First Legit Storm Potential of the Season Upon Us

[weatherwiz]

1 minute ago, eduggs said:

I'm not sure what you mean. I'm interested - fascinated really - why the AI models have been consistently west of their parent models for the past few days. I have not seen that behavior consistently or persistently earlier this season. And I offered one potential explanation for this specific situation. I reflexively tend to think the least snowy model will be correct, even if that's not scientifically sound.

Just busting. 

I'm with you, it is very fascinating to see why the two camps are drastically different. My opinion on AI model aside (since ultimately my opinion means nothing), this is a tremendous opportunity within the field of forecast modeling and obviously the only way to ever see the true value in AI and how helpful will be is by putting it to the test. If the AI score a coop here that would not only be a tremendous win in the AI department but it probably also further exposes significant weaknesses in the traditional modelling. 

On the other hand, if the AIs fail here, then at least to me, signifies the importance of physics and complex equations and that there is much more to the evolution of weather than just on how a 'similar match' evolved historically. 

[Typhoon Tip]

wait, what did i do ? 

wait first of all, we don't know precisely how ... and by defacto 'what' these AI (apparent marketing gimmicks )  are doing.  so how can we be sure about 20 or 30 years ago anyway? 

that's I pointed this out yesterday..  there's been no prospecti made easy to find - if at all - that answers the questions that everybody should be asking but no one is!  jesus... degradation of virtuosity and method on both side.   whereby any kind of advantages and disadvantages, circumstantially; basic modeling 101 stuff that has to be considered.   confidence intervals... methodologies.  nothing.   we can't say jack shit about them. 

I'm hugely displeased at deployment and anyone that uses them .... man, caveat emptor

[ariof]

11 minutes ago, weatherwiz said:

Another theory is always trying to find a way to justify why the snowiest models are correct

I believe this is called the Theory of DGEX

[ORH_wxman]

20 minutes ago, eduggs said:

A theory to explain the AI vs GFS/ECMWF battle could be the latent heat release - wave development feedback loop that Eric Webb mentioned a day or two ago. Shortwaves that tap Gulf moisture are susceptible to this effect. If this is biasing the global models too flat aloft, it could explain why the AIs are further west since they can bias-correct based on historical outcomes.

This is a pretty good theory. I wonder if the bias-correction also showed up (mostly successfully imho) in the “cutter” last weekend. A lot of OP models several days out kept showing big warm sectoring into New England but the AI models kept saying no dice and it would be much colder with wedging at the sfc. Much like what happens very frequently in the past where models get too warm-sector happy east of the Apps and north of about 41N. They turned out more correct. I think they were a touch too cold but closer to reality than the original OP runs…however, the biggest difference this time around is the OPs are still pretty far apart from AI. By the time we got inside 48h last weekend, the OP and AI guidance were mostly converged. 

[HoarfrostHubb]

2 minutes ago, ariof said:

I believe this is called the Theory of DGEX

"Raise a glass"

[wxsniss]

1 hour ago, ORH_wxman said:

I just am morbidly curious how long they will keep doubling down. 

Same.

In case people missed it and are interested... below is a thread for general discussion on AI guidance beyond the Jan 18 storm:

https://www.americanwx.com/bb/topic/62517-skynet-rise-of-the-forecast-machines-general-discussions-on-ai-guidance/

While I "do not have a lot of experience utilizing a lot of the AI guidance" (quoting Box AFD from this morning), some things I highlight in other thread relevant to Jan 18:

• AI models tend to have broad QPF swaths that are overdone in areal coverage and not supported by physics

• major weakness that is on display here: we have no idea why AI models do what they do...

Why is AIGFS so robust here? Is it seeing something that physics models are not? Are the legacy models too sensitive to smaller features (e.g. the other day I suggested feedback between vorticity and convection sweeping off southeast coast hampering cyclogenesis), or is AI guidance not sensitive enough?

If AI guidance changes, why did it change? Will that change be abrupt?

These models are totally nebulous.

Jan 18 is shaping up to be a great showdown.

10 hours ago, wxsniss said:

Most recently exemplified by this imminent showdown for Jan 18 where AI-GFS has been consistently and substantially more impactful than legacy GFS…

…which is correct TBD... and because lots of people are interested in this topic, I thought I’d start a thread to encompass general debates, updates, and verification tally of AI vs. legacy non-AI / physics-based forecasting guidance.

