Wednesday, March 31, 2010

Comparison of GHCN results

I promised some comparison plots from the code reported in the last post, but Zeke has made the task much easier. He posted a spreadsheet with all the data nicely arranged, and I added my own data - you can find the spreadsheet "Temp Comps_NS.xls" in the repository.

Below are some of the plots and discussion.

A summary of this series of posts is here.

Here is the global data from 1880 ( or 1881 in my case). Details of the sources are on Zeke's post. All data is set to zero average in 1961-90.

And here is a close-up of the period 1960-2009.

The agreement is good. This is encouraging, because the principle of the least squares method is very different, and bypasses all the merging of stations issues. Of course, the agreement of all the other reconstructions is also impressive.

Here is a plot with just the main indices GISS and Hadcrut:


  1. Nick,

    Looks like you have a small error in the Jeff Id/Roman M plot around 1966-67.

  2. Thanks, Zeke, I've fixed that glitch.

  3. Hey Nick, try a US only for your approach. Then we can add:

    1. zeke us only
    2. romid us only
    3. zeke us only
    4. giss us only
    5. residual analysis us only

  4. Steven,
    Yes, I hope the next version, which makes masking stations easy, will be done today (it's April 1 here). Then I'll do US.

  5. I'm working up a version of getting metadata into a R table if you want the code. I borrowed some of your stuff and JeffId stuff reading files in..

    I can cleaning it up and shoot it to you, I'm just learning R and using your code to really learn stuff so What's mine is yours.. literally.

  6. Oddly enough, our two approaches (despite their different methods) produced the most similar results:

  7. Hi Zeke, yes interesting. I'm about to post V1.1, which makes it easy to do all kinds of cases, so there will be more comparisons. One fly in the ointment - V1.1 gives slightly different results.

    And Steve, if you like to dig into this kind of code, I think v1.1 will be easier to make sense of.