I remember the NAO discussion. What was unclear to me (and still is) is how much does this improve on the forecasts the Met Office and Benfield Hazzard - UCL were producing 15 to 20 years ago? I thought they claimed similar skill levels. If that's the case it suggests recent developments will add little to forecast accuracy with a lead time of 6 to 9 months but may be useful when looking 24 months ahead.
Apples and oranges.
The original discussion centred around the misreporting of the r-value of 0.62 for this particular model as meaning thaf the Met Office were claiming that this model would get the direction of the NAO right 62% of the time. And as you argued back then, even a simple examination of the May tripole gives the right answer 2 times out of 3, so what's the big deal?
But as I tried to explain, that's not what an r-value is. It's no more than an arcane expression of how much of the real-life deviation in what you're modelling is actually successfully modelled. In this case the answer is 1 minus 0.62 squared expressed as a percentage, i.e. about 38%. It's nothing to do with how often it gets an outright positive or negative result the right way round. Nothing at all. It just expresses how much of the final atmospheric status which goes to make up the final NAO result each winter the model is getting right.
The simple way of avoiding this confusion would simply have been not to publicize the paper (and particularly as it was paywalled, so you had to be a pretty dedicated layman to access it). I can only imagine that they were eager to show that they were actually doing something with their expensive supercomputer, and had to press-release some sort of breakthrough.
Edited by user
20 October 2017 20:12:28
|
Reason: Not specified
2 miles west of Taunton, 32 m asl, where "milder air moving in from the west" becomes SNOWMAGEDDON.
Well, two or three times a decade it does, anyway.