EMC/HPC/MPC/NCO/CPC synergy meeting highlights 06/25/01
This meeting was led by Stephen Jascourt and followed a generic agenda denoted by the subtitles below. Attendees included Steve Tracton, Geoff DiMego, Hua-Lu Pan, Steve Lord, Ed Danaher, Mike Schichtel, Mike Bodner, Vern Kousky, Jim Hoke, and Ed Schoenberg.
1. IBM SP
No news was reported and nobody from NCO attended
2. Notes from EMC
a. Mesoscale Modeling Group: Geoff DiMego reported that the "spring" model upgrade package has been delayed and is now scheduled to be implemented on July 24. Hourly precipitation analyses to be used in the assimilation of precipitation data will become operational July 10. Meanwhile, the bundle is continuing to run in parallel mode and the results continue to look good.
The shortwave radiation is being worked on to properly reduce shortwave flux beneath clouds (right now surface fluxes on a cloudy day in the model are too similar to clear sky values). The scheme running in the global model would take around seven times longer in the Eta than what is currently running in the Eta, so shortcuts are being developed.
b. Global Modeling Group: Hua-Lu Pan reported that the minor fixes to the spring upgrade package are being tested, and as expected, the results show little change, with the tropics looking a little better. Also, assimilation of Quickscat scatterometer winds is being tested.
c. Short range ensembles (SREF): Steve Tracton reported that scaling for the spread in initial perturbations was increased to match the variance in analysis first guess error and that the Eta version used in SREF is being modified to match the changes being implemented in the operational Eta.
3. Input to EMC from Operational Centers
James Partain passes along his "big Thanks" to all those involved in coordinating and implementing the North Atlantic Hurricane Wave Model "in the midst of the backup, critical weather, etc!".
4. Scheduling Discussion / proposal to consolidate AVN/MRF medium-range models
Hua-Lu brought forth a Global Modeling Branch proposal to eliminate the MRF but extend the four daily AVN runs to 16 days, 7 days at full resolution and the remainder at much reduced resolution (as is done in the MRF now). The 0z AVN out to 16 days would then be available sooner than the MRF is now, and the 6z AVN to 16 days with all the late data that now gets into the MRF would be available roughly 3 hours after the MRF now comes out. This meets TPC requests for extended range guidance at 6z and 18z and allows building global ensembles four times per day. Looking down the road to a T254L64 global model and 12 km Eta model, each will take up more than half of the operational side of the SP. The MRF and the 6z Eta would collide, and this rescheduling both avoids that conflict and simplifies the product availability and computer schedules (they repeat every 6 hours). Notification of this change will be now sent to all of the NCEP service centers. Implementation is intended to coincide with the T254L64 implementation sometime next calendar year, giving plenty of advance notice.
Hua-Lu also noted that the T254L64 AVN will require around 4 minutes longer per day to run than the present AVN, so a 5-day forecast would be delayed around 20 minutes compared to presently. However this still falls around the same scheduled completion time as when NCEP models were running on the Cray C90, and the IBM SP has been much more consistently on time than the Cray.
Discussion was aided by participation from CPC and from HPC and MPC forecasters, including Mike Bodner, who just joined HPC from WFO Pleasant Hill. CPC is happy with the proposal. HPC and MPC appear happy with the proposal as well, particularly HPC with their possibility of having a second medium-range shift. However, HPC and MPC raised concerns about the AVN taking longer to run when the resolution is improved. This brings us back full circle to the discussions many months ago about whether to maintain fixed delivery times or allow earlier delivery times when new computers are running at less than capacity early in their lifetime. By squeezing requirements on the forecaster end, the acceptable model run time gets squeezed, causing problems down the road for running the better models for which the computer specifications were intended. Many factors are coming into play, including the field offices being required to use digital databases that generate worded products - it takes an hour longer now for the field to create the zone forecasts using the new technology. Additionally, the requirement for zones out to seven days has combined with a software "feature" that requires issuing/updating the entire days 1-7 package at the same time instead of morning issuance of short range and afternoon issuance of medium range forecasts. This is a factor driving times when HPC products are needed by the field. Also, MPC uses HPC medium range products in developing MPC products which are distributed to ships via radio fax at fixed times, preventing flexibility of HPC medium range schedule. HPC medium range forecasters issue their products roughly 80 minutes after the 12z AVN presently comes in.
5. Next Meeting Proposed Monday July 30, 2001 at noon in room 209.
Note however that the joint AMS conferences on Numerical Weather Prediction, Weather Analysis and Forecasting, and Mesoscale Processes will be held July 30-Aug 2 in Florida, and many people from this building may be attending.