high risk

I never buy the NAM for any reason. NAMNest thermals and sometimes the QPF, but otherwise, the 12km parent I ignore almost exclusively. The other problem is that the "total snowfall" maps are NOT predictions of what the NAM thinks will be on the ground. When the model's hydrometeors are not liquid (i.e. snow or sleet) as they reach the surface, the amount of liquid goes into a water equivalent bucket. That is what these sites grab and then apply either a 10:1 or Kuchera ratio. What the model thinks will be on the ground is the snow depth product. Unfortunately, that tends to run low in events like these we we've had a warm couple of days prior to the onset.

Certainly looks like the GFS has lost the idea of heavy snow during the midday hours Sunday.

I love the usage of the snow depth field, but two flaws of that product are that it tends to be too low when 1) it gets warm before an event 2) rates are legit

That's not true. The 06Z para GFS did have a low, but it was notably offshore and quickly trucked out to sea. The same is true for the 12Z cycle - it looks nothing like the ops GFS. And apologies for not being able to share graphics.

No. Still FV3-based.

Huge upgrade to the “regular” GFS coming later this year.

Very concerned that if much of the snow falls during the daylight hours Sunday, it will really struggle to accumulate in very marginal low-level temperatures. Rates will be critical to our chances of a measurable snowfall.

Most of the 12Z mesoscale guidance today showed that the organized snow would have a really tough time making it into the DC Metro area, and that seems to be playing out.

Maybe? I haven't looked at the mechanics of why this band is way out ahead of the actual arctic boundary. Another possibility is a few scattered bursts of snow right on the late night front....

Looks like roughly 3AM(west)-5AM(east) timing for the passage of the arctic front.

I understand the skepticism, but those forecasts were relying on intense radiational cooling late at night, and clouds/wind seemed to wreck those opportunities. This will be pure advection of an intensely cold air mass, and it's legit arctic air. I think this has a much better chance (and I'd say it's very likely) of single digits.

Indeed. Looks like mid 20s at midnight. In fact, it stays in the mid 20s through around 4am, and we then drop at least 20 degrees in the 4-5 hours following.

I gave 1/2 of a laughing emoji for the concept of this comment and 1/2 for the utter butchery of Louis' name

VERY well stated. The HRRR is generally pretty good with warm season convection (minus some flaws and the inherent challenges with modeling weakly-forced storms) but has never proven itself as a winter weather model.

Saturday looks brutal, with temps in the mid to upper teens and sustained winds in the 20-30mph with higher gusts.

With how cold surfaces have been, even a half inch of snow at night would likely cause problems on some of the roads, with the ungodly amount of leftover salt possibly saving us from a complete mess. School systems will have a complicated decision.

Was not a total miss. It had a very modest amount of snow about 24 hours earlier than the map you showed, with some sort of lead wave.

AI models don't do particularly well with QPF. They get a general shape of the field but struggle to resolve amounts and detail. I'd focus far more on 500 heights and SLP.

The mean has very limited value at this range, because a few snowy ensemble members can skew it. The probabilistic output is far more useful, but I'm not sure how easy it is to find that on the web.

I'm looking forward to those 4 hours above freezing on Tuesday!

This is exhibit A for the problem with using mean values from a large ensemble. There are a handful of GEFS and ECMWFE members with huge snowfall totals, so the mean value ends up as a couple of inches. But the 50th percentile map shows 0 for our area, and the chance of 1" of snow at KDCA is under 30%.

Absolutely. With that low pretty much guaranteed to bomb somewhere east of us, there is solid model agreement in a strong wind field over the Mid-Atlantic.

Ok, 3 things going on here: 1) The winter suite in the NBM is only updated at 01, 07, 13, and 19Z. So, if you're showing an 18Z cycle for this medium range case, it's as "stale" as you can have it. 2) In the medium range, the only inputs are the 50 ECMWF ensemble members, the 30 GEFS members, and the GFS. It doesn't even include the deterministic ECMWF run (don't ask). Some of the CMC members are now included in the parallel. Here is what you're going to hate: the NBM does not yet receive the 06/18Z ECWMF data. And at 19Z, the ECMWF ensemble has not yet arrived. So, the 18z cycle in question is still using the very snowy 00Z ECMWF ensemble. That's what's driving the big numbers. 3) One additional factor: bias correction is performed for QPF for the global model inputs, and the bias-corrected QPF is used by the NBM winter suite. The bias correction, based on the based 90 days of forecasted and observed QPF in this case is bumping up the QPF for the event, as it believes that the ECMWF ensemble has been running dry. So, you have some inputs with high raw precip values that are being further adjusted up. That's why the 18Z NBM looks so snowy. The 01Z cycle coming out soon will use the 12Z ECMWF ensemble and the 18Z GEFS; based on what I see in the QPF of those systems, my guess is the NBM snowfall will be a fair amount lower, although the bias correction might boost it a bit more than I expect.

Always like seeing NBM images here, but 1) this cycle reflects the 00Z Euro ensembles and not the less snowy 12Z 2) The NBM mean, like any ensemble mean, can be skewed by a handful of very snowy members. FWIW, the latest NBM cycle still keeps the probability of an inch or greater of snow this weekend at under 50%.

The NBM at this range gives heavy weight to the 00Z EPS, so this is exactly what I'd expect to see.