Thursday, March 29, 2012

Interactive climate plotting news.

A few items of news about the interactive climate plotter. Kevin Cowtan, in comments, mentioned that he has been working on a trend plotter for Skeptical Science, and that has now appeared. It emphasises the uncertainties of both trend and intercept, and shows graphically the range of fitted lines. He uses Tamino's near-AR(2) allowance for autocorrelation, which is an advance on the Quenouille adjustment that I used in the trend viewer. The Moyhu climate plotter is restricted to annual data, and so reports much lower significance (though little affected by autocorrelation).

He also has a post here on Hadcrut3, with an emphasis on land/ocean balance (or imbalance) as a source of bias.

The other news is that I've added some new things to the interactive plotter. You can now copy a rather long URL for linking. This will bring up the plotter in the state which you last created.

You can also embellish the plot with a legend and title. There is a current rather bland default title, but you can write your own.

I have also added the interactive plotter and the trend viewer to the permanent pages listed near top right. At some stage I'll reorganise the gallery.

Sunday, March 25, 2012

Autocorrelation, regression and temperatures

Temperature trends are much discussed, and less frequently their statistical significance. But there is a further question of what statistical significance means. It is tied up with the behaviour of the deviations from the trend, and assumptions that may have been made about them.

This became controversial a while ago with Eric Steig's analysis of Antarctic temperatures. Hu McCulloch wrote an informative post.

I've been thinking more about it as a result of the interactive plotter. I gave estimates for the uncertainty of trends based on OLS assumptions, but foreshadowed a correction for autocorrelation. The usual remedy, at least in climate blogging, is the Quenouille adjustment to dof that Hu mentioned, and which I used in the trend analyzer. But it does not adjust the actual trend, and I've long been curious about its adequacy.

Fortunately, for the interactive plotter I need not have worried. Global temperature data shows strong autocorrelation from month to month, but not much between annual averages, which is what I used exclusively in the interactive plotter. That's not to say that annual averaging is a free gain there - the averaging takes away more significance than autoorrelation correction would have.

In the process of thinking about it, and prompted by some recent comments from Kevin C, I looked at what Grant Foster had done in Foster and Rahmstorf (see Tamino). In the appendix he looks at GISS data from 1975, and somewhat adapts the Quenouille correction to allow for what seems to be more than lag 1 correlation. So I though that would be agood case to work on.

Thursday, March 22, 2012

Interactive JS climate plotter (update)

Next in the series of JavaScript/HTML 5 apps is an interactive plotter for climate annual indices. There's a whole lot of annual data from which you can click to put on a plot. There's an elaborate arrangement for handling different units, anomaly bases etc. You can slide curves around in the plot, and fit various types of curve by OLS regression.
In these respects it's a bit like WoodForTrees. The main difference is that it is client side - the code and data sit on your machine. This adds speed and on screen flexibility, but makes some facilities like Fourier analysis more difficult, and limits the amount of data that can be used (which is why it is annual only).
I'll give detailed instructions below the plot. To try it, just click on a data set. For the moment, there is a restriction to 7 curves with up to 3 sets of units. You can click on the pink and blue axis bars to move plots around. For more details, you can click with Ctrl pressed on any functional button to get info about it in the window bottom right.
I've added some new utilities. You can add a legend; just click on the plot area where you would like the legend to be. Clicking elsewhere will move it; clicking within the canvas but outside the plot area will make it vanish.
Below the graph, there is a long URL in small writing. If you copy that, it will reproduce the current graphs in a different browser (for linking)
There is also a window bottom left in which you can write your own title.
Update: There is an updated version as a maintained (but late) page here.

Sunday, March 11, 2012

February GISS up 0.05°C - comparison

GISS is rather early this month. It's up from 0.35°C to 0.4°C, a rise of 0.05°C. In contrast TempLS was down 0.05°C. It's the first time in the seven months I've been updating that they have diverged in sign, though the amounts are small. Details, including plots, of other indices, are here. Both satellite indices went down similarly to TempLS.

Here is the GISS plot:

And here is the TempLS spherical harmonics plot:

Friday, March 9, 2012

Feb TempLS temperature down 0.05°C

February 2012 temperatures down 0.05°C

The TempLS analysis, based on GHCNV3 land temperatures and the ERSST sea temps, showed a monthly average of 0.18°C, down from 0.23 °C in January. As last month, this is in line with satellite LT trends. There are more details at the latest temperature data page.

Below is the graph (lat/lon) of temperature distribution for February.

This is done with the GISS colors and temperature intervals, and as usual I'll post a comparison when GISS comes out.

Wednesday, March 7, 2012

Lindzen's misrepresentation at House of Commons

Prof Lindzen gave a presentation at the House of Commons in London on 22 February. Judith Curry ran a post that now has over 1200 comments.

Real Climate has noted what I think is quite a shocking misrepresentation. It is in this slide:

It's yet another in the series of claims that someone is "manipulating the record". That's a heavy accusation.  But as RC I think convincingly demonstrates, this is not a comparison of two versions of the same index. It is a comparison of the land only, met stations index “GLB.Ts.txt” in 2012 with the land and ocean index “GLB.Ts+dSST.txt” in 2008. And, as is well known, the land index has a higher slope. As they show, the land ocean index properly compared has changed very little over that period.

So did Lindzen just make a mistake? Well, as Hank Roberts has shown, it looks awfully like he just lifted it, without attribution (but that's not to say without permission) from the JunkScience blog, Feb 7th.