Well - where do I begin and how do I begin? If you were following the forecasts for this storm you may have noticed a theme; an increase in snowfall forecasts moving through the weekend with some significant increases in some spots Sunday and even Monday morning. Then Monday afternoon came...the 12z European forecast model made a significant jump south, then subsequently various other forecast models followed this theme. At one point, it was becoming a concern that even northern Connecticut may see nothing. This was a stunning turn of events. We've seen forecast models make big jumps 36-48 hours out, even 24 hours out, but 12 hours out, almost unheard of in this day in age of advanced modeling. During the day Monday, forecasts drastically started to change with significantly lowering snowfall totals across northern areas and increasing totals towards the south coast. Then Monday evening and overnight came...and well the rest is history.
So, what happened? Why did we see forecast model guidance make a near 180 Monday afternoon, make a significant shift south, only for the storm to end up a bit farther north? Is this a case of a horrific performance by our forecast models or was something just overlooked? The easy answer here is a mixture of both. But the answer to this is much more complicated and complex than just a one word answer. Below, we will revisit the storm from the perspective of our forecast models and then apply good ole meteorology. There is a phrase which sometimes gets tossed around in the field of meteorology, "Meteorology not modelology". What does this mean? In a nutshell, forecast is much more than just taking the output of a model(s) and saying, "this is my forecast". Forecasting is an art - its is a skill and the skill level is highest when a forecaster takes ample time to digest all the information and apply their knowledge of meteorology to create a forecast. A forecaster who is skilled in this department can develop a great understanding of what forecast models may have a better handle, where some forecast models may struggle more than others, and can sniff out trends before they even happen.
Below, we will look at the NAM, GFS, and European forecast models. Yes, there are many other models out there but I don't want to make this post any longer than it will turn out to be and the context of what is being discussed can be applied to other models. For a large part, this is more for visual referencing and I am using College of DuPage as my data source, therefore, I am able to be consistent with the outputs. Once inside 24 hours the HRRR will be incorporated as well.
Forecast Models - Initialized 12z Sunday, February 11, 2024
Here is the forecast 500mb vorticity field from Sunday morning, valid for Monday evening. We want to focus on the northern stream/southern stream energies (shown in the reference just below).
Notice there are some large differences between each model in the strength, structure, and placement of the southern stream energy with some structural differences within the northern stream energy. We can also note some differences within the Pacific-Northwest and northern Inter-mountain West region which likely had some influence on the downstream energies.
What is the importance with these northern stream and southern stream energies and why are these differences noteworthy? In the event of a stronger southern stream energy, this would increase the likelihood for the storm to track north, passing just south of southern New England, favoring heavy precipitation across much of the region. However, with the northern stream and associated confluence, there would be an extent as to how far north this storm could get. In the event of a weaker southern stream energy, the storm would be more likely to track well south of the region, with much lighter, if any precipitation. The northern stream was a bit more complex. In the event the two energies phased (which was becoming unlikely) or the northern stream had at least some interaction, this would help drive a more north track, increasing the potential for heavier precipitation across the region.
So, how did these differences at the 500mb level translate to the surface? While there are differences between all three models, notice how the European model is south and east with the western edge of the precipitation shield. This becomes important as we get closer to the event:
Let's fast forward to Tuesday morning and see what the 12z runs of the NAM, GFS, and Euro were forecasting in terms of QPF for early Tuesday morning. While there are certainly some differences, the one striking aspect of this is heavy precipitation across all of southern New England. This signal (amongst other factors) was increasing confidence for a significant region-wide storm:
This intense precipitation had substantial merit based on the following:
