Let's Talk Probabilities for the January 2015 Blizzard

If you're a typical individual, you probably had a very busy weekend with a rainy Saturday (on top of a couple inches of snow, gross!) and then a frantic Sunday packed with errands to get ready for the final week of January. If you hit the grocery store a bit early today you may have noticed that it was slightly more crowded than normal. If you hit the grocery store this evening, you're lucky to be back in the comfort of your own home after that stressful, chaotic affair! The news is definitely out that Long Island is in for a massive blizzard Monday through Tuesday.

I'm sure you have a few questions such as--
1) Where the heck did this storm come from?!
2) Are these forecast snowfall totals for real??
3) What cool data does Stony Brook University have to share for this historic event?

With this blizzard, it's all about probabilities. Probabilistic information is such an important part of weather forecasting. The atmosphere is so chaotic that predicting it perfectly is impossible (improbable, too!). That's where ranges of values (such as snowfall) and measures of confidence (such as lack of or overly so) really come in handy. Well, let's waste no time to answer the above questions!

Numerical Weather Prediction: Ensembles "Jazz Up" The Forecast

In previous posts, we've discussed numerical weather prediction. Weather models were showing a system moving from Canada through the Midwest US and to the East Coast bringing the possibility of some accumulating snow after the dud of a snow event we had on January 24th. However, few (if any), had this system strengthening close enough off of the coast to cause any major impacts. Well, that may have been true for what are called the "deterministic" models.

Once you have your initial conditions from observations and a weather model, you can run the model and look at the model output/your forecast. You may be happy with the model output from the GFS versus that of the NAM or you may not be. Most weather models are run four times daily at 00, 06, 12, 18 UTC or 7 PM, 1 AM, 7 AM, 1 PM EST. Each time they are run they take in new initial conditions to make their forecast. That's why there's a lot of run-to-run variability or changes in the forecast from the same weather model each time it is run. The single weather models are called deterministic models because they take observations just as they are and create a forecast. There is another method of modeling the weather that instead of providing one solution provides a whole ensemble of solutions and as such is known as ensemble system forecasting.

The basis behind ensemble forecasting is that there are a lot of errors with weather models, especially those that arise from the initial conditions. What if a weather station in Oklahoma is broken but no one notices and you input that bad data into your weather model? What if the weather balloon from Montana got hijacked by a goose and your data reflects that wild goose chase and you put it into your weather model? To account for this, ensemble forecast systems sometimes take the initial conditions and perturb them, or change them slightly, and then run the same weather model with this new input. Another method behind ensemble systems is to have the same input but change some of the ways that the model calculates key processes such as the formation of snow.

The output, or forecast, provides an envelope of likely solutions. It's like if you were to go to the doctor when you stub your toe. The best case scenario is that the pain will go away in ten minutes. The worst case scenario is that she'll have to amputate. Given that ensemble of solutions, you can understand what's likely and make a judgement which is usually somewhere in the middle which is known as the ensemble mean. Mean in the statistical sense, not the emotional sense, so it just means the average.

While the deterministic models may have been "out to sea" or weaker with this storm, the ensembles were showing the possibility (read: slight probability) of a stronger storm and closer to the coast. To provide an example, let's look at the Global Ensemble Forecast System (GEFS) run by NCEP. There are tools to see what atmospheric players are most important according to some ensemble systems that can impact the intensity or track of storms. One such tool is developed at the School of Marine and Atmospheric Sciences of Stony Brook University in collaboration with the National Weather Service to look into these key players so that forecasters can keep an eye on them and how they are shown in the weather models as they get closer to the storm event. Let's compare the mean (average) and spread (variations from the mean) for two forecast times-- 4 days and 2 days from 7 AM EST (12 UTC) on Tuesday, January 27th.

GEFS Sea Level Pressure Ensemble Mean and Spread 4 days out.

GEFS Sea Level Pressure Ensemble Mean and Spread 2 days out.

Using data from Friday morning, this particular ensemble was showing the average low of 994 mb located far off the coast with a clustering of ensemble members farther west towards the coast. While there was still uncertainty 4 days away, there's still the probability of a coastal low having more impacts over land.

Comparing that with Sunday morning's ensemble run, there is much less spread which shows that the ensemble members are converging on a solution (the truth?) as one would hope as an event gets closer in time. The ensemble mean is now showing a deeper (984 mb) low closer to the coast than a couple of days ago. By this time, all of the deterministic models have converged on a similar solution which increases confidence in that solution being what will happen. Keep in mind this wasn't the case just a few days ago!

