2026 Mid-Atlantic Severe Storm General Discussion

[mappy]

9 hours ago, 32º said:

I noticed the rotation also and did save a screen shot of storm relative velocity. 

Yes thank you! It had that rotation the whole time it cross York County with no warning on it until it was east of Stewartstown 

[WxUSAF]

Guidance is hot garbage for Thursday. 

[dailylurker]

39 minutes ago, WxUSAF said:

Guidance is hot garbage for Thursday. 

Just got done reading all the hype on Facebook lol. I just want rain. I don't care about severe. Hopefully that happens. 

[WxUSAF]

16 minutes ago, dailylurker said:

Just got done reading all the hype on Facebook lol. I just want rain. I don't care about severe. Hopefully that happens. 

Guidance is hot dry garbage 

[dailylurker]

39 minutes ago, WxUSAF said:

Guidance is hot dry garbage 

So just boring ass drought? Looks like it all misses to the north where they have a rain surplus lol

The 10 day gfs precip map really tells the story. Just incredibly boring. 

[high risk]

40 minutes ago, WxUSAF said:

Guidance is hot dry garbage 

Deep westerly flow here is never a good environment for widespread convection.  That said, the dew points aren’t progged to lower Thursday afternoon as much as I would have expected, and the RRFS is able to generate a few scattered storms.  And any storm in this kinematic environment would be interesting.  Still, the best chance of some rainfall might be early Friday as a weak wave moves along the front.  

[Kmlwx]

CSU MLP is pretty garden variety (15% in a big, broad area). CIPS is pretty unenthused. NCAR is still heavily bullish. 

I think most of us know this at this point is not going to be an outbreak in our region. But let's see how things evolve with Wednesday convection and timing and such. 

[Eskimo Joe]

40 minutes ago, Kmlwx said:

CSU MLP is pretty garden variety (15% in a big, broad area). CIPS is pretty unenthused. NCAR is still heavily bullish. 

I think most of us know this at this point is not going to be an outbreak in our region. But let's see how things evolve with Wednesday convection and timing and such. 

This is just my observation, but the AI stuff (NCAR) seems to over do convection is marginal environments. It's almost like there aren't enough events in it's database.

[Kmlwx]

34 minutes ago, Eskimo Joe said:

This is just my observation, but the AI stuff (NCAR) seems to over do convection is marginal environments. It's almost like there aren't enough events in it's database.

I've also noticed the "FengWu" ones seem to be more bullish than the other one. 

Maybe it's just not well calibrated or something?

[yoda]

FWIW, MRGL added on 1300z OTLK from SPC for just east of i95 corridor and the Delmarva for tonight... 0/5/5

.Delmarva/coastal Mid-Atlantic...

   Some strong/locally severe thunderstorms could occur tonight as
   convection develop near/north of a near-coastal/offshore surface low
   and northward-advancing warm front. Storms will tend to be elevated
   inland, but modest buoyancy and strong unidirectional shear through
   the cloud-bearing layer could yield some hail and/or gusty winds.

[yoda]

Updated morning AFD from LWX

KEY MESSAGE 2...A warm front lifting through the region brings
showers and thunderstorms overnight tonight.

High pressure will keep cool and dry conditions in the region
this morning. High temperatures this afternoon rise into the
upper 70s to low 80s for most with highest elevations in the
Alleghenies staying in the 60s. A warm front lifting through the
region combined with shortwave energy aloft, bring an increased
risk of strong to potentially severe thunderstorms in southern
MD. To accommodate this risk, SPC has much of the Chesapeake
Bay and portions of Southern MD in a marginal risk for severe
weather overnight. Any convection will be elevated in nature,
but adequate shear and lift may lead to gusty winds and hail
within stronger storms.

 

[JenkinsJinkies]

If this were winter Thursday would be getting a “time of death” post right about now.

[Kmlwx]

7 minutes ago, JenkinsJinkies said:

If this were winter Thursday would be getting a “time of death” post right about now.

Ehhh - with severe I'll keep watching until morning of. Always surprises since storms are smaller scale than snowstorms generally. 

[Eskimo Joe]

8 minutes ago, Kmlwx said:

I've also noticed the "FengWu" ones seem to be more bullish than the other one. 

Maybe it's just not well calibrated or something?

I've been in a few NOAA/NHC/FEMA webinars recently where the speakers have made observations where the AI guidance appears to not overhype, but rather really juice up convection in the warm sector. It's entirely possible these AI models need more events in their knowledge base before they can beat the traditional NWP. Just my observation. Others are welcome to chime in.

[Eskimo Joe]

1 hour ago, JenkinsJinkies said:

If this were winter Thursday would be getting a “time of death” post right about now.

Eh, this is a slight risk day. It's not like we were staring down a D3 MOD risk with a CIG2 hatch 60% or something that fizzled.

[Kmlwx]

12z CIPS guidance is a bit more enthusiastic. One analog that has been persistently showing up in some of the domains on CIPS is 6/21/21

[high risk]

18z NAM Nest is the first really encouraging run for Thursday.  It also seems to justify the outlook for tonight. 

[vortex95]

On 6/16/2026 at 8:47 AM, Eskimo Joe said:

This is just my observation, but the AI stuff (NCAR) seems to over do convection is marginal environments. It's almost like there aren't enough events in it's database.

That's the issue w/ AI models.  They can't handle outlier events or set ups well, so they underdo things.  Also, AI overdoing it is possible as you noted above.  Not a one-sided problem I bet (non-linear factors).