1. Surface low pressure passing south and east of the region (shown just above).
2. Mid-level/low -level pressure developing just south of the region and tracking northeast (shown just below).
3. Entrance region of the upper-level jet stream favoring upper-level divergence (not shown yet).
Each of these forecast models was developing a low pressure at the 700mb level around or just south of Long Island and tracking it east-northeast. This is nearly textbook for a band of extremely heavy precipitation to move right across the region:
Forecast Models - Initialized 12z Monday, February 12, 2024
Fast-forwarding to Monday's 12z runs the NAM and GFS were continuing to advertise heavy precipitation across much of southern New England in similar fashion to Sunday's guidance. The Euro, however, made a stride south. When this occurred, this started to raise some eyebrows:
Forecast Models - Initialized 18z Monday, February 12, 2024Forecast model guidance Monday afternoon continued with the theme of shunting the heaviest precipitation south, with some guidance barely getting heavier precipitation to the Connecticut-Massachusetts border. The Euro was leading the way with the south trend:
Forecast Models - Initialized 0z Tuesday, February 13, 2024 (Monday evening)
As Monday evening forecast model guidance was rolling in, a mere several hours before the onset of precipitation, forecast models continued with the more south track and a south shunt of heaviest precipitation. It was now becoming apparent along and north/west of I-84 very little snow was going to occur with the highest totals across southern Connecticut and into Rhode Island and perhaps Southeast Massachusetts:
The 0z NAM, GFS, and European forecast models had cut back significantly on the amount of precipitation which was expected to fall across southern New England. Based on these forecast totals, it was difficult to believe northern Connecticut (along and north/west of I-84) would see more than 3-4 inches with totals upwards of 6-10 inches across southern Connecticut into portions of Rhode Island and southeast Massachusetts. This was turning into one of the largest model busts inside of 24 hours in recent memory:
The ResultSo, what was the result? How did the storm fare. Below, is a compilation of snowfall totals received by the National Weather service:
What...in...the...heck...? What happened across central Connecticut? Forecast model guidance, just hours beforehand, had cut back on forecast precipitation so much, it was starting to become clear northern Connecticut may not even see a few inches with barely 3-4 inches for the Hartford area, yet over a FOOT occurred in a narrow swath from northeast Connecticut southwest through Hartford into Danbury with 7 inches as far north as Windsor Locks. What an absolute wild, bizarre turn of events. Within 24 hours we go from what looked to be a crushing from the Massachusetts Turnpike into central Connecticut to nothing from along the Massachusetts turnpike into northern Connecticut and barely more than a few inches into north-central Connecticut to a narrow strip of a foot-plus from northeast Connecticut through Hartford southwest into Danbury. How does this happen?
Theory
Did our computer forecast models let us down? Did they just perform one of the greatest busts in recent memory, and perhaps one of the greatest busts inside of 12 hours ever? Was something overlooked by forecasters? As stated in the opening, there is a combination of these two factors. There is no denying forecast models struggled, however, there were hints and signals this scenario was still likely to occur. So why were these missed? Below, I will present some of my theories and ideas. I want to state that this is just my opinion as a professional meteorologist and just because this is my opinion does not mean it is correct. Others may disagree and have a different viewpoint. That is totally fine, differing viewpoints is what drives this filed and is it what drives a forecaster to want to better themselves.
1) The use of model snowfall maps.
Don't get me started on these. One of these days I am going to dedicate a blog post to these so I won't provide much specifics about this right now. Since modeled snowfall maps (which aren't produced by forecast models, they're produced by vendors) have been introduced, I believe they have led to lazy winter weather forecasting. If you follow weather on social media, these maps are so widely used and tossed around, especially when they're spitting out large numbers.
For this storm, there was a lot of analysis going on by using trends in the snowfall output to justify changes within the storm. My simple answer here is - that is a terrible idea. If a forecast model (or models) are showing a decrease (or increase) in precipitation, a forecaster should be asking themselves, why that is and whether that makes sense.
2) Solely focusing on QPF maps.
