EMC/HPC/MPC synergy meeting highlights 7/31/00
This meeting was led by Peter Manousos and followed a generic agenda denoted by the subtitles below. Attendees included Dave Michaud, Jim Hoke, Steve Tracton, Jun Du, Carl Staton, Eric Rogers, John Ward, Bill Bua, Maxine Brown, Steve Schotz, Hua-Lu Pan, Steve Lord, Wes Junker, Ed Danaher, Keith Brill, Dave Reynolds and Steve Jascourt.
1. IBM SP
a. Current Status: John Ward reported that outside of small bug fixes, the IBM SP has no system problems. There has been some problem with the communication line to TPC (as it has been at 100% capacity for 60% of the time in the past week). The cause of this is currently being investigated. Disk space is getting low on the production side of the SP. Further, there are some upgrades that will have to be implemented before Phase II including running the GFDL Hurricane Model out to 5 days.
b. Implications of Eta 22km Becoming Operational: Dave Michaud reported that the running of the Eta at the planned horizontal resolution of 22km and 50 vertical levels out to 60 hours will have to be optimized with regard to disk space and node usage. As mentioned above, there is limited disk space on the production side of the SP. There is also a 192 node "fence" (which delineates node usage between the development side and the production side of the SP). Further, there is the requirement for the completion time of the Eta 22km out at forecast hour 48 to be no later than what was observed when the Eta 32km was run on the C-90 (or about 30 minutes delay compared to the Eta 32 km on the SP as currently operational). Therefore, tests utilizing different node allocation were conducted in order to attain the fastest completion times while not exceeding the 192 node "fence". The first test resulted in a delay of 47 minutes (at forecast hour 48) as compared to the operational Eta 32km at the same forecast hour. This was deemed unacceptable due solely to the time delay. The second test resulted in a delay of only 10 minutes, however this significantly exceeded the 192 node "fence". Therefore a third test was run such that a more optimal configuration barely exceeded the fence with the 60 hour production finishing 17 minutes later than what is currently observed. There was some discussion on further optimization (a fourth test) and results from this test should be available this week. It is expected that this fourth optimization will result in a finishing time of the 48 hour forecast delay on the order of 20 minutes (10 minutes earlier than on the C90) and within the 192 node fence.
2. Notes from EMC
a. Mesoscale Modeling Group: Eric Rogers reported that for about the past week the Eta 22km has been running with the satellite radiances incorporated into the 3DVar. Although, the output from these parallel runs have not been available routinely, they are expected to be by next week. The incorporation of the satellite radiances represents the last step of the parallel upgrade. If then, evaluations of the Eta 22km parallel by NCEP prove favorable, the fully upgraded parallel Eta 22km will be implemented as the operational model upon CAFTI approval. The CAFTI meeting for the Eta 22km is tentatively planned for August. However, since this date is driven by the Phase II implementation (previously planned in September and now delayed further into the fall of 2000), it is likely that the CAFTI meeting will be delayed as well.
b. Global Modeling Group: Hua-Lu Pan had three items to report -
1. Data Dump For Final Analysis. It was noted by the Global Modeling Group that the forecasts from the 12Z cycle were slightly worse in skill than those initialized at 00Z (particularly over the Southern Hemisphere). It was determined that the 06Z GDAS has only a 4 hour cutoff time and therefore much fewer data is incorporated into the analysis scheme. This is due to the satellite ground station not communicating with one of the data transmission satellites between 05 and10z (thus when it transmits the newer data first, the 06z data does not arrive until almost 12z). Therefore, it was decided to delay the cutoff time to 6 hours. This is not expected to have any impact on the SPs node usage.
2. Parallel Results: Hua-Lu reiterated the characteristics of the parallel Global Model which include
"Reduced Grid" (reduction in resolution toward the poles where it now reaches around 1 km because the longitude lines meet as they approach the poles), incorporation of 30 second orography, a fix to the Ozone scheme, and a modified quality control scheme. These changes will be implemented sometime this fall without a CAFTI meeting because they are considered sufficiently minor.
3. CMC data: It was reported that CMC grids (control and ensemble) are now routinely flowing into the WWB in part through the efforts of NCO USSB Luis Cano.
3. Status of the Ensembles
a. The focus of discussion centered around the Short Range Ensemble Forecasts (SREF). Steve Tracton reported that the SREF system is running routinely although is not operational. An improved color scheme has been employed on some products to help users more easily delineate between Eta and RSM members. New products have been added including QPF in 6 hourly increments, 700mb temperatures, and spread for 850mb winds. Jun Du reported that successful completion of a SREF cycle is dependent upon a number of outside factors including availability of nodes on the SP and the availability of the EDAS. The lateness of the SREF product suite availability is due to the SREF dependency on completion of the Global Ensembles Forecasts.
4. Input to EMC from Operational Centers
Outside of the recent Eta 22km evaluation submitted to EMC by HPC, there were no additional comments forwarded.
The decision made during the previous synergy meeting to add CPC to the mailing list for both meeting attendance and reception of highlights was reaffirmed.
6. Next Meeting Proposed Monday August 28, 2000 at noon in room 209.