Jump to content
  • Member Statistics

    17,606
    Total Members
    7,904
    Most Online
    ArlyDude
    Newest Member
    ArlyDude
    Joined

T1534 GFS .. GFS at 13KM ( along with other upgrades ) can be found HERE


Recommended Posts

  • Replies 62
  • Created
  • Last Reply

Great information. It looks like there is a lot information available on the net already if you google for "4DEnVar" including some journal articles on the AMS site...too bad I'll have to wait a couple years until they're free to access. Thanks!

Send me a PM with an email address.  I'd be happy to pass along my papers along with some publicly available presentations on the topic.

Link to comment
Share on other sites

  • 4 weeks later...

Dtk, will the GFS ensemble members have a different resolution when the GFS parallel becomes operational?

Not immediately.  I know that there is a plan to move to T574 Semi-Lagrangian from the current T254 Eulerian.  However, I have not seen the implementation schedule.

 

They will still feel the benefits (or negative impacts) from the T1534 from the analysis.  However, there will certainly be a window until they implement where the operational GEFS will not really be truly representative of the T1534 operational deterministic GFS....

Link to comment
Share on other sites

  • 3 weeks later...

I found this bulletin about the 13-km GFS upgrade on the NCEP NCO web page (this information has been edited)

 

full detailed message here:

 

http://www.nws.noaa.gov/om/notification/tin14-46gfs_cca.htm

 

Technical Implementation Notice 14-46 Corrected

National Weather Service Headquarters Washington DC

1055 AM EDT Tue Dec 23 2014

 

 

Subject:   Corrected: Global Forecast Systems (GFS) Update:

           Effective January 14, 2015

 

 

Effective on or about January 14, 2015, beginning with the 1200

Coordinated Universal Time (UTC) run, the National Centers for

Environmental Prediction (NCEP) will upgrade the GFS Analysis and

Forecast System as follows:

 

  - Changes to the model components

  - Increases in horizontal resolution

  - Addition of 0.25 degree gridded output

  - Addition of new product fields

  - Change to product naming convention

 

Link to comment
Share on other sites

I found this bulletin about the 13-km GFS upgrade on the NCEP NCO web page (this information has been edited)

 

full detailed message here:

 

http://www.nws.noaa.gov/om/notification/tin14-46gfs_cca.htm

 

Previous content removed. DTK fully addressed my concerns and also provided additional relevant information that was quite helpful.

Link to comment
Share on other sites

This is disappointing, as the parallel version continues to fare somewhat worse on all verification criteria (heights, slp, vector wind, zonal wind, meridional wind, and temperature) than the current version, both at 0z and 12z.

 

http://www.emc.ncep.noaa.gov/gmb/STATS_vsdb/

http://www.emc.ncep.noaa.gov/gmb/STATS_vsdb/12Z/

 

It appears that implementation on 1/14 will be more the result of the sunk cost fallacy (time and costs already expended) than the merits of making the switch given a persistent and broad-based performance deficit relative to the current version of the GFS. That the performance deficit has continued throughout December and into January suggests that this deficit likely is not a temporary phenomenon that will quickly reverse.

 

IMO, implementation should occur when the issues responsible for the performance gap are fixed and the parallel version has demonstrated consistently better performance.

Don, while the new system is not performing that great of for the past 30 days in the NH (it is doing much better in the SH), the developers have run something like 2.5 simulated years with the new system and on the while it is demonstrably better.  For example, here is a die off curve for the past year:

cordieoff_HGT_P500_G2NHX.png

 

The upper right panel is for all waves, and the difference between the 13km and the operational GFS is statistically significant for all lead times out to 6days.  Here is an example for Atlantic tropical cyclone track errors:

tracks_al.t.gif

Feel free to poke around the larger sample statistics from the various experiments here:

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhw14/

here

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhs13/

and finally, here

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhs12/

 

While the new package may not be as much of an improvement as people were hoping/expecting (including developers), it is certainly better than the current operational GFS for many variables, lead times, and metrics, when a large enough sample is considered.  Implementations are tricky....it is impossible to make everything better in all regions for all periods and lead times.

