Messages posted by: Bill Bua

Jonathan --

First off, I apologize for not responding until now. I'd not gotten a message saying something had been posted here.

Secondly, what got rid of the bull's eyes was making the radii of influence used to interpolate data from the individual single point MOS sites to the grid proportional to the density of the data. The less dense the data, the larger the radii of influence. This spreads the single point data further in sparse data regions, and reduces the "bull's eye" effect.

Hope that helps!

Bill Bua

This just forwarded to me from MDL in Silver Spring. The changes are expected to reduce the (what I would contend is) Gridded MOS tendency to analyze too close to single stations in areas of low-density data. Content of the e-mail is shown below:

1) On June 5th, starting with the 12Z cycle, we will be upgrading the GFS-based gridded MOS guidance. All of the elements will be updated to use a newer analysis scheme to eliminate excessive blotchiness and pixelation. We've also extended the coverage by about another 50 km around the CONUS. We are adding five new elements to the existing suite -- wind gust, opaque sky cover, 6- and 12-hr expected precipitation amount, and 24-hr snowfall amount. For more information, please see http://www.weather.gov/mdl/synop/tin/txt/tin06-87aab_gfs_mos_addl.txt.

Bill Bua

The new GFS is now running in real-time parallel on the MPP machines in Gaithersburg MD. The biggest changes are as follows:

1. New hybrid sigma-pressure coordinate system
2. Gridpoint Statistical Interpolation used for data assimilation (analysis done in grid point space, rather than in spectral/wave number space).

The graphics can be found at:

http://www.nco.ncep.noaa.gov/pmb/nwpara/analysis/

for the parallel, and just replace "nwpara" with "nwprod" in the URL above.

Bill

The NCEP MREF is about to be upgraded to 20 perturbations per cycle (+ a low-resolution control, for a total of 21 members in the ensemble). This new ensemble is running in parallel right now. Spaghetti graphics for this parallel can be seen at [url=http://www.emc.ncep.noaa.gov/gmb/wd20bb/EPSpara]this link[/url].

Chris --

Ignore the message on the other thread you set up. I'll find out who it is you need to talk to about how to set up getting the GMOS into your GFE.

Bill

Chris --

Do you have AWIPS Build 7.1? That's the build that is supposed to start "seeing" Gridded MOS.

Bill

Hi folks --

Questions that came up yesterday in Gridded MOS session that I wasn't sure of the answer to, and the correct answers, appear below. Thanks to Paul Dallavalle for help on most of these.

1. How big an area does the "lapse rate" calculation have to cover to get to the "60-100 stations"? Isn't this a problem where data is sparse? How does the area covered here related to the OI 30 gridpoint radius of influence?

A radial search is done around each of the 8,500 stations used for the single station part of Gridded MOS until either 100 stations is obtained or the search goes out to the maximum analysis radius (150 km or 30 grid boxes). The "60" figure is not a minimum, but rather a representative average of the number of stations that ultimately get used. In sparsely populated regions, this number will obviously be considerably less.

2. How about situations where synoptic situation might result in different lapse rates over different parts of the 60-100 station area, and averaging will give a bad result (e.g. cold-air damming, warm surge west of mts). Is averaging of marine layer and non-marine layer areas perhaps resulting in the relatively poor results for temperature in the GMOS graphics shown in the presentation from Monterey?

Yes to both. In general, the problem is that we are applying an "isotropic" (symmetric in all directions, in a circle) method to a "anisotropic" (not symmetric in all directions) situation. To deal with this, MDL is working on shrinking the search radius so that the assumption of symmetry/circularity of the region with similar "lapse rates" doesn't do as much damage in these circumstances.

3. In discussing the lapse rate determination process, I said that the data determines the lapse rates. The question arose: what about times when there is a diurnal flip (cooling with height daylight, warming with height during nighttime if no winds allow for radiational cooling)? How well will it do? Isn't it dependent on the single station MOS used for GMOS containing that information?

If the "diurnal flip" in lapse rates exists in the single station data being interpolated to the Gridded MOS 5-km grid, it will exist in the grid as well. This thus depends totally on the single station data accurately reflecting radiational cooling when winds calm down. I did find an example that seems to illustrate that this is working at least to some extent in central NY state, attached (see Elmira NY v. Binghamton NY).

4. How do I explain to my forecasters large differences between values from GMOS and MOS for the same location? (I suggested it was the result of different developmental samples between GMOS and MOS.).

According to Paul, this is not because of differences in times used for developmental samples. The analysis DOES make a difference, in that for any station, the nearest grid box to station is used. Additionally, buried deep inside is an "ensemble forecast", in that the previous cycle's forecast is included in the current cycle (2/3 weight given to current forecast, and 1/3 for prior forecast (12 hrs. old)). This is done for T, Tmax, Tmin, Tdew ONLY.

5. Is Tim Barker's MatchGuidance (here, traditional MOS) smart tool not as good as GMOS?

Paul stated that Gridded MOS is merely another approach we're trying to get gridded forecasts, not a value judgment on MatchGuidance.

