sci
• Kristin Larson, PhD
• Jim McCaa, PhD
• Eric Grimit, PhD
• Scott Eichelberger, PhD
• Jeff Yin, PhD
• Matt Garvert, PhD
• Mark Stoelinga, PhD
• Celeste Johanson, PhD
• Clark Kirkman, PhD
• Ken Westrick, MS
• Sara Harrold, MS
• Kyle Wade, BS
• Mark MacIver, BS
Civil Engineering
Pascal Storck, PhD
Bart Nijssen, PhD
Andy Wood, PhD
Amy Vandervoort, MS
Matt Wiley, MS
Paul English, BS
Other
Scott Otterson, PhD, ElecEng
Christian Sarason, MS, Oceanography
13 Ph.Ds, 5 MS, 3 B.S.
One more thing about
Ken…he did a study of
the lack of radar
coverage on the coast,
which was published in
BAMS. The impact
would be substantial:
more later!
Some Impacts on Research
• During the 1990s, my group
(especially Brian Colle) were
actively verifying precipitation in
MM5 and found clear
deficiencies….such as substantial
overprediction in terrain and to its
lee.
• Talked about this at depth with
Peter Hobbs (often while running)
and he noted that we lacked
proper data to tell what was
wrong.
• Peter suggested a field
experiment—and that became
IMPROVE. (Bob Houze, and I, were
also PIs)
Peter Hobbs
British Columbia
Legend
Washington
UW Convair-580
Airborne Doppler
Radar
Two
IMPROVE
observational
campaigns:
S-Pol Radar
Offshore Frontal
Study Area
BINET Antenna
Olympic Mts.
Olympic
Mts.
Paine Field
Univ. of Washington
NEXRAD Radar
Area of MultiDoppler
Coverage
Wind Profiler
Rawinsonde
Westport
WSRP Dropsondes
Special Raingauges
Columbia R.
PNNL Remote
Sensing Site
90 nm
(168 km)
Washington
Ground Observer
0
S-Pol
Radar
Range
S-Pol
Radar
Range
100 km
Portland
I. Offshore
Frontal
Study
(Wash. Coast,
Jan-Feb 2001)
Oregon
Terrain Heights
Salem
< 100 m
100-500 m
500-1000 m
1000-1500 m
1500-2000 m
2000-3000 m
Orographic
Study Area
Newport
> 3000 m
Rain Gauge Sites in OSA Vicinity
Santiam Pass
OSA ridge crest
Santiam Pass
Orographic Study Area
S-Pol Radar Range
SNOTEL sites
CO-OP rain gauge sites
50 km
Oregon
Medford
California
II. Orographic
Study
(Oregon
Cascades,
Nov-Dec 2001)
-20°C
-15°C
-10°C
-5°C
0°C
CV-580 flight track starting at the top at 2320 UTC 13 Dec 200, and ends
at 0310 UTC 14 December 2001
IMPROVE
• Resulted in a uniquely comprehensive
microphysical data set that is probably the
best in existence over terrain.
• Contributed to improvement in NCAR and
other microphysical schemes.
NCAR S-Band Dual-polarization “S-Pol” Radar
on the WA coast during IMPROVE
Revealed the great benefits of a
coastal S-band radar and with
Ken Westrick’s blockage mapping
led to:
The New Langley Hill Radar on the
Washington Coast
Langley Hill
Other Research Projects
• A number of other department research
projects have made use of the UW modeling
system.
– An example: the Olympics rainfall experiment:
Anders et al. 2007
MM5/WRF Driving ROMS: Coastal Ocean Model
(Parker Maccready, Oceanography)
The technology and experience of the regional
weather prediction effort was applied successfully
in regional climate simulations (Salathe et al…)
Data Everywhere
• In support of the nascent modeling effort in
the mid-1990s, we needed a lot more data
than was available as NWS/FAA airport sites.
• During that period a number of groups
established observing networks, whose data
was available over the nascent Internet (Puget
Sound Clean Air Agency, Schoolnet, etc.)
• Why not collect these networks, decode them,
and combine to create a dense mesonet?
• Mark Albright took this on.
NorthwestNet was born
Today
Over 72
different
networks
3000-4000
observations
per hour over
WA and OR
Others Noticed Mark’s Work: First, John
Horel, Utah leading to Mesowest
As Well as NOAA
Probabilities
• During late 1990s Brad Colman and I would go
back and forth about resolution versus
ensembles. Should we run down to 4 km or
go to 12 or 20 km ensembles.
• We tried resolution first.
• At the same time, there was a project to
evaluate various global models (Lynn
McMurdie led, Brett Newkirk, graduate
student). As part of this project we gained
real-time access to major global modeling
systems.
UW Mesoscale Ensemble System
• In 1999, Eric Grimit began building the UW
Mesoscale Ensemble System (UWME) that
used all the global model output to drive
MM5 at 36 and 12 km resolution…perhaps the
highest resolution ensemble system at the
time.
• Shortly, joined by Tony Eckel, and together
they made rapid progress in developing the
system and post-processing the output.
