SR SSD 2000-03
Summary of NCEP Production Suite Review
For the past few years the National Centers for Environmental Prediction (NCEP) has conducted an internal review of its production suite each December. Representatives of the NWS regions and the various NCEP centers are invited to participate in the review. This year's review was conducted on December 9 and 10 at the NOAA Science Center in Camp Springs, MD.
Welcome. Louie Uccellini, NCEP Director, welcomed the participants and commended the NCEP Central Operations staff for their outstanding performance in the wake of the Cray C90 fire on September 27th. In addition to the delivery of the Class VIII IBM SP computer, NCEP also purchased a Silicon Graphics Origin 2000 computer this year to support climate model research. This machine has already proved to be valuable in supporting research and development activities at NCEP.
Louie noted the frustration concerning the inability of getting the current full model resolution output to the NWS field offices. He also observed that neither the current or planned communications bandwidth would be sufficient to support the 50-level 10-km Eta model which NCEP expects to run in the near future. Louie announced that NCEP is creating a Model Utilization Team to ensure the most effect use of the NCEP model output, and plans to conduct a Workstation Eta Users' Workshop to support the use of that model in universities and NWS field offices.
Environmental Modeling Center Overview. Steve Lord, Acting Director of the Environmental Modeling Center, also commended the Central Operations staff and reviewed the agenda for the two-day meeting.
IBM SP Status. John Ward, NCEP Operations Center, announced that the Cray C90 has been officially declared beyond repair and that talks are underway with a museum to accept the machine for display. The IBM SP has been successfully moved from Federal Office Building Four in Camp Springs, MD to the Camp Mabry computing facility in Bowie, MD. Much of the code for the NCEP production suite has been ported to the new machine and it should become operational on January 17th. The NCEP Cray J90 computers are scheduled for retirement in April. At that time, there will no longer be an operational backup to the IBM SP computer. NOAA's Forecast Systems Laboratory (FSL), the U.S. Navy's Fleet Numerical Oceanography Center (FNOC), and the Air Force Weather Agency (AFWA) will be the backups for numerical model guidance. The Department of Commerce, recognizing the economic value of the NCEP model guidance, is considering acquisition of a backup computer for the IBM SP.
The current model IBM SP is called Phase I. In September the IBM will be replaced by an upgraded machine (Phase II). As a result, there will be a two to three month freeze on model changes during the overlap period (September-November) while the new machine is tested and code is ported to it.
Global Modeling. Hua-Lu Pan, Global Modeling Branch Chief, noted that research into model improvements had been rather limited during the past year since the Cray C90 was already running at full capacity. Model changes during the year included incorporation of the NOAA-15 radiance observations in the analysis system in March, which improved the analysis particularly in the Southern Hemisphere. The CAFTI-approved increase in the resolution of the global spectral model to T170L42 will be implemented in January on the IBM SP.
Hua-Lu reported that the skill of the AVN/MRF dropped after the C90 fire since the data assimilation could only be run twice a day on the backup computer. Previously, the data assimilation ran four times per day. He noted that improvements in data assimilation have been most responsible for the model's skill in the short-range forecasts, while model bias becomes a problem after about Day 5. He observed the ensemble forecasts are particularly useful for the extended ranges. The Global Modeling Branch has added a number of model statistics and diagnostic pages to their Web site (http://www.emc.ncep.noaa.gov/research/global2.html , or use the link on the SSD Weather Links page).
The Global Modeling Branch's plans for the coming year include:
* implementation of the T170L42 resolution increase in January;
* extension of the 0000 and 1200 UTC AVN forecasts to 120 hrs to support the Tropical Prediction Center and the issuance of Day 1-5 quantitative precipitation forecasts;
* evaluation of a potential increase in the vertical resolution of the model to 64 levels for possible implementation in early fall;
* attempt to reduce the model's warm bias over the western U.S.;
* reduce the model's wet bias in the spring and summer precipitation forecasts;
* improve the accuracy of the prediction of tropical precipitation in the 5-15 day range;
* test and implement a prognostic cloud condensate scheme;
* improve the spread of the ensemble guidance, with an eye to making the ensemble output and official product (pending CAFTI approval);
* test a hybrid vertical coordinate based on the Weather Research and Forecast (WRF) model design.
