date: Thu May 8 13:42:39 2008
from: Tim Osborn
subject: Re: Model Choices
to: Steve Jones
Hi Steve; not sure if I'll be in the office on Friday -- I might be, so if you're on campus
anyway, then feel free to come and check, but don't make a special trip. I'll be here Mon
and Tue, if that's any good? Tim
At 13:54 07/05/2008, you wrote:
Hi Tim,
I've managed to extract all the data I need from Ron Miller's stuff, and
performed a few stats analyses on the results. There's some interesting
outcomes, which should enable us to choose a few models to use for the
main part of the project.
I've attached an Excel spreadsheet containing the stats from the models,
together with two graphs.
The first shows the 2000-2100 detrended variance for individual models
with a given number of 21st Century runs, which may or may not allow us
to select some models based on variance.
The second graph shows the Least Squares Regression divided by the
1900-1970 standard deviation (Miller's data was scaled to have a zero
mean for this period). For each model, both the 1950-2100 (red) and
2000-2100 (blue) trends are shown. I thought it would be interesting to
see how the trend varies for different time periods
(accelerating/decelerating/constant), and hence be able to choose some
models based on those characteristics.
My next step will be to look into the more detailed characteristics of
the NAM in order to determine what kind of things are likely to give a
reasonable comparison between models and observations - I'll be reading
up on this in the next couple of days.
Are you around either tomorrow or Friday to have a chat about which
models would be best to use, and what kind of things I might look for in
the next analysis stage? If so, let me know (any time except late
morning on Thursday is good).
Thanks,
Steve.