QPF maps are an incredible forecasting tool. While forecast models can struggle greatly with QPF, assessing QPF output and combining this with an assessment of the situation is a major part of forecasting snowfall accumulations or rainfall accumulations. Like the snowfall maps, however, QPF trends were being analyzed to visualize the trends but again, there needs to be a, "does this make sense" thought process behind this.
3) The importance of Fundamentals
Forecasting is an art. Being a good or great forecaster requires knowledge of forecast models, knowledge of historical weather and patterns, and an understanding of how the atmosphere works. A great forecaster isn't afraid to be wrong, learns from their mistakes, and a great forecaster understands there is much more to a forecast than being right or wrong. A forecast that is correct, but for the wrong reasons is not correct. Forecasting is much more than looking at the output of a forecast model and then rolling out some thoughts. Creating a great, accurate requires a great deal of time.
I saved this point for last because I wanted to elaborate into greater detail on this aspect. Ultimately, applying fundamentals to the forecasting process, even when forecast model guidance was decreasing the precipitation totals may have prevented the panic of reducing snowfall totals across much of the region (outside of along and north of the Massachusetts turnpike). Above, we looked at how the NAM, GFS, and European forecast models were cutting down precipitation totals and shifting the highest totals southwards. But what we didn't look at was whether this made sense.
Anyone who understands southern New England snowfall climatology understands that our heaviest snows and biggest storms tend to occur when low pressure at the surface takes a northeast trajectory and passes through the latitude/longitude of 40°N/70°W. Additionally, when low pressure centers develop just southwest of Long Island and strengthen as they move northeast this favors the heaviest banding of snow to impact the region. What was just described here within this paragraph is exactly what forecast models were indicating multiple days out. This is why, especially as we got closer, the storm was being hit harder and forecasts for snow were increasing. There was increasing confidence and consistency in this exact scenario occurring.
So, what happened in these last 24 hours. Did something drastically happen with the surface low pressure track and development/track of the low pressure centers at 850mb and 700mb? Let's take a look.
0z NAM - Initialized Tuesday, February 13, 2024 (Monday Evening)
850mb Vertical Velocity and Height - 15z Tuesday (Left) 18z Tuesday (right)
Vertical velocity is a measure of how quickly air is rising or sinking. The more positive the value, the more rapidly air is rising, the more negative the value, the more air is rapidly sinking. This chart also displays the height level (in meters) the 850mb level can be found. I've outlined where the NAM had developed the 850mb low and where it was tracking. If you look at the vertical velocity values, you'll notice extreme upward vertical motion (rising air) right within the center of the low pressure. Scrolling back up to the precipitation field, you'll see this matches very closely with where the 0z NAM had the heaviest precipitation. Going back to fundamentals, we would expect or anticipate the strongest upward vertical motion to be northwest of the low. I've outlined where this would be expected in purple:
700mb Vertical Velocity and Height - 15z Tuesday (Left) 18z Tuesday (right)At the 700mb level, the NAM was not forecasting a closed off low pressure at 15z, however, by 18z a low was closing off just south of Cape Cod. We continue to see the highest vertical velocities very close to the low center. I've once again outlined where you would expect the strongest vertical motion to be in purple:
0z GFS - Initialized Tuesday, February 13, 2024 (Monday Evening)
850mb Vertical Velocity and Height - 15z Tuesday (Left) 18z Tuesday (right)
The 0z GFS was similar to the NAM at the 850mb level with 850 low development south of Long Island and tracking northeast. Like the NAM it was pegging the greatest vertical velocities towards the center of the low. Based on 850 low development and track you would expect the greatest lift to be north and to the west:
700mb Vertical Velocity and Height - 15z Tuesday (Left) 18z Tuesday (right)
At the 700mb level some noticeable differences versus the NAM and if you recall, even during the 0z run the GFS was more north with the heaviest precipitation and that matches up better with what you'd expect based on the low tracks:
0z Euro - Initialized Tuesday, February 13, 2024 (Monday Evening)
850mb Winds and Height- 15z Tuesday (Left) 18z Tuesday (right)
Unfortunately, there is no vertical velocity plot for the European forecast model on College of DuPage, however, we can still view 850mb and 700mb low via height field overlaid with wind field. While the Euro is farther south with the 850mb low, based on the track you would still expect heavy precipitation at least back to the I-84 corridor:
700mb Winds and Height - 15z Tuesday (Left) 18z Tuesday (right)
Similar to the 0z NAM, the European model was not developing a closed low pressure at 700mb by late Tuesday morning, however, it was by early Tuesday afternoon. Based on the development and track, you would expect the strong upward vertical motion north and west of the low and right across southern New England.
