IBM demonstrates regional weather
forecasting system at the Supercomputing 1998 conference
The capabilities developed and utilized for the Olympics, AMS97, CMA, SC97
and the 1998 IBM Stockholder's meeting have since been employed in other
operational forecasting settings. For example, at the Supercomputing
'98 (SC98: November 7 - 13, 1998 in Orlando, FL), this capability
was replicated in the IBM booth as part of the conference's technical exhibition.
The system was adapted to the Orlando area as shown below. For this
and any of the subsequent images, you can view a higher-resolution version
by simply clicking on it. You can also interact with this map via
a scene in PanoramIX or simplified VRML.
New 24-hour mesoscale forecasts were produced at 8 km resolution in
a region roughly 800x800 km in extent during the conference. The
computation took place on eight 160 MHz P2SC thin nodes of a much larger
SP that was in the booth. One 160 MHz P2SC thin node was used for
I/O. Two workstations (an IBM RS/6000 43P-260 and an IBM Intellistation
M-Pro) and one laptop (IBM RS/6000 860) were available in the booth to
interact with the model and analyze results. Due to a number of logistical
difficulties, raw observations and access to data assimilation via LAPS
for the pre-processor step were unavailable at SC98. Therefore, RAMS
was initialized with the results from the ETA synoptic scale model from
NCEP, which are computed at
32 km resolution, but sampled at 80 km for public availability. These
same data were also used for boundary conditions for the model. A
24-hour forecast over this domain required only 80 minutes
to complete on the eight SP nodes. This improved performance was
due to the incorporation of a more efficient microphysics package developed
by FSL and additional optimizations, including those from the latest version
of the FORTRAN and C compilers for AIX. Compared to the system used
to support forecasting at the Olympics, this was less than half the time
of the computing time for a forecast one third longer, over a domain 40%
larger with only one-fourth the number of thin nodes (i.e., an order of
magnitude faster).
Output from RAMS every 10 minutes of forecast time were provided for
browsing visualization. Animations were produced routinely as the primary
mechanism to evaluate the model. However, this Data Explorer-based
application was significantly enhanced over the earlier implementations
to include integrated flyover (key-frame animation), and creation of new
products for web distribution besides animations and image "snapshots".
Image-based rendering of three-dimensional scenes via PanoramIX
and geometric descriptions of three-dimensional scenes via VRML
incorporating simplified
geometry were available. The system also permitted simple tracking
of the simulation, so that the interactive tools could be utilized while
the model was running. If there were problems in the model run, then the
execution could be terminated and the model restarted with new input observations.
In addition, interactive visualization applications for the analysis
of post-processed model results were available operationally, which
are illustrated below.
The following image is from one of the animations
produced during the conference. The 145-frame animation
shows the remnants of Hurricane Mitch after it passed over the Florida
peninsula. (The animation can also be viewed
at higher resolution, but the file is three times bigger.) Although
this event occurred the week before the conference, data to initialize
the model was accessed during the meeting to provide additional examples
to demonstrate. The image shows a terrain map, pseudo-colored by
predicted total precipitation overlaid with coastline, state, county and
river maps for 1 PM on November 5. Notice that the amount of rainfall
is quite significant (several inches). Predicted winds are illustrated
by arrows, colored by speed. A circular pattern is visible as a signature
of the hurricane. Even though the storm is dissipating, the wind
speed is still fairly high (> 50 mph). The clouds are visualized
as a white, translucent isosurface of cloud water density. Inside
the cloud surface to the northeast are cyan surfaces of predicted reflectivity.
These correspond to rain bands generated by the hurricane. Some major
cities in the area are also marked. This particular time step can
also be examined via a flyover animation,
simplified VRML geometry and a PanoramIX
scene. (The flyover animation can
also be viewed at higher resolution, but the file is about two times bigger.)
Other results from the same model run are shown in the following image
to illustrate more of the structure of the hurricane. An 145-frame
animation shows these details. (The animation
can also be viewed at higher resolution, but the file is three times bigger.)
The terrain surface is still pseudo-colored by total rainfall, but the
winds are now shown as streamlines with directional arrows. Cloud
water density is no longer presented, but the reflectivity is at a higher
threshold to better illustrate the rain bands. In addition, local
surface pressure highs and lows are marked on the map with an "H" and "L",
respectively. This particular time step can also be examined via
simplified VRML geometry and a PanoramIX
scene.
The next examples are for model runs during the
conference period. The first image is for 1 PM on Tuesday, November
10. The local terrain (or lack thereof) is visible, overlaid
with state, county, river and coastline maps. The topography is colored
by predicted temperature. The surface is also overlaid with arrows
which point in the direction of winds and are colored by speed (white is
30-35 mph). Predicted clouds are illustrated by white, translucent
surfaces. The motion of clouds during the day and the cooling of
the land at night can be seen in an 145-frame animation.
(The animation can also be viewed at higher
resolution, but the file is three times bigger.) This particular
time step can also be examined via a flyover
animation, simplified VRML geometry
and a PanoramIX scene.
The next two images are derived from a run the following day, which
accurately predicted a front moving through the panhandle area of Florida,
and the resulting precipitation. The first sample image shows the
area being forecasted and some of its major cities at 1 PM on Wednesday,
November 11. The topography is colored by predicted precipitation,
where blue regions illustrate up to two inches of total rainfall.
