Where traditional physics-based models might struggle, AI models can excel. WeatherMesh-5c spotted the NE forecast change earlier than other major model suites, becoming the first to abandon the Northeast warmup and correctly resolve the storm's big-picture evolution.

No long thread from me on this topic today. I’m going to be stepping back for a bit to focus on my personal well-being. This is just an update, not an ask for sympathy. I appreciate everyone’s understanding.

…and how exactly do you plan on focusing on weather research & operations for seasonal & severe weather prediction if you get rid of the supercomputers & aircraft that make said research possible www.nsf.gov/news/intent-...

Update on the status of PolarWx: The server hosting PolarWx has been at maximum capacity for a while now, on top of several issues with the server's memory usage that arose in the last few weeks. This has significantly degraded the performance of the website & map production.

Seeing this floating around, and want to say this directly: Tomorrow's fire conditions are NOT comparable with Marshall Fire (12/30/21). Wind speeds, longevity, soil moisture, RH: not the same. We should be vigilant tomorrow. But stop with the Marshall Fire comparisons. #COwx

Before you complain about how incompetent/evil/etc Xcel is for preemptively shutting down power, consider a few things: (1) People make these calls. Their families & friends are all affected by it. Schools. Hospitals. It’s not a decision to make lightly by any means.

Preliminary verification of the forecast map from last night vs. interpolated snow observations from LSRs so far:

Random probabilistic rant: 80% chance is not the same thing as 100% chance. Using a weather example, if 80% of members in a well calibrated ensemble show snow, even if consistent run to run, it doesn’t mean that snow is 100% guaranteed to occur, and in 20% of cases would lead to a false alarm.

The new AI-GFS outperformed the NWP (FV3 core) GFS for tonight’s storm — while the GFS (loop in next post) did not show much if any precipitation for areas now expecting to receive >4” until the very short range, the AI-GFS had a more amplified and temporally consistent signal.

Based on latest observations, and short-term adjustments on the synoptic scale, below is my final snow map for tonight into tomorrow morning. I suspect the max snow totals may be in northern MD & SE PA, where 6-10" of snow are likely with heavy snow rates.

I’ve got a suspicion there may be a snowy surprise in store for parts of PA/NJ/Long Island/southern New England tonight — overnight guidance came in less zonally oriented & more north with a tropopause vortex in Ontario, leading to a less suppressed height field downstream:

You may have seen a "GraphCast-GFS" model on some real-time websites - development on that model at NOAA has now led to the "AI-GFS" model which is scheduled to become operational on December 17th. Output from the AI-GFS model is now available on PolarWx: polarwx.com/models/?mode...

Here's something you don't see every day - a 987 hPa low over the Great Lakes with snow all the way down to eastern PA & northwest NJ. Two things make this possible: - Very deep antecedent cold airmass - lack of subtropical airmass origin

Liquid Glass is arguably one of — if not the — worst UI designs Apple has ever added to their iOS (at least in terms of features that can’t be turned off)

Follow-up analysis [🧵]: the same person who has already spread rampant disinformation about many topics from weather to politics has chimed in on this subject with a different take, so I thought I'd dig into the flaws in this analysis & overall caution in interpreting incidents like this:

It's easy to think of AI models as superior in the synoptic-scale in extended lead times vs. typical NWP models - and statistically speaking this is on average the case - but AI models are not perfect. Case in point -- AIFS completely missed an upcoming potential major cold outbreak:

I know this isn’t an easy topic for some, but if you put aside any religious/political beliefs and strictly approach it from an analytical POV, this is a clear cut example of news articles’ failure to reconcile critical inconsistencies & sufficiently address relevant context.

[thread 🧵] This analysis is strictly concerned with whether the incident described is antisemitic or not. Not everyone reads beyond a news article’s headline, so let’s start by comparing the Harvard Crimson (left) vs Boston Globe (right) — very different framing of the same story:

I haven’t read much about this so I don’t know if it was financial motives or something else that drove the university to cut the program, but sometimes a program’s impact can extend far beyond what its size or cost might suggest.

One of these (AI generated) things is not like the others

Slightly technical analysis on the discussion re: Tuesday's snowstorm 🧵 While it is technically correct to say that the progressive upper-level flow & lack of a high pressure to the north are unfavorable for an I-95 snowstorm, reality is often more nuanced & context-dependent.

AIFS Ensemble plots are now available on PolarWx for a limited number of map projections: polarwx.com/models/?mode...

Sneak peek of the next addition to PolarWx - AIFS ensembles including estimated precipitation types:

An easy way to identify an AI-generated map is unrealistic or missing county and state borders. Something that looks right at a quick glance until you take a closer look.

