Explaining the forecast bust

Although our forecasts were lower in snow totals for the NYC area than some other outlets at the time they were issued, we still feel like we could have done a better job. Apologies go out for a missed forecast. But, as scientists, we always try to do our best to analyze what went wrong, and how this could help improve forecasting down the road. We did have lots of ideas and gave some warning about the uncertainties with the storm and the beginnings of the south trend in this article, as well as giving a brief explanation of why the storm was trending south in this article, but we did not reflect these thoughts strongly enough in our forecast maps. We will begin dissecting the several aspects that went into the forecasts and how everything changed.

1) Previous computer model forecasts not only consistently showed a major snowstorm, but they made sense given the general pattern they were forecasting.

Sometimes when computer models forecast a snowstorm, they may come up with what we call a “pulling a rabbit out of a hat” solution, where lots of features luckily and coincidentally fall into place, in a general pattern that does not necessarily support these features falling into place. That does not mean snowstorms are impossible in those scenarios, but they are less likely.

This time around, however, there were many favorable features, at least initially. A relatively potent southern stream shortwave ejecting into an Arctic airmass, and pieces of energy from the northern stream of the jet stream were diving down and interacting with the southern stream, helping to pull more moisture northward.

The 00z Friday European Model run showed quite the favorable pattern for a major snowfall for New York City. But small changes with the handling of the Polar Vortex and associated energy led to exponential shifts in the eventual track.

The 00z Friday (2/28) European Model run showed quite the favorable pattern for a major snowfall for New York City. But small changes with the handling of the Polar Vortex and associated energy led to exponential shifts in the eventual track. Image credit goes to the WSI Model Lab.

The above is the 500mb forecast from the European Model generated overnight Friday morning, valid for Monday morning. It showed the initially strong piece of energy in the West ejecting eastward and interacting with a piece of energy rotating around the back side of the Polar Vortex. This pulled the system a bit north and helped moisture to blossom. Additionally, the Polar Vortex to the north was situated perfectly to provide a strong temperature gradient between the warmth to the south and the cold to the north. The Polar Vortex was also elongated from southwest to northeast, which allowed heights to rise out ahead of the storm system, instead of fall. Additionally, partially because of the elongation, the lobe of energy on the south side of the Polar Vortex was not compressing the flow in New England — its strongest influences stayed in Canada.

The storm did not have to generate its own cold air, and despite the system gradually weakening as it headed eastward, there was a beautifully-placed strong temperature gradient throughout the country to make precipitation blossom. The phasing wasn’t a result of a rogue piece of energy — it was energy that was well-sampled from the northern stream. The pieces were there, and the evolution made sense. Here is what that European Model run ultimately showed at the surface:

Friday morning's European Model run showed a major snow event, with over five tenths of an inch of liquid equivalent precipitation in a 6-hour period alone. Image credit goes to the WSI Model Lab.

The 00z Friday (2/28) European Model run showed a major snow event, with over five tenths of an inch of liquid equivalent precipitation in a 6-hour period alone. Image credit goes to the WSI Model Lab.

That run of the European Model gave the entire area over a foot of snow, with snowfall rates exceeding 1″/hr between 7:00am and 1:00pm on Monday. Of course, that did not even come close to verifying.

2) Just because the models were consistently showing snow, that does not necessarily mean that the setup for snow had a lot of room for error. Subtle changes to the pattern led to exponentially different results. 

Often times, when a model consistently shows a certain solution, it increases the chances that it will happen, since if a constantly changing atmosphere is yielding the same results, that gives those same results much more room for error to occur. However, this setup was an exception, because it relied so heavily on the exact orientation of the Polar Vortex. Thus, just because some models were very consistent in showing snow, that does not mean they were necessarily consistent with their handling of the Polar Vortex. Initially, a few small tweaks to the way the models were handling the Polar Vortex did not mean as much, since models were still showing the northern and southern stream interacting. This explains why they were so consistent. However, once we headed towards Friday and Saturday, models reached their “threshold” with how much they could change their handling of the orientation of the Polar Vortex, yet still show snow. Once that threshold was crossed, rapid southern shifts occurred.

Below is what we deemed the “threshold” run. This was Friday’s 12z European Model run, which was the very run that followed the one we just showed.

Friday's 12z European Model run, which was the run that immediately followed, already started with unfavorable trends for snow. Image credit goes to the WSI Model Lab.

Friday’s 12z (2/28) European Model run, which was the run that immediately followed, already started with unfavorable trends for snow. Image credit goes to the WSI Model Lab.

