0.75 correlation for a forecast seems like good evidence against overfitting.
Perhaps I should have suggested running your model past some 'local' statisticians rather than a climatologist; Steve Venner and Nick Stokes spring to mind.
Have you a link for "Clarke's estimated value of K=2.2 rad/yr^2"?
Eureqa looks dangerously powerful. Do you think it merits an intro in the software pages of the wiki?
I've never tried implementing a 2nd derivative in a spreadsheet. Is an averager the smoothing function?