WRF Software Testimonials

I am a new user of WRF. I had little difficulty porting to our IA32 and IA64 Linux clusters and running ideal and real (including benchmark) test cases. I am quite impressed by the volume and quality of the documentation and by the quick response of the WRF help desk.

Art Mirin

Leader, Scientific Computing Group

Center for Applied Scientific Computing

Lawrence Livermore National Laboratory

The WRF software has been a valuable component of the DoD High Performance Computing Modernization Program's Applications Benchmark Test Suite for the past two years. Portability of the software has been excellent, and the "configure" tool makes WRF easy to install on virtually every modern HPC platform. These characteristics of WRF, as well as its excellent scalability in parallel environments, enable a very useful comparison of HPC platforms of interest to the DoD.

Tom Oppe, Applications Programmer/Analyst

Engineer Research and Development Center

Major Shared Resource Center (ERDC MSRC)

My first exposure to WRF was in a project where I was tasked with coupling it with an ocean model running on a different grid, and providing it to several different users, each with different needs (different physics options, different dynamical cores, and different boundary conditions and time/length scales). We were running on several different platforms, including the new (at the time) Itanium-64 clusters, and coupling across the TeraGrid. No small feat!

We found that WRF's modular software structure was both easy to understand and well-documented. The user community was helpful when we did need a hand. Questions about the physics and what variables needed to be passed to the ocean model using different dynamical cores were answered by researchers at NCAR whose expertise is unparalleled, and questions about the software design and how best to pass data in parallel were also well addressed by the WRF software engineers.

In all, we've had a good experience with WRF, and continue to enjoy not only it's support as a leading research tool by the atmospheric science community, but its well-designed software structure, too.

Christopher Moore

Research Scientist, Oceanography

University of Washington/JISAO/NOAA-PMEL

My first experience on WRF software infrastructure was in the GRAPES project, which aimed to develop a new numerical weather model for Chinese Meteorological Administration. It appeared to be state-of-the-art as it included advanced mechanisms for both distributed and shared memory parallel processing.

The WRF software infrastructure was used as the driver layer of the GRAPES model. The user interface is well defined and it is very easy to understand. Variables could be introduced easily with the registry mechanism, and almost all the functions we need was available. It greatly eased the effort to write the parallel code.

The software is quit portable, I installed the model on IBM sp, and several Intel based clusters without any problem and the performance was quit encouraging.

Zhiyan Jin

Chinese Academy of Meteorological Sciences

The entire WRF development team should be congratulated on bringing a flexible, portable, and high performance NWP system to the meteorological community in such little time. I am particularly impressed by the portability of the software to so many different systems and the high computational performance on diverse computer architectures. I am optimistic that we will continue to see improvements in forecast skill and basic understanding of the atmosphere since the efforts of many people can be focused on a common software platform.

Jim Tuccillo

Meteorologist and Sr. Applications Analyst

Linux Networx

In general, I have been very pleased with the ease with which we have been able to download the WRF software and run it on our computers at GFDL, as well as the ease with which we have been able to introduce new physics packages into the model. Scientifically, we have been impressed with its skill as a cloud-system-resolving model, in which we run WRF at the resolution of individual cumulus clouds and their associated systems.

Leo J. Donner

Geophysical Fluid Dynamics Laboratory/NOAA

The WRF software infrastructure appears to be state-of-the-art insofar as it includes advanced mechanisms for automatically generating source-code during compilation. This radically facilitates modification of the code for addition of new physics packages by the user.

Moreover, such mechanisms allow new arrays and variables to be introduced easily, with their declaration and interaction with history/restart/input files being coded automatically. The software appears to perform well on a variety of platforms.

The documentation was very helpful, concise and lucid, and the support from "WRFHelp" was excellent.

Vaughan Phillips

Geophysical Fluid Dynamics Laboratory/NOAA

Our group is using WRF-chem for research applications that require several hundred prognostic scalars. It has been relatively straightforward to port chemistry and aerosol modules into the WRF framework and to add code to permit aerosol-radiation-cloud feedback processes. NCAR has been very supportive in this effort and has made several modifications to the registry that reduced the amount of code needed for hundreds of scalars and made simulating atmospheric chemistry more efficient.