The surface is also overlaid with arrows which point in the direction of
winds and are colored by speed (white is 30-35 mph). Predicted clouds
are illustrated by white, translucent surfaces. A cyan colored surface
is a forecast of a rain band corresponding to the front moving across the
region along with the pattern of winds. The model accurately predicted
the location and time of this precipitation when compared to the actual
observations. This can be seen in a 145-frame animation.
(The animation can also be viewed at higher
resolution, but the file is about 2.5 times bigger.)
This particular time step can also be examined via a flyover
animation, simplified VRML geometry and
a PanoramIX scene. (The flyover
animation can also be viewed at higher resolution, but the file is
about two times bigger.)
The next image is of the same time, but shows surface temperature as
color-filled contours instead of total precipitation. The surface
wind data are now illustrated as streamlines with directional arrows.
The front can be seen in a 145-frame animation.
(The animation can also be viewed at
higher resolution, but the file is about 2.5 times bigger.)
This particular time step can also be examined via simplified VRML
geometry and a PanoramIX scene.
The final example is for one of the last model runs completed during
this experiment example image is for 1 pm on Thursday, November 12.
The local terrain (or lack thereof) is visible, overlaid with state, county,
river and coastline maps. The topography is colored by predicted
temperature. The surface is also overlaid with arrows which point
in the direction of winds and are colored by speed (white is 30-35 mph).
Predicted clouds are illustrated by white, translucent surfaces.
The motion of clouds during the day and the cooling of the land at night
can be seen in an 145-frame animation.
(The animation can also be viewed at higher
resolution, but the file is three times bigger.) This particular
time step can also be examined via a flyover
animation, simplified VRML geometry
and a PanoramIX scene.
After each RAMS execution, all of the results
are collected and reorganized into a form that can be used by standard
meteorological analysis tools as provided by NWS (e.g., AWIPS), FSL and
others. This post-processed data were made available for interactive
three-dimensional visualization and analysis via a Data Explorer-based
viewer application. This includes all of computed variables from
the model, but at hourly resolution unlike the browser application that
worked with a subset of variables but at six times the temporal resolution.
Here is a sample image and animation created
with this application for a forecast initiated on Thursday, November 5
at 12Z UTC (7 AM EST). These data are generated from the same model
run that show the demise of Hurricane Mitch, that was
illustrated with results created by the browser application.
A surface variable (maximum reflectivity) has been selected for display
as pseudo-color, which is overlaid on a topographic map. Rivers (blue),
state boundaries (white) and coastlines (black) are draped on the surface.
An upper air variable (ice content) has been selected for display via surface
extraction. The surface at 10-3 kg/m3
is requested in translucent white, which corresponds to the boundary of
cloud ice. Another field (rain content) has been selected to show as a
vertical slice, which is pseudo-color contoured. Any of the three-dimensional
fields available from the model can be visualized with either of these
methods. The upper air wind data can be seen along two vertical profiles,
which are specified interactively, and via streamribbons. The direction
of the model wind field along these "virtual sounding" is shown via vector
arrows. Both the arrows and ribbons are pseudo-colored by horizontal
wind speed. The length of the arrows also corresponds to the horizontal
speed. Points along the profile are used as seeds for the streamribbon
integration. Each profile is realized as a pseudo-colored tube, which
is contoured by the variable selected for isosurface realization (i.e.,
ice content). The tubes are only partially visible in this image
and the animation because the model does
not predict any ice at those locations. The heavy rainfall and the
structure of the hurricane as it dissipates can be seen with these techniques.
Here is a another sample image and animation
created with the RAMS viewer for a forecast initiated on Wednesday, November
11 at 00 UTC (7 PM EST on November 10).
A surface variable (heat index) has been selected for display as pseudo-color,
which is overlaid on a topographic map. Rivers (blue), state boundaries
(white) and coastlines (black) are draped on the surface. An upper
air variable (rain content) has been selected for display via surface extraction.
The surface at 8.5 x 10-1 kg/m3
is requested in translucent tan. The surface is not visible in this
image, but will appear during the animation.
Another field (specific humidity) has been selected to show as a vertical
slice, which is pseudo-color contoured. Any of the three-dimensional
fields available from the model can be visualized with any of these methods.
The upper air wind data can be seen along three vertical profiles, which
are specified interactively, and via streamribbons. The direction
of the model wind field along these "virtual sounding" is shown via vector
arrows. Both the arrows and ribbons are pseudo-colored by horizontal
wind speed. The length of the arrows also corresponds to the horizontal
speed. Points along the profile are used as seeds for the streamribbon
integration. The profile is realized as a pseudo-colored tube, which
is contoured by the variable selected for isosurface realization (i.e.,
rain content).
To evaluate these model results, it is useful to compare
them to actual observations as well as
other model results.
Consider this image showing the prediction of surface temperature made
by the ETA model at 32 km resolution over roughly the same domain as the
RAMS runs discussed herein. It is for 11 AM EST, although earlier
than the corresponding image generated for RAMS.
The animation does cover this period. (The
animation can also be viewed at higher resolution,
but the file is three times bigger.)
lloydt@watson.ibm.com