* inclusion of more kinds of observations in the data assimilation; and
* continued interactions with NCEP and field forecasters.
Mesoscale Modeling. Geoff DiMego, Mesoscale Modeling Branch Chief, reviewed the activities of his branch. Echoing the frustrations in providing full model resolution output to the field mentioned earlier by Louie Uccellini, Goeff emphasized that his branch is not responsible for data and product distribution!
Accomplishments of the Mesoscale Modeling Branch during the past year included the correction in the mass-wind balance in the Eta 3D variational analysis (3DVAR) system, the elevation of the model output soundings (BUFR files) to production status, the fortuitous(!) development of a procedure to get the best AVN boundary conditions for the backup (80 km) version of the Eta running of the Cray J90, the successful migration of the RUC2 and Eta model code to the IBM SP (just before the fire) and the elevation of the hourly RUC2 and the four time per day (0000, 0600, 1200 and 1800 UTC) Eta-32 to production status in November after the move of the IBM SP to the Camp Mabry facility.
Goeff noted that the skill of the 0000 and 12000 UTC backup 80 km Eta (which ran on the Cray J90) while less than that of the Eta-32, was better than the skill of the NGM. In contrast, the skill of the Air Force Weather Agency MM5 model, which was substituted for the off-time Eta runs was no better than that of the NGM.
Now that the Eta-32 is running on the IBM SP, all four model runs are six hours apart (0000, 0600, 1200 and 1800 UTC), produce guidance out to 48 hours, and use the 12 hr Eta data assimilation scheme (EDAS) for the first guess for the 3DVAR analysis.
The Mesoscale Branch has re-run a Wasach MT wind storm case. A change from Eta to sigma coordinates produced improved guidance, while a non-hydrostatic run of the model did not. The Branch also successfully tested the assimilation of WSR-88D precipitation estimates.
The Mesoscale Modeling Branch's plans for the coming year include:
* extension of the 0000 and 1200 UTC runs of Eta-32 forecasts to 60 hrs in the spring, with an ultimate goal of producing forecasts to 84 hrs in response to request from the Storm Prediction Center and the Hydrometeorological Prediction Center;
* an resolution increase to ~22km/50 levels by August, with a possible increase to ~10km/70 levels by December 2001;
* extension of the model domain to that previously covered by the 48 kn "Early Eta;"
* improvement of the precipitation physics package;
* assimilation of cloud and precipitation observations, satellite radiances and WSR-88D radial velocities;
* implementation of high-resolution (10-15 km) regional Eta models over Alaska, Hawaii, Puerto Rico and the U.S. Virgin Islands;
* development of a movable, nested Eta-8 model for local threats assessment runs; and
* implementation of a 10 member, 40-50 km resolution Eta/Regional Spectral Model short-range ensemble.
Goeff showed some slides produced by Stan Benjamin (NOAA Forecast Systems Laboratory) outlining planned improvements in the RUC-2 model during the coming year. These include an increase in model resolution to 20 km (with a further increase to 15 km in 2001), replacing the optimum interpolation analysis technique with a 3D variational analysis system, and various improvements to the model physics. Goeff noted that increasing the RUC2 output to hourly files out to 12 hrs from every model run could only be accomplished if the code could run in the assigned time slot.
The NGM model was moved to a Cray J90 after the C90 fire. Since the J90's speed is only one-fifth that of the C90, the NGM currently takes about one hour longer to run and has no regional data assimilation system. Goeff noted an test NGM forecast system (using Eta initial conditions over North America) is currently running on the IBM SP. The EMC plans a six to eight week test to compare current MOS to that generated by the IBM SP version of the NGM. Should that test be successful the IBM SP, as early as February, could be capable of producing NGM output as timely as what was produced before the C90 fire. This would also allow one complete year of overlap of NGM MOS with the replacement AVN MOS being developed by TDL.
