high risk

The point, though, is that the heaviest rain and flooding potential are indicated by all guidance to arrive after schools would have dismissed at the regular time (and bus routes would have ended).

I'm not seeing the big wind threat for most of the area (except southern sections and areas along the Bay) until dinner time and especially later. Temperatures will be slow to rise tomorrow; I think we have to get into the low to mid 50s before we can really mix, and that's not going to happen until much later in the day or early evening.

Looking forward to seeing what this looks like on the NAM Nest once we get into range.

The 12 km NAM has a nice depiction of the intense, forced rain band that will accompany the front later Tuesday evening. This will potentially be our greatest chance to mix some of the stronger winds just above the surface down to the ground. I'm very modestly hopeful of hearing thunder.

For future reference, that's the 12 km NAM, as also known as the parent. The 3 km NAM Nest is usually labeled as either the 3km or the nest.

Totally agree about the low-level temps. The EC and Canadian bring much of the I-95 corridor into the upper 30s by Saturday evening, but the NAMs this evening keep that area around 33. With the north-northeast wind direction shown on most guidance, a quick rise into the upper 30s while precip is falling seems unlikely.

The NAM 10:1 maps look huge even along the I-95 corridor, because even after the precip type maps show rain, the model microphysics are generating sleet which is going into the snow water equivalent tally.. Here is a sounding from the "rain" area: That's heavy sleet. Not saying that the thermal profile will be correct, but it explains the generous 10:1 totals.

That's fair. The depth field is underdone in events with warm ground going in, marginal temps, and good rates. But it's a strong alternative to 10:1 maps when you have a lot of sleet (or other mix) or poor rates with marginal temps. In events that argue for rates overcoming temps, a blend of the 10:1 and snow depth, with a solid lean towards the depth, often works well.

Thanks for this example! WB must be making their own determinations about how much QPF is going into each precip type. As I mentioned, the only "accounting" in the actual models (RAP/HRRR excluded) is a snow water equivalent bucket that counts whatever fraction of the falling precip is snow or sleet. It's "good" in that the WB approach wipes out the sleet and also the accumulations when the profile doesn't support pure snow, but they're trying to match precip the accumulates over a longer period of time with the instantaneous precip type, so that won't always work very well. Ultimately, the RAP/HRRR do things the right way: they have separate buckets for each precip type and output each of them. It removes the guesswork. This approach will be used in the new RRFS, and the GFS is going to adopt it too.

Assuming you're talking about 10:1 maps? That's really odd, as the models generate a water equivalent of snow field in mm that should simply be converted to inches and multiplied by 10. The only thing that they could be doing is trying to make their own determinations about how much of a period of QPF is snow, but that would involve a lot of assumptions on their end that would not work well. I'd be very interested in seeing an example.

Good to see you here @usedtobe! I'm also a huge proponent of the snow depth product in these marginal temperature events, but I like to throw in the caveat that it will be underdone in these type of events if the rates are really good.

Do you have an example? The 10:1 maps should be the same everywhere. The snow depth maps should be too, except that while Tropical Tidbits only counts positive changes in depth (i.e. they ignore when the amount goes down due to melting or compacting), while others may count any change. Even the Kuchera method should, in theory, yield the same result across sites.

That's because the code that accumulates snow is inside the model and is driven by microphysics, and the code that generates precip type is outside of the model and is driven (mostly) by the temperature profiles.