Assessment
While forecast model guidance continue to shift the axis of heaviest precipitation south and subsequently lower precipitation totals, based on how forecast models were continuing to evolve the low-levels and mid-levels of the storm, this did not make sense. Forecast models were placing the greatest vertical velocity values close to the low pressure centers, which was a vast change from earlier runs where forecast models had the strongest vertical velocity values moving across southern New England and northwest of the low pressure tracks.
There was another signal in place for enhanced upward vertical motion across southern New England. Forecast models had southern New England in the right entrance region of the upper-level jet, a favorable quadrant for rising motion:
Verification
As Tuesday morning arrived, many were likely surprised as majority of Connecticut was experiencing not only heavy snow, but portions of central Connecticut were into some prolific snow with rates exceeding 2 inches per hour! What the heck happened? SPC mesoanalysis showed intense 850-700mb frontogenesis traversing across Connecticut, which is exactly what you'd expect based on the forecast low tracks:
So the question now remains, why was forecast model guidance continuing to shift south with the axis of heaviest precipitation and why did they back off the the heaviest precipitation north and west of these low pressure tracks? Unfortunately, that is not something I have much of an idea on, however, forecast models can underdo precipitation north and west of the low tracks. As we saw with the 0z guidance, forecast models were pinging the greatest upward motion very close to the low centers. Typically with these bands of heavier precipitation induced by developing and strengthening low pressure centers, as the low pressure centers develop and mature, a band of heavy precipitation will develop well north and west of the low pressure centers, sometimes as far as 50-75 miles north and west. In situations where these lows become stronger, the banding will then collapse closer to the low pressure center. In this situation, the forecast models were bypassing the initial step here and just developing everything in close proximity.
Ultimately, all forecasters should have held firm with original forecasts and not go into panic mode just because forecast models were cutting back on QPF and because model snowfall maps were indicating less snowfall. While this certainly should have been a flag, or a cause for concern, a more thorough and detailed analysis should have been conducted to determine why that was occurring and whether it made sense. It is also very likely the evolution of the northern stream energy, southern stream energy, and their interaction ended up being different than forecast models anticipated 12 hours out. These may have all ended up evolving in a way favoring a bigger hit.
Below, was my forecast for this storm (top) versus what occurred. One could say all I had to do was just shift this down some miles and tighten the gradient. If I had more time/energy I would have made an updated map but l would have not made those changes. I would have went significantly less for Connecticut and maybe do 4-7 inches southern and 2-4'' northern so it would have been much worse if I did something so this map makes the forecast better than it would have been. I would have been JUST AS GUILTY for lowering totals and just taking the reduced precipitation totals and running away. I did also scream bust Monday evening.
For anyone involved in weather forecasting, there is no doubt this event was a tremendous learning experience. We can't forget the basic fundamentals of forecasting and sometimes we need to rely on our knowledge and understanding of the atmosphere over computer forecast models. While computer forecast models are excellent and allow us to do our jobs, they are not perfect and we must remember, there is a recent they are called model guidance.
This was also a great test in communicating weather and communicating uncertainty, especially as we get within hours of a weather event occurring. We live in a day and age where everyone expects perfection and everyone wants pinpoint details days out. Unfortunately, we must remember it doesn't always work out that way.
