Advanced Computing Branch
Michael Kraus, Acting Chief
Objectives
The mission of the Advanced Computing Branch is to enable new advancements in atmospheric and oceanic sciences by making modern high-performance
computers easier to use and by exploiting high-speed networks and Web technologies to utilize distributed data management and distributed computing.
Modern parallel supercomputers, typically composed of commodity off-the-shelf components, offer a less costly alternative to traditional vector supercomputers
for the fast, efficient production of numerical forecasts. However, they are more difficult to use. The branch has developed software that simplifies the porting
of numerical geophysical models from FSL, other NOAA/OAR laboratories, the National Centers for Environmental Prediction (NCEP), and other organizations
to modern parallel computing architectures. The culmination of this development is the Scalable Modeling System (SMS).
Using SMS, parallelism is added to a Fortran program by inserting directives in the form of Fortran comments. SMS then automatically translates this source
code into parallel source code, inserting calls to SMS subroutines that perform interprocess communication and other parallel operations as needed. Since the
directives are comments, a single source code can be maintained for both serial and parallel machines. Also, automatic source code translation allows complexity
to be hidden from users to a greater degree than more traditional subroutine-based approaches.
The SMS subroutines form a software layer between the prediction model’s source code and Message Passing Interface (MPI), the industry standard for
interprocessor communication. This layered approach provides SMS users with ease of use, minimal impact to their source code, portability, and high
performance. Source codes that include SMS directives are fully portable to most high-performance computers, Unix workstations, and symmetric
multiprocessors (SMPs). SMS subroutines provide high-performance scalable I/O supporting both native and portable file formats. Also, data ordering in
files is independent of the number of processors used. Further, since parallel operations are implemented as a layered set of routines, machine-dependent
optimizations have been made inside SMS without impacting the model source code. SMS also supports many user-specified optimizations. For example,
the execution of redundant computations to avoid time-consuming interprocessor communication will reduce run times in some cases. SMS also provides
tools to assist in testing and debugging of parallel programs.
The following atmospheric and oceanic analysis and prediction models have been parallelized using SMS: Quasi-nonhydrostatic (FSL), Rapid Update
Cycle (FSL), Local Analysis and Prediction System (FSL), Regional Ocean Modeling System (Rutgers University/UCLA, Pacific Marine Environment
Laboratory), Global Forecast System (Central Weather Bureau, Taiwan), Typhoon Forecast System (Central Weather Bureau, Taiwan), NALROM
(Aeronomy Laboratory), Princeton Ocean Model (Environmental Technology Laboratory), Hybrid Coordinate Ocean Model (Los Alamos National
Laboratory/University of Miami), Eta (NCEP), and a coupled POM-ice model (NASA Goddard Space Flight Center). Computer architectures supported
by the SMS include the IBM SP2 and IBM Power4 Cluster, Cray T3E, SGI Origin 3000, Sun E10000, Linux clusters (Intel OA32 and IA64 and Compaq
Alpha), and other Unix workstations and SMPs.
Accomplishments
During 2002, the branch continued development and enhancement of the functionality and portability of SMS, and updated documentation with each new
release. The most significant newly developed features are additional runtime debugging tools, support for more flexible decompositions, and support for
most Fortran90 syntax. Training on SMS was provided for scientists from NOAA and other organizations including CIRA and NASA.
SMS was used to parallelize a convection code supporting a grant for air quality, a coupled POM-ice model for NASA’s Goddard Space Flight Center, and
the Large-Eddy Simulation (LES) cloud model for the Pacific Northwest National Laboratory (PNNL).
The branch continued its collaborative efforts to support the development of the Weather Research and Forecast (WRF) model, and became involved in related
efforts such as the Joint Modeling Testbed (JMT). The JMT, a plan to provide a testbed for weather models which would span weather laboratories, is superseded
by the Developmental Test Center (DTC), which is being replaced by the WRF System Test Plan. The design and implementation of the WRF model's I/O API
was completed. The Standard Initialization was modified to provide the capability of performing output in the WRF I/O API. The WRF regression test was
ported to Jet and IJet, and the compile time of WRF on IJet was cut in half.
Support continued to be provided, as needed, for the parallel RUC and Quasi-Nonhydrostatic (QNH) models, for users of the High-Performance Computing
System (HPCS) at FSL and for the HPCS management team, regarding hardware and software upgrades.
The TeraGrid was studied and a proposal was written for NOAA HPCC grant funding to explore the feasibility of running an ocean model and an atmosphere
model on different machines coupled over the TeraGrid. The proposal was funded.
A paper on "The Scalable Modeling System: Directive-based Code Parallelization for Distributed and Shared Memory Computers" was written and accepted
for publication by the Journal of Parallel Computing.
Projections
Plans for the Advanced Computing Branch during 2003 include:
- Use SMS to parallelize other atmospheric and oceanic models as needed. Continue to develop and enhance SMS and to port it to new computer
architectures. Continue to support users of SMS and of FSL’s HPCS. Provide SMS user training as needed.
