January 2026 Short/Medium Range Thread

[GBOVolz]

Supposedly the Euro AI has been trained on 80 years of ERA 5 data 

I haven’t read much about the AI mods… what calibrations did they make to the euro AI for it to show a different scenario from the original euro? Do we know?


.

[Bigbald]

I have noticed some mets getting on board for upper east tn, fairly unusual this far out.

[Weatheriscool]

It is amazing that the Weather Channel literally has their forecasts pegged to the GFS; last night it had snow for my area Friday night Saturday and Saturday night, now, maybe an inch Friday night where before it was 1 - 3 for Friday night; 5 - 8 for Saturday, 1 - 3 for Saturday night 

[tnweathernut]

Next 24-48 hours of model runs will be pivotal in determining what this looks like coming out. (i.e. one system vs. several smaller waves of overrunning moisture- Miller A/Miller B and what the orientation the incoming high will take on). Pretty confident a nice winter storm will scoot across, but in the mid-south there's always a chance the warm air across the gulf surges further north than modeled.

[1234snow]

6z Euro was just a thing of beauty (except for the ice). How it unrolls the Baja low is almost perfect with how it has the shortwave moving through the Dakotas.

There is going to be ice somewhere though. I think that is unavoidable with an overrunning setup.

[tnweathernut]

37 minutes ago, Holston_River_Rambler said:

Supposedly the Euro AI has been trained on 80 years of ERA 5 data 

This would actually make sense.  I've been watching winter storms forecast across the mid-south turn into Kentucky maulers (80% of the time) for over 50 years now.  

[Wurbus]

6z GEFS was a beauty. Most of northern East TN in the 6+ mean. Some huge dogs showing 20"+ skewing it some though. 

[Carvers Gap]

So, I have been digging through lots of verification for the AIFS.  The one thing I notice over and over is that the EPS scores equal or better than the AIFS at days 4-5.  At days 4-5 all deterministic models and AI models are fallible...sometimes very fallible in regards to dew point, windspeed, and temp-C.  I Most papers that I have read this morning strongly recommend using a blend of models - EPS, Euro IFS, and AIFS.   I have provided a link to one paper below.  If you favor ensembles, NWP, or AI...there is probably something in that article for you.

This will be a good test, but by no means is it the final test.  Modeling is only as good as its programming and quality/quantity of data input.  All modeling is much better inside of d4.  If I was going to rank models...it would still be pretty much the following at d5....EPS, Euro AIFS EPS, Euro or AIFS.  I would feel much better about the AIFS if it wasn't in the same camp as the UKMET which has low verification scores at this range and also with the AIGFS.  But again, this will be an interesting comparison for sure!  

Right now, the AIFS is on the north side of guidance w/ the GFS suspiciously on the south side.  My guess would be the solution is somewhere in between.  However, I am reminded that most folks were skeptical that a chess bot(not AI) could defeat a grand master.  The chess bot is now better.  But in this case...pretty much all numerical modeling is a chess bot.  We are just trying to see which chess bot is better.

Here is the article I found out of Australia in regards to comparing AIFS vs IFS from AMETSOC.

https://journals.ametsoc.org/view/journals/aies/4/4/AIES-D-25-0037.1.pdf

Interesting quote...

However, as with NWP models, the autoregressive nature of AIFS means that small errors are amplified over the forecast period, which causes AIFS (and HRES) to fall behind ENS in accuracy as the lead time increases.

It is well established that blending models generally im- proves the accuracy relative to the best individual forecast [see e.g., Vannitsem et al. (2021)], and therefore, it is a common approach in operational weather forecasts. 

 

 

 

[tnweathernut]

2 minutes ago, Carvers Gap said:

So, I have been digging through lots of verification for the AIFS.  The one thing I notice over and over is that the EPS scores equal or better than the AIFS at days 4-5.  At days 4-5 all deterministic models and AI models are fallible...sometimes very fallible in regards to dew point, windspeed, and temp-C.  I Most papers that I have read this morning strongly recommend using a blend of models - EPS, Euro IFS, and AIFS.   I have provided a link to one paper below.  If you favor ensembles, NWP, or AI...there is probably something in that article for you.

