Thursday, February 01, 2007

The worst climate science paper ever of all time anywhere

I've an important announcement to make. I have just spent the past few days looking of what may well be the worst climate paper yet produced. You heard that right – the absolute worst. This paper is so poor it makes Khilyuk and Chilingar (2006) look like Einstein’s special theory of relativity in comparison.

So anyway, I decided to go all Climate Audit on this paper partly to get a handle on the solar influence on climate that denialists are always on about (there is actually some decent literature on solar forcing, though obviously this paper isn't part of that), and partly because it is so shockingly bad it’s humorous in the way troma movies are.

Firstly, some background. In a rather silly post from NZ denialist Vincent Grey over at Jen Marohasy, a fellow by the name of David Archibald popped up, spouted the usual shill crap, and the directed readers to his relatively recently published paper in Energy & Environment.
You asked for solar - climate papers. Look no further than my own effort:
http://www.lavoisier.com.au/papers/articles/Archibald.pdf
Now, Lavoisier is a crazed denialist group based in Australia so the fact that the article was reprinted there didn’t bode well.

Archibald goes on to claim his paper is peer-reviewed. Ian Castles informs me that E&E really does have a peer-review process and he has, in fact, reviewed papers for them himself.

But in this case, I’m not so sure. Actually, that’s a bit of an understatement.

So, what is so wrong with Archibald, D.C. (2006) Solar cycles 24 and 25 and predicted climate response?

To begin with, there are some, ahh, editorial and grammatical errors.

Pick what is wrong with these two paragraphs at the top of page 31.
Badalyan, Obridko and Sykora’s projection of solar cycle 24 maximum of approximately 50 is shown in figure 1 with solar cycle activity back to the end of the Maunder Minimum. Solar cycle 25 is also expected to be weak. The rise in amplitudes prior to the Dalton Minimum mimics the rise in amplitudes from the late nineteenth century to the end of the twentieth century.

Badalyan, Obridko and Sykora’s projection of a solar cycle 24 maximum of approximately 50 is shown in Figure 1 with solar cycle activity back to the end of the Maunder Minimum. Solar cycle 25 is also expected to be weak. The rise in amplitudes prior to the Dalton Minimum mimics the rise in amplitude from the late nineteenth century to the end of the twentieth century.
Yes, Archibald has written the same paragraph twice. It doesn’t appear to be a printing error either as each of the paragraphs are slightly different (i.e. projection of solar cycle 24 maximum verses projection of a solar cycle 24 maximum).

The Dalton Minimum is spelt Delton below. There’s a space missing in …..experienced a2.0C decline.

Quibbles maybe, but basic errors such as these indicate little effort was put into the writing and no effort was put into the review. E&E editorial staff clearly don't read the papers they publish.

In the introduction Archibald gives a backgrounder on some of the literature linking solar cycles to climate. Mentioned are the correlations between the solar cycle amplitude and temperature, solar cycle length and temperature etc. It’s fairly shoddy and ignores the literature which refutes the idea that these correlations indicate that solar changes are the main driving force behind recent climate change. I don’t intend to look at the introduction in depth here.

What I do intend to cover is the ‘original work’ part of the paper.

To determine a temperature baseline for predicting response to solar cycles 24 and 25 (we’re currently in 23), Archibald takes a startling approach. Instead of using world-wide temperature data, only data from the US mainland is used. Additionally, Archibald decided that only data from rural meteorological stations should be used to avoid the urban heat island effect. Fair enough, you may say. But the catch is, he chose just 5 stations out of the hundreds and hundreds available! Not only did he only choose 5, all 5 were within several hundred miles of each other in South Eastern USA!

The possibility of these stations being representative of anything other than the small local region they covered is non-existent.

There must be something special about the chosen five. And there sure is – they show lower temperatures in the latter half of the 20th century compared to the first half. This actually forms one of the major conclusions of the paper!



The conclusion is that for the current and recent temperature record for all of North America:
The profile indicates that temperatures remain below the average over the first half of the twentieth century.
What Archibald forgets to mention is that most met. stations across the US and indeed most of the world show that it is warmer in the latter half of the 20th century than the first. This trend is evident in non-heat island affected rural stations, as well as urban stations.

Archibald has clearly showed a regional phenomena and falsely claimed it is representative, when even the NASA GISStemp database he has used shows many, many stations where this isn’t the case. All are ignored and the reason why isn't given (though I think we can hazard a guess).

I find it simply unbelievable that Archibald would make such an obviously wrong claim that is based on insufficient non-representative data. But he does, and it’s there for all to see.

Moving right along.

Archibald then decides to predict what the temperature response to solar cycles 24 and 25 will be. To do this he first hypothesizes that temperature is responding to solar cycle amplitude (in this case, the number of sunspots per year). Instead of 5 meteorological stations, this time he decides to use only 1, De Bilt in the Netherlands!

Archibald produces this strange graph, with only 1 data point per solar cycle verses average annual temperature. A liner regression (I’m assuming) is performed, but there is no mention of the slope of the line or its R^2 value. Ignoring this, Archibald claims:
By expanding the time interval studied to 1705 to 2003, a good correlation of temperature and solar cycle amplitude is evident. This is shown in Figure 4 which demonstrates a correlation between solar cycle amplitude and annual average temperature at de bilt, Netherlands.




Here’s the thing, temperature data for De Bilt and sunspot data are available are available for every year from 1705. Why not see if they correlate? Here’s why:



The correlation is very weak (the slope of the line) and the predicative value of the model is also very weak (the R^2) value. Unsurprisingly, it shows that other factors are more important in driving temperature in De Bilt.

So what data points did Archibald use? To be honest, I can’t work it out. It appears that he has taken the maximum solar amplitude value for each cycle and plotted it against the temperature for that year only.



As you can see, some of the data points are the same, but some aren’t. One doesn’t like to suggest that he’s fudged the data so I’m sure there’s a good explanation ;)

This time the slope’s slightly better but the predictive value is even worse. And by ignoring nearly all the available data, the graph is meaningless anyway.

What’s interesting is what Archibald hasn’t shown, as usual. Here’s the smoothed solar cycle data from 1706. You’ll note maximum amplitude of each solar cycle has been getting smaller since 1956.



Here’s De Bilt’s temperature plotted over the same period (5-year floating average shown).



You’ll notice temperature has been sharply increasing since 1960-odd, which is exactly the opposite of what should be happening according to Archibald’s model. Larger solar cycle amplitude should equate to higher temperature, only that hasn’t been the case for some 50 odd years. He has, of course, completely ignored this, maybe becasue it's likely due to that evil conspiracy, AGW.

Archibald shows another graph, which I couldn’t be bothered re-creating, as I’ve really no idea what he’s done, that purportedly shows that solar cycle length displays an even better correlation with temperature than solar cycle amplitude (again, no slope and no R^2).



Here it gets weird. Archibald goes on to conclude that based on solar cycles 24 and 25 with predicted amplitudes of around 50 (hardly anyone is actually predicting this, most predictions are much higher), the correlation equation from his second graph shows a declining temperature of 1.5C in the US!. But his second graph has a correlation based on solar cycle length, not solar cycle amplitude, so it isn’t applicable! He then claims in the next paragraph that the 1.5C figure actually comes from the temperature response to cycles 5 and 6, and not the correlation I guess.

I’ll fire off an email to Archibald and hopefully he’ll drop by to explain how I’ve got it all horribly wrong and defend his paper. Or maybe not.

Anyway, onwards and upwards........some real science is on its way; IPCC AR4 part I.