Ensemble Modeling – A Great Tool!

The atmosphere is always moving. It’s infinitely complex. It’s a fascinating fluid that can provide meteorologists with amazement, humility, and a whole lot of fun! To figure out how the atmosphere will evolve over a given time we use computer modesl.  In order to do this we need to take a snap shot of the atmosphere to give a that model a set of initial conditions.

Since it is impossible to sample every point (both in the horizontal and vertical) in the troposphere our initial conditions will never be perfect. Weather balloons, surface observations, and increasingly remote sensing do provide us with a good first guess! Errors in that first guess amplify with time due to the chaotic nature of the partial differential equations that the model solves. Other issues, such as the imprecise parameterization of variables, also lead to issues with forecast accuracy.

Ensemble modeling is a powerful tool that tries to alleviate the issues associated with the initializing a model.  We do this by “tweaking” the initial conditions for each ensemble member and then run the model for each different member with different sets of (reasonable) initial conditions. When all the ensemble members are reasonably close with a 48-hour or 72-hour forecast, for instance, predictability is high and there’s a low sensitivity to small changes in the initial conditions. When ensemble members are all over the place, however, we can infer there is a high degree of sensitivity to the initial conditions and the forecast uncertainty is high.

Every storm is different. Some storms are well forecast and modeled and forecast confidence is high. Other times there is a huge spread in our ensembles and the forecast confidence is low. When ensemble members agree with their respective operational run things are in great shape!

Enter in the Monday night/Tuesday storm. If you were to look at the 15z Short Range Ensemble Forecast (SREF) members you’d be ready for a big snowstorm! Here’s a plot of the probability of >4″ of snow (the plot is created by looking at how many members produce >4″ of snow at a given grid point and assigning a percentage).

15z SREF 69-hour >4" of snow (in 12 hour) Probability

15z SREF 69-hour >4″ of snow (in 12 hour) Probability

So… yeah… about that 70 percent chance! While the vast majority of SREF members are quite amped up and quite wet most of our other models are quite blah (except, of course, the NAM)! The 12z European, UKMet, GGEM, and GFS produce very little snow over Connecticut. Maybe an inch or two!

So what’s the deal? It’s important to understand the difference between the different models and the forcing behind what’s producing the precipitation. Let’s start with the synoptics.

Since the NAM is the most robust let’s look at why. First here’s pressure on the 2 PVU surface (dynamic tropopause). The orange lines are isotachs on the DT. The image below is 500mb absolute vorticity and the orange shading is upward 700-500mb vertical motion.


So what’s going on? We have a strong PV anomaly that’s heading toward the coast. The PV anomaly is strengthening as it heads east. The PV forcing is strongest off the east coast of New England. In addition, the low static stability thanks to the moist/mild air over the Ocean favors upward motion. Using a conceptual jet streak model we can also see forcing for lift. I marked the 2 jet streaks that have developed downstream of the PV anomaly. One over Nova Scotia and another south of Nantucket. The left exit region of the latter and right entrance region of the former are right over eastern Massachusetts and the adjacent Atlantic Ocean.

Here’s the 18z NAM QPF and 1000-500mb Thickness / SLP valid 12z Tuesday. Under that area of strong synoptic forcing an inverted trough/NORLUN signal has developed with a localized maximum in QPF.

18z NAM - 66 Hour Forecast

18z NAM – 66 Hour Forecast

A combination of strong synoptic-scale forcing and mesoscale forcing along that inverted trough result in heavy QPF values. So why do the other models not have this? While all the models are reasonably close with the synoptics they have dramatically different QPF forecasts.

Some of our models are referred to as global models and others as mesoscale models. The mesoscale models tend to have a finer resolution than their global counterparts and they’re also non-hydrostatic (i.e. don’t assume a balance between vertical pressure gradient force and gravity). The NAM and its ensembles (the SREF) also have different terrain and convective parameterization schemes designed to resolve mesoscale details.

So what are we to do? This event is driven by synoptic and mesoscale phenomenon. While this is true of all events it’s particularly true of these kind of setups where mesoscale boundaries are absolutely vital. We’ve seen storms like this where only miles separate 3″ of snow from 15″!

In general a few things to note about these inverted trough/NORLUN deals:

  • I’d like to see more QPF from the global models. Stronger synoptic forcing and better inflow off the ocean would be a start.
  • These things always seem to trend north and east with time! When models have something over central New Jersey 72-hours out more often than not it ends up over the north shore of Boston by verification time!
  • Simple geography tells you that eastern New England is more vulnerable to these NORLUNs as they “stick” out to the east more. Pressures lower and the trough forms as the PV anomaly is able to interact with the low static stability over the ocean. 
  • Mesoscale models can hint at the potential for super-big totals but are notoriously poor with figuring out where exactly this will be. Global models are tend to catch on and show something.
  • Where the NORLUN does setup it can dump! Normally the airmass is quite cold and many times the strongest omega is near the top of the boundary layer up through 700-ish mb intersecting with the dendritic growth zone.

This is something to watch over the next few model runs. While the NAM and SREFs are honking a big event I want to see the Euro jump on board. Until it does there’s no reason to play up the snow threat though these things are always a pain in the butt to forecast.

Think snow!!!!


Light Saturday Snow

In this winter even a light snow event is enough to garner some extra attention! The setup that we have is actually quite close to developing a fairly sizable snowstorm but the pieces will not come together just right this time.

Here’s the GFS 500mb height/vorticity forcast for 1 p.m. Saturday.

I circled a “vort max” or “shortwave” off the east coast. Think of this as a piece of energy in the mid levels of the atmosphere. To the west there is a large and relatively impressive trough digging through the Great Lakes. These 2 features will remain separate. They will not phase into one. That means we’re left with a relatively large but diffuse area of lift in the atmosphere offshore as opposed to a concentrated area of strong lift – which is how we get our biggest storms.