I keep hearing, "AI will get better."  Well, isn't that the case for all types of modelling w/ time?  And the notion that AI is improving so fast acting like it will solve all problems?  All the AI advancement in the world is meaningless if you don't have many outlier events/analogs for it to pull from!  And since outlier events are rare, progress in AI forecasts should be very slow.  I would argue physics-based models will advance faster as in situ/remote observations get better globally for improved model initialization, along w/ better direct simulation of the atmosphere.

[Eskimo Joe]

8 hours ago, vortex95 said:

That's the issue w/ AI models.  They can't handle outlier events or set ups well, so they underdo things.  Also, AI overdoing it is possible as you noted above.  Not a one-side problem I bet (non-linear factors).

I keep hearing, "AI will get better."  Well, isn't that the case for all types of modelling w/ time?  And the notion that AI is improving so fast acting like it will solve all problems?  All the AI advancement in the world is meaningless if you don't have many outlier events/analogs for it to pull from!  And since outlier events are rare, progress in AI forecasts should be very slow.  I would argue physics-based models will advance faster as in situ/remote observations get better globally for improved model initialization, along w/ better direct simulation of the atmosphere.

Let's be honest for a moment, a lot of managers and "senior leaders" are excited about AI because it means less staff.

Focusing on weather, the only advantage I've seen from the AI models so far is they come out faster. I haven't seen any meaningful performance in accuracy beyond D3 as a customer of the weather enterprise. That includes the private sector and government. I've been in a few private sector test groups with AI guidance for "perfect long range forecasting", and the verification scores beyond D7 are laughable. I don't need a 15 day forecast, I just need 95% confidence out 72 hours!

[WxWatcher007]

Will just say that AI has been really promising with regard to tropical forecasting. Exceptional actually, imo. 

[Kmlwx]

12 minutes ago, WxWatcher007 said:

Will just say that AI has been really promising with regard to tropical forecasting. Exceptional actually, imo. 

I wonder if that's a factor of the larger nature of tropical systems and them being more susceptible to the larger scale upper-level pattern(s). With severe, so much is dependent on much smaller scale factors that are below the "resolution" of a broad AI model that might just look at the overall pattern. 

[WxWatcher007]

10 minutes ago, Kmlwx said:

I wonder if that's a factor of the larger nature of tropical systems and them being more susceptible to the larger scale upper-level pattern(s). With severe, so much is dependent on much smaller scale factors that are below the "resolution" of a broad AI model that might just look at the overall pattern. 

I think so. From what I’ve seen AI has done very well with broader steering and genesis window forecasting at medium and longer range. Less so with intensity but Melissa was a win last season. 

[Eskimo Joe]

Don't look now, but the HRRR looks like it's trying for at least some storms tomorrow.

[Eskimo Joe]

7 hours ago, WxWatcher007 said:

Will just say that AI has been really promising with regard to tropical forecasting. Exceptional actually, imo. 

Agreed. The Euro and GFS AI models seem to be proficient at picking what systems will spin up and when. I'm very much excited about that.

[Stormchaserchuck1]

16 hours ago, Eskimo Joe said:

Let's be honest for a moment, a lot of managers and "senior leaders" are excited about AI because it means less staff.

Focusing on weather, the only advantage I've seen from the AI models so far is they come out faster. I haven't seen any meaningful performance in accuracy beyond D3 as a customer of the weather enterprise. That includes the private sector and government. I've been in a few private sector test groups with AI guidance for "perfect long range forecasting", and the verification scores beyond D7 are laughable. I don't need a 15 day forecast, I just need 95% confidence out 72 hours!

I notice the CPC doesn't even use AI models in their long range forecast, and I can verify that they don't have much skill at that range. EPS is probably still the way to go

Saying "AI is the way to go" just makes you look smart imo, not a huge leap from previous models

[vortex95]

1 hour ago, Stormchaserchuck1 said:

I notice the CPC doesn't even use AI models in their long range forecast, and I can verify that they don't have much skill at that range. EPS is probably still the way to go

Saying "AI is the way to go" just makes you look smart imo, not a huge leap from previous models

This happens *every* time new technology becomes available.  It is sold as the best thing since sliced bread w/ lofty claims.

I recall in the 80s it was said "in 10-15 year advancement in computer modelling will make human forecasting obsolete!'  Right, how did that work out?

In fairness, computer modelling advancement has removed a lot of manual work a human used to do, but this is a *good* thing actually.  Why?, b/c the shear volume of wx data now is enormous and keeps increasing. We need faster computers and AI to help sort and manage it all (do the grunt heavy lifting) so the human forecast can provide the insight/wisdom/understanding -- which AI does not have -- to give concise and clear messages to all.  That's one reason why IDSS is so big in the NWS.  Forecasts have become very good, but the real challenge is *communicating* it all effectively in this information overload and social media world. 

So I would not worry about AI "taking over," so to speak, anytime soon.
 

[vortex95]

13 hours ago, WxWatcher007 said:

I think so. From what I’ve seen AI has done very well with broader steering and genesis window forecasting at medium and longer range. Less so with intensity but Melissa was a win last season. 

We have a lot of climo for TC tracks globally, so AI has a lot to draw from, so it should do well here.  Intensity, not so much.  There are still so many things we can't model well for the stronger TCs.  The inner cores of the intense ones are basically mesoscale features. RI occurs often in such a short time frame and it is hard for physics-based models to handle.

For decades, we have seen a steady improvement in TC tracks, but very little for intensity until the 2010s, when significant improvement occurred.  So again, AI should do well for TC tracks.  And TC tracks often are latitude/longitude dependent, so this variable is a solid base indicator for AI to use.