An update passed along to Premium members earlier explained the number of reasons why the forecast failed so badly, and why all forecasts were so far off for the region. We wanted to pass this along to the public as well to stimulate discussion on the issue, even though you’ve likely read articles about it already. There is also a Clipper impacting the area Thursday night into Friday, and the first forecast for that (while out for Premium members) will be freely released tomorrow evening. Below is the discussion:
First, The Weather Channel has organized accumulations here. One thing I want to point out (that was poorly forecast by much modeling guidance) is the extreme gradient between Long Island and Southwestern Connecticut. Massapequa due south of Stamford saw a foot and a half whereas Stamford saw around 7 inches per current reports. That’s almost a foot difference across such a small area, just a testament to the extreme banding and gradient with this storm. Islip as well broke the 20 inch mark before the storm is even done, while due north in Bridgeport snowfall amounts in the 5-9 inch range are being reported. This crazy gradient was picked up by our modeling guidance in the final hours of this storm but not much before then, and is just a testament to the extreme variations within this storm. Even in Connecticut, we are seeing reports of close to/over 2 feet of snow in New London County with less than a foot in New Haven. That heavier band set up just east and did deliver the promised amounts, just not quite over where it was expected.
What we had in this storm is what we call a “Miller B” storm formation. In this, an Alberta Clipper was diving down into the Ohio River Valley and approaching the southeast. It then was going to transfer energy to the base of an upper level trough offshore, where a new low pressure center was going to develop and rapidly intensify. While intensifying and being fed energy from the jet, this storm was going to wobble northeast and then get “captured” by lower pressure aloft in the atmosphere that would guide it back to the west before the storm would eventually make its way out into the ocean. From the beginning, I remained concerned about this storm because a Miller B has the highest variability in a forecast. Models are notoriously poor handling that energy transfer and the placement of that secondary low, and that is part of what can lead to forecast busts as large as the “blizzard that wasn’t” in March 2001 and this storm here. From the beginning I expressed caution about uncertainty, but clearly not nearly enough. These Miller B storms can sometimes be the strongest of all as they strengthen so incredibly rapidly, (though this wasn’t a classic one) but they can also lead to the largest busts. This one had a much smaller energy transfer than usual, potentially meaning there were added intricacies that much of our weather model guidance was unable to pick up on. Thus, from the beginning there were potential issues with this storm.
Issues arose on Saturday and into Sunday when two distinctive model camps began to emerge. The GFS and CMC weather models remained staunchly east, not quite as strong with the storm but fairly significant. The ECMWF weather model continued to show an all out blizzard, and it has performed as one of the most accurate models. The NAM weather model, one of the least accurate, agreed with the ECMWF, as did many other short-range models. Later on Sunday, the NAM weather model flipped east with the other models, before coming back early on Monday to agree with the ECMWF model. Through the storm, the ECMWF model was unwavering with its endorsement of a major blizzard, and almost every major forecasting bust I have seen has come on the heels of the ECMWF weather model. What happened is that the ECMWF/NAM modeled the preliminary part of the storm well, so many took their solutions verbatim and forecast those to the public, ignoring some of the other weather model guidance that continued to trend to the west.
This is a victory for numerical weather guidance as well. The RGEM weather model (regional Canadian) did very well as it predicted proper banding placement, though still being a little overdone. The HRRR weather model did very well too, showing from the beginning on its 15-hour runs that the storm would be east. The model ended up being too far east and did trend to the west (as expected), but not enough to save the region. The real forecasting error came at the expense of an undue belief in the European weather model. The NWS forecast was entirely predicated on that model verifying almost perfectly, and when they’re calling for a historic blizzard and 24-36 inches, I can say it’s hard not to call for at least some very impressive snowfall totals, even if not to that level. There were plenty of warning signs from other weather model guidance that an east trend could occur, but these models did not handle the last storm well, which was one of many reasons they were generally discarded. I do have a policy through a storm (however unpopular) to consult all weather model guidance and attempt to use accordingly. That’s one of the reasons why in updates I referenced in Premium emails and updates those two model “camps” and talked about how one gave the region 6-12/12-24 inches and the other gave the region over 2 feet. No matter how a weather model does with one storm, it can very easily fail you on the next, and putting all of your eggs in one basket can lead to an epic bust like the one we just saw.
No doubt we were extremely close to a record storm. The upper level low at 500mb that I had been watching closed off about an hour to an hour and a half later than the time I projected it needed to for the worst conditions to move into New Haven and Fairfield County. Had that upper level low closed off sooner, it would have pulled the surface low a little west like it was supposed to and we would have seen much heavier precipitation for a longer amount of time. Simply, at the 500mb level in the atmosphere I like to look at so closely, a defined low pressure center did not form early enough to “capture” that surface low, and we ended up just escaping from what could have been a much stronger blizzard.
There are some other factors here, too, that caused many to overestimate this storm. Temperatures were very cold throughout, both at the surface and in the atmosphere. This caused many to forecast high-ratio snow that should be able to overcome lesser liquid amounts by stacking quickly. They forecast ratios of 15:1 or 20:1 in spots which was frankly irresponsible. Due to the nature of the storm and the gusty winds, dendrite crystals were likely going to be shattered aloft, resulting in the very fine, powdery snow that did fall. We saw this exact same thing happen in a storm a year or two ago, where the snowfall amounts busted because the ratios were not what were promised. Those forecasting the higher-ratio snow were irresponsible and let the hype get away with them, and it is in those scenarios that people saw the 3+ foot amounts with the storm. Don’t get me wrong, in the strongest banding I saw it possible, and I saw average ratios of up to 12:1 possible in the storm as it ends as higher ratio snow, but some of the reporting on ratios was a little outrageous, and the snow ratio also seemed to bust entirely in some areas as well.