This brings us to the first question, "Where the heck did this storm come from?!" Using the ensemble sensitivity analysis tool developed at SBU, the key feature from that would impact the location and intensity of the coastal low development 4 days later was the intensity (or amplitude) of a shortwave, or weak low pressure system, developing lee of the Canadian Rockies and amplifying the existing trough behind the departing Saturday storm that had yet to even happen. Now that this system has formed, its strength using such tools as satellite imagery can be assessed and compared with what models were showing to see whether it is weaker/stronger or in a different location.

GOES-East Water Vapor Image from 7:15 EST on January 25th  (0015 UTC 26 Jan)

Ready the Plows... or the Front Loaders?

Now that we are closer to the event, snowfall forecasts are everywhere! So that brings us to the second question-- "Are these forecast snowfall totals for real?" As of tonight (Sunday night), the intensity and placement of the low pressure center is almost certain to be in the "sweet spot" for significant snow accumulations on Long Island.

NCEP Forecast low tracks for various times denoted by the colors and marker shapes. 

In a previous post, we've discussed why snowbands love Long Island. The feeling is mutual, of course! Once a coastal low forms, it usually strengthens and moves northeastward along the coast up towards New England. The broad area of precipitation that is usually light in nature can organize itself into narrow areas of intense snowfall, known as a snowband. A snowband can be a few miles wide and a few tens to about a hundred miles long-- and like most things they come in a lot of different shapes and sizes. When a coastal low forms, there is typically warm air to the southeast of it and cold air to the northwest of its center. Air moves in a cyclonic or counterclockwise direction which acts to create fronts, or boundaries between the distinct air masses. A snowband that is set to impact Long Island typically forms north of the warm front where warm, moist air is ascending over a wall of colder, more dense air. As the low strengthens, this warm air is vigorously churned counterclockwise towards the west and the snowband pivots to the northwest of the surface cyclone center. As the low matures the snowband can sometimes be found to the west of the surface cyclone center. You can see the positions of the snowband relative to the surface low in the following schematic:

The recipe for a very strong snowband includes strong winds clashing together to the northwest of a low pressure center that extends a few miles up into the atmosphere, sufficient moisture so that there is plenty of water vapor to condense into clouds and provide a lot of precipitation, and just like with summertime thunderstorms there should be a little bit of instability or a region of air in which a pocket of air will be less dense and able to rise rapidly. A lot of research has been completed investigating these snowbands and a Stony Brook University alumnus, David Novak, published many peer-reviewed papers about those that occur in the Northeast. He looked at many storms and determined the general ingredients needed to cook up a great snowband as summarized in the following figure.

The ingredients for a mesoscale snowband modified from Novak et al. 2004.

The two locations highlighted by the stars indicate the preferred locations for snowbands. In the red shading are regions of frontogenesis, or the creation (i.e. genesis) of fronts or the clashing of two air masses. The region enveloped in a scalloped line shows a region of deformation or where the winds have a component that are converging which enhances frontogenesis. While there may be snow all around the surface low for miles and miles it shouldn't be as heavy as within the regions of snowbands indicated by those stars-- all thanks to frontogenesis. Let's talk more about frontogenesis and not just because it's a really fun word to say.

Frontogenesis, or the formation of fronts, means that the atmosphere is unbalanced. The fact that there is a temperature gradient (that's what a front is) means that the atmosphere is a bit unstable and wants to even out that temperature gradient. How can it create a uniform temperature distribution if it has to fight against the really strong horizontal winds that keep making the temperature gradient stronger? To combat the imbalance, the atmosphere initiates a vertical wind circulation because when you can't go out you go up, right? The atmosphere induces a frontogenetical circulation that causes warm air to rise and cool on the warm side of the front (cooling down the warmer air) and cold air to descend and warm on the cold side of the front (warming up the colder air). This warm, rising air is usually incredibly moist so there's plenty of water vapor to create deep clouds that produce heavy snow and with that the snowband is formed. This circulation is usually very small because the temperature gradient at the front itself is very narrow in width so the snowband is very narrow as well. A summarizing schematic is provided below that shows a horizontal view of the snowband (yellow star) to the northwest of the surface low with the counterclockwise winds in the upper-right corner above a cross section showing the vertical circulation that causes a lot more snow to fall on your buddy's house (yellow star) versus your house (cozy cottage to the west or mansion to the east depending on your preference).

So why doesn't the weather forecaster know exactly where the snowband is going to set up and instead calls for a large amount of snow for the whole region? As you are now aware of, snowbands are very sensitive to particular ingredients with the most tricky being the location of the surface low pressure center. If the low travels a bit farther off of the coast then so will the snowband and the East End of Long Island may get hit a lot worse than NYC. If the low tracks closer to NYC then perhaps NJ will get hit harder. See the pattern?