Notice how the Polar Vortex became a bit less elongated, as the arrows indicate a more north-to-south orientation. This helped to compress the flow a bit more, which shunted the southern stream wave further south. This in turn led to less interaction between the northern and southern stream. In fact, it was pretty close to showing almost complete separation. This run still showed pretty decent snow for our area, but considering how close it was to showing complete separation between the northern and southern stream, we were only a very small tweak with the Polar Vortex away from there being no snowstorm at all. That’s why we called this run the “threshold” run, since this was the absolutely least elongated the Polar Vortex could have been while still showing snow.

Below, we will see the pattern that actually verified.

Here is the pattern that actually resulted for the storm. Note how different this image is from what had previously been forecast. Image credit goes to the WSI Model Lab.

Here is the pattern that actually resulted for the storm. Note how different this image is from what had previously been forecast. Image credit goes to the WSI Model Lab.

Note how the orientation of the Polar Vortex completely changed. It is not even close to the initially elongated shape we previously saw, as its orientation is almost due north-to-south. This significantly compressed the flow, which can be partially illustrated by how far south and east that southern lobe penetrated — it reached Northern New England. This compressed the heights even further out ahead of the storm, so there was no room for the storm to turn northward. Additionally, this squeezes the strongest temperature gradient further south instead.

Arguably more important is how the compression of the Polar Vortex led there to be no interaction between the northern and southern stream — they remained mostly entirely separate. This allowed the northern stream to run out ahead of the southern stream, instead of interact with it. When two pieces of energy phase, the trough gets enhanced, which means the heights out ahead of the storm also rise. But when the northern stream misses the phase and runs ahead of the southern stream, the heights get compressed ahead of the storm, which prevents any height rises. Thus, not only was the northern stream not aiding the southern wave, but it was actually inhibiting its progress. Once the threshold for the Polar Vortex orientation was crossed, these were the results we saw.

3) Several other factors such as climatology, the seasonal trend, and how models usually handle the Polar Vortex were being misinterpreted. 

The 00z Friday GFS Model run also showed heavy snow for the entire region. Image credit goes to the WSI Model Lab.

The 00z Friday (2/28) GFS Model run also showed heavy snow for the entire region. Image credit goes to the WSI Model Lab.

As we head into March, temperatures tend to warm, particularly in the Southern States, due to the stronger March sun angle. This intuitively means that storms should trend north, given that warm air advection can often be a trigger for storms moving northward. Or, at the very least, this can mean that cold air should not be able to penetrate far enough south to suppress a storm, given residual warmth to the south.

However, this is particularly flawed reasoning, since it only quantifies the past, and does not quantify the present. It already goes against climatology to have such a strong piece of the Polar Vortex parked in Southeast Canada. The pattern supported a large pool of cold air that goes against climatology, so using climatology to justify a north                                                                               trend does not make sense.

The 06z Friday run of the GFS Model was so far north that it actually gave rain/snow issues to lots of Central NJ and southward, and even heavy snow for Boston. Image credit goes to the WSI Model Lab.

The 06z Friday (2/28) run of the GFS Model was so far north that it actually gave rain/snow issues to lots of Central NJ and southward, and even heavy snow for Boston. Image credit goes to the WSI Model Lab.

Additionally, the seasonal trend all winter had been for storms to trend northwest within the last three days, at least for the most part. We are firm believers that seasonal trends can certainly have some weight in a forecast, and we will admit that part of the reason why we did not significantly lower our totals was that we thought the storm would creep back to the north in the last two days — definitely not back to the original major snowfall solutions, but enough for a light-to-moderate event. The north trend never happened. The south trend never stopped.

Lastly, models in the past have overestimated the strength of the Polar Vortex, which is why sometimes cold outbreaks do not live up to their initial hype, and part of the reason why so many storms — not just in this winter — have trended north within the last few days of modeling. But as mentioned before, the issue regarding the storm was not the strength of the Polar Vortex, but the orientation of the Polar Vortex. This means that it mattered very little if the Vortex ended up being weaker than modeled, because it was already in a very unfavorable orientation for major snowfall to reach the New York City area.

All of this spells not only a rapid south trend in the models, but a complete forecasting nightmare. The Washington, D.C. area ended up getting a moderate snowstorm out of this, as Dulles Airport received 4.9″ of snow, and some suburbs received 6-8″. Parts of Atlantic County, Ocean County, and Cape May County, NJ received in excess of 6″ of snow, but no one north of southern Monmouth County, NJ, received 3″. Even Philadelphia only received 3.4″.

We live and we learn.

 

 

 

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Join the discussion...