You can see the impact of that by looking at the sea level pressure prog for the same time.

Notice how diffuse the low is (i.e. not tightly packed/wound up)? This shows that this storm is going to have trouble getting organized until it reaches or passes our latitude. Not good for a big storm.

That said there will be enough lift and convergence to produce a widespread swath of snow across southern New England. Many locations will see 1″-3″ of snow. It’s possible that if the storm winds up organizing faster and closer to the coast we could see more impressive snow as a comma head develops.

There is one other thing to watch out for but at this point seems unlikely. The 00z and 06z NAM is developing an inverted trough from the low offshore back to New York City and southern Connecticut. This essentially maximizes low level convergence and

produces a narrow, but intense, band of heavy snow. At this point the NAM is the only model showing this so it’s hard to get too excited about it but it’s something to watch.  The 2 panels on the left show the 06z NAM at 1 p.m. Saturday with a relative QPF max over eastern Long Island and southern Connecticut with the lower panel showing why – a band of convergence stretching north from the Atlantic toward Connecticut.

For what it’s worth the 03Z SREF guidance shows relatively robust probabilities for >4″ of snow across parts of southern New England. The blue shading is >25% chance while the green is >50% chance. At this point I think these numbers are a bit high but there is a chance that this storm trends a bit more impressive for some of the reasons discussed above.

Freak October Snowstorm Becoming More Likely

Hello, Snow!

This evening’s snow is a rare enough event. It looks as if the greater Hartford area has picked up measurable snow in October for the first time since 1979 and for only the 4th time in the last 100 years. We will need to wait to hear from the weather observer at Bradley Airport to get the official word.

This may be just a prelude for a significant and high impact winter storm that appears to be brewing for the weekend. All day I’ve spoken to meteorologist friends who have all said the same thing, “I’ve never seen anything like this.”

Obviously things can change and a variation in the track could mean only a little bit of snow. As of right now, however, the model consensus is a major winter storm that has the potential to bring down trees, tree limbs, and power lines given the amount of foliage still on the trees.

As we’ve seen tonight even though it’s October the sun angle is quite low and given a heavy enough snow rate snow can accumulate on the grass and on the roads. The GFS and the NAM computer models along with the Euro show substantial snow totals (most certainly “plowable”) even down to the shoreline.

For entertainment purposes here’s a look at the >12″ snow probability from our SREF guidance suite. The blue shading is 25 percent chance. Obviously this isn’t what I’m forecasting but this shows just how incredibly bizarre, rare, and unusual this storm could be.

By midday Friday we should have a better handle on this storm. At this point it looks like this storm has the potential to be unlike anything we’ve seen in recent memory.

Thursday Evening Snow

Most areas in Connecticut haven’t seen a frost and most certainly haven’t seen a hard freeze! All that will change Thursday as a burst of cold will move south as a departing storm heads into the Atlantic.

18z NAM 3-Hour Precipitation Forecast 00z Friday

While the NAM computer model drops temperatures fast enough to change rain to accumulating snow just about everywhere the GFS is much milder. As is typical the NAM Is likely too cold and the truth will lie in the middle.

At this point it seems like a slushy coating to an inch of snow is possible in the northwest and northeast hills Thursday evening. For the valley locations some places may see some wet snowflakes mix in but that’s just about it. One of the differences in the model is that the stronger solutions develop stronger northerly winds at the surface which is able to tug colder air to the coast much faster. This will be something to watch.

15z SREF 1" Snow Probability

The 15z SREF snowfall probabilities are pretty similar to previous runs with about a 50% chance of >1″ of snow in the higher elevations of Litchfield County. The green shading is probability over 50% and red is over 75%. As you can see there is a low probability of >1″ of snow in some of the lower elevations but at this point that is rather unlikely.

Even a slushy accumulation could make roads slippery as temperatures drop below freezing in the hills. Watch for some black ice as well early Friday morning.

As for Saturday’s storm… stay tuned! At this point it looks like a rain/snow mix glances by Connecticut but there are some indications the storm could be closer to the coast and much more powerful.

Major Snowstorm on the Way (Again)

It’s been a snowy start to the winter for sure. Though the western part of the state has cashed in the most in recent storms I think everyone gets hit hard by this storm and right now I’m forecasting 8″-14″ of snow.

The SREF (short range ensemble forecast) probabilities show greater than an 80% chance for 8″+ of snow. That’s extremely impressive. The 00z NAM is going bonkers with this storm (as usual) dropping 15″ or more of snow statewide. This is likely overdone (a good rule of thumb is to cut the NAM forecast by about 25%) but can sometimes signal the potential for the storm to really go to town.

700mb NAM 36 Hour Forecast

One thing to watch here is how quick and how close to the coast the mid level low forms. If the mid level low really goes bonkers along the Jersey Shore like the 00z NAM does than I expect 12″-20″ of snow across the state. If, however, the mid level low takes a bit longer to organize and scoots a bit east I think the heaviest snow will wind up northeast of Connecticut (just west of Boston?). This seems like the most likely scenario at this point. The 00z GFS continues the trend from previous GFS runs with a track a solid 75 or 100 miles east of the overly-amped NAM.

Wherever the cold conveyor belt starts to rip someone in southern New England is going to get hit very hard and probably pick up 18″ or so. At this point I think 8″-14″ is a reasonable forecast for most areas though I wouldn’t be surprised if those numbers are bumped up in some areas, particularly northeast of Connecticut.

One additional note. Unlike the last storm blizzard conditions are unlikely. The heaviest wind and snow will likely be out of phase so even though snow totals will likely be more impressive than the blizzard after Christmas the wind will not arrive until the heaviest snow ends.