January 7th, 2017 Snow Storm – A Forecasting Miss?

If you live anywhere along the coast from Maryland up to Connecticut you probably spent your Saturday doing more digging out than Friday’s forecast might have led you to believe. The storm’s center was “west” of what was originally forecasted (in other words the storm ended up passing closer to our coastline), and that’s what brought those heavier snow bands farther inland than expected. This is not the first time a forecast didn’t get snow totals entirely right. In fact, our area is VERY prone to getting “the bread and milk” without needing it. So I wanted to explain whether the forecast for this storm was effective and some of the discrepancies among the models leading up to the storm. This won’t be a light read, but if you’re interested in the weather, this is for you!

Before diving into this past weekend’s storm, it is important to keep in mind that snowfall is much more difficult to forecast than rain. Quite frankly, no one cares about exactly how much rain an area gets if they don’t have meteorological or agricultural interest. People are impacted by how long rain lasts and whether there is a little or a lot. That’s a lot easier to forecast than snow. It takes a LOT less moisture to drop an inch of snow than an inch of rain (usually 1″ of rain = 10″ of snow, with even greater ratios occurring during colder weather ). With snow anyone who has to take a step outside is immediately observant of the amount of snow that fell and are impacted by each and every inch. There is so much more at stake with snow storms and snow totals are much more finicky to pin down perfectly.

How Bad Was The Forecast For Our Storm This Weekend?

On January 26-27, 2015, New York City and Philadelphia were both literally shut down because the forecast called for 2 to 3 feet of snow. Only 8-10 inches fell in New York City and Philadelphia saw even less than that. With the storm in 2015, an experimental probability map existed which plotted the percentage a certain amount of snowfall would accumulate. Following that devastating forecast bust, the National Weather Service in Philadelphia (Mount Holly) and New York City (Upton), found a better way to communicate the probability of a snow event to the public. Instead of sending out one snowfall map that plots a range of totals across the Tri-State area, winter forecasts by the NWS now consist of three maps: the forecasted snow range, the lowest potential amounts, and the highest potential amounts. This way the public has access to the National Weather Service’s thinking and can know what the best and worst case scenarios are for each event. When a storm threatens and the chance for error is high due to the storm’s dynamics, the best case scenario will look very different from the worst case scenario. A storm with high confidence would have more similar best and worst case scenarios.

With that history in mind we go back to this weekend’s storm. Unlike in 2015, this weekend’s storm dumped more snow than forecasted for the city. Let’s take a look at the NWS’s official forecast for snow on Friday before diving into some weather models.

Figure 1 shows the official snowfall forecast on the evening of January 6th, just 12 hours before snow started to fall. The forecast called for 1 to 2 inches for New York City, as much as 5″ in the worst case scenario, and as little as 0″. The storm was communicated as one with uncertainty, but was considered to be a minor winter event for the city with the greatest impacts out east on Long Island.

The actual snowfall totals were 5.5″ for Newark, 5″ for the city, 7.5″ for JFK, 10″out in eastern Long Island, and 5.3” in Bridgeport. The official forecast was not far off for eastern Long Island, but a definite miss for everyone west of the Hamptons. However, I do not think this was a “forecast bust” at all, and in fact it was a great example of the strength in the NWS’s new way of visualizing snow storm forecasts. The worst case scenario map was nearly spot on for this storm. While not perfect for every part of the Tri-State, the worst case scenario forecast map closely mirrored what was seen with our storm this weekend.

Figure 1: National Weather Service’s Snowfall Forecast as of January 6th, Approximately 12 hours Before Event

15895616_1529192040443439_3876839837777650304_o

Not so fast though! A forecast is only as good as the message it sent. So how well was Figure 1 received by the public? Facebook is a good place to get a taste of that. (Keep in mind that millions of people are being forecasted for, and I am only showing a few Facebook comments so this is no way meant to be representative of everyone). In Figure 2, check marks are placed next to comments that indicated an understanding of the potential for greater snowfall totals. An x mark was placed next to comments that did not indicate an understanding of this potential. Most comments recognized the fact that the storm could either trend west of current forecasts and bring more snow to the area or trend east and bring less snow. Overall, this snowfall graphic was well received; however, the winter storm advisories and warnings that were posted for individual counties were met with much greater confusion. Even in this chain of comments, one person indicated confusion about warning wording. On a post about warnings, people were questioning why they were “only in an advisory” given how dangerous conditions were during the afternoon hours. Watches and Warnings are constantly being reworked because of public response (or failure to respond) in certain situations. Watches and warnings are often a topic of research because even a good forecast goes bad if it is not communicated properly.

Figure 2: Facebook Comments Received On NWS New York’s Facebook Page Regarding Figure 1

facebook-commentsThe forecast for this weekend was not perfectly accurate, but from a communication stance it did an effective job alerting people of the uncertainty in the forecast and the potential for heavy snowfall. A “forecasting bust” does not properly communicate the potential outcomes and leaves the public completely confused as to what happened when the event does not match what they were told. That was the case with the snow storm in January 2015, it was definitely not the case this weekend.

The Meteorology Behind the January 7, 2017 Storm

The GFS (Global Forecasting System) model is a long range, lower resolution model meaning that it projects weather solutions farther out in time and on a global scale compared to short range models like the SREF (Short Range Ensemble Forecast) and NAM (North American Model). Each of these models have their strengths and weaknesses, but the GFS seems to consistently miss out on the key details when it comes to forecasting coastal storms. So where do these models go right and where do they go wrong with this storm? Let’s take a look at what they projected for Saturday Evening, January 7th, back on Thursday Night, January 5th, (36 hours before the storm).


