On Fri, 25 Sep 2009 11:38:53 +0100, "Will Hand"
wrote:


"Norman" wrote in message
...
John Dann wrote:

On Fri, 25 Sep 2009 08:47:53 +0100, "Will Hand"
wrote:
Put simply - chaos in action!

If I might be permitted a small hobby-horse aside here, I do wish this
word chaos had never become associated with atmospheric modelling. To
my way of thinking, this behaviour is not chaotic at all, it's
sensitive, or super/hyper-sensitive if you will (to the starting
conditions). Chaotic - to me - implies irrational or unknowable, which
is not what's happening here. Once you can define the starting
conditions with sufficient accuracy then the models can indeed run
with a rational outcome and so aren't chaotic.

JGD


I think the word chaos is used because so many of the important processes
taking place in the atmosphere are at scales that are orders of magnitude
smaller than can be resolved by the numerical models, at least the global
and
regional models. If it can't be adequately modelled then it does appear
chaotic

Norman


Chaos is a mathematical theory. Very simple looking equations can be
chaotic, i.e. miniscule changes in starting numbers can produce vast
differences very quickly. The Navier Stokes equations used to model the
atmosphere, I don't think are themselves chaotic, but they have no precise
analytical solution and so approximations have to be made using grids of
data. This introduces chaotic behaviour. Finer grids are not the ultimate
answer as errors can amplify more quickly then. There is a limit to
deterministic predictability with NWP, there will also be a limit to
statistical (ensemble) prediction as well, nobody is sure where those limits
are but they are there just the same. Some say 10 days, some say further
ahead before you get into climatology and dominant signals forced by
physics. HTH.

Will

Hi Will,

I am not sure why you used the term statistical and ensemble together
here as ensembles of dynamic models are used in a probabilistic
framework. Is this what you were calling statistical models as it is
a term that is more often associated with empirical models e.g. those
that directly correlate rainfall over a region with SSTs etc.

Andy