 

[TauntonBlizzard2013]

I’ve found the AI models to be way too wet this winter. Consistently overdoing QPF

[weatherwiz]

5 minutes ago, ORH_wxman said:

This is a pretty good theory. I wonder if the bias-correction also showed up (mostly successfully imho) in the “cutter” last weekend. A lot of OP models several days out kept showing big warm sectoring into New England but the AI models kept saying no dice and it would be much colder with wedging at the sfc. Much like what happens very frequently in the past where models get too warm-sector happy east of the Apps and north of about 41N. They turned out more correct. I think they were a touch too cold but closer to reality than the original OP runs…however, the biggest difference this time around is the OPs are still pretty far apart from AI. By the time we got inside 48h last weekend, the OP and AI guidance were mostly converged. 

I just find it really difficult to believe the traditional guidance would totally miss this...I mean this would be up there with one of the biggest busts in quite some time. If any of the big dawgs were showing a big hit I would perhaps feel a bit differently but what's really the likelihood of guidance being this wrong? Like lets say we didn't have AI guidance yet...how many of us would actually think there is a legit shot? 

I suppose though its still just far enough away to we could see big movements but we would have to see that tonight.

[ORH_wxman]

Just now, weatherwiz said:

I just find it really difficult to believe the traditional guidance would totally miss this...I mean this would be up there with one of the biggest busts in quite some time. If any of the big dawgs were showing a big hit I would perhaps feel a bit differently but what's really the likelihood of guidance being this wrong? Like lets say we didn't have AI guidance yet...how many of us would actually think there is a legit shot? 

I suppose though its still just far enough away to we could see big movements but we would have to see that tonight.

We’ve had some hard west trends on scrapers before in the final 48 hours (2/5/16 is one that really stands out…1/7/17 as well)….that wouldn’t be terribly unique. But it hasn’t happened in a while. 

 

[dendrite]

9 minutes ago, wxsniss said:

Same.

In case people missed it and are interested... below is a thread for general discussion on AI guidance beyond the Jan 18 storm:

https://www.americanwx.com/bb/topic/62517-skynet-rise-of-the-forecast-machines-general-discussions-on-ai-guidance/

While I "do not have a lot of experience utilizing a lot of the AI guidance" (quoting Box AFD from this morning), some things I highlight in other thread relevant to Jan 18:

• AI models tend to have broad QPF swaths that are overdone in areal coverage and not supported by physics

• major weakness that is on display here: we have no idea why AI models do what they do...

Why is AIGFS so robust here? Is it seeing something that physics models are not? Are the legacy models too sensitive to smaller features (e.g. the other day I suggested feedback between vorticity and convection sweeping off southeast coast hampering cyclogenesis), or is AI guidance not sensitive enough?

If AI guidance changes, why did it change? Will that change be abrupt?

These models are totally nebulous.

Jan 18 is shaping up to be a great showdown.

 

I’d rather compare the upper levels rather than QPF. It seems like they throw QPF further back than you would expect from the shortwave, but sometimes as you get close the shortwaves trend more amplified. So are the AIs sometimes right for the wrong reasons?

[weatherwiz]

Just now, ORH_wxman said:

We’ve had some hard west trends on scrapers before in the final 48 hours (2/5/16 is one that really stands out…1/7/17 as well)….that wouldn’t be terribly unique. But it hasn’t happened in a while. 

 

The biggest bust I can think of (in the opposite direction) was February 2018? (don't remember the date). It looked like a good chunk of SNE was inline for like 12-18" but the confluence to the north screwed us and Long Island got smoked. Southern CT barely ended up with more than a few inches.  

[EastonSN+]

6 minutes ago, weatherwiz said:

The biggest bust I can think of (in the opposite direction) was February 2018? (don't remember the date). It looked like a good chunk of SNE was inline for like 12-18" but the confluence to the north screwed us and Long Island got smoked. Southern CT barely ended up with more than a few inches.  

That was March of 2018.

[weatherwiz]

Just now, EastonSN+ said:

That was March of 2018.

Ahh yes...thought it was later in the season. That was a brutal, brutal bust. 

[SouthCoastMA]

Remember the storm modeled on or around 3/17/17 that was showing 12"+ for a good part of New England, then completely evaporated? I think it was withing 48-72..I know not the same idea but that was a pretty drastic change. 

[CoastalWx]

6 minutes ago, SouthCoastMA said:

Remember the storm modeled on or around 3/17/17 that was showing 12"+ for a good part of New England, then completely evaporated? I think it was withing 48-72..I know not the same idea but that was a pretty drastic change. 