Link to comment
Share on other sites

Don, while the new system is not performing that great of for the past 30 days in the NH (it is doing much better in the SH), the developers have run something like 2.5 simulated years with the new system and on the while it is demonstrably better.  For example, here is a die off curve for the past year:

cordieoff_HGT_P500_G2NHX.png

 

The upper right panel is for all waves, and the difference between the 13km and the operational GFS is statistically significant for all lead times out to 6days.  Here is an example for Atlantic tropical cyclone track errors:

tracks_al.t.gif

Feel free to poke around the larger sample statistics from the various experiments here:

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhw14/

here

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhs13/

and finally, here

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhs12/

 

While the new package may not be as much of an improvement as people were hoping/expecting (including developers), it is certainly better than the current operational GFS for many variables, lead times, and metrics, when a large enough sample is considered.  Implementations are tricky....it is impossible to make everything better in all regions for all periods and lead times.

Thanks DTK. I greatly appreciate this information and it is reassuring.

 

I hope you had a great Holiday Season.

Link to comment
Share on other sites

Don, while the new system is not performing that great of for the past 30 days in the NH (it is doing much better in the SH), the developers have run something like 2.5 simulated years with the new system and on the while it is demonstrably better.  For example, here is a die off curve for the past year:

cordieoff_HGT_P500_G2NHX.png

 

The upper right panel is for all waves, and the difference between the 13km and the operational GFS is statistically significant for all lead times out to 6days.  Here is an example for Atlantic tropical cyclone track errors:

tracks_al.t.gif

Feel free to poke around the larger sample statistics from the various experiments here:

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhw14/

here

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhs13/

and finally, here

http://www.emc.ncep.noaa.gov/gmb/wd20rt/vsdb/prhs12/

 

While the new package may not be as much of an improvement as people were hoping/expecting (including developers), it is certainly better than the current operational GFS for many variables, lead times, and metrics, when a large enough sample is considered.  Implementations are tricky....it is impossible to make everything better in all regions for all periods and lead times.

how do the die off curves look compared to the ecmwf?
Link to comment
Share on other sites

It will be interesting to see how it fairs with East Coast cyclogenesis once we get an actual storm to develop in that region. The complicated s/w interactions are what the current GFS had some problems with. Anecdotally, I did notice the 13km GFS did well on a few events wrt QPF. 

Link to comment
Share on other sites

The para GFS has been dramatically better in the AK region where our weather patterns are dominated by strong diabatic lows moving in from the NPAC (especially this fall/winter), complex wave-wave, trough-trough, trough-wave mergers, and complex terrain interactions with the synoptic pattern (for instance, the effects of the AK Range on cold air advection patterns and associated 500 mb (upper air) height field adjustments, etc. As a whole, the para GFS has been notably better and closer to the ECMWF in most cases (when glancing at the NCEP mag para site).

Link to comment
Share on other sites

Working on getting my internal scripts changed in time for the upgrade. Sadly the height anomaly maps will be kicking the bucket for a few weeks to months due to a change in the available variables until I roll out a more comprehensive set of anomaly maps, but other than that all of the changes should be positive (such as increased resolution).

Link to comment
Share on other sites

The para GFS has been dramatically better in the AK region where our weather patterns are dominated by strong diabatic lows moving in from the NPAC (especially this fall/winter), complex wave-wave, trough-trough, trough-wave mergers, and complex terrain interactions with the synoptic pattern (for instance, the effects of the AK Range on cold air advection patterns and associated 500 mb (upper air) height field adjustments, etc. As a whole, the para GFS has been notably better and closer to the ECMWF in most cases (when glancing at the NCEP mag para site).

Can you PM me or send me an email with this information?  Although the decision brief is already done, I'd like to pass this along to the developers and get it on record.

 

Thanks.

Link to comment
Share on other sites

Working on getting my internal scripts changed in time for the upgrade. Sadly the height anomaly maps will be kicking the bucket for a few weeks to months due to a change in the available variables until I roll out a more comprehensive set of anomaly maps, but other than that all of the changes should be positive (such as increased resolution).

I am not seeing the GFS maps on your sight at all right now.

Link to comment
Share on other sites

Twisterdata doesn't plot the full 13km information from the new GFS. It looks like it has 25km grid scale information plotted. College of Dupage may be plotting this at 25km, similar to Twisterdata.  I wonder if these web sites (or others) are having a problem with the computing resources needed to fully process the 13km GFS GRIB files (or NETCDF or whatever.)

Instantweathermaps also has been dealing with the 13-km GFS alright. Note to Instantweathermaps developer: you may want to check the 2m-Temperature maps.