This will be a good test, but by no means is it the final test.  Modeling is only as good as its programming and quality/quantity of data input.  All modeling is much better inside of d4.  If I was going to rank models...it would still be pretty much the same at d5....EPS, Euro AIFS EPS, Euro or AIFS.  I would feel much better about the AIFS if it wasn't in the same camp as the UKMET which has low verification scores at this range and also with the AIGFS.  But again, this will be an interesting comparison for sure!  

Right now, the AIFS is on the north side of guidance w/ the GFS suspiciously on the south side.  My guess would be the solution is somewhere in between.  However, I am reminded that most folks were skeptical that a chess bot(not AI) could defeat a grand master.  The chess bot is now better.  But in this case...pretty much all numerical modeling is a chess bot.  We are just trying to see which chess bot is better.

Here is the article I found out of Australia in regards to comparing AIFS vs IFS from AMETSOC.

https://journals.ametsoc.org/view/journals/aies/4/4/AIES-D-25-0037.1.pdf

Interesting quote...

However, as with NWP models, the autoregressive nature of AIFS means that small errors are amplified over the forecast period, which causes AIFS (and HRES) to fall behind ENS in accuracy as the lead time increases.

It is well established that blending models generally im- proves the accuracy relative to the best individual forecast [see e.g., Vannitsem et al. (2021)], and therefore, it is a com- mon approach in operational weather forecasts. 

 

 

 

Good find.  Thanks for your research.

[Carvers Gap]

The fun question is weather the AIFS will outperform its own ensemble at this range.  The ensemble is quite different for our forum area.  One thing that many models have missed over time is that when we have a hp funneling cold down west of the Apps, many models will often miss that the actual surface is below freezing.  Depending on the sequence that this arrives, the scenario exists that cold air could get trapped in the eastern valley.  That is why the model difference below is important.  The AIFS has changed quite a bit since 18z w/ a more amped cutter while the AIFS EPS has remained fairly steady.  The ensemble mean (rightly or wrongly) smooths out that amped look.  If the mean is correct, then the heat transport northward(moisture) will be less and cold air at the surface could be trapped.  If the AIFS is correct....liquid drops for almost everyone but NW TN.

[Weatheriscool]

4 minutes ago, Carvers Gap said:

The fun question is weather the AIFS will outperform its ensemble at this range.  The ensemble is quite different for our forum area.  One thing that many models have missed over time is that when we have a hp funneling cold down west of the Apps, many models will often miss that the actual surface is below freezing.  Depending on the sequence that this arrives, the scenario exists that cold air could get trapped in the eastern valley.  That is why the model difference below is important.  The AIFS has changed quite a bit since 18z w/ a more amped cutter while the AIFS EPS has remained fairly steady.  The ensemble mean (rightly or wrongly) smooths out that amped look.  If the mean is correct, then the heat transport northward(moisture) will be less and cold air at the surface could be trapped.  If the AIFS is correct....liquid drops for almost everyone but NW TN.

That is a huge difference in temps for places such as central and northern Alabama, almost 30 plus difference

[GBOVolz]

So I asked Mike Witcher about the AI mods. He said the euro AI will initiate the same as the euro OG but the AI mod leans heavily on climatology and that’s why we are seeing a cutter from it. He said as of now he feels like this is a snow event for East Tennessee, except for the southern Valley, which might get warm nose a little.


.

[Carvers Gap]

I mean when looking at the AI scenarios above...either looks plausible, right?  We have seen both scenarios play out.  I think we need to really watch trends w/ hp to our north and how the Baja slp evolution plays out.  I do think we could see a truncation as the energy involved gets better sampled in the northern stream.  No matter how good programming is...data input quality is just as important at the programming in the model.

[GBOVolz]

I mean when looking at the AI scenarios above...either looks plausible, right?  We have seen both scenarios play out.  I think we need to really watch trends w/ hp to our north and how the Baja slp evolution plays out.  I do think we could see a truncation as the energy involved gets better sampled in the northern stream.  No matter how good programming is...data input quality is just as important at the programming in the model.

I feel pretty confident that as long as we don’t see a cutter,IMO, we should not have temperature issues.


.

[tnweathernut]

3 minutes ago, GBOVolz said:

I feel pretty confident that as long as we don’t see a cutter,IMO, we should not have temperature issues.


.

I certainly hope Mike is onto something, but i'm never confident 4+ days out.  I guess that's just me and 40+ years of following these things.

[Carvers Gap]

10 minutes ago, GBOVolz said:

I feel pretty confident that as long as we don’t see a cutter,IMO, we should not have temperature issues.