Aviation Weather Center. Fred Mosher, Aviation Weather Center (AWC) SOO, reviewed the operations of the AWC. He noted that the AWC is a member of the Federal Aviation Administration-funded Model Analysis and Enhancement Product Development Team and has been active in developing aviation-related enhancements to the NCEP operational product suite. Among the enhancements are the implementation of FSL-developed improvements to the RUC2 and the implementation of the Stoelinga-Warner visibility algorithm in the Eta and RUC2. (This algorithm parameterizes visibility as a function of the forecast mass concentration of rain water, snow, cloud water and cloud ice in the lowest level of the model.) The AWC worked closely with the EMC in the improvements to the Eta model output soundings (BUFR files) which included the production of model soundings for all TAF sites, the addition of cloud water, cloud ice, cloud-base pressure and low level visibility to the BUFR files, and the Web-based meteograms generated from the BUFR files. (See http://www.emc.ncep.noaa.gov/mmb/research/meteo2.html.)
The AWC has also been designated as one of six experimental test beds mentioned in the NWS Strategic Plan to accelerate the infusion of new science and technology into the forecast process. The AWC would like to hire a scientist to perform aviation-based model diagnostic studies for feedback to the EMC.
Fred noted that one advantage of MOS-like text messages compared to the model output soundings is that the forecast elements are presented in the same units (e.g., cloud bases and amounts in hundreds of feet and visibility in fractions of miles). In the discussion that followed it was noted that converting the model output to such a format would not require much work. For example, CWSU Fort Worth already produces text tables from the RUC2 output.
Fred's wish-list for the EMC included: the inclusion of drizzle drop size distribution in the models to support aircraft icing forecasts; the accurate inclusion of remotely-sensed cloud and precipitation into the models; improved diagnostics of gravity-wave triggers to support turbulence forecasts, and the need for two- to six-hour forecasts of thunderstorms that include the continuation of existing convection and the initiation of new convection.
Storm Prediction Center. Steve Weiss, SPC forecaster, reviewed the performance of the models during the May 3 tornado outbreak in Oklahoma and Kansas. He noted that neither the model guidance nor the observed data gave a clear indication of the onset, location or evolution of the storms that occurred. Steve noted the need for a better understanding of the relevant physical processes and an improved specification of the atmospheric structure prior to severe weather events. He suggested an experiment in which radiosondes would be launched at six hour intervals throughout the months of April to June in order to develop a research database for use with potential numerical model improvements.
Hydrometeorological Prediction Center. Peter Manousos, Hydrometeorological Prediction Center (HPC) SOO, offered the HPC's perspective to the EMC's annual review. He noted that the HPC has assumed the role of liaison between EMC and the NWS field offices with regard to the interpretation of model performance. The HPC is aware that their narrative discussions are highly valued by the field forecasters. Peter reminded everyone that the HPC has developed a Web site (http://www.hpc.ncep.noaa.gov/mdlbias/modelbias.html) which includes subjective and objective summaries of model biases. The site includes an interactive page to display the average bias (forecast - observed) over the past five or ten days (cycles) for a user-specified selection of model, cycle, forecast hour, and parameter. The bias maps provide an objective method of discerning model bias over the given time periods.
Peter noted the need for a more consistent manner of objectively assessing model performance. As an example, he noted the current process does not capture pattern-dependent biases. The HPC is investigating the use of Allan Murphy's mean square error concepts (see the December 1988 issue of Monthly Weather Review) to identify pattern-specific model biases which may allow forecasters to make skillful changes to the model guidance.
Peter summarized the planned expansion of the HPC product suite. This includes:
* six hourly continuous field QPF (with the 0" isohyet included);
* Day 3 and Day 4 & 5 (combined) QPF maps;
* a winter weather product;
* Day 6 & 7 surface features maps
* possible maps of expected sensible weather (including precipitation type, chance of thunderstorm and cloud cover) for Days 3 through 5;
* issuance of twice per day model diagnostic discussions; and
* a transition towards the production of gridded products for ingest directly by the field offices AWIPS Interactive Forecast Preparation System.
