Numerical Weather Prediction Questions Answered

[WxUSAF]

Great stuff. 

[Bob Chill]

Dtk, I've heard the phrase convective feedback issues about million times over the last 10 years. I think I have a better grasp now with seeing odd precip shields with coastals at range. Especially the gfs.

Are models really prone to this or is it more a weenie scapegoat for not seeing what we want and pointing fingers?

[usedtobe]

Dtk, I've heard the phrase convective feedback issues about million times over the last 10 years. I think I have a better grasp now with seeing odd precip shields with coastals at range. Especially the gfs.

Are models really prone to this or is it more a weenie scapegoat for not seeing what we want and pointing fingers?

Bob, I'll give my opinion though I'm not the expert DTK is on models.  Convective feedback can occasionally occur but is way overused.  The problem used to be really bad back in the days of the LFM when it could substantially alter the surface pattern.  The scale of the convection was considerably smaller than the grid size so the use of parameterization schemes was not really to produce the right amount of rainfall but to keep the latent heat from really messing up the mass fields. The less successful the cnvective scheme, the more it would play an adverse role on the pressures, winds, low level convergence.  The convective schemes have gotten more sophistcated so they don't have that problem near as often as they used to. 

 

Now most of the time when the phrase is used it's because someone doesn't understand why the models are doing what they are doing.   I think, there was a run of the NAM during the lead up to the blizzard when it had a similar 500h pattern to all the other models and then somehow jumped the low north over Delmarva with a precip max and had an impact on making the soundings over DC warmer than the other models.  I think that was a case of convective feelback but more often than not, the term is just thrown out there to explain non linear evolution of storms that we don't quite understand. 

 

As DTK alluded to,  there as you go down in resolution you get into a zone where the scale is between the scale where you can explicitly handle convection (1Km) and resolution that is a little lower than what you want for using a convective parameterization scheme but is still too large to explicitly handle convection

 

Anyway, those are my thoughts.  DTK can correct me or elaborate. 

[Jebman]

Best thread EVER on American Weather. I am loving it with all my heart!

[dtk]

Dtk, I've heard the phrase convective feedback issues about million times over the last 10 years. I think I have a better grasp now with seeing odd precip shields with coastals at range. Especially the gfs.

Are models really prone to this or is it more a weenie scapegoat for not seeing what we want and pointing fingers?

Bob, I'll give my opinion though I'm not the expert DTK is on models. Convective feedback can occasionally occur but is way overused. The problem used to be really bad back in the days of the LFM when it could substantially alter the surface pattern. The scale of the convection was considerably smaller than the grid size so the use of parameterization schemes was not really to produce the right amount of rainfall but to keep the latent heat from really messing up the mass fields. The less successful the cnvective scheme, the more it would play an adverse role on the pressures, winds, low level convergence. The convective schemes have gotten more sophistcated so they don't have that problem near as often as they used to.

Now most of the time when the phrase is used it's because someone doesn't understand why the models are doing what they are doing. I think, there was a run of the NAM during the lead up to the blizzard when it had a similar 500h pattern to all the other models and then somehow jumped the low north over Delmarva with a precip max and had an impact on making the soundings over DC warmer than the other models. I think that was a case of convective feelback but more often than not, the term is just thrown out there to explain non linear evolution of storms that we don't quite understand.

As DTK alluded to, there as you go down in resolution you get into a zone where the scale is between the scale where you can explicitly handle convection (1Km) and resolution that is a little lower than what you want for using a convective parameterization scheme but is still too large to explicitly handle convection

Anyway, those are my thoughts. DTK can correct me or elaborate.

The quick and dirty version: This is probably something that was a much bigger issue long ago when models were much coarser and convective schemes less mature. So I would go with "weanie scapegoat" when things don't look as they should. The atmosphere is very complex and never fits the smooth, conceptual models that we think it should.

The longer response: Wes's comments are pretty much spot on. To elaborate, the real issue I have is in the phrase itself "convective feedback". By design, the convective parameterization is supposed to "feed back" to the model grid. That's the whole point. Without it, the model would be left with all sorts of unphysical, unstable profiles and precipitation would be totally messed up. Now, are there occasions where there is too much convection firing and the scheme is overactive? Sure. Are there cases where the convection is happening in the wrong spot? Sure. Is the scenario that Wes described possible? Absolutely.

Deep convection causes substantial changes to a profile in the real atmosphere. Models are trying to mimic this. If strong convection is firing and there is substantial latent heating (and upward moisture flux), the model has to respond in order to restore some sort of quasi-balance. The generation of vortices is something that can really happen in the atmosphere.

I think that two of the issues with present day physics, even mature deep convection schemes, are

1) Many of the schemes were initially developed for, and are most appropriate in, the tropics; and

2) Most model physics are done column wise-independently.

With respect to point 2, this means that a single grid point can have deep convection fired while all surrounding grid points may not. Now, typically there will be some feedback to neighboring points through advection, etc.; but at the fidelity of the model physics time step, every vertical column is treated independently. A lot of work has been done using single column models to develop, test, tune, and calibrate these schemes, but there is some evidence that this may not be the best way to treat certain processes.

If any of you are familiar with MetEd, there is a decent material on this topic in the "How Models Produce Precip and Clouds" Module.

