Someone on the other side made a good point this morning - if we could re-analyse the charts and (if they existed) ensemble data for Dec 1963, Feb 1978 and Jan 1987 for instance, we'd probably see just the same amount of uncertainty and spread leading up to those event, if not more. And actually the 2010 episode has many a wobble and variation even 5 days out. I completely accept that the scripting of NWP and the associated processing power is exponentially better than it was just 20 yrs ago, but subtle changes to starting data are introduced for a reason - because the starting data itself has error bars and is often sparse in areas where it's needed.
The skill in medium term forecasting is not in the interpreation of one chart from one model, but in looking at all available data and deriving a blended solution that (at the time) is most likely, with the caveat that it will change from hour to hour. I find the intense scrutiny of every chart entertaining, but in scientific and practical terms it is not the best approach.
Originally Posted by: nsrobins