Peter's wish-list for the EMC included: running the Eta model out to 84 hrs; adding more members to the NCEP ensemble forecasts at higher resolution to support the medium-range forecasts; access to any short-range ensemble output to support QPF; and access to ECMWF and UKMET model output beyond 120 hrs to support Day 6 and Day 7 forecasts.
Tropical Prediction Center. Richard Pasch, Tropical Prediction Center (TPC) Hurricane Specialist, reviewed the 1999 tropical storm season. He noted that Irene, Jose and Lenny all formed after the Cray C90 fire. Richard reported that the tracks of this year's storms were slightly more predictable (based on the skill of the CLIPER model) while the intensities were somewhat less predictable (based on the skill of the SHIFOR model). However, the skills of the official TPC track and intensity forecasts were both better than average, indicating real improvement in the intensity forecasts. The official forecasts typically had a leftward track bias.
The TPC forecasters report the spin-up of spurious tropical vortices in the global spectral model were much less of a problem this year as compared to last year. Tests of a coupled ocean-atmosphere version of the GFDL model during the hurricane season indicated improved storm intensity guidance, but no improvement in the track guidance.
The NOAA Gulf IV aircraft flew 23 synoptic flow missions during the season. Data collected from those missions were used to initialize the numerical guidance models and to quantitatively assess synoptic-scale features related to steering currents and intensity changes.
Ocean Modeling Branch. D. B. Rao, Ocean Modeling Branch (OMB) Chief, reviewed the mission of his branch. The OMB has four program elements: marine meteorology, ocean waves, coastal ocean forecasting, and sea ice. Operational products produced by the OMB include: ocean surface wind analyses and forecast fields, "perfect prog" wind forecasts for several coastal and Great Lakes points, open ocean for forecasts, and ship superstructure icing forecasts. Many of the products are available through the OMB's Web site (http://polar.wwb.noaa.gov). He noted the OMB is in the process of rebuilding its ocean wave model Web pages, in preparation for the operational implementation of the new NOAA WAVEWATCH III global and regional wave models.
Global wave forecasts have been produced for the last five year using the WAM model twice a day our to 72 hrs. About a year ago, the OMB began producing wave guidance for the East Coast and the Gulf of Mexico based on the cycle-4 version of the WAM model. The grid resolution of ECG is 1/4 by 1/4 degree. Ocean surface winds (also at a height of 10 meters) used to drive the model are obtained from the Eta model. The model is run twice daily at 0000 and 12000 UTC with guidance generated out to 48 hrs.
Experimental coastal visual range guidance charts are being posted on the OMB Web site (http://polar.wwb.noaa.gov/coastal.fog/Welcome.html). The lowest layer and 950 mb products are provided as post-processed fields direct from the Eta-32 model. These products are valid all year for the coasts of the contiguous United States, Alaska and Gulf of Alaska. Three contours are given: seven n mi (the maximum range that can be seen from the bridge of a ship under normal conditions), three n mi (of interest to mariners and must be reported in offshore and coastal forecasts by National Weather Service Forecast Offices), and 0.5 n mi (this range is the WMO definition of dense fog and can be used as a guide to forecast fog, assuming meteorological conditions warrant such a forecast). The 950 mb visual guidance is presented to give the viewer a means of determining what may be causing the reduction in visual range at the model lowest layer.
The OMB has been very active in the assimilation of remotely sensed ocean winds and waves using satellite retrievals. A new, third generation global wave model, called NOAA Wave Watch 3 (NWW3) has been developed from scratch by the OMB. The global version has 1.25x1.00 degree resolution, uses winds from the AVN model, and can compute seas and swells at 25 frequencies and 24 directional bands out to 72 hrs. The NWW3 shows overall improvement over the WAM, with particular improvements in wave guidance throughout the tropics and away from storm tracks in the mid-latitudes. The NWW3 runs twice a day and will replace the WAM this January once the IBM SP becomes operational. The NWW3 model output is already broadcast on the AWIPS Satellite Broadcast Network (SBN) and display software will be included in AWIPS Build 5. The implementation of two new regional wave models, covering the northwest Atlantic and the Gulf of Alaska/Bering Sea, respectively, has also been approved by CAFTI.