- Continue to participate in the WRF System Test Plan, which will include the porting of the NCEP Verification and Post Processing software to an
FSL computer. In collaboration with the WRF community, develop a requirements document for a WRF Portal. Publish results in conference proceedings
and journals.
- Optimize the RUC code for the IBM Power4 Cluster.
- Support ITS procurement activities, beginning this spring, for acquisition of FSL’s next HPCS. Benchmark suites will be created for the procurement.
The feasibility of running a coupled model over the TeraGrid for the HPCC will be demonstrated by developing a prototype coupled model and running it
at FSL and PNNL, coupled over the Teragrid. The branch will produce a report documenting what we accomplished and the lessons learned, both about
how to use the TeraGrid as well as possible shortcomings of the TeraGrid itself.
- In collaboration with the International Division, support the Taiwan Central Weather Bureau in their upcoming procurement of an HPCS.
- Help NOAA/ETL parallelize a RAMS model using SMS, and help NASA Goddard parallelize its Ocean/Ice Ecosystem model using SMS.
Return to Top of Aviation Division Section
Forecast Verification Branch
Jennifer Luppens Mahoney, Chief
Objectives
Verification is the key to providing reliable information for improving weather forecasts. As part of FSL's involvement with the Federal Aviation
Administration (FAA) Aviation Weather Research Program (AWRP), the Forecast Verification Branch develops verification techniques, mainly
focusing on aviation weather forecasts and tools that allow forecasters, researchers, developers, and program leaders to generate and display
statistical information in near real time using the Real-Time Verification System (RTVS).
In adhering to related goals in the FSL Strategic Plan, the branch strives to maintain a strong verification program by working closely with other
agencies, such as the National Centers for Environmental Prediction (NCEP), National Weather Service (NWS), and the National Center for
Atmospheric Research (NCAR) Research Applications Program. The technology developed through these close interactions can benefit all agencies
by building and strengthening the verification programs.
The branch is involved in a variety of national programs such as the International H2O Project, the Coastal Storms Initiative (CSI) program, and
projects relating to fire weather. Another task involves development of verification techniques for evaluating precipitation forecasts, a capability
that has been used to support local-scale numerical modeling efforts at FSL. Other important activities include serving as co-lead of the AWRP’s
Quality Assessment Product Development Team and lead of the Collaborative Decision-Making Weather Applications Verification Subcommittee.
Real-Time Verification System (RTVS)
In support of these verification efforts, scientists throughout FSL collaborate with scientists at NCAR and the NWS Aviation Weather Center (AWC) to
develop the RTVS as a tool for assessing the quality of weather forecasts. RTVS is designed to provide a statistical baseline for weather forecasts and
model-based guidance products, support real-time forecast operations, model-based algorithm development, and case study assessments. To this end,
RTVS was designed to ingest weather forecasts and observations in near real time and store the relevant information in a relational database management
system (RDBMS). A flexible easy-to-use Web-based graphical user interface allows users quick and easy access to the data stored in the RDBMS. Users
can compare various forecast lengths and issue times, over a user-defined time period and geographical area, for a variety of forecast models and algorithms.
The RTVS has become an integral part of the AWRP by providing a mechanism for monitoring and tracking the improvements of AWRP-sponsored forecast
products. RTVS will run operationally at the AWC providing feedback directly to forecasters and managers in near real time.
Verification Methods
The branch is an active participant, in collaboration with NCAR, in developing and testing state-of-the-art verification methods, with an emphasis mainly
on aviation and precipitation forecast problems. New techniques have been developed for convection, icing, turbulence, ceiling and visibility, and precipitation.
Many of these techniques are applied to aviation forecasts that have been deemed "unverifiable." Nevertheless, the development and implementation of these
verification methods are leading to a better understanding and improvement in the aviation forecasts.
Accomplishments
This year extensive verification activities supporting the transition of the Integrated Turbulence Forecasting Algorithm (ITFA) were completed. The results
were used in the FAA/NWS decision process to transfer the algorithm from an experimental phase to fully operational weather product that will be supported
by NWS. The ITFA algorithm (known as the Graphical Turbulence Guidance Product) will be available to NWS Aviation Weather Center forecasters and
others to be used as a forecast guidance product for evaluating where turbulence may occur within the atmosphere.
From 30 July – 1 August 2002, a workshop entitled "Making Verification More Meaningful," cosponsored by FSL and NCAR and funded by the
AWRP program, brought together an international group of researchers and operational meteorologists and hydrologists. The workshop focused on the
development of advanced diagnostic verification approaches, operational and user issues, observational concerns, and verification of ensemble forecasts.
