Draggable graphs

In the climate plotter a large amount of climate data could be plotted on an adjustable scale. There were bars on which you could click to move and expand the graph vertically and horizontally. Curves with different units could be moved independently. This is all based on the HTML5 canvas.

The bars were a bit clunky. I've been experimenting with mouse dragging. I had thought it would be slow, but it isn't. I'm planning to use it routinely in plotting, and for the climate plotter, and to post the code that enables it. Entering different data is easy.

Here's an example which I'll include in the monthly data tracking. It looks like just the last three years of monthly index data, similar to a graph that is currently shown. But it's backed by data back to 1850, which you can see by dragging back, and shrinking the scale if you want.

Jaxa Arctic Sea Ice - active polar plot

It's the time of year for tracking ice. I have kept current plots and tables of Jaxa daily reports. This year, there will also be daily updated movies of high resolution Arctic SST, both last 50 days and full year.

But I wanted to update the plotting methods using some of Javascript active features. I've made a polar plot, where the ice extent is represented as distance from the centre, and time progresses by angle, just as with a clock. Various parts can be magnified. When it has settled, I'll include it in the current page.

April GISS Temp down by 0.08°C

GISS LOTI went from 0.58°C in March to 0.50°C in April (back to February level). TempLS was fairly steady, satellites mixed.

Climate of the Past fails Fourier test

This is a belated post. I'm writing about a paper by Ludecke et al which was accepted in February by the EGU online journal "Climate of the Past". Eli wrote about several aspects, including data quality and how the paper made it to acceptance. Tamino gave a definitive mathematical takedown. Primaklima has a thread with some of the major local critics chiming in.

So what's left to say? And why now? Well, Ludecke had a guest post at WUWT a few days ago, promoting the paper. While joining in the thread, I re-read the online discussion, and was surprised at the lack of elementary understanding of Fourier analysis on display. Surprisingly, the guest post was not well received at WUWT, at least by those with math literacy.

I expect that notwithstanding this negativity, the paper's memes will continue to circulate. It comes from EIKE, a German contrarian website. And they have been pushing it for a while. Just pointing out its wrongness won't make it go away.

So here my plan is to redo a similar Fourier analysis, pointing out that the claimed periodicities are just the harmonics on which Fourier analysis is based, and not properties of the data. Then I'll do a similar analysis of a series which is just constant trend; no periodicities at all. Ludecke et al claim that their analysis shows that there is no AGW trend, but I'll show the contrary, that trend alone not only gives similar periodicities, but is reconstructed successfully in the same way.

TempLS global temp unchanged in April

The TempLS monthly anomaly for April 2013 was 0.408°C, compared with 0.405° in March. UAH showed a fall of 0.08°C.

Emulating Marcott et al

Update - error found and fixed. Agreement of 2nd recon below is now good.

I have now posted three "elementary" emulations of Marcott et al. Each, in its own way, was a year by year average (here, here, and here). As such, it missed the distinctive feature of Marcott et al, which was the smoothing produced by Mante Carlo modelling of the dating uncertainty. So here I am trying to grapple with that.

A story - I used to share a computer system with a group of statisticians. It worked well (for me) for a while - they used Genstat, GLIM etc - no big computing. But then some developed an enthusiasm for Monte Carlo modelling. I've had an aversion to it since. So here, I try to use exact probability integrals instead. There are clearly big benefits in computing time. Marcott et al had to scale their MC back to 1000 repetitions for the Science article, and I'm sure they have big computers. My program takes a few seconds.

A reconstruction is basically a function F(d,t,e) where d is data, t time, and e a vector of random variables. MC modelling means averaging over a thousand or so realizations of e from its joint distribution. But this can also be written as a probability integral:
R(t)=∫F(d,t,e) dΦ(e)
where Φ is the cumulative distribution function.

MC modelling means basically evaluating that integral by the Monte Carlo method. For high dimensionality this is a good alternative, but if the variables separate to give integrals in one or few variables, MC is very inefficient.

The aim here is to separate the stochastic variables and integrate analytically in single variables. The scope is just to generate the recon - the CI's will wait for another day

March GISS Temp up by 0.10°C

This might be old news - my automatic system had again stopped automatically checking GISS. Anyway GISS LOTI went from 0.49°C in February to 0.59°C in March. Most other measures were fairly steady. Maps below.