12/31/2007

Folks;

We seem to be at that point again when many of our users have forgotten what are the differences between NAM and GFS and between their respective data assimilation systems, NDAS and GDAS. This comes up every time there is a big difference between the two solutions (I hear some of you asking "... like this doesn't happen EVERY day?") People then want to know why and I'm sure most everyone has an opinion and I've heard quite a few. Sorry to say, many of them are just not accurate or are based on an unwarranted assumption. So here, in all their glory, are ALL the differences laid out in table form in the attached MS Word file and on the web at http://www.emc.ncep.noaa.gov/mmb/mmbpll/misc/NAM+GSI-vs-GFS+GSI.html .

Table 1 has a column labeled "Priority" in which I've listed a year when we hope to have brought the two systems together in the particular attribute or when we have ruled it out as a contributor. The column listed "Possible Test Strategy" was left off the public html version.

Table 2 lists the model differences and this table has no Priority column. This is due to the fact most of our modeling efforts in the next couple of years will be devoted to two major projects: 1) tuning the 'explicit' 4 km nested version of the NMM for inclusion in NAM in the 2009-2010 time frame and b) converting over to the NCEP Environmental Modeling System (NEMS) that Steve Lord detailed at the last two NCEP Production Suite Reviews (see http://www.emc.ncep.noaa.gov/research/NCEP-EMCModelReview2007/Progress_on_Future_NCEP_Production_Suite.dec2007.pdf)

As I mentioned in my presentation at the NCEP Production Suite Review, we continue to see situations of poor NAM forecasts being improved when we rerun them using GFS initial conditions. We have examined Table 1 or ones like it for years trying to isolate the critical differences between NDAS and GDAS. Even now, though both use the same analysis code in the Gridpoint Statistical Interpolation (GSI), we are still seeing cases where rerunning with GFS initial conditions improves the NAM forecast. Please remember it is not possible to run NAM off GFS initial conditions in the Production Suite at this time. It is, however, our intention to evolve towards a Production Suite where NAM and GFS are run concurrently as Steve stated.

This Winter we have decided to run a near-real-time parallel where we return to a partial-cycling strategy in the parallel NDAS. In this we use GFS as the source of atmospheric fields at the T-12 starting point of each NDAS but continue to use the cycled land-surface states of the previous NDAS. This is running now as NAMY. We are using the NAMEXP as the NAMY's control. The NAMEXP is the new NAM/NDAS upgrade bundle being tested for a planned February 2008 implementation. See my remarks at http://www.emc.ncep.noaa.gov/research/NCEP-EMCModelReview2007/EMC-MMB-ProdSuiteRev.Dec07.pdf or check out the parallel change logs for information on the NAMEXP. TIN 07-96 has already been issued on this upgrade bundle and can be found at http://www.nws.noaa.gov/om/notif.htm. This partial cycling approach could be run in current production suite and we will consider it for the next NAM/NDAS bundle in late Fall 2008 if we see consistently higher scores. However, we have not given up on the idea of independent NAM cycling and we have GSI development resources working on several of the items in Table 1. Please don't give up on us yet.

As usual, if you have any questions or comments - I'm all ears.

Geoff DiMego