The final forecast ended up being 10-20 inches in Westchester, 12-24 inches in Fairfield, and 16-28 inches in New Haven County. Totals are listed below by county according to the NWS:
RYE 11.0 930 AM 1/27 PUBLIC
HARRISON 11.0 1000 AM 1/27 PUBLIC
MAMARONECK 10.5 1020 AM 1/27 PUBLIC
MOUNT VERNON 10.1 500 PM 1/27 PUBLIC
YONKERS 10.0 1100 AM 1/27 PUBLIC
SCARSDALE 10.0 100 PM 1/27 PUBLIC
HARTSDALE 9.9 400 PM 1/27 PUBLIC
EASTCHESTER 9.8 1100 AM 1/27 PUBLIC
DOBBS FERRY 9.8 1115 AM 1/27 PUBLIC
LARCHMONT 9.5 900 AM 1/27 PUBLIC
BRONXVILLE 9.5 900 AM 1/27 PUBLIC
SOMERS 6.0 1200 PM 1/27 TRAINED SPOTTER
MOUNT KISCO 6.0 900 AM 1/27 PUBLIC
ARMONK 5.7 800 AM 1/27 PUBLIC
HASTINGS-ON-HUDSON 5.5 1101 AM 1/27 PUBLIC
WHITE PLAINS 5.3 900 AM 1/27 TRAINED SPOTTER
KATONAH 4.5 1050 AM 1/27 PUBLIC
PEEKSKILL 2.4 630 AM 1/27 TRAINED SPOTTER
NEW HAVEN COUNTY…
GUILFORD 17.0 1030 AM 1/27 TRAINED SPOTTER
MERIDEN 12.5 100 PM 1/27 CT DOT
MILFORD 11.2 930 AM 1/27 PUBLIC
WEST HAVEN 11.0 1112 AM 1/27 PUBLIC
NORTH HAVEN 9.5 940 AM 1/27 TRAINED SPOTTER
OXFORD 9.0 800 AM 1/27 TRAINED SPOTTER
BEACON FALLS 9.0 100 PM 1/27 CT DOT
NEW HAVEN 8.5 100 PM 1/27 CT DOT
SOUTHBURY 6.5 100 PM 1/27 CT DOT
BETHANY 6.0 1100 AM 1/27 TRAINED SPOTTER
WATERBURY 5.5 100 PM 1/27 CT DOT
WESTON 14.0 1200 PM 1/27 PUBLIC
NEW CANAAN 12.3 100 PM 1/27 CT DOT
DARIEN 9.5 100 PM 1/27 CT DOT
NORWALK 8.0 1100 AM 1/27 PUBLIC
DANBURY 7.0 500 PM 1/27 TRAINED SPOTTER
FAIRFIELD 6.7 1200 PM 1/27 TRAINED SPOTTER
GREENWICH 6.5 115 PM 1/27 PUBLIC
SHELTON 6.3 1250 PM 1/27 PUBLIC
RIDGEFIELD 6.0 900 AM 1/27 TRAINED SPOTTER
BRIDGEPORT 6.0 100 PM 1/27 NWS COOP
BETHEL 5.0 632 AM 1/27 PUBLIC
BROOKFIELD 3.6 700 AM 1/27 TRAINED SPOTTER
Going through the data, it becomes clear that the storm busted. A few things stood out to me, though. The first was the consistency of very high amounts right over the CT border in Westchester County, along with the fact that some regions did report over a foot of snow, showing again how extreme the banding was where others areas received around 6 inches. Some of these reports were from the middle of the storm still, so this is not entirely final, but anything after 11 AM is essentially finalized. Going over into New London County, we actually see accumulations upwards 2 feet in some areas with around 2 feet of snow throughout. It’s there that the banding set up I expected across most of New Haven County and part of Fairfield County, the same banding that the NWS said today was “near impossible to predict ahead of time.” So, just in case you didn’t believe me about what a close call this was, you can see how really we were just spared.
Finally, forecasters exuded stronger confidence than existed in many forecasts; Premium email updates through the storm portrayed exactly what a difficult forecast it was, but NWS forecasts did not. A combination of over-reliance on one key piece of forecasting guidance and a misrepresentation in forecast confidence helped lead to one of the biggest busts ever in New York City blizzard-forecasting history. From the meteorological perspective, as frustrating as it is to miss a forecast like that, much good can come from it as well. Now, we will see less of a reliance on individual weather models and an abundance of caution in future events. And maybe we can get additional funding to upgrade some of our computer models to perform even better. The recently American upgraded model, the GFS, actually performed fairly well this storm. It surprised many after previously not being worthwhile in such a short range, and so that is something else we have learned from the storm.
We thank you for bearing with us through such a challenging forecast, and look forward to serving you through future storms. Please stay tuned for the latest, as 2 winter storm threats have already appeared in the coming 7 days.