Interactive graphic: how much snow fell vs. forecast

A tongue-in-cheek forecast for Snowquester, which — in reality — may have been more effective than many forecasts provided. (Brad Panovich)

The best forecast for Snowquester was one we could not issue with a straight face, and one most Washingtonians would have ridiculed: Rain, sleet, and/or snow likely — heavy at times — with snow accumulations of 0-14 inches.

As most people appreciate, weather forecasting is not an exact science. In certain circumstances, which present themselves often in the D.C. area, an error of one or two degrees can be the difference between bare ground and a crippling snowstorm.

Related: Forecasts and probabilities by Joel Achenbach

The D.C. region sits in the murky transition zone between the mountains and the ocean, where contrasting flows of air collide. This juxtaposition presents a stormy quandry — which will win: the Arctic air from the northwest or the mild, marine air from the east?

To help answer these kinds of questions, we use computer models — complicated programs that ingest millions of data points and process thousands of mathematical equations. These models are usually very accurate and have led to incredible strides in weather prediction in the last several decades.

But for certain problems, these models have their limitations. Deciphering the exact location of the rain-snow line is one of those problems.

In forecasting Snowquester, we examined multiple models from the United States, Europe and Canada, and most suggested substantial amounts of precipitation and temperatures just cold enough for it to fall as snow. In fact, as the event drew closer, the models generally trended colder with their predictions.

But there were some warning flags.

The European model, which is the most accurate (in general — not always; performance can vary from storm to storm), suggested that while temperatures at high altitudes were cold enough for snow, near the ground the flakes could melt. Not only that, it simulated about half the amount of precipitation as other models.

Link: Second rate U.S. numerical weather prediction: Why you should care

A big snowstorm in D.C. requires not only the cold air, but also substantial precipitation. If one of these ingredients is missing, your snowstorm prospects have a problem. If both are missing, it’s game over.

But forecasters, including ourselves, National Weather Service and broadcast media, were influenced by the other models, some of which predicted a double-digit dump of snow on the District. We saw forecasts — on the order of 5-10 inches or so — as a fair compromise between the European shutout and the American, Canadian clobber.

Some critics have fairly raised the question: why are forecasters relying so much on models? They pointed to near 50-degree temperatures the day before the storm and the fact we were predicting snow accumulating with temperatures above freezing. These are legitimate arguments and, believe me, most meteorologists considered them. But we also were able to point to snowstorms from the past were heavy snow fell under exactly these circumstances. If it precipitates hard enough, the rate at which the snow is falling can overcome milder ground temperatures. While this failed to occur in D.C. in Snowquester, locations near Fredericksburg and Richmond — to our south — received 4 to 8 inches of snow as heavy precipitation bands materialized.

Given the problems with the modeling as well as the more intuitive reasons to be skeptical of snow prospects, the most critical error we and every single forecaster in this region made was the failure to communicate the uncertainty in the forecast.

We actually did this reasonably well two to three days prior to the storm. For example, on Sunday, we presented a “lower probability” scenario (given a 30 percent chance) in which “any snowfall accumulations around the city would be limited to an inch or so on grassy surfaces,” and we displayed a complementing graphic. On Monday, we reiterated that there was bust potential:

The bust potential is highest in the immediate D.C. area due to marginally cold enough temperatures for snow. There’s about a 20-30 percent chance it becomes too warm and precipitation is not sufficiently heavy for accumulating snow

But we watered down some of this uncertainty information the day before the storm — a major communication error.

Our headlines and lead paragraphs, for example, overstated our own confidence in snow materializing.

Take Tuesday’s headline: “High-impact, heavy snowstorm to hammer Washington area tonight and Wednesday” Or, this sentence from Wednesday morning’s forecast post opener: “This storm could turn out to be one of March’s top-10 biggest in and around the nation’s capital.” These were over the top and inadequately qualified statements.

Accumulation map we issued the night before Snowquester

To be fair, our snow accumulation map did include confidence levels, which conveyed only “medium confidence” in the 5-10 inch forecast for the District.

But, in the spirit of looking at ways to improve, here’s a revised version of the same graphic which would have more effectively communicated the uncertainty.

Map we wish we had issued: Improved communication of uncertainty (Dan Stillman)

Note that this graphic highlights the bust potential and includes some explanatory text.

Communication of uncertainty is something the entire weather forecasting community should strive to improve.

Few broadcast meteorologists communicate uncertainty information particularly well (Bob Ryan at WJLA being an exception). The National Weather Service did not qualify its 8-10 inch snowfall forecast.

Related: Washington Snow Storm: Really “Wrong”? Why? (Bob Ryan, WJLA)

Chris Strong, the warning coordination meteorologist at the National Weather Service forecast office in Sterling, Va., said his office is working on a pilot project to include more uncertainty information with forecasts.

“We have a few concepts we are developing that we have just begun to test with some of our core partners,”Strong said.

Consumers of weather information must also accept that weather forecasts will not be perfect and that there will be cases where we cannot, with high confidence, predict the amount of snow that will fall in their backyard.

Link: Your responsibilities as a weather forecast user

One of the reasons, as we get closer to the onset of the storm, that we drop some uncertainty information is that some readers want to know the bottom line, without qualification. They view scenarios and percentages as “cop-outs.”

Ultimately, there has to be a sweet spot, where we can effectively communicate uncertainty concisely and effectively while also presenting a most likely forecast. We’re constantly working to find that and came up short in this last case.

We would value any feedback on how to improve.

Related article: D.C. Snow No-Show A Lesson in Forecasting Uncertainty (Climate Central)