SR SSD 2000-07
Changes to the NCEP Nested Grid Model (NGM)
1. Rationale for Changes
At 1200 UTC on March 15, 2000 a series of changes to the initialization procedure for the NGM were made. Since the fire which disabled the NCEP Cray C-90 computer last September 27, it was necessary to run the NGM on a Cray J-90 which is 3-4 times slower than the C-90. This resulted in NGM products being available about one hour later than before the fire. This delay is significant, because NWS field forecasters and outside customers are still heavily dependent on timely NGM products, including NGM-based MOS guidance. In addition, there are no funds to maintain the Cray J-90 in operations beyond March 2000, so it was necessary for NCEP's Environmental Modeling Center (EMC) to convert the NGM to run on the new IBM-SP computer.
Conversion of the entire suite of codes which ran on the C-90 (including the Regional Optimum Interpolation (ROI) analysis) using IBM's Message Passing Interface (MPI) was estimated to take two person-years of work. Such an effort was neither feasible nor cost-effective. To satisfy both the users' requirements for a timely NGM forecast and NCEP's requirements for a maintainable NGM system, it was decided to make the following changes:
- NGM initial conditions will be obtained from the Meso Eta analysis over North America, and from a 6-h AVN forecast over the rest of the Northern Hemisphere, instead of from the hemispheric ROI analysis.
- Since the Eta analysis does not extend beyond the North Pole into Europe and Asia, the high resolution NGM C-grid was modified to be similar in size to the Eta computational grid.
2. Advantages and Disadvantages
Use of the Eta analysis/AVN forecast to initialize the NGM will have advantages and disadvantages which will cause improvement or degradation over the previous Cray NGM system. A discussion of the expected advantages and disadvantages follows.
Current and Future Advantages...
1) Since the NGM is initialized from the Eta analysis, it can start one hour earlier on the IBM than it did on Cray J-90. Although the NGM runs slower on the IBM than it did on the Cray C-90, the earlier start time on the IBM will compensate for this, and NGM products from the IBM will be available at about the same time as before the C-90 fire (about 50 min sooner than during the interim operations on the J-90).
2) The Eta analysis uses more types of data than the Regional Optimum Interpolation analysis; for example, aircraft temperatures, WSR-88D VAD winds, SSM/I oceanic wind speeds, surface wind observations over land, and precipitable water from the GOES and SSM/I sounder.
3) The Eta analysis will continue to be improved by algorithm changes and by new data sources such as satellite radiances, assimilation of observed precipitation and cloud, and WSR-88D radial winds.
4) Continued improvements are expected in the AVN forecast suite.
5) The IBM version of the NGM could be run well into the future to provide early MOS guidance and to provide EMC with a baseline for the measurement of Meso Eta skill.
1) The smaller inner-most grid stops at the North Pole.
2) Occasional problems with the Eta analysis in the Pacific may adversely impact the NGM forecast on the IBM. This could be the result of fewer observations being available to the Eta analysis due to the early (T+70 min) data cutoff time.
3) Use of a 6-h AVN forecast instead of an ROI analysis to initialize any part of the NGM grids outside of the Eta computational grid. At the early data cutoff time of the Eta there will be few if any observations available for an analysis in the Eastern Hemisphere.
3. Results of Testing Prior to the Change
Prior to implementation of the changes, EMC performed two parallel tests of the IBM version of the NGM: a real-time test which started at 1200 UTC on December 16, 1999, and a warm season retrospective test for July 1999. EMC's quantitative skill scores (fit to rawinsonde data and 24-h quantitative precipitation forecasts scores) for these tests can be found at http://sgi62.wwb.noaa.gov:8080/NGMSTATS/ . The Techniques Development Laboratory (TDL) ran the NGM MOS package from from the parallel NGM forecasts, and these results can be found at http://www.nws.noaa.gov/tdl/synop/ngmcafti.htm.
During the July 1999 test, the EMC scores and the TDL MOS verifications showed small differences in the quality of the parallel NGM when compared to the operational NGM. The QPF skill from the parallel NGM was slightly better in the eastern U.S. and slightly worse in the western U.S., while the fit to rawinsonde data was generally better for all variables.
For the current cool season test, there were larger differences in quality between the operational and parallel NGM, with degradation more common at upper levels than in the warm season test. Greatest differences were seen in winds (worse above 700 mb), temperature and heights (comparable or better below 400 mb, but worse above due to a growing warm bias), and surface temperature (better at at 12-h and 24-h range but slightly worse at 48-h). The QPF skill was comparable or slightly better in the eastern U.S., but worse in the western U.S.
TDL produced average verification statistics from 700+ sites in the CONUS and Alaska for both the operational and parallel NGM. For the two NGM runs the errors were comparable through the 12 to 36-h period, with slight degradation (4% for the cool season temperature) for the parallel NGM's 60-h MOS temperature forecasts.
Further details are discussed in NWS Technical Procedures Bulletin No. 464, "Changes to the NCEP Regional Analysis and Forecast System (RAFS): Initial Conditions for for the Nested Grid Model," which can be found at http://sgi62.wwb.noaa.gov:8080/NGMTPB/.
A summary of the March 1, 2000, presentation on NGM changes made to the NWS Committee on Analysis and Forecast Techniques Implementation (CAFTI) is also available online at ftp://ftp.ncep.noaa.gov/pub/emc/wd20er/ngmcafti6/index.htm.
4. Note to Forecasters
Users should not expect to see an exact match between the Eta analysis and the NGM 00-h fields derived from the Eta analysis in all situations for the following reasons:
1) The NGM implicit normal mode initialization is performed on the interpolated Eta analysis before the NGM model integration.
2) Vertical resolution difference: The operational Meso Eta model has 45 vertical levels, while the NGM has 16 levels. Certain features (such as narrow jet streaks or sharp vertical moisture gradients)which are resolved in the Eta analysis may be not be captured with the same detail when interpolated to the thicker NGM vertical layers.
3) Model terrain height differences: The 32-km Eta model terrain height is derived from high resolution 30 sec data over the CONUS, while the NGM uses much smoother terrain based on R40 spectral resolution. Thus, there are many regions with significant differences in model terrain height. During the Eta-to-NGM interpolation, the first step is to recompute the Eta analysis surface pressure based on the NGM terrain height. If the NGM terrain height is less than the Eta terrain height, one must extrapolate the wind, temperature, and moisture from the first atmospheric Eta layer to get values at NGM levels below the Eta model terrain.
4) Forecasters who use the Meso Eta and NGM forecast hourly station profiles packed in BUFR format should be aware the station data in the Eta model is taken from the model grid point closest to the station location, while the NGM performs a bilinear interpolation from the four grid points closest to the station. Therefore, the 00-h Eta and NGM profiles may not always agree even for stations close to sea level, especially in regions with strong gradients.