PolarWx PSA: As some of you alerted me recently, there has been a bug recently with the temperature & geopotential height anomaly plots on my site. This has now been fixed as of last night's 0z cycles. Speaking of which - a colder than average pattern is coming up for the NE US:

A story in 3 parts of how disinformation spreads… from the same person who falsely claims to be a hurricane expert. The ages old advice of “be careful of who you get your information from” holds true.

Random note no one asked for: the “country of origin / based in” feature Twitter added for accounts may sound good in theory, but can be easily misused between photoshop, browser HTML editing via the inspect tool, a parody account, etc — which is ripe for fake info to spread

Fact check: False — signed, an employee at a weather balloon company

While I know I don’t owe public explanations, I do want to apologize for the delay in bringing back the real-time stratosphere plots and full model map suites to my website. I resolved numerous issues with the hosting server recently, but it will still take some time.

Move over NAM, we just got HRRR'd

Well, this is a first. My Mac crashed because I tried pasting a json dictionary into the terminal

I may have just witnessed my first aurora from the light pollution of northeast NJ no less

Gap in the cloud cover is growing over north central/NE NJ. This might be our opportunity

My view of the aurora from cloud & light pollution filled NJ

The NYC metro got its first snowflakes of the 2025-26 winter this morning! Alongside our first cold outbreak of the season, residual lake effect snow showers in long fetch WNW flow reached the area, with a quick dusting in spots.

And another one:

Thanks to @franklinjamesl.bsky.social and @pppapin.bsky.social for help in building this product — in the continued absence of JTWC public a-deck data, I’ve reproduced interpolated model guidance plots for tracks and winds. These are live on PolarWx for all active storms: polarwx.com/tropical

If you prefer using interpolated "early-arriving" model tracks for tropical cyclones (last cycle's models interpolated to this cycle, then compared against the official JTWC/NHC forecast made using these guidance), I've replicated this method on my site: polarwx.com/tropical/?st...

With the addition of HAFS-A and HWRF, I've now independently reproduced most model tracks for global tropical cyclones & typhoons that would typically be in an a-deck file, given that these have not been available for storms in JTWC's domain this year. polarwx.com/tropical/?st...

The East Coast will get its first preview of winter weather next week as a transient but deep cold airmass traverses the region — widespread well below average temperatures are expected for a couple days: (1/3)

Swing… and a (huge) miss Be careful with interpreting AI model run-to-run consistency as a sign of high confidence & likelihood — FNV3 was insistent on exceptional rapid deepening up to the last minute when it failed to materialize:

While recognizing there are different views on whether a category 6 is needed for hurricanes, my personal view on this boils down to a simple phrase: “just because you can, doesn’t mean you should”

So far this year, the Gulf has been very quiet with only one short-lived tropical storm (Barry). Assuming this holds (which is likely), this would be the quietest Gulf season since 2015, and the first season without US Gulf coast tropical cyclone impacts since 2014.

This is part venting as a NJ resident in election season & part plea for help — is there any way to get myself unsubscribed from political operatives’ phone lists that I never subscribed to in the first place & am getting spam calls and texts from different numbers too many times a day

Multi-panel view of the last few days of Hurricane Melissa: ↖️ GOES-19 infrared brightness temp ↗️ GOES-19 visible satellite ↙️ Hurricane hunter planes & flight paths ↘️ Recon-derived flight level wind swath ⬇️ Estimated minimum pressure from recon dropsondes

Given higher than usual confidence of rapid intensification (RI) from analyses of an exceptionally favorable environment & with the aid of newer hurricane models (HAFS) & Google DeepMind, NHC was able to skillfully predict Melissa's RI days in advance.

A look at TDR observations of Melissa over the last few days shows just how impressively persistent and compact its inner core was - while the lack of eyewall replacement cycles helped Melissa intensify right up to landfall, it also kept its wind field fairly small.

The last hurricane hunter flights into Melissa have departed before its landfall in Jamaica. As always kudos to them for their work powering through especially powerful storms like this to collect invaluable data.

Thousands of lightning flashes have been detected in the eyewall of Hurricane Melissa in the last couple days, oftentimes completely enveloping the eyewall - the Enveloped Eyewall Lightning Signature - a phenomenon found in the strongest tropical cyclones on the planet.

Wow. This is easily the most ridiculous dropsonde I’ve ever seen. 188 knot mean winds in the low-levels with gusts over 250 mph Absolutely scary and historic hurricane headed into SW Jamaica this morning