Saturday, April 17, 2010

The Math description of TempLS

I've put up a file called TempLS.pdf (about 80 kb) on the document repository. It sets out the least squares theory, and goes through the analysis code explaining the implementation.

It may already be a little bit obsolete. It's likely that V1.4 will solve the design system by the conjugate gradient method. The matrix is well conditioned, and the method seems reliable, and at least five times faster, making inversion a matter of  10-20 seconds.


  1. Nick,

    Nice idea, especially for novices like me. Think the file 103Kb now :]

    Model for global temperatures
    "on station s, year s and calendar month m."

    Shouldn't that be - on station s, year y and calendar month m?

  2. Steve,
    Thanks, you're right with that typo.

    I'm about to post V1.4 (a few hours). I think you'll find some things useful. Apart from speed, it gives you maps of the regions you are doing with the included stations marked with dots. It also gives a plot of how the number of stations in your sample varies with year.