.

I just don't know at this point.  

[Carvers Gap]

5 minutes ago, tnweathernut said:

I certainly hope Mike is onto something, but i'm never confident 4+ days out.  I guess that's just me and 40+ years of following these things.

Maybe one of the worst head fakes(here) was snowmageddon.  All those pretty 30" clown maps moved to DC.  Well, the worst headfake was CLT area forecast to get 3-4' right up to and during the storm....not even sure the grass got covered.  

So yes, I want the synoptics right before jumping on board fully.  I don't like that the Baja low has a different iteration w/ each run.  I don't know which camp is gonna be right about the high which will likely roll into the northern Plains.  Is it 1040-1042 or is it 1050+?  Definitely important things still to sort out...

[GaWx]

2 hours ago, Carvers Gap said:

The 6z AIFS Euro EPS is almost identical to the 6z Euro…looks like they may have fixed the algorithm for the ensemble at wxbell.  

Hey Carvers and others,

 WxBell definitely didn’t fix the Euro AI ensemble snow maps algos.

 These two members (#10 and #22) with multiple inches of snow way out in the Gulf are from the hour 84 6Z Euro AIFS, when temps are near normal (70s):

[tnweathernut]

The most important feature I'm watching is the high pressure that enters the US around hour 100-110.  

[Carvers Gap]

7 minutes ago, GaWx said:

Hey Carvers and others,

 WxBell definitely didn’t fix the Euro AI ensemble snow maps algos.

 These two members (#10 and #22) with multiple inches of snow way out in the Gulf are from the hour 84 6Z Euro AIFS, when temps are near normal (70s):

Precip and temp maps above are good though, right?

[Carvers Gap]

At 93, the GFS does not want to budge w/ weaning that big hp. 1051 sitting in North Dakota.  Might come north a hair, but that maybe will yield yet another suppressed solution?

[Holston_River_Rambler]

1 hour ago, GBOVolz said:

I haven’t read much about the AI mods… what calibrations did they make to the euro AI for it to show a different scenario from the original euro? Do we know?


.

Not sure. My (probably wrong) running hypothesis is that it reviews thousands of scenarios at initialization and then runs with a "given a certain set up "X," "Y" will then happen 90% of the time."

Lower resolution helps this approach. 

[Holston_River_Rambler]

1 hour ago, tnweathernut said:

This would actually make sense.  I've been watching winter storms forecast across the mid-south turn into Kentucky maulers (80% of the time) for over 50 years now.  

That's where I'm at as far as hope goes now. The 6z AIFS is a much more common winter storm scenario than something like the 1988 Jan southern slider. 

[Carvers Gap]

Well, the 12z GFS just doubled down on that big high.  That might end up in Cuba.

[Holston_River_Rambler]

46 minutes ago, GBOVolz said:

So I asked Mike Witcher about the AI mods. He said the euro AI will initiate the same as the euro OG but the AI mod leans heavily on climatology and that’s why we are seeing a cutter from it. He said as of now he feels like this is a snow event for East Tennessee, except for the southern Valley, which might get warm nose a little.


.

Thank you for asking him about that. 

[Carvers Gap]

Very suppressed solution incoming by the 12z GFS at 111...stepped away from the 6z AIFS.  No idea if right, but that would be a massive shift to get that solution to cut.  Heck, I just want it to come back north on that run.  Haha.

[Carvers Gap]

We need the 12z GFS to get a lil' bit of AIFS in its veins. 

[GaWx]

9 minutes ago, Carvers Gap said:

Precip and temp maps above are good though, right?

Great question. I haven’t investigated Euro AI ensemble precip/temp maps closely, but a “quick and dirty” check doesn’t reveal to me any obvious algo issues with those. So, my focus on these posts remains fully with the snow maps.

[Carvers Gap]

At 120, the 12z GFS is south of its 6z run.    Honestly, that is probably about where we want that model at this range.  As a good friend noted, those over-running events tend to trend north.  I agree.

[Vol4Life]

Just now, Carvers Gap said:

At 120, the 12z GFS is south of its 6z run.    Honestly, that is probably about where we want that model at this range.  As a good friend noted, those over-running events tend to trend north.  I agree.

The timing of the onset precipitation seems a little later too.  Of course, that is likely due to suppression.