The OMB is continuing development, with a number of partners, of a prototype coastal ocean forecast system for the U.S. coasts. A Coastal Marine Demonstration Project (CMDP) was conducted during June-July 1999, in which experimental forecasts of surface winds, waves, visibility, fog, upper layer currents and salinities, and sea surface temperature were provided through a vendor's Web site. Responses indicated the forecasts met the users' needs at good-to-better ratings 74% of the time. Phase II of the CMDP will take place in February-March 2000, including a test of the Coastal Ocean Forecast System on the new NCEP Origin 2000 computer.
The OMB has developed techniques to include winds from NASA's QuikSCAT satellite into NCEP's data assimilation systems. (The QuikSCAT mission is a "quick recovery" mission to fill the gap created by the loss of data from the NASA Scatterometer (NSCAT), when the satellite on which it was flying lost power in June 1997. Launched on June 19, 1999, the SeaWinds instrument on the QuikSCAT satellite is a specialized microwave radar that measures near-surface wind speed and direction under all weather and cloud conditions over Earth's oceans.)
The Ocean Modeling Branch's plans for the coming year include:
* implementation of the global NWW3 and regional models on the IBM SP.
* develop methods to provide wind fields generated by NCEP's operational GFDL hurricane model to the Northwest Atlantic regional wave model.
* assist in implementation of a coupled ocean-atmosphere version of the GFDL model for hurricane prediction.
* test the impact of QuikSCAT satellite wind retrievals on global data assimilation and forecast models.
* initiate steps to produce high-resolution regional wave forecasts for the U.S. West Coast and the Hawaiian islands.
* make further improvements to model physics.
Marine Prediction Center. Bob Kelly, Marine Prediction Center (MPC) forecaster, noted that most of the MPC's products are delivered directly to their large non-governmental user community either by radiofax or via the MPC's Web site (http://www.mpc.ncep.noaa.gov/). The MPC also provides guidance to coastal WFOs. The MPC operates four desks, providing regional and high seas forecasts for both the Atlantic and Pacific oceans. Special guidance products are provided upon request, such as six hourly sea and swell forecasts the MPC produced in support of the recover of the EgyptAir plane.
The MPC forecasters have found the NWW3 model guidance to be very useful and also make use of the QuikSCAT winds, the Eta-based wave model and the recent improvements to the AVN model. The MPC's experience with the "BlastUp" conference call system to coordinate with field offices has been positive.
Climate Modeling Branch. Ming Ji, Chief of the Climate Modeling Branch (CMB), reviewed the products and planned changes in the seasonal climate forecast system. To support the Climate Prediction Center's seasonal forecasts, seasonal climate anomalies are produced by an atmospheric general circulation model (AGCM). The AGCM is forced by predicted SSTs from the NCEP coupled ENSO forecast system and runs out to seven months in the future. There are 18 AGCM integrations with different initial conditions and the predicted seasonal anomalies are for the 18-member ensemble mean. Predicted anomalies are computed with respect to the 1961-90 AGCM climatology to remove any model bias.
The NCEP Climate AGCM has been developed to be portable, scalable and reproducible for climate applications. Based on the MRF model, the AGCM presently runs at a T42L28 resolution. Plans are to increase the resolution to T62L28 within the next year. The current model does not handle the Pacific-North American (PNA) pattern well, nor does it reproduce observed decadal precipitation variations.
The OMB optimum interpolation (OI) sea surface temperature (SST) analysis is produced weekly on a one-degree grid. The analysis uses in situ and satellite SSTs plus SSTs simulated by sea-ice cover. Before the analysis is computed, the satellite data are adjusted for biases using the method of Reynolds (1988) and Reynolds and Marsico (1993).
The current CMB coupled ocean-atmosphere model coupled (designated as CMP14) covers just the Pacific Basin. The CMB will soon expand to a global ocean domain, include corrections for salinity variability (which can be as large as those due to SST variability) from the NOAA Tropical Atmosphere Ocean (TAO) buoy array and satellite altimetry observations.