Yeah every model kept coming west. Gfs of course back then was lagging. 

[weatherwiz]

Quite the difference in that nrn stream on the 18z nam at 51hr compared to the 12z run lol 

[TauntonBlizzard2013]

NAM is definitely west through 48.

 

**** I don’t know how much of an effect that will have on the solution up here 

[TauntonBlizzard2013]

NAM still whiffs, but it’s a closer whiff lol 

[weatherwiz]

Looks like its still chasing convection 

[SouthCoastMA]

1 minute ago, TauntonBlizzard2013 said:

NAM still whiffs, but it’s a closer whiff lol 

a closer version of ass. 

getting the  ready for the 18z AI vs OP battle

[CoastalWx]

Nam gets some light rain and snow into SE areas 

[weathafella]

NAM is well west of 12z which was pushing the low to Iceland.

[eduggs]

46 minutes ago, weatherwiz said:

I just find it really difficult to believe the traditional guidance would totally miss this...I mean this would be up there with one of the biggest busts in quite some time. If any of the big dawgs were showing a big hit I would perhaps feel a bit differently but what's really the likelihood of guidance being this wrong? Like lets say we didn't have AI guidance yet...how many of us would actually think there is a legit shot? 

I suppose though its still just far enough away to we could see big movements but we would have to see that tonight.

They (GFS/ECMWF) aren't that far apart at this point, are they? What, maybe 50-75 miles with the heavy stuff offshore? The ECMWF OP in particular has shifted NW with the heavier stuff.

As many people have mentioned, these are the kinds of setups that can and sometimes do shift NW in the last 48 hours. The outcome is extremely sensitive to minor changes in the shortwaves near the Gulf and also the Lakes. Presumably the AI models are correcting for those cases somehow. But that sensitivity is a double edged sword. I can also see how the AI models might overcorrect if their training datasets aren't perfectly representative or well matched to the current setup. 

[weathafella]

It actually seems to invert as qpf spreads throughout 75% of sne.

[weatherwiz]

The NAM is pretty garbage but again not really a model to use for when dealing with phasing so all we can hope for is the GFS how some improvements 

[weatherwiz]

1 minute ago, eduggs said:

They (GFS/ECMWF) aren't that far apart at this point, are they? What, maybe 50-75 miles with the heavy stuff offshore? The ECMWF OP in particular has shifted NW with the heavier stuff.

As many people have mentioned, these are the kinds of setups that can and sometimes do shift NW in the last 48 hours. The outcome is extremely sensitive to minor changes in the shortwaves near the Gulf and also the Lakes. Presumably the AI models are correcting for those cases somehow. But that sensitivity is a double edged sword. I can also see how the AI models might overcorrect if their training datasets aren't perfectly representative or well matched to the current setup. 

No they really aren't. I mean when dealing with phasing and relying solely on phasing, minuscule changes with how the energies evolve and interact can have significant differences on surface evolution. Traditional guidance is still struggling to get a good handle on the northern energy and even with how the southern energy evolves. It wouldn't take much but I presume we really need to start seeing a consensus towards positive ASAP 

[eduggs]

Just now, weatherwiz said:

No they really aren't. I mean when dealing with phasing and relying solely on phasing, minuscule changes with how the energies evolve and interact can have significant differences on surface evolution. Traditional guidance is still struggling to get a good handle on the northern energy and even with how the southern energy evolves. It wouldn't take much but I presume we really need to start seeing a consensus towards positive ASAP 

I meant the GFS & ECMWF vs. the AI models. Considering the size of the model domain, the camps aren't that far apart. The difference just feels really big considering the local sensible weather outcome. I feel like there have been many 75 mile shifts in precip. shields over the past 20 years with coastal storms within 48 hours.

[ORH_wxman]

Ray will be excited that he gets IVT snow on the 18z NAM 

[WinterWolf]

Just now, ORH_wxman said:

Ray will be excited that he gets IVT snow on the 18z NAM 

Could that IVT be a precursor to a jump west next cycle by the NAM? Almost like not knowing where to put the low..so that’s the intermediate solution?

[Go Kart Mozart]

12 minutes ago, eduggs said:

I meant the GFS & ECMWF vs. the AI models. Considering the size of the model domain, the camps aren't that far apart. The difference just feels really big considering the local sensible weather outcome. I feel like there have been many 75 mile shifts in precip. shields over the past 20 years with coastal storms within 48 hours.

Remember folks, verifications show that AI has its greatest value beyond 4 days.  I suspect the physics-based models will win and win big here.