GFS getting major upgrade, rolling out in November

maquiladora

Just in time for the winter. :pac: The GFS model is getting one of it's biggest ever upgrades. Version 12 will roll out this November, and the single biggest upgrade is the change of resolution for 23km to 13km. This will nudge the GFS ahead of the current ECMWF's global model which runs at 16km.

This higher resolution output will extend to a longer period, out to 240 hours, like the ECM does now.

There are lots of other quite technical upgrades and tweaks that have been made to the model too, including the input of high resolution daily sea surface temperatures and ice coverage etc. Full list of changes here.

Hopefully all these changes makes a more reliable model overall. The higher resolution will certainly help with giving a more accurate picture for the type of intense low pressure systems we saw last winter.

No word on an exact date of when it will launch yet.

Iancar29

Of what i know though ECM is gettin another upgrade then in the New year which will bring it past the GFS one more

maquiladora

Iancar29 wrote: »

Of what i know though ECM is gettin another upgrade then in the New year which will bring it past the GFS one more

I know they have tested 10 km and the long term goal is to have it down to 2.5 km by 2025, which is insane for a global model.

Nabber

Is there much human calculations done with each run?

Or is it just a variable change to induce another likely out come?

maquiladora

Nabber wrote: »

Is there much human calculations done with each run?

Or is it just a variable change to induce another likely out come?

There are no human calculations with NWP, it requires supercomputers to crunch the numbers. But of course the output is not much use without skilled forecasters to interpret it and apply their own knowledge and experience to it.

Variable changes are part of ensemble forecasting.

Duiske

With this new and improved GFS model we'll be able to watch as that sub-zero Siberian air headed towards us disappears on the 120z, instead of the usual 96z.

M.T. Cranium

The other thing is, you can resolve it down to 5 mm but if energy cycles last 7-10 days and have not started at time zero, then they won't get into the supercomputer.

People need to realize that advances in forecasting will come from a blend of technology and theory. It's possible, I suppose, that the theoretical will creep in by accident if these cycles are hidden away in tiny increments of data.

piuswal

M.T. Cranium wrote: »

The other thing is, you can resolve it down to 5 mm but if energy cycles last 7-10 days and have not started at time zero, then they won't get into the supercomputer.

People need to realize that advances in forecasting will come from a blend of technology and theory. It's possible, I suppose, that the theoretical will creep in by accident if these cycles are hidden away in tiny increments of data.

it seems to me that the real challenge is developing the mathematics that allow the theory to be implemented. Though there is work to be done on theory as well.

Satellite data was avail for many years before the mathematical techniques were developed that allowed the satellite information to be utilised in the physical models.

Additionally, as you go to finer and finer grids, the assumptions that are made for coarser grids have to be modified, e.g. they have had to start using a non-hydrostatic model instead of a hydrostatic model in Met Eireann in recent years.

Likewise the algorithms that are used to describe various process have to be refined. The surface descriptions have to be changed as they get to finer and finer resolution; how urban areas are distinguished from rural areas, how the different types of surfaces are represented etc.

The computer power available will continue to increase inexorably and really is quite a separate issue. The challenge is in the maths and physics.

Another challenge is in the initial data. Satellite data now tends to dominate but there is still a need for ground based data, among other things, to validate or benchmark satellite data and ground based data is very asymmetrically available though made of the gaps have been filled in in recent decades by ocean buoys, both drifting and moored and by constant level balloons (I'm assuming there still being used but have not checked in quite some time).

maquiladora

Meteociel will have the new GFS running in parallel within the next two weeks. Keep an eye out for it.

alfa beta

Looks like the UKMO are upgrading too - cue the battle of the supercomputers!!

http://www.metoffice.gov.uk/news/releases/archive/2014/new-hpc

maquiladora

GFS is due to upgrade on Dec 9th so I'm guessing we should see the parallel runs starting around Nov 9th.

There is also a new hi-res German model called ICON which runs to 77 hours available here : http://meteociel.fr/modeles/icone_cartes.php

hatrickpatrick

GFS on meteociel is live. Click " [ Run parallèle NEW ! ]" on the bottom line of options above the GFS charts. Click the same spot [ Run opérationnel ] to switch back to the original.

maquiladora

hatrickpatrick wrote: »

GFS on meteociel is live. Click " [ Run parallèle NEW ! ]" on the bottom line of options above the GFS charts. Click the same spot [ Run opérationnel ] to switch back to the original.

Sooner than I expected!

maquiladora

New GFS is now out in 0.25° scale on Meteociel.

http://www.meteociel.fr/modeles/gfs-para/royaume-uni/pression/3h.htm

Iancar29

Not bad !

maquiladora

Iancar29 wrote: »

Not bad !

Yeah. You can see the difference the higher resolution makes.

Oneiric 3

maquiladora wrote: »

Yeah. You can see the difference the higher resolution makes.

A definite difference alright on comparison.

Old GFS:



New GFS:

RED L4 0TH

Notice up on meteociel.fr that the GFS parallel run won't be available from the 3rd to the 8th December. Will the old version be taken down after the 8th I wonder?



http://www.meteociel.fr/modeles/gfse_cartes.php?ech=6&code=0&mode=0&runpara=1

maquiladora

RED L4 0TH wrote: »

Notice up on meteociel.fr that the GFS parallel run won't be available from the 3rd to the 8th December. Will the old version be taken down after the 8th I wonder?



http://www.meteociel.fr/modeles/gfse_cartes.php?ech=6&code=0&mode=0&runpara=1

It's been pushed back until January 7th now. I'm sure it'll still run in parallel until it becomes the new operational. That will replace the old GFS.