What will soon be the nation’s flagship computer
model for weather prediction and research is getting a dedicated cyberspace
where scientists can put it through cutting-edge experiments. The Weather
Research and Forecasting (WRF) model—in the works since the late
1990s, and scheduled for operational use starting late next year—is
the first client of the new Developmental Testbed Center (DTC). A joint
effort of NCAR and NOAA, the center opens this autumn at NCAR’s
“The idea of the DTC is simple,” says Robert Gall. “It’s
a place where you can go and try out new ideas in numerical weather
prediction without interfering with forecast operations.” Gall
takes the reins as founding director of the testbed after more than
a decade as head of NCAR’s Mesoscale and Microscale Meteorology
Division. He’ll continue as lead scientist for the U.S. Weather
Research Program (USWRP).
Though located at NCAR, the center will be an autonomous entity, with
much of the computing done by scientists at a distance. Its founders
are starting off with about $600,000 of USWRP seed money while working
to get longer-term support. The ultimate goal is about a dozen full-time
staff and a strong visitor program, with an annual budget on the order
of $5 million.
Testing a step earlier
Until now, the models used in daily forecast operations have been tested
mainly at NOAA’s National Centers for Environmental Prediction
(NCEP). Even after a model in development is frozen—its code mostly
finalized—it gets another full year of rigorous test-driving at
NCEP before National Weather Service forecasters can use it routinely.
The sheer size and scope of WRFcall for extraordinary testing. With
a horizontal resolution of 1 to 10 km (0.6–6.2 miles), the model
will generate mesoscale forecasts so detailed thatthey will resemble
radar images. WRFis also designed to mesh easily with models of air
chemistry and other specialized areas.
The new DTC will allow a wide range of new methods and model components
to be thoroughly checked out before they are moved closer to operational
use. The center will also maintain the code for WRF’s various
permutations and keep an archive of each day’s forecasts, totaling
200 to 300 terabytes of data per year. Another task is to explore the
best means of verifying the model’s performance, especially in
predicting individual thunderstorms and other features that are omitted
or more crudely predicted in current models.
A dual approach
In addition to the Boulder center, a parallel Operational Testbed Center
(OTC) is in the cards, this one to be affiliated with NCEP, which is
based in Camp Springs, Maryland. In the research and operations worlds,
“we’re dealing with two very different communities,”
says Steven Koch (NOAA Forecast Systems Laboratory), who worked closely
with Gall in planning the DTC.
The idea is that, just as a new medication is tested first for safety
and then for efficacy, WRF variants will go through two evaluation hoops.
The first is in the Boulder-based DTC, where a promising new approach
will be checked for its robustness in a variety of seasons, climates,
and locations, with the code revised as needed along the way. It would
then migrate to the OTC, where the code is frozen and checked for any
bugs or biases that might show up only in real-time, 24/7 use.
The operational tests will help WRF serve a variety of users. In April,
the U.S. Air Force, Navy, and Federal Aviation Administration formalized
an agreement with NOAA and NCAR to support WRF development. “The
military has very different requirements and different missions [from
civilian forecasting],” says Nelson Seaman (NCEP). “It’s
not interested in whether it rains and spoils somebody’s picnic.
It’s more concerned about things like the transportability of
trucks through mud or the refractivity of radar signals.”
Seaman, a former military forecaster and a long-time professor of meteorology
at Pennsylvania State University, came on board in late 2002 as the
national program manager for WRF. He believes an operational testbed
will be vital for taking code from the developmental testbed and tailoring
it for various users. In the past, he says, “A research lab would
develop code and it’d be up to two years before you could get
To help keep the testbeds working in sync, Seaman hopes to see a few
NCEP staff stationed long term at the DTC. “You’d want them
to be thoroughly familiar with the operational setup at NCEP but also
able to talk with researchers. It’s like building a bridge with
one crew on one side of the river and one on the other. Where they meet
is those DTC and OTC people sitting next to each other.”
The big freeze
WRF developers have been in high gear lately, finalizing the ensemble
that will be frozen and tested for WRF’s operational debut in
October 2004. This six-member ensemble will include three variants of
a core WRF code developed at NCAR. Three other variants will based on
core code from NCEP that is somewhat different in numerical structure
from the NCAR code.
The uncertainty within these ensembles will be handled in a novel way.
Most ensemble modeling on the global scale adds tiny perturbations to
the initial conditions for each member. These represent “initial-state
uncertainty”—the limits to our weather observing system—and
they allow the ensemble to portray a broader range of forecast possibilities.
But scientists are still exploring new methods for handling the physics
in high-resolution weather models, where growing computer power allows
finer and finer model grids. To capture that uncertainty, a few members
of the WRF ensemble might include variations in their model physics
instead of in their initial conditions.
All this testing and evaluation will give forecasters a much broader
range of WRF options a year from now, according to Seaman. “It’s
very exciting work.”
by Bob Henson
WRF nails Isabel
As Hurricane Isabel churned toward North Carolina in mid-September,
configuration of WRF, oriented to reveal the tropical cyclone
in detail, produced real-time forecasts. Up to twice a day, the
model generated 120-hour forecasts at10-kilometer resolution,
as well as 48-hour outlooks using a 4-km grid (as shown in the
reflectivity image at right). WRF brought Isabel’s inner
features into sharp focus. “Even at 10 kilometers, the model
produced realistic structures, including spiral bands and an eye,”
says NCAR’s Jordan Powers. When Isabel was nearing the coast,
the WRF outlooks produced central pressures near 950 millibars
(about 28.00”), and the model realistically depicted the
hurricane’s large eye region. After WRF becomes an official
operational model in 2004, its planners hope to see it coupled
with ocean and/or wave models for even better forecasts of hurricanes.
Real-time access is available via the NCAR/WRF