But it's not perfect, and that's especially true when it comes to snow. Meteorologists, like us, care about every detail of a snow forecast, and they still sometimes struggle to get it right. This is clearly the case with Friday night's forecast. We are not at fault for lack of trying; The tools we have are not good enough yet. Ask Boston meteorologists how they felt about a forecast of 7 to 13 inches of snow last Monday when only 0.1 inch was measured in the end. There was sound logic and sound effort to these predictions, but the storm has subtly changed.
A similar dynamic played out here Friday night, in a less dramatic way.
Our Capital Weather Gang forecast called for about 1 to 5 inches of snow across the Washington region from south to north. Just a trace of up to a few inches fell. The forecasts weren't wrong everywhere, and it's not the worst miss ever, but it is among the most notable events of the past few years.
Here is the amount of snow that fell in some selected locations:
- Damascus, Westminster and Elkridge: 3.0 inches
- Gaithersburg: 2.8 inches
- Loftsville: 2.5 inches
- BWI Marshall Airport: 2.2 inches
- Rockville and Laurel: 1.8 inches
- Crofton: 1.5 inches
- Silver spring: 1.4 inches
- Herndon: 1.2 inches
- Dulles Airport: 1.1 inches
- Fairfax: 1.0 inches
- Northeast Washington: 0.8 inch
- Springfield: 0.5 inch
- Reagan National Airport: 0.1 inch
The amounts missed the mark as the snow arrived later, left earlier and fell lighter than we expected.
Why did we make a mistake? In short, the models we used as forecasting tools were simulating significant amounts of precipitation and the heaviest amounts were shifting northward. On Friday morning, when we analyzed about six different models, all of them supported the expectations we had made.
But problems started to appear in the afternoon. Ironically, about 15 minutes before the National Weather Service issued a winter storm warning for our western counties, the European computer model — which received the highest skill score of all the models — indicated that no location in the region would see snowfall by warning criteria, Which is ridiculous. 5 inches or more. Its projections indicate that if each flake was stuck, amounts would range from about 1 to 4 inches from south to north.
We were reluctant to reduce our expectations based on the European model alone. Although it is the most accurate model on average, it is not the best performer in every storm. Snow forecasts were slightly lower than other models. But it is a harbinger of a sharp decrease in quantities in other patterns during the afternoon and evening.
The big warning sign was that the high-resolution NAM model — which is among the most reliable in the short term — had cut its forecast snowfall by 50 percent between simulations released Friday morning and Friday evening.
The decline in the NAM pattern prompted us to lower our overall forecast at 9:30pm on Friday. Importantly, another high-resolution model known as HRRR also consistently showed massive amounts of snow since the morning — which was further evidence that more aggressive snowfall forecasts — such as the forecast of 4 to 6 inches of snow from the US model (GFS) — It will happen. Be mistaken.
Of course, the adjustment we made at 9:30pm to lower amounts was not enough. Actual snow totals were on the low side of our revised forecast. When the forecast gets worse, it is psychologically difficult to turn it off completely, and there are still some models (such as the US model) that support a forecast of more snow.
All this talk about models might raise the question: Why do forecasters like us rely on them so much? It's simple – they are the best weather forecasting tools we have, and without them, weather forecasts would be set back many decades. We get a lot of blizzard forecasts very close to the right, largely because of the models. However, it is guidance, not gospel. Skilled forecasters analyze models and apply their experience and knowledge to adjust for their errors and biases. Have we not done enough with Friday night's forecast?
One can make this criticism but without appreciating our thought process. Weather systems that come in from the west like Friday night are often deprived of moisture, especially after they pass over the Appalachian Mountains. Should we have been more skeptical of typical simulations of at least 4 to 6 inches of snow? Temperatures were also only slightly cold enough for snow to fall. High temperatures on Friday reached 45 to 50 degrees and were still hovering around 40 at 10 p.m. just two hours before the rain started. Shouldn't we take that into consideration?
Trust me when I say we took these two factors into consideration. For these reasons, I wrote the following in our summary of these expectations:
Amounts could end up at the lower end of the forecast or in our “depression” range if precipitation intensity is slightly lighter than simulated and temperatures are on the warm end of the forecast. The risk of avalanche is highest in downtown Washington and to the south, where temperatures above freezing can eat away at large amounts. Also, if the snow moves more quickly than expected, that will also reduce the total. Some models show only a few hours of moderate to heavy snow.
—Cross-Community Action Group
Although we knew that humidity could be limited and temperatures were marginal for accumulation, there were two meteorological components that we thought could overcome these factors. First, the models were showing the possibility of strong “frontalization” – the process in which colliding air masses create a local front and a narrow area of particularly heavy snow. Second, they showed cold air at about 5,000 feet above the ground. Our thinking was that these two factors would work together to cool the air and eliminate snowfall.
The frontal generation area ended up shifting to the north — and parts of east-central Pennsylvania were covered in up to 14 inches of snow. One of the most difficult aspects of forecasting snowstorms is knowing where very localized bands of heavy snow will form, because models are bad at this. In the days leading up to this forecast, models consistently showed strong potential for such a range over Washington. I just converted.
I hope this perspective has provided a useful lens, under the hood, into our forecasting process. The goal of this review is not to apologize for our predictions but to make clear that incorrect snow forecasts like these are still just a fact of life. The National Weather Service and TV meteorologists predicted similar or even more amounts than we expected. But, every now and then, despite our best efforts, we will fail to achieve the goal because the tools we have are not good enough.
In such situations, the best we can do is continue to try to make the best possible predictions and express uncertainty. I believe we did that in this case, although, with the benefit of hindsight, we should have done so more forcefully.
Despite these missing forecasts, there are many exciting developments in weather forecasting that, over time, will make these snow predictions less likely. For example, rapid advances in artificial intelligence and machine learning show great promise for improving predictions, but this technology is still in its infancy.
