Louis Uccellini, Director of NCEP and president-elect of the AMS, visited Happy Valley tonight. In his speech to PSUBAMS he discussed the evolution of forecast models and how those models performed in two recent notable winter storms.
Uccellini says that the goal of any good computer model is to contain and minimize errors, which is why ensembles are helpful and increasingly being used. He cited the February 10, 2010 snow storm that rocked Washington, DC as an example of a storm where the ensembles performed quite well several days in advance, leading to an excellent forecast.
It was a different story just before Christmas 2010, however. The potential was there for a major winter storm to strike areas from New York to Boston just after Christmas, but operational models were waffling back and forth and the ensembles were somewhat divergent. With so much uncertainty, most forecasters bought into the European model which sent the storm more or less out to sea. Uccellini admitted that the “European is the best model in the world”, so that approach didn’t seem completely unreasonable.
On Christmas Eve, just before the storm was set to impact the Northeast, the European stayed wide-right. However, the American GFS model shifted dramatically west, a solution that would bring crippling snow to the Northeast. Modelers sent out a discussion saying that GFS run was fraught with initialization errors, and that it should be taken with a grain of salt. In hindsight, it was actually a decent projection of what was coming. When I asked him about this statement, which likely ended up delaying most meteorologists in buying into a big storm situation, he said the issue would have been better described as an ‘analysis’ issue as opposed to an initialization issue. He went on to say that RAOB data helped to clarify the situation.
In the end, forecasters failed to truly warn of the impending blizzard until 24-36 hours before the storm. By that time it was too late for many emergency managers to prepare as necessary, especially in New York. For full storm plans to be put in place, a forecast with decent confidence and consistency is required 2-4 days out. The result was chaos in the Big Apple.
The bottom line, according to Uccellini, is that ensembles can often better predict storm tracks. And, he says, if the ensembles fail to provide consistency on a storm track, it at least gives forecasters a better understanding of the uncertainty involved, which can then be passed along to the user for better planning.
One final point presented was the fact that winter storm forecasts are typically made with more confidence during El Nino patterns as opposed to La Nina patterns since the sub-tropical jet is weaker and more variable in the latter case. That leads to a cross polar flow, and phasing troughs that are hard to predict.