|
Experimental ARPS Products for Florida
Click on the links below to display the forecast products.
Pages are best viewed with Internet Explorer 6.0 or Netscape
6.2 or later.
When viewing loops, click Refresh or Reload to ensure latest data.
*SLF = Shuttle Landing Facility, Runway 15
Note: These are experimental products and the timely
receipt of products is not guaranteed. Not intended for protection
of life or property.
About the SMG run of the Advance Regional Prediction System
SMG is currently running ARPS Version 5.1.5.
The Applied Meteorology
Unit (AMU) performed a study to recommend possible hardware configurations
to support mesoscale modeling using the ARPS
model. In order to best utilize the ARPS forecast model for SMG�s
operational needs, it was determined that SMG needed to take advantage
of the Message Passage Interface
(MPI) features of ARPS (the �arps_mpi� program). SMG purchased
a Beowulf LINUX PC Cluster to support the modeling effort (see Table 1
for the cluster�s hardware specifications). In addition to the network
interface described in Table 1, SMG also purchased a Dolphin Wulfkit Interface
and the SCALI software needed to utilize the faster communications of the
Wulfkit interface. The cluster is currently running the Red Hat Linux
7.3 operating system.
Table 1. Beowulf Linux PC Cluster Specifications
| |
Master Node |
Slave Nodes |
| Quantity |
1 |
17 |
| CPU |
Dual Intel Pentium III 1.26 GHz |
Dual Intel Pentium III 1.26 GHz |
| RAM |
1 GB |
512 MB |
| Hard Drive |
120 GB |
120 GB |
| Network |
1 GB Ethernet |
100 Base T |
The original goal for the implementation of the ARPS forecast model
was to provide a 12-hour forecast every 3-hours at 3-km resolution over
most of Florida. However, testing of various ARPS configurations
on the SMG cluster indicated that this goal was not achievable. SMG
is currently configured to produce a 9-hour forecast every 3-hours at 4-km
resolution over the same domain as the ADAS
analysis. This forecast run takes nearly 3 hours to complete.
In order to obtain a forecast in a timely enough fashion, it was determined
that the �arps_mpi� runs have to use 35 processors, with 7 processors in
the x-direction and 5 processors in the y-direction. These parameters,
along with many other forecast parameters, are all configurable in the
ARPS input file that is read in by the �arps_mpi� program. The ability
to rapidly reconfigure the ARPS model run is a significant improvement
of ARPS 5.x over previous versions that forced the selection of grid dimensions
at compilation.
The ARPS model runs currently use the "warm-start" ADAS initialization.
For example, the 03 UTC ARPS model run uses the 0300 UTC ADAS as the initialization,
the 09 UTC ARPS model run uses the 0900 UTC ADAS, etc.
Long Range Goals
-
Increase the domain, resolution, and frequency of the ARPS model. (Need more cluster!!)
-
Configure the ARPS model to run over one or both of SMG's other CONUS landing
sites: Edwards AFB, California and/or Northrup Strip White Sands
Space Harbor, New Mexico.
-
Produce comparisons of SMG ARPS model runs with MLB's
ARPS model runs.
-
Produce "real time" verification of the ARPS model with surface observations
and possibly radar and/or satellite data.
Acknowledgments
SMG appreciates the invaluable assistance of Jon Case (NASA KSC/AMU),
Dan Weber (CAPS), Ming Xue (CAPS), Keith Brewster (CAPS), Kelvin Droegemeir
(CAPS), Mike Eilts (Warning Decision Technologies), John Carpenter (WDT),
and Gene Bassett (WDT). The implementation of the prototype ADAS
would have been impossible without Jon Case�s hard work. CAPS and
WDT provided SMG advice and assistance in understanding the ARPS model
and developing a usable operational configuration.
Mention of any private organization or
person does not constitute an official government endorsement of
a product or service.
Contacts
For questions, comments, or feedback regarding SMG's ARPS model efforts,
send email to SMG Webmaster.
|
www.nws.noaa.gov 
