Jump to content
  • Member Statistics

    17,609
    Total Members
    7,904
    Most Online
    Vesuvius
    Newest Member
    Vesuvius
    Joined

January 2020 Discussion


Torch Tiger
 Share

Recommended Posts

3 minutes ago, MetHerb said:

I thought that the GFS moved to better hardware as part of an upgrade a couple of years ago?

https://www.noaa.gov/media-release/noaa-kicks-off-2018-with-massive-supercomputer-upgrade

If only they could improve the physics too...

This.

I always wonder with so much emphasis now into higher resolutions if that is playing a roll into some of the model chaos we've encountered these past few years. There have been a tremendous amount of wavering between forecast models...even inside 24-36 hours. I don't have much knowledge with model physics or a background, however, from my understanding to incorporate higher resolutions and incorporate some mesoscale features there needs to be different sets of equations added into the model parameterization? Anyways, while it's great to be able to do things at such high resolution if the physics are being adjusted accordingly or improvements being made I would think doing things at higher resolutions could hurt more than help?

Link to comment
Share on other sites

10 minutes ago, weatherwiz said:

This.

I always wonder with so much emphasis now into higher resolutions if that is playing a roll into some of the model chaos we've encountered these past few years. There have been a tremendous amount of wavering between forecast models...even inside 24-36 hours. I don't have much knowledge with model physics or a background, however, from my understanding to incorporate higher resolutions and incorporate some mesoscale features there needs to be different sets of equations added into the model parameterization? Anyways, while it's great to be able to do things at such high resolution if the physics are being adjusted accordingly or improvements being made I would think doing things at higher resolutions could hurt more than help?

It's why we have/use Ensembles.  With the increased resolutions, you need to have the different permutations to smooth out a mean.

Link to comment
Share on other sites

21 minutes ago, weatherwiz said:

This.

I always wonder with so much emphasis now into higher resolutions if that is playing a roll into some of the model chaos we've encountered these past few years. There have been a tremendous amount of wavering between forecast models...even inside 24-36 hours. I don't have much knowledge with model physics or a background, however, from my understanding to incorporate higher resolutions and incorporate some mesoscale features there needs to be different sets of equations added into the model parameterization? Anyways, while it's great to be able to do things at such high resolution if the physics are being adjusted accordingly or improvements being made I would think doing things at higher resolutions could hurt more than help?

Something like this...  

It's about 'fractal logistics' ... when to tamp down emergent forcing -vs- when to keep them.  

I had this discussion with Ekster years ago at the SNE conference, pre - NAM years... wrt to that meso model, and we were surmising that as the grid got more and more discrete, there seemed to be a correlation to that erratic noise, and continuity distraction .. where the model seemed to be doing worse.  There was a time back in the mid 1990s during the ETA years, where if one was privy to the ETA's dependable biases, you could correct for those on the fly and it was actually a pretty darn good model when doing so..  It seems the grandfather of the present day child was a bit better, doesn't it -

This could be happening with these large numerical model clusters, too, as their grids - in some cases ...- are getting close to competing with meso's and then yes, different convection sequencing and hydrostatic balancing ...all that factors in.   I mean the Euro is not the NAM considering these latter physical make-ups.  

But as scales shrink ... think of it this way - above a certain threshold of finite processing ... you are advantage by the 'natural smoothing' of noise.  Like letting the nature of the physical equations take care of the deviance,  and then what falls out ( no pun intended ) has that normalization baked in.  However, if you get 'too discrete' ..that actually exposes the physical equations to actual empirical assumptions, and then they have to conserve those assumptions out in time.  You end up blowing up falsities in some case...when you're lucky, you don't - that's when the NAM scores that coup that it does ever 20th storm. Heh.  

I'm starting to sense or wonder if we are getting the Global numerical model technology ( in general) operating to a state where it too 'considers too much' ?   

The Euro is supposedly ahead of the tech curve on this with there '4-d variable' system - which is basically ( as I have read ...), not the same thing as managing chaos down at the fundamental emergence, more so what to do with the noise once it's already effecting the run - which is good and better than not doing it. And it shows at < D4.5 or so.  But even the Euro is seemingly prone at times to "giga motions" and running away with false emergence from time to time.  

  • Like 1
Link to comment
Share on other sites

1 minute ago, Baroclinic Zone said:

It's why we have/use Ensembles.  With the increased resolutions, you need to have the different permutations to smooth out a mean.

Absolutely...

however, (an obviously) with that comes limitations and why of course sometimes ensembles are not always the best way to go...even in the medium range. If the mean is too smoothed then the ensemble may "eliminate" a key feature having significant ramifications of the forecast output. 

Link to comment
Share on other sites

10 minutes ago, JC-CT said:

 

 

Right...yay for getting a faster shit

lol!