This is where probabilities can come in very handy! Using our good friends, ensembles forecasting systems, we can determine the probability of exceeding a certain snowfall threshold. The NOAA/NWS Weather Prediction Center (WPC) has online tools to plot such data. Using 57 total members from several different models (more info here), the WPC creates its probability plots. Here is the probability of accumulating greater than or equal to 18 inches of snow in a 24-hour period between 7 PM EST Monday and 7 PM EST Tuesday:

WPC 24-hour probability of snowfall accumulating greater than or equal to 18 in by 7 PM EST Tuesday.

Most of Long Island is shown as seeing >50% probability of at last 18 inches of snow (the highest amount that is available to be plotted via WPC) with a maximum of >80% over central CT. Those are high probability values given the number of different ensemble members that go into that product! 

Looking at a more local scale, the National Weather Service Upton, NY Forecast Office has issued their storm total snowfall ranges, but we'd also like to highlight some experimental graphics that use probabilities. The storm total forecast as of Sunday night is showing between 24-36 inches of snow for all of Long Island. Some experimental graphics show the minimum expected storm total snowfall (6 - 9 inches) or the 10th percentile amount and the maximum expected storm total snowfall (29 - 35 inches) or the 90th percentile amount. These graphics can be accessed here. 

Now it's one thing to have snow, it's another thing to have a blizzard. The NOAA/NWS Gray, ME Forecast Office put together a great graphic to summarize blizzard conditions. It's the combination of winds > 35 mph, fast snowfall rates, and blowing snow creating white-out conditions. A Blizzard Warning has been issued by our local NWS Forecast Office from 1 PM EST Monday to Midnight on Tuesday forecasting for snowfall rates of 2 - 4 inches per hour and wind gusts to 65 mph. The blizzard conditions are resulting from the rapid development, or bombing out, of the low that is forecast to deepen (strengthen) > 24 mb in 24 h. The models vary on its forecast intensification, but there will be enough of a pressure gradient between it and the high to our north to create the potential for wind gusts almost approaching hurricane-force! For example, the SBU GFS-WRF is showing sustained winds of >35 kts (> 40 mph) at Montauk Point on Monday night!

Montauk Point Time Series of Simulated Wind Speeds and Directions from the SBU GFS-WRF initialized 12 UTC 25 Jan.

Given this probabilistic information, it makes sense that Stony Brook University cancelled its first two days of classes (Monday and Tuesday) while the system is still hanging out near West Virginia! However, we will need to pay attention to any snowbands that develop north and west of the surface cyclone center that may determine who gets the maximum amounts and sees the most treacherous conditions. 

Play with the snow, from the comfort of your own home! 

Now that we've determined that this storm developed within the available weather models within the past couple of days, there is potential for possibly (probably!) historic snow accumulations especially under any snowbands that develop, we can get to our third question, "What cool data does Stony Brook University have to share for this historic event?"

We still have our vertically pointing radar (MRR) on the roof of Endeavour Hall on campus that can be used to compliment conventional radar data from Upton (KOKX) as well as show fall speeds of the hydrometeors (just snow for this event). Joining it up there this year is a brand-new instrument that actually takes pictures of snowflakes in free-fall! The Multi Angle Snowflake Camera (MASC) was developed at the University of Utah by Dr. Tim Garrett and is supported by both Dr. Brian Colle of Stony Brook University and Dr. Sandra Yuter of North Carolina State University. Mark Lang of SBU and Dr. Matthew Miller of NCState implemented a website that will show real-time images of the snowflakes falling and being captured by the camera! Here's an example of some from Saturday's (brief) snow event:

MASC image from 6:05 PM EST January 24th

Pretty cool, right? (Well, cold, in this case!) To conclude, this storm will likely be remembered as the "Blizzard of 2015" and is an example of why forecasting using ensemble forecasting systems is incredibly important as well as paying attention to the potential development of any smaller-scale (mesoscale) details such as snowbands. The MASC is now available to provide a real-time eye into the storm. The main point with a storm like this is to exercise extreme caution to stay safe and then we can all appreciate this storm for its intensity in safety. Best of luck in the blizzard! You probably won't forget it anytime soon! 

Main links used in this post:

- Stony Brook University Ensemble Sensitivity Analysis: http://dendrite.somas.stonybrook.edu/CSTAR/Ensemble_Sensitivity/EnSense_Main.html

- Stony Brook University WRF Ensemble: http://itpa.somas.stonybrook.edu/LI_WRF/

- NOAA/NWS New York, NY Forecast Office Winter Weather Page: http://www.weather.gov/okx/winter 

- MASC Real-time Images at SBU: http://itpa.somas.stonybrook.edu/MASC/snowflakes_test.html