6. Some folks wanted to know which specific stations were being used in their CWA for GMOS. Is there a place where that is stored? Can I get the info?

MDL is working on parsing out the list of Gridded MOS sites (the "GMOS dictionary) for each WFO now, as requested at an ISST meeting over the summer. We'll set up a webpage (or MDL will) once that information is available.

Let me know if you need further information or have more question.

Thanks for your interest in Gridded MOS!

Bill

Rob and others --

I removed my examples from the string because I found out that that particular NAMY run (that looked much better than the NAM with its 2-m temps) was started from different initial conditions from usual, because the 18 UTC NDAS (the NAM Data Assimilation System) failed and the NAMY for 12 UTC 9 July was started from interpolated [b]GFS[/b] initial conditions, so it wasn't a clean comparison. Not only that, but the NAM overamplified a trough to create a 570-dm 500-hPa low over southwest lower MI while the NAMY had something that just progressed along; probably the right answer on the 2-m temps was the result of the initial system being weaker in the NAMY as the initial condition was interpolated from a coarser model.

That being said, the cool bias seems to be seen quite a bit in the NAM. One of the culprits that has been discussed is too little vertical mixing. Rest assured the problem *is* being worked on, and that there are some possible improvements in the works. Timing on getting anything implemented, however, is unknown at this time.

Bill

All --

The Gridded MOS is now being run on an experimental basis over the full CONUS.

Check out

http://www.weather.gov/mdl/synop/gmos.html and click on "experimental MOS graphics as part of NDFD" in the middle column.

Training on Gridded MOS will be forthcoming before summer's end.

Bill

All --

I'll add this topic for the forum, even though the event took place the day before the NAM WRF NMM became operational.

I've attached some graphics as an example that came up in a discussion I had with some forecasters on Tuesday; how well did the NAM NMMWRF do compared to the NAM Eta with the flooding rains in Houston TX on 6/19. These are from the [url http://wwwt.emc.ncep.noaa.gov/mmb/ylin/pcpverif/daily/]Meso Modeling Branch precip verification webpage[/url]. Note that there really isn't much of a qualitative difference between the two, though it would appear that the NAM WRF-NMM had a little bit better intensity and placement. The Hi Res window versions of the WRF (with the NCAR/ARW/EM dynamics and the NCEP NMM dynamics) for the central region (not attached, but in the web page) produce prodigious amounts of precipitation; in fact, TOO much of it, but that would be a big heads up for the forecasters.

If you check out the webpage, you'll see that the GFS had the most precip. and the most consistent and accurate forecast as far out as 72 hours!

Bill

Ray --

The Mesoscale Modeling Branch maintains verification statistics on its operational and parallel model runs (here the Eta versus WRF):

http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/mmbverif.namplls/

Near-surface and upper air statistics are kept. Additionally, precipitation can be compared over 24-hour periods at:

http://wwwt.emc.ncep.noaa.gov/mmb/mmbpll/mmbverif.namplls/

Bill

Steve --

The NAMX has been tending to create spurious tropical cyclones, and it would appear that this will be a feature of this model. I suspect it's the result of the non-hydrostatic nature of the model (buoyancy is predicted, at least at the 12-km/resolved scale of the model). Retrospective runs from last summer handled the development of Hurricanes Dennis, Katrina, and Rita much better than the NAM, and really "went to town" on their deepening into major hurricanes.

The spurious cyclone problem was also a problem with the GFS (and still is to a small extent). I would expect to see some more false alarms out of the NAMX when it's put into operations, but since it's not really used for hurricane guidance (a Hurricane WRF is under development, however, that'll be tuned to handle hurricanes [b]specifically[/b]), it's not a show-stopper for implementation. You just have to know to look out for these false alarms.

If the GFS also develops a tropical system, I'd suggest that would be a good way to confirm development.

Bill

It looks as though the problem has been solved. See the message that got posted from Dave Novak in Eastern Region. Interesting case, too!

Hi All,

Noticed large QPF differences from the 0600 UTC 2 May 2006 runs of the NAM, parallel NAM, and GFS in the northeast U.S. In particular, the NAM completely missed the precipitation backing into New England associated with a retrograding inverted trough, while the parallel NAM and GFS did well (at 6 h forecast projections). This seems to be further evidence of the improvements in the initial condition scheme implemented in the parallel NAM (probably moreso than the model dynamics change to WRF given the large differences at such a short forecast projection). As soon as the relevant images can be attached, I will do so.

Dave/Bill Bua

Joe --

This is a MOS question that I'll refer to the Techniques Development Lab (TDL). They're the ones that develop MOS.

However, at day 6 or 7, it probably is synoptic variable fluctuations in the model that are resulting in what you are seeing. I *think* the same equations are used for ALL GFS MOS, no matter when the forecast cycle starts. What did the large scale flow look like for each of those forecasts?

Bill