 

But Maue is saying it's beating the American models.  I don't believe any of those stats anyway....they're all skewed imo to try and prove each ones point.  This model is better at this; this model is better at that etc etc....????   Twist it anyway you want is the name of the game with that stuff it seems.

  • Confused 1
Link to comment
Share on other sites

2 minutes ago, WinterWolf said:

lol!

 

But Maue is saying it's beating the American models.  I don't believe any of those stats anyway....they're all skewed imo to try and prove each ones point.  This model is better at this; this model is better at that etc etc....????   Twist it anyway you want is the name of the game with that stuff it seems.

I don't believe EMC is is skewing stats. They wouldn't be showing the sub-par performance of the GFS model generally if they were. Maue is definitely agenda-driven. 

Link to comment
Share on other sites

1 minute ago, Typhoon Tip said:

Something like this...  

It's about 'fractal logistics' ... when to tamp down emergent forcing -vs- when to keep them.  

I had this discussion with Ekster years ago at the SNE conference, pre - NAM years... wrt to that meso model, and we were surmising that as the grid got more and more discrete, there seemed to be a correlation to that erratic noise and continuity distraction.   

This could be happening with these large numerical model clusters, too, as their grids - in some cases ...- are getting close to competing with meso's and then yes, different convection sequencing and hydrostatic balancing ...all that factors in.   I mean the Euro is not the NAM considering these latter physical make-ups.  

But as scales shrink ... think of it this way - above a certain threshold of finite processing ... you are advantage by the 'natural smoothing' of noise.  Like letting the nature of the physical equations take care of the deviance and the what falls out ( no pun intended ) has that normalization baked in.  However, if you get 'too discrete' ..that exposes the physical equations employed to actual empirical assumptions, and then they have to conserve those assumptions out in time.  You end up blowing up falsities in some case...when you're lucky, you don't - that's when the NAM scores that coup that it does ever 20th storm. Heh.  

I'm starting to sense or wonder if we are getting the Global numerical model technology ( in general) operating to a state where it too 'considers too much' ?   

The Euro is supposedly ahead of the tech curve on this with there '4-d variable' system - which is basically ( as I have read ...), not the same thing as managing chaos down at the fundamental emergence, more so what to do with the noise once it's already effecting the run - which is good and better than not doing it. And it shows at < D4.5 or so.  But even the Euro is seemingly prone at times to "giga motions" and running away with false emergence from time to time.  

This is a fantastic post! 

Obviously a significant aspect to model errors is now everything is initialized...but as you know just b/c an initialization is incorrect doesn't mean the entire model run should be tossed. But from what I've always understood is that the Euro's initialization scheme is far more advanced than ours and that makes a significant advantage in the accuracy of the model.

I have thought if sometimes whether the models consider just too much...so in the end you're just including more variables and just relying on the mathematical equations to solve them and by running ensembles you derive numerous different outputs and just look for whatever the clustering is...good idea in sense but perhaps not as great as one would think. 

I would be interested in the scores but my wager is we've seen so much changes (even with large-scale) inside of 24-36 hours that even short-term forecasting has become rather erratic. Just think of how many storms (going beyond winter weather) have a high level of uncertainty until we're inside of 6-12 hours.

Link to comment
Share on other sites

Just now, OSUmetstud said:

That is a poor trend on the GFS as of late to be the worst verifying model at 500mb at 24 hours over north america. It appears to be right around the upgrade. 

That's a major problem...verification of 500 should not be that far off at 24-hours out. 

Is there a seasonal trend too within these graphs? Especially the first one....much more accurate scores up to 120-out during the summer and less accurate in the winter...which makes sense

Link to comment
Share on other sites

Just now, weatherwiz said:

That's a major problem...verification of 500 should not be that far off at 24-hours out. 

Is there a seasonal trend too within these graphs? Especially the first one....much more accurate scores up to 120-out during the summer and less accurate in the winter...which makes sense

Yeah, but RMSE is only one statistical measure. There's less variability in 500mb heights during the summer so thus RMSE is lower. But in terms of model skill, winter has more skill. Wavelengths are longer and disturbances are stronger and more trackable. Also, less convection. 

Link to comment
Share on other sites

Just now, OSUmetstud said:

Yeah, but RMSE is only one statistical measure. There's less variability in 500mb heights during the summer so thus RMSE is lower. But in terms of model skill, winter has more skill. Wavelengths are longer and disturbances are stronger and more trackable. Also, less convection. 

I was actually just reading about that not long ago. Think it was within one of moduels on MetEd...was quite interesting to read about it.

Link to comment
Share on other sites

13 minutes ago, mahk_webstah said:

I think we had 90% for 3 and similar for 6, but that was perhaps for a whole 10 or 15 day period.  I feel optimistic about this storm, although that doesn't mean shite.  Anything in the pipeline after this?/

In the 10 days that I can see for the euro, this is it... there might be something forming around D12

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...