In addition to the current class of schemes, there is some pretty neat work on developing more "scale-aware" parameterization and using mini cloud resolving models within coarser model grid boxes (so-called "super-parameterizations").

[H2O]

Great thread and i love learning how things work so many thanks.

 

But I do have to ask(because I have a reputation to uphold), when will the CRAS get an upgrade?  I'm really hoping this becomes the model of choice

[Wonderdog]

Are models calendar sensitive?

[MN Transplant]

Are models calendar sensitive?

 

In that they know where the sun is?  Yes.

[Wonderdog]

In that they know where the sun is?  Yes.

Can you elaborate? Or do I need to check my sarcasm meter?

[midatlanticweather]

This seems too good to keep in the MA Forum! Maybe if it was all summarized it could be a great Blog entry. Great stuff in here!

[MN Transplant]

Can you elaborate? Or do I need to check my sarcasm meter?

 

They know what the solar insolation is (W/M2 at the top of the atmosphere).  Really, for the rest they have the observations.  Land surface cover, sea surface temps, temperatures through the column, etc.  The physics of the system does not change over the seasons.  So, if you are asking whether the models in particular need to know that it is summer to form pop-up thunderstorms in the south, no. 

 

(I'm pretty sure that is right, I'm not a modeler)

[CoastalWx]

I'll have to set aside time for this thread after skimming. Good stuff DTK.

[cae]

Could someone please explain what the FIM and NIM models are?

[dtk]

Could someone please explain what the FIM and NIM models are?

The FIM (Flow-Following Finite-Volume Icosohedral Model) is an experimental global model under development at NOAA/OAR/ESRL. In addition to using a grid made up of hexagons, it utilizes an adaptive hybrid sigma-isentropic vertical coordinate.  In other words, the vertical layers are terrain following near the lower boundary and then on potential temperature surfaces aloft.  Technical details regarding the formulation can be found here: http://ruc.noaa.gov/pdf/FIMpaper-Bleck-etal-MWR-2015.pdf.

 

The NIM is a nonhydrostatic modification to the FIM. This effort has occurred separately from the NWS/Environmental Modeling Center developments.  However, many of the FIM/NIM runs utilize the NCEP GFS physics and/or initial conditions.

 

Despite what some folks may have heard, the FIM is not a potential replacement for the GFS.  There is a forward looking project under way called the Next Generation Global Prediction System (http://www.nws.noaa.gov/ost/nggps/), from which some initial tests have been carried out in an effort to decide on a future dynamic core for the NCEP global model.  A report from the initial set of tests can be found here (http://www.nws.noaa.gov/ost/nggps/dycoretesting.html).  The initial set of tests have narrowed things down to two potential candidates:  NCAR MPAS and GFDL FV3.  I believe that the final decision is to be made sometime in the next several months.  That's right, the spectral dynamic core of the GFS is going to be replaced with either MPAS or FV3 in the future.  I don't recall the official timeline, but I believe that it is set to occur by 2019.  

[cae]

Thanks.  It looks like the next generation weather model will be non-hydrostatic.  In practice, what are the advantages of a non-hydrostatic model?  Are any of the currently operational global models non-hydrostatic?

[dtk]

Thanks.  It looks like the next generation weather model will be non-hydrostatic.  In practice, what are the advantages of a non-hydrostatic model?  Are any of the currently operational global models non-hydrostatic?

Most of the operational global models are hydrostatic, ECMWF and GFS included.  I think that the Met Office Unified Model is probably run non-hydrostatically, but I'm not entirely sure.  It certainly has a non-hydrostatic option given the resolutions that they run that core for their regional applications such as the UKV, etc.

 

Essentially, hydrostatic models are making approximations about vertical balance.  In atmospheric dynamics, a scale analysis of the governing equations shows that this is an appropriate approximation for phenomena that have larger length scales than vertical depth.  Hydrostatic balance is one of the fundamental balances/assumptions used to diagnose synoptic scale weather.  In simplest terms, vertical motion is generally treated as a diagnostic variable through the continuity equation, and is not something that is actually in the model prognostic equations.  This has implications since vertical motion is not being advected or doing advecting (see previous bits on what parameterizations do to help compensate for this).

 

For phenomena with small spatial scale, large vertical accelerations, etc., the hydrostatic approximation becomes less valid.  If you are trying to explicitly predict boundary layer processes, convection, etc., having non-hydrostatic equations is extremely important.  For the current global models, these things are generally handled by parameterizations.  When the spatial resolution starts to get down to regions for which some of these processes are resolved by the model, the need for relaxing the hydrostatic assumption is paramount. 

 

In simple terms, non-hydrostatic models include vertical velocity in the actual prognostic equations.  Vertical motion can be advected (and do advecting), buoyancy is directly accounted for including pressure perturbation effects, etc.

[cae]

Thanks again dtk.

[isohume]

Excellent thread, dtk. However, I have a feeling folks will be back to model-bashing in no time.   

[ers-wxman1]

I would like to say thank you to whoever started this thread. There is good information in here and it is a breath of fresh air compared to recent posts in other threads. It's hard to make a meteorological analysis or provide a new way to look at forecasts and models without somebody making a mockery out of it or belittling without proper understanding.

[clskinsfan]

Since this winter season is basically over. How happy or unhappy are the developers with the performance of the new GFS? Are there any changes in the works before next winter?