Link to comment
Share on other sites

Twisterdata doesn't plot the full 13km information from the new GFS. It looks like it has 25km grid scale information plotted. College of Dupage may be plotting this at 25km, similar to Twisterdata.  I wonder if these web sites (or others) are having a problem with the computing resources needed to fully process the 13km GFS GRIB files (or NETCDF or whatever.)

Instantweathermaps also has been dealing with the 13-km GFS alright. Note to Instantweathermaps developer: you may want to check the 2m-Temperature maps.

In my case I highly suspect the only problems (read: delays on the order of tens of seconds, also far-more-substantial delays on the GEFS) are on the end of NOAA's bandwidth. I'm going to try more tests and upgrade to gigabit on my end fairly soon to make sure it's not my bandwidth that's causing problems (I suspect that for the GEFS it might be since I'm capping out at 100Mbps quite often)... after some pretty hefty optimizations (which will be added to the other models soon), CPU is really a non-issue, and the downloads aren't totally keeping up, so I'm pretty sure it's bandwidth-related. I've temporarily(?) removed the rudimentary error-checking I had in place (which involved downloading everything twice), so a very small number of maps may end up corrupted... so far that hasn't been any more of an issue as in the past, so I'm thinking there was a placebo effect involved. Also thanks for noticing the T2m thing; I recently stopped using the 0.117-degree (formerly 0.205-degree) files due to how slowly they were being uploaded to the NOAA server (I may re-add them later as a premium feature when I implement specific regions of the US) and forgot to change the 2m temps and snow depth to generate with the normal files. That has been fixed and should be updated with the 18Z run.

Link to comment
Share on other sites

Now is there any site that has eastern us 13km maps?  

 

Given the grib files that are generated, I doubt it.  I think that .25 degree files are the highest resolution grib data that are available for most (aside from some of the stuff that shows up in flux files which are on the model native grid....this includes the surface fluxes, precipitation, and other 2d fields.....not any of the upper air or 3D quantities).  One can always create full resolution plots from the native spectral files, but I doubt anyone is doing so.

Link to comment
Share on other sites

Given the grib files that are generated, I doubt it.  I think that .25 degree files are the highest resolution grib data that are available for most (aside from some of the stuff that shows up in flux files which are on the model native grid....this includes the surface fluxes, precipitation, and other 2d fields.....not any of the upper air or 3D quantities).  One can always create full resolution plots from the native spectral files, but I doubt anyone is doing so.

 

Are those available on the NCEP server?

Link to comment
Share on other sites

Are those available on the NCEP server?

I'd have to poke around, but it looks like only for the first 24 hours.  They are called "gfs.t${cyc}z.sf${fhr}" where cyc is the initialization cycle, and fhr the forecast hour.  I think that only the "sflx" files are available at native resolution in grib, everything else is .25 degree lat-lon or coarser.

Link to comment
Share on other sites

I'd have to poke around, but it looks like only for the first 24 hours.  They are called "gfs.t${cyc}z.sf${fhr}" where cyc is the initialization cycle, and fhr the forecast hour.  I think that only the "sflx" files are available at native resolution in grib, everything else is .25 degree lat-lon or coarser.

 

Cool, thanks!

Link to comment
Share on other sites

Old, unless they upgraded without upgrading the output or issuing a TIN (which I find very unlikely). From what I've heard the new one will be in late 2015 or early 2016; this is extremely preliminary though as there hasn't even been an eval notice issued.

 

The ensemble mean is more defined with areas of 500 mb ridges and troughs after the operational upgrade.

But the NOAA article didn't get into anything more about a full ensemble upgrade beyond what was mentioned

a few weeks ago.

 

http://www.noaanews.noaa.gov/stories2015/20150105_supercomputer.html

 

ahead of this upgrade, each of the two operational supercomputers will first more than triple their current capacity later this month (to at least 0.776 petaflops for a total capacity of 1.552 petaflops). With this larger capacity, NOAA’s National Weather Service in January will begin running an upgraded version of theGlobal Forecast System (GFS) with greater resolution that extends further out in time – the new GFS will increase resolution from 27km to 13km out to 10 days and 55km to 33km for 11 to 16 days. In addition, theGlobal Ensemble Forecast System (GEFS) will be upgraded by increasing the number of vertical levels from 42 to 64 and increasing the horizontal resolution from 55km to 27km out to eight days and 70km to 33km from days nine to 16.

 

More defined ensemble mean after operational upgrade

 

Before upgrade

 

 

After upgrade

 

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

  • Recently Browsing   0 members

    • No registered users viewing this page.

×
×
  • Create New...