The workshop included 9 invited presentations, 20 contributed presentations, and 10 poster presentations. Access to the presentations can be obtained
at http://www.rap.ucar.edu/research/verification/ver_wkshp1.html. Some of
the main conclusions presented include 1) measures of forecast quality are not equivalent to measures of forecast value; 2) there is a large loss of
information — and consequently, economic value — associated with use of nonprobabilistic forecasts; 3) pilots and other aviation personnel are in
need of clearly defined weather information; 4) scaling issues need to be taken into account in verification studies; 5) new object- or field-based verification
approaches show promise for providing more useful information than the traditional verification methods; observational uncertainty limits how well quality
can be measured; and 6) additional educational opportunities regarding statistics and verification should be made available through atmospheric science
curricula, short courses, and Web-based material.
In support of the CSI project, RTVS was modified to include verification of temperature, relative humidity, and wind forecasts. Meteorologists at the
Jacksonville NWS Forecast Office and at Florida State University will be using these results to determine the usefulness of local-scale modeling on the
accuracy of weather forecasts. This ground-breaking work on the impact of local-scale modeling will help shape the NWS modeling activities in the future.
During the IHOP project, statistical results were produced by RTVS for four high-resolution models. The impact of the LAPS Hot-Start technique was clearly
indicated in the standard statistics produced by RTVS. In addition to the standard approaches, an object-oriented approach for diagnosing errors in
precipitation forecasts was implemented into RTVS (Figure 69). The results generated using this diagnostic approach brought new insights into the accuracy
of precipitation forecasts produced by a variety of high-resolution numerical models. For instance, the phase and orientation errors produced in the forecasts
could be clearly identified.
Figure 69. An example of an object-oriented approach to diagnosing errors in
precipitation forecasts (Ebert and McBride 2000).
Verification of Eta 12-hour
QPF for a precipitation accumulation threshold of 0.25 inches. The left image is
an Eta 12-hour QPF issued
at 1800 UTC 12 June 2002 and valid at 0600 UTC
13 June. The right image is the NWS Stage IV precipitation analysis for the 12-
hour
period ending 0600 UTC 13 June. Statistical scores are given below.
Several verification exercises, supporting the work of the AWRP Product Development Teams, were conducted throughout the year. Specifically, a convective
exercise was held from 1 March – 31 October 2002, during which numerous high-resolution convective forecasts were evaluated over the northeastern United
States. Automated forecasts for convection were compared to human generated forecasts so that the strengths and weakness of each could be evaluated.
Throughout the exercise period, feedback was provided on RTVS through graphical displays and statistics to the algorithm developers and AWC forecasters.
In meeting the needs of the AWC forecasters, graphical displays depicting forecasts of turbulence, observations, and statistics were developed and implemented
on RTVS for evaluation. Examples of the turbulence displays are shown in Figure 70; a plan and vertical view of the turbulence forecasts, PIREPs, and lightning
data are shown on the graphic. The graphic is updated in near real time so that the information can be used during the forecast MetWatch activity. Figure 70a (top)
shows the AWC human-generated forecasts (e.g., AIRMET) and Figure 70b (bottom) shows the automated Integrated Turbulence Forecasting Algorithm (ITFA).
Forecasters can use these graphics to distinguish differences that may occur between the two types of forecasts. For instance, the depth of turbulence produced
by the two forecasts is quite different, as indicated by the heights of the solid bars shown on the vertical panels of Figure 70a/b. The graphic is updated in near
real time and made available to AWC forecasters so that the information provided on the graphic can be used to update the forecasts as needed.
Figure 70. a, top) Plan view and vertical depiction of the Aviation Weather
Center turbulence forecast and the
verifying observations. The plan view
shows the active turbulence forecasts, lightning, and voice pilot reports
(PIREPs) at the PIREP locations. The PIREP number corresponds to the
numbers shown in the plan view. The "X" axis
is simply the PIREP index
allowing forecasters the ability to investigate and interrogate the raw
forecasts and
observations. b, bottom) Same as above except for ITFA
(Integrated Turbulence Forecasting Algorithm).
Projections
The Forecast Verification Branch will continue with real-time objective intercomparison exercises for turbulence, icing, convection, and ceiling and visibility
in support of the AWRP. New verification capabilities will be developed for aviation forecasts produced by the NWS Center Weather Service Units, and
support is ongoing for FAA and NWS activities such as providing input into the development of a national verification program and the WRF numerical
modeling efforts. In addition, the branch will explore verification techniques that address questions pertaining to flight operations, and participate in
developing a verification program for the Graphical Area Forecast; an advanced graphically produced weather forecast that would be used directly by
airline pilots. The RTVS will be enhanced to include advanced diagnostic verification approaches that will provide users with the ability to partition errors
into errors that are associated with the phase, orientation, and displacement of the forecasts as compared to the observations. Extensive evaluations of the
Forecast Icing Potential (FIP) algorithm and the Current Icing Potential (CIP) algorithm will be completed and provided to the FAA/NWS Aviation
Weather Technology Transfer Board for its consideration to operational status within the NWS. Finally, staff will continue to develop verification tools,
through RTVS, that provide immediate and useful feedback to forecasters.
Return to Top of Aviation Division Section
