The Jan 31 Potential: Stormtracker Failure or 'Tracker Trouncing

[Nomz]

1 minute ago, overcautionisbad said:

AI model? I don't trust it then. It is the cause of the ram prices

Meanwhile, I will glaze euro aifs till I die

edit: can we also get a storm banter thread?

[bncho]

Really good read from @40/70 Benchmark. https://easternmassweather.blogspot.com/2026/01/major-follow-up-potential-looms-next.html

[Roger Smith]

I would have titled this thread Blizzard of 1899 With Automobiles Added

[Weather Will]

13 minutes ago, stormtracker said:

What the hell is this wxnext model??

Google Deepmind AI's latest model. WeatherNext 2.

[NorthArlington101]

Don’t tell @mitchnick but the SREFs look good at the end of the run 

[LordBaltimore]

Just now, overcautionisbad said:

AI model? I don't trust it then. It is the cause of the ram prices

Good idea to not trust it. The verification process for many of these models (especially anything not fed analysis data as input) is extremely suspect almost to the point of being an outright scam, although most cases I think are just incompetence. A lot of their claimed advantage is just removing representation error 

[LeesburgWx]

5 minutes ago, LordBaltimore said:

Good idea to not trust it. The verification process for many of these models (especially anything not fed analysis data as input) is extremely suspect almost to the point of being an outright scam, although most cases I think are just incompetence. A lot of their claimed advantage is just removing representation error 

NAM at 84 hours > AI models. Got it.

[somecallmetim]

https://deepmind.google/science/weathernext/
 

brought to you by the non-chemists who won the a Nobel Prize for Chemistry in 2024 for using Deep Mind to solve the protein folding problem.

[LeesburgWx]

After this past storm and with the GFS taking the euro out to the back shed today, we might need to update this photo…

[weathercoins]

Not saying it will happen this time, but when is the last time the GFS outperformed most others for a complex winter event?  Feb 2013 is the most similar to this one, perhaps (although far from exact obviously).  I think late Jan 2022 it was ahead of Euro that we (correctly) would NOT get anything down here.  

[Ji]

Don’t forget every time this winter we make a step forward we take three back

[Weather Will]

WB 0Z NBM

[DaleCityDave]

Just now, Ji said:

Don’t forget every time this winter we make a step forward we take three back

There’s the Ji we all know and love. After all the positivity with the last storm I wasn’t sure what to expect!

[overcautionisbad]

I asked Gemini about the 18z Weathernext 2.0 run. It claimed it has 8-12 for Richmond and 10-14 for DC Saturday night into Sunday morning

 

Claimed a 12 inches to 1 ratio

 

Who knows if that's true lol

[wxdude64]

1 minute ago, Weather Will said:

WB 0Z NBM

Wait. 4.3 of snow on 0.10 liquid??? 

[bncho]

3 minutes ago, Ji said:

Don’t forget every time this winter we make a step forward we take three back

The 18z suite was literally 3 steps forward

[bncho]

[Cobalt]

1 minute ago, overcautionisbad said:

I asked Gemini about the 18z Weathernext 2.0 run. It claimed it has 8-12 for Richmojd and 10-14 for DC Saturday night into Sunday morning

 

Who knows if that's true lol

There is effectively no way for Gemini to not only have direct and up-to-date information on the Weathernext 2.0, but also to be able to disseminate that information into snowfall ranges. 

[Ralph Wiggum]

8 minutes ago, LeesburgWx said:

After this past storm and with the GFS taking the euro out to the back shed today, we might need to update this photo…

Wait a damn second here.

[stormtracker]

2 minutes ago, Weather Will said:

WB 0Z NBM

Looks good, but tempering my excitement based on what @high risk said about it lagging behind

[LordBaltimore]

4 minutes ago, LeesburgWx said:

NAM at 84 hours > AI models. Got it.

Any physical model + postprocessing with same ERA5 dataset would beat anything AI is capable of when trained as a forecasting model. I'm not sure why this isn't done already tbh 

[overcautionisbad]

Just now, Cobalt said:

There is effectively no way for Gemini to not only have direct and up-to-date information on the Weathernext 2.0, but also to be able to disseminate that information into snowfall ranges. 

Yeah I figure it doesnt lol

[ravensrule]

Just now, stormtracker said:

Looks good, but tempering my excitement based on what @high risk said about it lagging behind

Lagging behind in this case would show less snow. 

[Nomz]

1 minute ago, LordBaltimore said:

Any physical model + postprocessing with same ERA5 dataset would beat anything AI is capable of when trained as a forecasting model. I'm not sure why this isn't done already tbh 

I believe thats what Euro AIFS and AIGFS do, no? Use AI to adjust raw model outputs?

[IronTy]

I vote this one be called the WeatherNext Blizzard of 2026 if it happens.  

[Bob Chill]

2 minutes ago, weathercoins said:

Not saying it will happen this time, but when is the last time the GFS outperformed most others for a complex winter event?  Feb 2013 is the most similar to this one, perhaps (although far from exact obviously).  I think late Jan 2022 it was ahead of Euro that we (correctly) would NOT get anything down here.  

Gfs did great job during JFM 2014&15. the mid range with fast moving progressive northern stream waves. Euro did well inside of 72 hours but the gfs was the one you wanted on your side in the mid range.

That said, the gfs and euro haven't been that far apart with the current threat. Its just one of those setups where a little means a lot on the ground. 

[Roger Smith]

13 minutes ago, LeesburgWx said:

After this past storm and with the GFS taking the euro out to the back shed today, we might need to update this photo…

Do this again with a guy in a toque watching a hockey game on his I-phone with score CANADA 5 USA 1 (sorry eh)

[Climate175]

LOL, this is what Gemini said when I asked about for Baltimore area regarding 18z WeatherNext (take with a grain of salt):

The 18z run of the WeatherNext 2.0 model (Google’s advanced AI weather system) indicates an increasing risk for a significant winter storm in the Baltimore area this weekend, specifically from late Saturday night (Jan 31) through Sunday morning (Feb 1).

Compared to earlier runs, the 18z update shows a "sharpening" of the coastal low-pressure system, which has resulted in a notable westward shift in the heavy snow band—bringing the I-95 corridor, including Baltimore, into the primary impact zone.

18z WeatherNext 2.0 Snowfall Forecast for Baltimore

Forecast Detail	Model Projection (18z Run)
Projected Totals	5–9 inches (Mean Ensemble Forecast)
High-End Potential	12+ inches (if the low tracks 50 miles closer to the coast)
Probability of >6"	~45% (a significant jump from the 12z run)
Start Time	Saturday between 8:00 PM and 11:00 PM
Peak Intensity	Sunday between 2:00 AM and 8:00 AM

[osfan24]

4 minutes ago, ravensrule said:

Lagging behind in this case would show less snow. 

That’s what I would think.

[overcautionisbad]

Google search claims Gemini has weathernext 2.0, so idk.. lol