NY Times
article:

By early 2013, the European model had nearly 10 times the raw
computing capacity of the Global Forecast System, or G.F.S., which is run by the
National Weather Service. There were other problems, too, and the cumulative
effect was obvious and irrefutable: The G.F.S. was doing worse than it rivals,
and it played out in high-profile cases, like Sandy.

The upgraded G.F.S. prevailed over the European model in the blizzard that
largely missed New York. Many of the forecasts confidently calling for high snow
totals in New York were banking on the European model, or even outright
disregarded the G.F.S. But despite its early victory, the upgraded G.F.S. is
still behind.

Perhaps the biggest shortcoming is in data assimilation — the process of taking
all of the available data and building an initial description of the atmosphere.
The model runs from that, but a perfect model of the wrong atmosphere will yield
a wrong answer.

“It is clear that our initializations are inferior,” Mr. Mass argued. “That’s
the real problem. They have a lot more people and have taken a more
sophisticated approach.” Differences in initialization were probably at play in
the different forecast for Joaquin. “There’s a subtlety that the European center
is getting right that we’re not.”