Table of Contents

1.1   WRF EMS: What is it?
1.2   Why Should You Care?
1.3   What if I Need Help?
1.4   Summary of SOO/STRC WRF EMS Features
1.5   How much computer do I need?

 

1.1 WRF EMS: What is it?

The National Weather Service's (NWS) Science and Operations Officer (SOO) Science and Training Resource Center (SOO/STRC) Weather Research and Forecasting (WRF) Environmental Modeling System (EMS) is a complete, full-physics, numerical weather prediction (NWP) package that incorporates dynamical cores from both the National Center for Atmospheric Research (NCAR) Advanced Research WRF (ARW) and the National Center for Environmental Predictions' (NCEP) Non-hydrostatic Mesoscale Model (NMM-WRF) releases into a single end-to-end forecasting system. Nearly all the capabilities of the individual NCEP and NCAR releases are retained within the STRC EMS; however, installation, configuration, and running of the NCEP and NCAR versions has been greatly simplified to encourage its use by NWS forecast offices and the university community. No compilers are necessary as statically-linked x32 and x64 binaries are provided for both distributed and shared memory Linux systems. The SOO/STRC WRF EMS is easy to run on most Linux workstations; it should be possible for those with limited modeling experience to have the system installed and running in less than 1 hour.

 

1.2     Why Should You Care?

The NWS SOO Science and Training Resource Center version of the WRF modeling package was developed to promote the local use of numerical weather prediction models in the Weather Forecast Offices (WFOs) and to achieve the following goals set by the SOO Science and Training Resource Coordinator (SOO STRC);

1.     To improve the knowledge and use of NWP forecast models and issues at the local level;

2.    To advance the forecasting process through an improved understanding of mesoscale atmospheric processes and the use of non-traditional diagnostic tools, and;

3.    To increase participation among the WFOs and other agencies in developing and executing NWP studies to examine local forecast problems.

 

Running the WRF EMS locally at the WFOs will serve to provide;

1.     NWP guidance to WFO and River Forecast Center (RFC) forecasters at temporal and spatial scales not available from operational data sources;

2.    A powerful tool for studying local forecast problems and historically significant weather events;

3.    An alternative to the configuration and physics of operational systems;

4.    A means to develop and test new diagnostic forecast techniques, and;

5.    A method of training forecasters on NWP-related issues.

 

1.3     What if I Need Help?

For NWS forecast offices, the SOO/STRC WRF EMS modeling package is fully supported by the NWS Science and Training Resource Coordinator. I will assist your office in setting up and running the model and getting data to the forecasters. Binaries are provided for those who don't have a compiler. Note that the package is always under development, so please feel free to send me your suggestions for improvement. Support is also available to other government agencies upon request. Non-governmental agencies, such as the academic and commercial sectors, are also welcome to use the WRF EMS; however, support will be limited and up to the discretion of the NWS National SOO Science and Training Resource Coordinator.

 

1.4     Summary of SOO/STRC WRF EMS package features

a. The WRF EMS is a complete, full-physics, NWP package that incorporates dynamical cores from both the NCAR ARW and the NCEP NMM WRF packages into a single end-to-end forecasting system.

b. The WRF EMS is very easy to install and configure. Users should be ready to run simulations within 30 minutes of DVD install.

c. No compilers are necessary. The WRF EMS includes pre-compiled binaries optimized for 32- and 64-bit Linux systems running both shared and distributed memory environments. The MPICH executables are also included for running on local clusters across multiple workstations.

d. The installation script automatically creates a WRF user account, configures the machine for both shared and distributed memory binaries, and installs sample crontab entries for real-time forecasting.

f. Auto-updating capability has been integrated into the WRF EMS. When an update or patch becomes available, it is downloaded and installed automatically by your system.

g. The STRC WRF EMS includes a complete, preconfigured benchmark case for use with the ARW and NMM cores in order to compare your workstation performance with others.

h. The STRC WRF EMS is designed to give users flexibility in configuring and running model simulations, whether it is for local research efforts or making regular real-time forecasts.

i. For real-time forecasting, the system has the capability to minimize the impact of missing initialization data by incorporating multiple "fail-over" options that include alternate servers, data sets, or initial forecast hour.  Should a run fail, the WRF EMS will send e-mail to the user alerting that there was a problem.

j. There are no namelists to edit or modify with the WRF EMS. All model parameters have been reorganized and documented in easy to read configuration files that contain default settings for each dynamical core.

k. The WRF EMS system allows for acquisition of a large variety of data sets for model initialization via NFS, FTP, and HTTP. The package is semi-intelligent, in that it will determine which data sets from different model runs are available for ingestion at a given time. Also, different data sets may be used for initial and boundary conditions.

l. At the user's request, the WRF EMS will calculate an appropriate time step for the model dynamics and various physics schemes.

m. The WRF EMS support 1-way nesting with the WRF NMM core and 2-way nesting with the ARW core.

n. The WRF EMS package provides support for real-time post processing of the WRF forecast files. The user may process and view a forecast while the model is running.

o. The post processor supports a wide variety of display software including AWIPS, BUKIT, NCL, GRADS, GEMPAK, and NAWIPS.

p. The WRF post currently can output fields on 47 pressure levels including 2, 3, 5, 10, 20, 30, 50, 70, 75, 100 to 1000 every 25, and 1013 mb.

q. Processed forecasts may be sent to remote systems via FTP, cp, SCP, or rcp depending upon the user's need.

 

1.5     How much computer do I need?

The answer to this question depends on whether you will be running the model for research or real-time purposes. For real-time users, you need a computer with a premium placed on a fast 2 CPU system and at least 2GB of physical memory. For research purposes, speed is not as critical since you are not up against dead lines to get the model forecast completed.

If you are planning on processing WRF forecast files concurrent with the model running, I strongly recommended that you use a separate, inexpensive Linux workstation with more than 512mb of memory. Attempting to post process forecast files on the same workstation as the model will cause the performance of the system to degrade significantly.

I have executed many benchmark runs with both the NMM and ARW cores of the WRF on various single workstation 2CPU platforms (see below). The data from these benchmark tests show that the NMM (non-hydrostatic) core runs almost as fast as the original WS Eta, and at least twice (2x) as fast as the ARW core.

Machine #1	Machine #2
2x INTEL Xeon (3.2 GhZ)	2x AMD Opteron 248 (2.2 GhZ)
1.5GB physical Memory	2.0GB physical Memory
PGI Compiler version 6.0-8	PGI Compiler version 6.0-8
Red Hat Enterprise 4	Fedora Core 3
DM                -   Distributed Memory environment (mpich 2.7)
SM                -  Shared Memory environment (Open MP)
STRC WRF EMS Benchmark Case
Grid Dimensions:	5005 Horizontal grid points x 45 vertical levels
Simulation Length:	24 hours with 3 hour output frequency
Grid Spacing:	15km
Time step:	90s    (ARW)  36s  (NMM) (yes, ARW is >2x NMM)
Cumulus Scheme:	Kain-Fritsch

	Machine #1		Machine #2
Core	ARW	NMM	ARW	NMM
SM	41m 16s	14m 50s	30m 02s	12m 05s
DM	33m 01s	14m 44s	25m 50s	12m 46s