Now would be a good time for a few quick reminders about model snow accumulations: - Most of the American models (I'm not entirely sure about our international friends) generate a water equivalent of snow. It's then up to the user to apply a reasonable snow-to-liquid ratio. Some sites compute one (i.e. Kuchera), but that tends to be somewhat generous. 10:1 is the most commonly displayed field, but while it's a nice one-size-fits-all approach, it's not always applicable to the temperature profile. The key here is that if you're looking at a GFS 10:1 map in a marginal temperature situation, don't say that the GFS shows 6-8 or whatever amount. It doesn't. Mentally adjust that "starter map" as needed. - Along those lines, the water equivalent of snow is not always a measure of how much snow can accumulate. Sleet goes into the same bucket as snow, so an all sleet event will still show large water equivalents, and a 10:1 map looks absurd. Even an event with 1" of liquid that consists of a 50-50 mix of rain and mangled snowflakes will show 0.5" of snow water equivalent. If you use a 10:1 map, you get 5", even though the model never really thought that anything would stick. - Those issues with sleet or snow that won't accumulate due to warm ground or marginal temps are largely handled by the snow depth field which is the actual amount that the model thinks will be on the ground. 10:1 and snow depth maps look amazingly different in sleet events, as they should. But they also often look different in events with marginal temps or very warm ground. The snow depth maps are usually good, but they do tend to underestimate when the ground is quite warm heading into an event or if big rates overcome marginal temps. In those events, a blend of the 10:1 and snow depth maps is often a good call. - RAP/HRRR actually predicts accumulations (with a temperature-dependent SLR) that account for compacting and melting, although they're not displayed on Tropical Tidbits. - The NBM computes SLRs for each member with snow, and the system overall uses a lot of inputs (including ensembles) and is calibrated, so it's worth a look. Remember, though, that it won't respond immediately to big shifts in guidance. Knowing the limitations of the products you're examining helps to limit the confusion, frustration, and ugly analysis. Thanks for coming to my TED talk.

The last GEFS upgrade (Fall 2020) switched the system to the FV3 model core and among many improvements, the spread was increased (as intended), and the extra spread was largely meaningful spread. That said, the GFS had a fairly significant upgrade (to version 16) after the GEFS upgrade, and the GEFS has not received the corresponding update. Overall, it still feels to me like the GEFS responds to GFS trends more than it should, consistent with the observation by@WxUSAF that there seems to be an element of follow the leader.

GFS temps are at or above freezing for a good chunk of the event. Snow of course can absolutely accumulate at those temps with good rates, particularly if it falls after dark Saturday night, but 10:1 is likely to be too high for the SLR.

This isn't how standard NWP works. Historical data plays no role: you have equations that represent all relevant processes, you create an initial state based on observations, and you integrate the model. Historical data *does* factor in for AI/ML approaches, but if you're just looking at a standard GFS forecast map, it doesn't know anything about years past.

Worth noting that the two CAMs that are run to 60 hours (NAM Nest and HiResWindow FV3) both show a burst of snow showers very early Tuesday for those of us on the east side of the Potomac.

The GEFS was indeed upgraded to use the FV3 core in 2020! https://www.emc.ncep.noaa.gov/users/meg/gefsv12/ It doesn't, however, have the changes that were made to the GFS in Version 16 in 2021.

Before midnight? I guess it depends how you define "many of us NW of 95", but it looks to me to show that occurring just after midnight. The HREF is just averaging the HRRR, NAM Nest, and Hires Windows - the most recent 2 cycles of each.

Yeah, I'm the biggest advocate for the depth maps for events with warm ground and marginal temps, but those maps can run low IF the rates are legit (and it helps if the snow falls at night)

NDFD is the gridded NWS forecast

I'll chime in, the answers (opinions??) are complicated and nuanced. The HRRR is still the flagship CAM, but it has primarily been tuned for deep convection, and I hesitate to use it much for winter weather, especially with cold air damming events. The HRRR also hasn't been updated in way too long, because all efforts are on RRFS, which hasn't come together as quickly as planned. The NAM Nest, IMHO, is still very good for precip type, although it runs a bit on the cold side and is wet, but there isn't currently a CAM that does better with cold air damming. There are 3 HiRes Windows (ARW, ARW2, and FV3) - they are all operational, but I don't love them for winter weather. The FV3 HiRes Window is particularly bad for thermal profiles. If I had to pick a CAM for winter, I'd go with the NAM Nest, but I'd be weary of its synoptic errors beyond 36-48 hours and realize that it runs a bit on the cold side.

If we're talking about Northern Hemisphere Day 5 anomaly coefficient at 500 mb, which is generally considered the standard metric for synoptic performance, the GFS has been solidly in 4th place for most of this year, trailing the ECMWF, UKMET, and CMC.

For the love of God, please stop showing the 10:1 maps. This type of event with marginal temps and warm ground screams for the positive snow depth product.