The case of the disappearing snow: A stormy week — for the computers
The new, improved U.S. forecast model and others had foreseen major snow for Tuesday. What happened?
The Philadelphia region might see its first snowflakes on Tuesday. But what happened to that threat of a major snowfall? How did it devolve into a prosaic forecast of harmless white fireflies appearing in the air?
The alleged offender is a familiar one, particularly as winter nears: the computer. While last week represented a welcome respite from the traumas of tornadoes and mass power outages, it was a wild one in the world of long-range virtual weather.
Early in the week, computer models were hinting at the region’s first snow of the season for last Thursday. It rained. Then they got into a virtual smackdown over a potentially more serious snowfall due this Tuesday.
The recipe of temperature contrasts and moisture looked promising: A record-cold air mass oozing across the continent — something the models correctly latched onto days in advance — encountering an ocean on simmer, with sea-surface temperatures several degrees above normal.
For the newly upgraded U.S. Global Forecast System, or GFS, which became operational in June, the threat provided an opportunity to watch how it would handle a Northeast winter-type storm threat.
The result? The early returns suggest the improved GFS is unlikely to deplete the atmosphere’s well-stocked treasury of secrets.
Nor is any other model, even the vaunted one operated by the European Centre for Medium-Range Weather Forecasts — the Euro, long viewed within the meteorological community as the gold standard.
It is too early to pass judgment on just how much the American model has improved or how much ground it might have gained on the European.
But in the view of meteorologists such as Glenn Schwartz, the dean of local on-air forecasters, the GFS has a lot of catching up to do. “The Euro is still dramatically superior to the new GFS,” Schwartz said last week as he and other meteorologists struggled to make sense out of the conflicting model guidance.
“It was a really tough forecast,” Niki LoBiondo, a meteorologist at AccuWeather Inc., said Monday.
In the long-range, the Euro fared better with its Tuesday forecast. The GFS went as high as 7.4 inches for Philadelphia; the Euro’s top figure was 2.6.
That was better, but it also was about 2.6 inches more than anyone now expects.
The takeaway: When it comes to advance snow threats, if you don’t like the forecast — wait.
Why is this so hard?
All computer models have problems.
They rely on observations that capture the “initial condition” of the atmosphere, then, using the laws of physics, they calculate how it has changed in the previous six hours and how it might change in the next six.
Unfortunately, the atmosphere is a chaotic, three-dimensional gas that thinks it’s a fluid. That entire sloshing mass is loosely attached to a planet spinning at 1,000 mph and hurtling through space at 67,000 mph. The amount of solar energy reaching the planet’s surface changes every instant.
The atmosphere is imperfectly observed. Most of the planet is uninhabited. About 70% of the Earth’s surface is water, and less-developed countries have priorities other than building weather stations.
While satellites have made a difference, the gaps are so significant that forecasters run “ensembles” — multiple models tweaked to account for what’s missing and possible errors — to come up with likely outcomes.
Small errors can and do snowball over time.
Are the Europeans really better at this than the Americans?
“The new GFS does well sometimes,” said Schwartz, “but also has times that it is far inferior.”
Independent meteorologist Ryan Maue recently shared an analysis showing that the Euro was outperforming the new GFS in forecasting weather-driving upper-air patterns.
Tony Gigi, a former National Weather Service meteorologist who helps run the phillywx.com discussion board, said the recent accuracy measures for the new GFS are lower than those of the old GFS at the same time last year.
» READ MORE: The 'European's' Joaquin forecast was better
Not that the Euro has been perfect. While its scores were better than the American model’s, they weren’t as good as they were last year.
The National Weather Service said it couldn’t grade the performance of the improved GFS vs. the old one because it no longer runs the previous version.
Bob Larson, senior meteorologist at AccuWeather Inc., said he wasn’t ready to pass judgment on the upgrade.
“I’d like to give it more time,” he said. “I would need a larger sample size.”
So what’s a meteorologist to do?
Schwartz and his counterparts never rely on one model. They look at all the reliable ones, including the big three: the Euro, the GFS, and the well-regarded Canadian.
Schwartz said the Euro and GFS ensembles offer “the most accuracy,” but even the Euro sometimes has problems.
In late January 2015, the Euro had Philadelphia entombed under 30 inches of snow. What followed was an all-time bust, a forecast so bad that the local National Weather Service chief publicly apologized for it.
» READ MORE: Flakegate' and the incredible shrinking snow totals
100% chance of uncertainty
For now, expect models to bicker with each other, and even themselves.
The model runs last week for this Tuesday’s likely nonevent were classic examples.
Last Monday morning, the Euro had a forecast of 0.2 inches for this Tuesday in Philadelphia; Monday evening, it was up to 1.2 inches; Tuesday morning, 2.6; Tuesday evening, 1.6; Wednesday morning, 1.2; Wednesday evening, 1.6; Thursday morning, 0.1; Thursday evening, 0.0.
The GFS numbers for the same intervals were 3.5 Monday morning, then 0.0, 0.0, 6.0, 7.4, 0.8, then 3.9 on Thursday morning.
The Canadian went from 11.8 Monday evening to 2.1 on Wednesday morning.
Said Gigi, “Collectively, all the models are struggling.”