SR SSD 2001-03
01/2001

Technical Attachment

Summary of Selected Presentations
from the
Year 2000 Review of the NCEP Production Suite

Welcome and Introduction. Louie Uccellini , NCEP Director, noted that the NCEP Strategic Plan 2000-2005 had been completed and distributed to all NWS field offices. He indicated that NCEP was actively collaborating with external groups on areas such as the Weather Research and Forecast Model (WRF), global climate models, the Workstation Eta and short- and extended-period ensemble forecast systems. He stated that NCEP plans to do a better job using satellite data and is working with NCAR and NCDC to provide improved access to model output archives.

Overview. Steve Lord, director of the Environmental Model Center (EMC), announced that Ralph Peterson had been appointed the Deputy EMC Director. Steve outlined the NCEP Strategic Goals in the NCEP Strategic Plan. Steve also reviewed plans for the establishment of a Joint (NASA/NOAA/OAR) Center for Satellite Data Assimilation and a National Test Bed for Numerical Weather Prediction. He noted that about 58% of EMC funding is "soft money" provided by external agencies and even foreign countries.

IBM Status. John Ward (Chief, Production Management Branch) reported on the status of the conversion to Phase II of the IBM SP computer. Phase I of the IBM (896 processors with 256 Gb memory) was a five-fold increase in computer power over that of the Cray C-90, while Phase II (2048 processors with 1024 Gb memory) is a six-fold increase in power over that of the Phase I system. The Phase II system will be comprised of two equal pieces. One set of nodes will be used for production while the other set of nodes will be used for development.

The first set of nodes has been installed and all the model code has been ported to it. It was declared operational on December 7th. The Phase I equipment is scheduled to be removed at the end of December. The second (development) set of nodes will be installed by February. (By the way, delivery of the successor to the IBM SP Phase II is targeted for August 2002.)

John announced that a Web page is now available (http://www.ncep.noaa.gov/NCO/PMB/nwprod/prodstat/) that shows the current status of all the model runs comprising the NCEP Production Suite. Field offices should notice that the model output is now available up to 20 minutes sooner than it had been with the Phase I computer. Once the new computer is fully operational and all the code has been ported to it, John expects to establish -- with input from the regions -- target completion times for all products which the Network Control Operations will then be expected to meet. He indicated that NCEP will utilize more computer resources (i.e., mode computer nodes) as needed to implement model improvements rather than deliver the model output later as improvements are introduced.

Global Modeling Branch. Hua-Lu Pan, Chief of the Global Modeling Branch, reviewed the improvement made to the global models (AVN and MRF) during the past year. These included the increase in model resolution to T170L42 (~75km 42 layers), the extension of the 0000 and 1200 UTC AVN forecasts to 126 hours, the increase in the resolution of the first 60 hours of the ensemble forecasts from T62 to T126 (~105 km), and the introduction of the tropical cyclone vortex relocation producer.

Plans for the coming year include a smoothed mean sea level pressure product (field office input has been solicited), the production of forecast freezing level as done by the Mesoscale (Eta) model, the introduction of a new precipitation initialization procedure, the transition to full operation of the Pacific Winter Storm Reconnaissance program (seehttp://sgi62.wwb.noaa.gov:8080/ens/target/wsr2001.html), extension of the T126 portion of the ensemble forecasts to 84 hours, implementation of a prognostic cloud water/ice scheme, a change to the convection scheme, improvements to the tropical storm relocation scheme, improvement to the model software to allow it to be run on a variable number of processors.

The Global Modeling Branch will also test a T254L64 version of the global model that is targeted for implementation during the second quarter of 2002. In the long term, NCEP is working with NASA, NCAR, GFDL and MIT to obtain support for a project that will allow the research and operational models to share the same model infrastructure (similar to the WRF mesoscale modeling system).

While there are plans to extend the 0000 and 1200 UTC runs of the AVN model to support forecasts for Days 3 to 5, there are no plans to extend the 0600 and 1800 UTC runs. As Hua-Lu noted, "How many times per day should one update the forecast guidance for Days 3 to 5?"

Mesoscale Modeling Branch. Goeff DiMego, Chief of the Mesoscale Modeling Branch, reviewed the staffing (nine government scientists, seven contractors and eight visiting scientists) and activities of his branch.

Goeff noted that the NGM model code was ported to the IBM SP last March, albeit with a change in the analysis system. The new analysis system uses initial conditions from the Eta analysis over North American and from the 6 hr AVN model forecast elsewhere. While the verification results of the new system were mixed, he felt they were defendable since the model output and associated MOS products are now delivered 50 minutes earlier than when it was run on the Cray C90 computer system. Goeff noted that improvements in the Eta analysis system, and the AVN forecast model, have resulted in incremental improvements in the NGM model output over time. Although previous plans called for the termination of the NGM model runs on April 1, 2001, Goeff announced that the NGM will continue to run until the full suite of off-time (0600 and 1800 UTC) AVN-based MOS products have been developed.