On the January 5th evening model run, the NAM had the storm intensify more rapidly than the GFS’s solution. It’s important to consider how quickly a storm is intensifying because that would mean a smaller distance between the big snow totals and areas with no snow at all. In addition to a more rapidly intensifying storm, the NAM had the storm trending slightly closer to the coast and stronger than the GFS. As of Thursday night the disagreement between the NAM and the GFS had to do with the storm’s positioning and general intensification, which would play a large role in getting those snow bands on shore in the tri-state area.

The GFS had absolutely no snowfall for New York City on Thursday night from this storm. The NAM had a good handle on the storm with 4-6 inches of snow for the city predicted 36 hours out. Not bad! (Keep in mind in the graphics below, the GFS snow map is a 24 hour total and the NAM snow map is total snowfall – which is why there is so much snow west of the coast with the NAM model). The SREF graphic is the “spaghetti” looking one. Each line of spaghetti represents a slightly different initial starting condition. As you can see, the results were ALL over the place on Thursday for LaGuardia Airport in Queens. It’s not right to put out a forecast of 0″ to 12″, but if you only forecasted with models, that would have been in the cards! Instead most meteorologists stayed conservative with snow totals given the sharp cut off between the big snow and no snow with this storm.

Models On Thursday Night (36 hours before the storm)

GFS central pressure and radar valid Sat. Evening

gfs_mslp_pcpn_frzn_neus_8

NAM central pressure and radar valid Sat. Eveningnamconus_ref_frzn_neus_40

GFS 24 hour snow total up to Sat. Evening

gfs_asnow24_neus_6-copy

NAM total snow total up to Sat. Evening

nam4km_asnow_neus_21

SREF “Plumes” snow fall total for LaGuardia Airport in Queens.

screen-shot-2017-01-08-at-2-56-45-pm-copy


Lets see how things changed on Friday, 12 hours before the storm’s onset….

The GFS and NAM continued to move the storm slightly farther west than the previous run, which hinted at the final outcome of the storm: a closer track to the coast and more snow making its way on shore. Another change from Thursday night was the storm’s intensification. The NAM has a stronger storm off the coast than the GFS’s projection, but the GFS has the storm intensify more rapidly than the NAM. Basically, the storm is looking even healthier in both models during the Friday night run than it did 24 hours earlier at same forecasting period of Saturday night.

In terms of the snowfall totals, the GFS finally has snow bands making landfall – although 2 inches for the Tri-State area was still way too small (even taking into consideration the ratio of snow to liquid being higher than 10 to 1. (1)) The NAM did not change its projected totals much from the run 24 hours prior, still about 4″-6″ for New York City. The SREF plumes have a range from 2″-10″possible. These three models remained inconsistent even the night before the storm’s arrival, although at least they were trending in the right direction. It is amazing how a forecast can have such great variation despite the computer power available today, but it happens when every mile really counts!

(1) Aside: Colder temperatures mean a higher water to snow ratio. The general rule is that 1″ of rain equals 10″ of snow. During periods of cold weather like this weekend, the liquid to snow ratio would be closer to 1″ to 15″ because snow flakes become “branchier” when it is colder so they stack onto each other easier. That means it takes even less moisture to really rack up snow totals, but the good news is you get a nice and fluffy snow that is easier to shovel!

Models on Friday Night (12 hours before the storm)

gfs_mslp_pcpn_frzn_neus_4nam4km_ref_frzn_neus_24
gfs_asnow24_neus_1

nam4km_asnow_neus_9screen-shot-2017-01-08-at-2-57-20-pm-copy

———————————

Finally we’ll take a look at where the center of the storm actually was Saturday evening. We know the short range models, the NAM and SREF, did a much better job with snowfall than the GFS, but how did the GFS stack up with the short range models when it came to storm development and positioning?

Actual Surface Analysis from Saturday Evening according to the Weather Prediction Center (WPC)

screen-shot-2017-01-08-at-7-17-00-pm

It turns out the storm tracked just slightly west of the NAM’s projection and at least 50 miles west of the GFS’s projection from the model runs on Friday night. The GFS did a better job handling the storm’s central pressure and was more closely matched to the storm’s actual rate of intensification than the NAM’s solution, although the storm fell short of the official criteria for “bombogenesis” contrary to the GFS’s projection. (2)

(2Aside: Bombogenesis is a fancy way of saying a storm is rapidly intensifying. The official criteria for a storm to be considered “undergoing bombogenesis” is a drop in central pressure of 24 mb in 24 hrs. That means it is one very healthy storm!

My final thoughts? This was a complex storm. The forecast was conservative in hindsight but not irresponsibly so. This storm was not well behaved! There were a lot of changes occurring at the surface and upper levels that models never had a real consensus on. The only agreement in the models was that the storm was going to contain snow squalls, rapidly intensify, and the data continued to favor a more westerly track. The NAM has lately proven to be a great resource, even in some of the most ridiculous circumstances, so it is not a model that any meteorologist or weather enthusiast should ignore this season! There are still many more weeks of winter left so it will be interesting to see if these model “quirks” continue if another storm follows a similar pattern this year. In the meantime, mild weather is on the way so most of the snow that fell this week will be just a memory in a few days. We’ll have a break, before we have to worry about any more snow!

AC