Was No. 1& Still Is No. 1
There has been a lot of talk in the weather world in the past few years about the accuracy of the European vs. U.S. computer models. We’ve been talking about this in the weather biz since the 90s, but it exploded publicly after Superstorm Sandy in 2012.
It was a clear “battle”. The main U.S. model, the GFS, kept predicting that Sandy would curve to the right and head harmlessly out to sea. Every 6 hours, when a new model run came out, the solution was similar. At the same time, the European model was predicting an unheard of sharp left turn, causing Sandy to slam directly into the U.S. East Coast. Every 12 hours, when a new model run came out, the solution was similar. Unfortunately, the Euro was right. (The National Hurricane Center, part of the National Weather Service, did get the forecast right a couple of days later, and gave plenty of warning that Sandy would hit and become a major disaster.)
They Don’t Win ALL the Time
Over-simplification is common in trying to communicate science to the public, and that can sometimes lead to myths. One of them is that the European model is always better than the GFS. It isn’t. A perfect example was last winter, when the Euro predicted a 1- to 2-FOOT snowstorm in places like Philadelphia and New York City, while the GFS suggested only a few inches. We all seemed to “take the bait” and go with, or lean toward the Euro. I remember going for a fraction of the Euro forecast, but that was still too much. Even the National Weather Service went with it. (That was unusual, since the NWS has been stubborn in acknowledging the Euro’s superiority in the past). And this time, the GFS was right: there was very little snow.
Of course, that single non-storm (it did develop, but too late for us) is now used anytime someone points out the superiority of the Euro. Of course it’s not perfect. The overall, average skill score is only appears to be a little better. But it’s the big events that are by far the most important in weather forecasting. Frankly, I don’t care if the GFS was better for a weak tropical storm in 2013 (as some in the NWS have pointed out, in trying to defend their model). I care that it was the only model that was right this summer, predicting that dangerous Hurricane Joaquin would track out to sea and NOT make a deadly left turn similar to Sandy. That was a big one. And it’s no surprise to me that the Euro has recently even beaten the specialized hurricane forecast models on storm tracks.
So Why Is the Euro Better?
Many in the U.S. were embarrassed when the Euro beat the GFS badly for Sandy. It even got Congress whipped into a lather, and they approved tens of millions to upgrade the computers at NOAA -- the agency that runs the National Weather Service. NOAA just announced that their Supercomputer is “running at record speed,” and that it “secures the U.S. reputation as a world leader...”
“A” world leader, but not “THE” world leader...
How do I know that? You can call it “chaos,” “The Butterfly Effect,” or simply “poorer initialization.” This gets complicated, but the basic story is: “garbage in, garbage out.” OK, it’s not garbage -- it just isn’t as good as the way the Euro does it.
The reason weather forecasts can never be perfect is that we can never perfectly input the exact current conditions all over the world -- at all levels of the atmosphere. We’d need to measure every cubic inch of the earth! MIT meteorologist Edward Lorenz is credited with recognizing how chaos theory applies to weather forecasting. His famous paper in 1972 was titled: “Predictability: Does the Flap of a Butterfly’s Wings in Brazil Set Off a Tornado in Texas?” So, something as seemingly small and trivial as a butterfly flap could lead to changes in weather patterns around the world days, weeks, or months later. Small initial errors lead to bigger and bigger errors over time.
So, the more accurate a computer model can get the current conditions, the more accurate the model is likely to be. Sure, there are dozens of other factors, but it’s like giving the Euro a 10-yard head start in a 100-yard dash. It’s possible to win, but awfully hard.
So, Why Can’t We Catch Up?
I won’t bore you with the details, but the Euro uses something called a “4D-VAR,” also known as “four-dimensional variational data assimilation.” Observations are taken not at a single time, but over a period of several hours. For some reason, this costs 10 TIMES as much to run. Since the GFS is just a part of NOAA’s many computer models, changing the GFS would apparently mean everything else would have to change, too. And they are just not ready -- or willing -- or able to do that. So they’re stuck with “3D-VAR”, and a not-as-good product.
Until NOAA can start the race even with the Euro, it can’t catch up. Every time the GFS improves (and it is WAAAY better than it used to be), the Euro improves, too. And forecasters continue to look at the Euro and its products with a little bit more attention than the GFS.
(If you really want to get some more technical details, there’s a great article in “Physics Today” on it.