Jerome Fast

Pacific Northwest National Laboratory

My research involves the study of atmospheres of other planets and moons. Instead of writing our own atmospheric model for research purposes, we decided to adopt and modify WRF to be both a global and a more general planetary atmospheric model. Making WRF into a global model involved modifying parts of the dynamic core, and the cleanly-written and very modular WRF made dealing with modification of the dynamical core easy. The code is clear, easy to understand, and flexible for a variety of purposes, including ours, which the model designers probably never anticipated! The flexible, modular format of the model code made the additional modifications necessary to produce a planetary atmospheric model easy as well. The notes intended for terrestrial scientists to implement their own physics parameterization schemes were clear and useful for us as well in designing schemes appropriate to other atmospheres.

Attending the WRF support meetings has also been an enormous help, and everyone involved in the project has been helpful and quick to respond to any questions we have had. All in all, it is a pleasure to work with the WRF model and the people associated with it.

Anthony Toigo

Research Associate

Cornell University

Ithaca, NY

From a Cray X1/X1E performance perspective, the flexibility inherent in the WRF software architecture makes it very easy to use the Cray X1/X1E vector supercomputer efficiently. The array storage order and loop structuring allows effective use of both shared memory X1/X1E streaming features over bands of latitudes and fast vector hardware along the length of each latitude. Using simple runtime options, WRF allows these parameters to be easily tuned for each forecast domain.

Peter Johnsen

Meteorologist, Applications Engineering

Cray, Inc.

WRF is readily ported to a variety of architectures due to it's supported makefile and configuration menus. The input format and problem setup are straightforward, allowing experiments of a wide variety of domain sizes and physics options to be easily conducted. The parallel scaling is excellent on both SMP and clusters, even for modest sized problems.

Mark Straka

National Center for Supercomputing Applications

I have used the WRF I/O infrastructure to build a GRIB (version 1) input/output module. The defined and documented I/O API that is part of WRF made this task possible. Because the interface was so well documented, I was able to write the entire module without any input from the WRF developers. My opinion is that it is extremely important to have well-defined and documented API's within a software architecture such as WRF so that it can be easily extended. So, thank you!

Todd Hutchinson, WSI

The success of this model is in the seamless introduction of object-oriented architecture in scientific computing. The registry mechanism of the model provides a sophisticated dynamical memory management facility for different types of variables and also making it flexible to add new schemes. The layer and tile framework used in the WRF parallelisation has eased the efforts to port the model on our hybrid architecture based supercomputer PARAM PADMA.

Browser based code documentation is a boon to the modelers to study the code.

Amit Kesarkar

Computational Atmospheric Science Group

Centre for Development of Advanced Computing (CDAC)

Pune, India

The WRF IO framework makes it easy to add data formats to the software without messing IO calls inside the model. It also provides a way for other models to adopt IO modules easily.

Muqun “Kent” Yang

National Center for Supercomputing Applications

The WRF software architecture from the initialization to the actual model has been truly well designed for easy use and easy modification. We are using WRF/Chem for air chemistry studies and air quality forecast in Latin America also incorporating our own chemistry modules. Due to its modular structure these modifications have been very easy to implement. In case of any problem the support has been quick and accurate. Good stuff....

Rainer Schmitz

Forschungszentrum Karlsruhe, IMK-IFU

University of Chile

The WRF build infrastructure simplified our development of tools for parallel post processing of WRF output files in two ways: by simplifying the configure and build process of WRF, and by abstracting I/O functionality into a single place where it can be studied free of any simulation code. WRF is an example of how software organization can vastly reduce the complexity of programming and maintaining not just the model itself, but up/down-stream applications as well.

Jace Mogill

ROTANG

I support several weather forecasting and climate simulation codes on SGI systems so that our end-users can be productive with the least amount of trouble. I find WRF to be one of the easiest packages to build and use of the many I deal with in my work.

Dr. Gerardo Cisneros

Scientist

SGI

What I find most impressive about WRF is its portability. My colleagues at FSL and I have run WRF on various machines and it always installs with minimal problems.

Jacques Middlecoff

Forecast System Laboratory/NOAA

If you would like to provide a testimonial on the WRF Software for inclusion here, please contact John Michalakes at michalak@ucar.edu .