Among the changes made to the Meso Eta model during the past year Geoff mentioned the extended range of the 0000 and 1200 UTC runs to 60 hours, an increase in the number of model output soundings (in BUFR format) and meteorograms (http://www.emc.ncep.noaa.gov/mmb/meteograms/index.html), improved storm motion parameterization, the reinstatement of WSR-88D VAD winds into the analysis system, the provision of 20 km (surface) and 40 km (upper air) model output on AWIPS, the increased domain and model resolution to a grid spacing of 22 km and 50 vertical levels, and the release of the Workstation Eta model code.

Goeff also mentioned the increased amounts and expanding access to model verification statistics (see, for example: http://www.emc.ncep.noaa.gov/mmb/research/meso.verf.html and http://www.emc.ncep.noaa.gov/mmb/research/nearsfc/nearsfc.verf.html)

The next release of the Workstation Eta model code (expected in the next month) will include two dynamics options (hydrostatic and non-hydrostatic) and two vertical coordinate system options (step-mountain Eta and terrain-following sigma). A later release (expected in the spring) will include a version of the Eta three dimensional variational analysis (3DVAR) system. The 32 km tiled model output -- lost after the Cray C90 fire last year -- will soon be restored for use as initial conditions for Workstation Eta model runs.

Geoff noted the Workstation Eta 3DVAR system will use NCEP's high-resolution data files as input. All observations from local sites/mesonets should be sent to NCEP via AWIPS for inclusion in those files. The result will be a higher-resolution analysis than is now possible with the current coarse initial fields used to start the workstation Eta model runs. Prospective modelers should remember that, without supporting upper air information, surface observations that significantly depart from the model's first guess fields will have little impact on the analysis and subsequent forecasts.

Research into potential model improvements based on the January 25, 2000 storm (http://www.emc.ncep.noaa.gov/mmb/research/blizz2000/ and http://sgi62.wwb.noaa.gov:8080/BLIZZFCST/) has indicated that the assimilation of observed precipitation into the model analysis improved the forecasts everywhere, while the computationally expensive use of 4DVAR improved the forecasts on the U.S. East Coast during the event, but degraded the forecasts on the West Coast. NWS field offices are invited to review and comment of the format of this Web site (http://www.emc.ncep.noaa.gov/mmb/research/blizz2000/) as a potential template for the presentation of model performance during subsequent weather events. Does it contain the information field offices need, presented in an appropriate format, to adequately assess model strengths and weaknesses?

Plans for the Meso-Eta for the coming year include an extension of the 0000 and 1200 UTC runs to 84 hours the week of January 15th, and upgrade to the land-surface model and the assimilation of precipitation in the Spring, the implementation of high-resolution "Threats" run on selectable domains starting in June, and an increase in model resolution to ~12km grid spacing with 60 vertical levels late in the year. At that time, the frequency of the Eta data assimilation system will be increased from three-hourly to hourly, the cloud and precipitation schemes will be upgraded, and WSR-88D radial velocities will be assimilated.

Long range plans include testing of the WRF model with the goal of having it replace the Meso Eta model in 2004, and continued increase in the resolution of the mesoscale model from a 12 km grid spacing in 2001 to 20 km in 2003, 8 km in 2005 and 6 km in 2007.

Goeff also relayed information from Stan Benjamin (NOAA's Forecast Systems Laboratory) concerning upcoming plans for the Rapid Update Cycle (RUN) model. These include an increase in resolution to a 20 km grid spacing, replacing the optimum interpolation analysis with a three dimensional variational technique, implementation of new versions of the MM5 microphysics and Grell convective parameterization schemes, the addition of a two-layer snow model, improved land-use and soil type data, and the extension of the forecast duration to at least nine hours for the intermediate hourly runs, continuing the 12 hourly runs every three hours.

Ocean Modeling Branch. D. B. Rao, Chief of the Ocean Modeling Branch, reviewed the four program elements of his branch. They are: Ocean Surface Wave Forecasting, Marine Meteorology, Ocean Analysis and Forecasting, and Sea Ice Analysis and Forecasting.