Computer
Infrastructure:
Linux DualProcessor
Clusters
“Ensemblers”
Eric Grimit (r ) and
Tony Eckel (l) are besides
themselves over the
acquisition of the new 20
processor Athlon cluster
UWME
Precipitation
UW MURI
(Integration and Visualization of Multi-source Information for Mesoscale
Meteorology: Statistical and Cognitive Approaches to Visualizing Uncertainty )
• The Department ensembles were in the center
of an interdisciplinary proposal to DOD to
create an end-to-end probabilistic prediction
system.
• Included UW Atmospheric Sciences, UW APL,
UW Statistics, UW Psychology.
• Big project, big money, big results.
Bayesian Model Averaging (BMA)
A standard approach to statistical inference in information.
Gaining wide acceptance in the weather prediction community for combining
discrete members of a forecast ensemble to produce a calibrated, predictive PDF.
[c.f. Raftery et al. 2005, Mon. Wea. Rev.]
PROBCAST
Regional Probabilistic Data Assimilation and
Forecasting (with Greg Hakim’s group)
The Long Term Future of the UW Modeling
Effort
• Based on a 64-member ensemble of forecasts at
36 and 4 km grid spacing. WRF model and DART
Ensemble Kalman Filter (EnKF) System
• Every three hours assimilate a wide range of
observations to create 64 different analyses.
• Then we forecast forward for 3 hours and then
assimilate new observations.
• Thus, we have a continuous cycle of probabilistic
analyses.
EnKF Ensemble Forecasting System
• We can run ensemble of forecasts forward to
give us probabilistic forecasts for any period
we want. Now doing 24h, four times a day.
• Planning to go to a 1-hr cycle and to use more
observations (e.g., more surface pressure
obs).
Mean and Spread of Analyses
4 km analyses
UW Regional NWP Effort Serves As
A Model
• Based on the success of the UW NW modeling
and consortium approach, a number of others
tried to clone or duplicate it.
• A major example: the U.S. Forest Service
FCAMS (Fire Consortia for Advanced Modeling
of Meteorology and Smoke) effort:
FCAMMS
Should the Department and
College Build Another Modeling
Center?
Center for Region Climate
Simulation and Impacts (CRCSI)
This Idea has the Professor
Tom Ackerman Seal of Approval
Luncturam Instituti Semper
Why a regional climate simulation
and impacts center?
Society needs to know the
local implications of global
warming for key
adaptation/infrastructure
decisions that are being made
now.
– Example: The City of Seattle was
about to spend ¼ billion dollars on
new drainage pipes that would be
used for nearly a century. What
diameter? Would short-term
precipitation become more
Why dynamical downscaling is
required
• GCMs have insufficient resolution to define
key regional weather features (e.g., major
terrain features), resulting in large simulation
errors that make their results undependable
for local decision making.
PCM
Cold
Wave
Under
Global
Warming
12 Feb 1990
PCM and ECHAM-5 Driving Small
Domain MM5: Crazy Cold Waves
Seattle 2080 Under Global Warming
Why Dynamical Downscaling?
• Weather regimes may
change, so statistical
downscaling may not
be appropriate.
– Changes in stability
and moisture
advection out of the
SW during summer.
And local mesoscale interactions can produce nonlinear interactions that are hard to statistically
downscale
1990s to 2050s
Temperature Change
Change in Winter Temperature (degrees C)
Difference between
MM5 and ECHAM5
Change in Winter Temperature (degrees C)
Snow melt on terrain produces banded structures
Other Challenges
• Must run multiple GCMs to get some handle
on uncertainty. Requires Ensemble dynamical
downscaling.
• Must use sophisticated post-processing to
insure the ensemble of dynamically
downscaled GCMs runs is properly calibrated
using contemporary periods.
– One approach: Bayesian Model Averaging
Secondary Models
• Need the ability to run hydrological models,
air quality models, coastal ocean models, and
others to examine the implications of the
changed statistics of a new regional climate.
DHSVM Hydrograph
From WRF driven by
A large-scale model
Creating a Center
• The UW, more than any other university, has the
essential components of such a center.
• Much of the technology developed for local
weather modeling can move over to this new
entity.
• Could be a flagship facility of the College of the
Environment.
• Establish a consortium of local, state, regional
and Federal agencies, AND private sector firms,
that will support the effort.
The New Center
• The Center will not only aid local stakeholders
but would develop the technology that could
be used in any location.
• And perhaps it could lead to new spin-off
businesses:
4-Tier
Corporation
We have developed a multidisciplinary regional environmental
prediction center…we can do the
same for regional climate.
Thanks
•
•
•
•
•
•
•
•
•
David Ovens, principal modeler
Rick Steed
Mark Albright
Jeff Baars
Neal Johnson
Phil Regulski
Harry Edmon and David Warren
Andrew Sattler
… and several others (Brian Colle, Jim Steenburgh,
Pascal Storck, Ken Westrick)
• Consortium members
The End or ….