Climate Prediction Center. Rusty Martin, Climate Prediction Center (CPC), reviewed the operations of the CPC. Highlights for the CPC during 1999 included the implementation of the Operational Threats Assessment (http://www.cpc.ncep.noaa.gov/products/predictions/threats/) and Drought Monitor (http://enso.unl.edu/monitor/index.html), the successful Winter and Tropical Storm seasonal outlooks, record skill scores during the year for the 6 to 10 Day Temperature Outlooks, and the ongoing skillful La Niña outlooks. The CPC 6 to 10 Day charts (http://www.cpc.ncep.noaa.gov/products/predictions/6-10_day/6_10day1.gif) are the most popular pages on the CPC Web site.
Rusty noted that CPC would appreciate more input from the NWS field offices concerning the Threats Assessment, since the CPC forecasters are not always aware of significant local impacts.
During the coming year the CPC plans to start issuing probabilistic forecasts for Days 3 to 5, an experimental Heat Index Outlook for Days 6 to 10, and temperature and precipitation outlooks for Days 8-14.
MOS 200 Plans. Although the Techniques Development Laboratory (TDL) is not a part of NCEP, by popular request of the participants, Paul Dallavalle, TDL Synoptic Scale Techniques Branch Chief, gave a review of TDL's planned implementation of MOS 2000 during lunch on Friday. Much of the information in Paul's presentation is available on the TDL Web site:
Development of a new model output statistics (MOS) package based on the Aviation (AVN) run of NCEP's Global Spectral Model is underway. Unlike the current AVN MOS package, the new AVN MOS guidance will be analogous to the current NGM MOS guidance in terms of weather elements and forecast intervals with many of the forecasts being valid at 3-h intervals. Forecasts from the AVN MOS will be produced for projections of 6 to 72 hours after 0000 or 1200 UTC and will be generated for over 1000 sites. Plans are to add more forecast sites, weather elements, projections and both 0600 and 1800 UTC cycle guidance in later developments.
The TDL is also developing MOS guidance based on the Medium-Range (MRF) run of NCEP's Global Spectral Model. This expanded guidance will be issued for the same sites at the AVN-based MOS, will include new forecast elements and some revised definitions of some elements. MOS based on the Eta model is also planned, but only for "storm-related" elements such as probabilities of thunderstorms and severe weather.
The TDL expects to implement partial messages from the new AVN- and MRF-based MOS in February and complete AVN-based messages in April, to allow a one year overlap with the old NGM-based MOS. Complete MRF-based messages are planned for October, along with partial guidance messages from the 0600 and 1800 UTC AVN runs. The NGM-based MOS messages will be terminated in April 2001, pending CAFTI approval.
Model Output Flow to AWIPS. Ward Seguin, AWIPS Program Office (APO) Systems Engineering, reported that the APO is developing a process to add products and display capabilities in fewer than six months. This initiative will be presented to the NWS Corporate Board during their February meeting.
Together with contractors from Raytheon, Ward presented the preliminary results of a study to determine the current AWIPS Satellite Broadcast Network (SBN) usage and to predict the impact on network performance by the inclusion of future products. The study is based on program requirements and analyses of product logs on both the transmitting and receiving ends of the SBN. Preliminary results suggest the SBN may reach capacity with the addition of radar imagery and full resolution model output to the current data stream. A lively discussion followed concerning the development and communication of field requirements to the APO.
Community Weather and Climate Modeling Activities. Steve Lord, Geoff DiMego and Hua-Lu Pan presented summaries of new and on-going collaborative activities among NCEP, other Federal agencies and the university community. These include a satellite data assimilation project, the Weather Research and Forecast (WRF) model (http://wrf.fsl.noaa.gov/), and a Community Model Infrastructure Working Group.
Steve Lord outlined a proposal for the establishment of a national test bed for numerical modeling. Such a site would serve as a test bed and community resource to quickly take advantage of research advancements through technology infusion, and also potentially serve as a backup for the NCEP operational computer. Steve will propose this as an initiative for the FY2002 budget.