He noted the Wavewatch III model (http://polar.wwb.noaa.gov/waves/), which became the operational model last March, is the best possible model in the world. The global version of the model is run on a 1 x 1.25 grid from 78S to 78N twice daily out to 126 hours, with nested regional models for Alaskan waters and the Western North Atlantic (0.25 x 0.25) run for the same time frames. All versions use the Global Data Analysis System, sea ice analysis, and the AVN model forecasts for atmospheric forcing, with only limited model output currently available on AWIPS. Plans are to develop an Eastern Pacific regional wave model in the coming year

The Ocean Modeling Branch has been testing hurricane generated wave forecasts from the 2000 hurricane season and hopes to propose the operational implementation of these forecasts to the NWS Committee on Analysis and Forecast Techniques Implementation (CAFTI) in the spring.

The operational Marine Meteorology products produced by his branch include Open Ocean and Coast fog (http://polar.wwb.noaa.gov/coastal.fog/Welcome.html), Vessel Superstructure Ice Accretion Guidance, and perfect prog coastal wind forecasts (http://polar.wwb.noaa.gov/cofs/Welcome.html).

Rao presented results of the use of a neural network system to include SSM/I (Special Sensor, Microwave Imager) satellite data in operational marine wind analyses (http://polar.wwb.noaa.gov/winds/Ssmi/). Future plans for his branch include the inclusion of Quikscat satellite-derived winds through the use of neural networks.

The Coastal Ocean Forecast System (http://polar.wwb.noaa.gov/cofs/Welcome.html), currently run on an experimental basis, is intended to forecast the physical conditions of the coastal oceans in real time to support ship routing, bay and estuary models, environmental hazard models, and monitoring and management of the coastal environment. One of the challenges of this system is to accuracy locate and predict the position of the Gulf Stream.

Climate Modeling Branch. Ming Ji, Chief of the Climate Modeling Branch, discussed changes and plans for the seasonal climate forecast system. Much of the numerical guidance for the Climate Prediction Center's monthly and seasonal outlooks is produced by the Climate Modeling Branch.

Routine ensemble seasonal forecasts from the Atmospheric Global Climate Model (AGCM) were resumed after the Cray C90 fire in April 2000, and the model was upgraded in August. Nine month forecasts of the tropical Pacific sea surface temperature (SST) are made four times each week using weekly ocean initial conditions and various atmospheric conditions. (See, for example: http://www.emc.ncep.noaa.gov/research/cmb/sst_forecast/images/cmb.SSTfcst_nino34.gif)

Plans for the coming year and beyond include upgrading the AGCM from T42L28 to T62L28, implementation of the global Modular Ocean Model (MOM-3) for routine weekly ocean data assimilation, improving seasonal forecasts by including varying land surface initial and boundary conditions, the implementation of a coupled ocean-atmosphere model, and the inclusion of other models in the seasonal forecast ensemble system.

Engineering Assessment of AWIPS Capability to Support ETA-12. Representatives from the AWIPS Program Office (APO) reviewed the existing Satellite Broadcast Network (SBN) bandwidth and field site computing resources with an eye to how they might be impacted by the distribution of model output from a 12km Eta model.

They noted that gridpoint product decoding and storage is done on-site by the AWIPS Data Servers, and these servers already have extremely high CPU and memory utilization, especially during receipt of the current Eta model output. This high utilization affects critical AWIPS functions and can result in delayed or dropped text products. They noted that improving system performance is the top priority of the APO.

In addition to the planned increase in resolution of the Eta model, during the coming year the number of radar products sent over the SBN will be increased from eight to 22 products/radar/volume scan, output from the RUC-2 will be added, as will output from the 0600 and 1800 UTC AVN model runs. These products must be supported with the existing SBN and site computing capacity.

The NOAA Forecast Systems Laboratory (FSL) has written software to decode and display high resolution (~10 km) grids for LAPS, QPF and national radar mosaics. However, additional AWIPS software development is required before these products could be displayed using AWIPS. An increase in the AWIPS data server disk space would also be needed.

The APO is reviewing budget initiatives to increase the bandwidth of the SBN by adding an additional T-1 channel, the use of data compression (e.g., GRIB-2), and the migration of the AWIPS communication processors to faster, less expensive LINUX hardware. The APO is also investigating the feasibility of "on-demand" distribution of radar and model output products.

Even with these proposed upgrades it will not be possible to distribute and process the full model output from a 12km Eta model. The NWS will have to identify which Eta grids should be distributed at full spatial and temporal resolution, and which grids can be distributed at a lower resolution.

Other Presentations. Presentations were also given by each of the NWS regions and the NCEP service centers (the Aviation Weather Center, the Climate-, Hydrometeorological-, Marine-, Storm- and the Tropical Prediction Centers) concerning how they use the model output and their needs and requirements from the Environmental Modeling Center.