The lure of solar forcing

It’s obvious.

The sun provides 99.998% of the energy to the Earth’s climate (the rest coming from geothermal heat sources). The circulation patterns of the tropical Hadley Cell, the mid latitude storm tracks the polar high and the resulting climate zones are all driven by the gradients of solar heating as a function of latitude. So of course any significant change to solar output is bound to affect the climate, it stands to reason! Since we can see that there are changes in solar activity, it’s therefore just a question of finding the link. Researchers for over a century have therefore taken any climate records they can find and searched for correlations to the sunspots, the solar-cycle length, geomagnetic indices, cosmogenic isotopes or smoothed versions thereof (and there are many ways to do the smoothing, and you don’t even need to confine yourself to one single method per record). At the same time, estimates of solar output in the past are extremely uncertain, and so there is a great deal of scope in blaming any unexplained phenomena on solar changes without fear of contradiction.

Astute readers will notice that there is a clear problem here. The widespread predisposition to believe that there must be a significant link and a lack of precise knowledge of past changes are two ingredients that can prove, err…., scientifically troublesome. Unfortunately they lead to a tendency to keep looking for the correlation until one finds one. When that occurs (as it will if you look hard enough even in random data) it gets published as one more proof of the significant impact that solar change has on climate. Never do the authors describe how many records and how many different smoothing methods they went through before they found this one case where the significance is greater than 95%. Of course, if they went through more than 20, the chances of randomly stumbling onto this level of significance is quite high.

The proof that this often happens is shown by the number of these published correlations that fall apart once another few years of data are added, cosmic rays (which are modulated by solar activity) and cloudiness for instance.

Sometimes even papers in highly respected journals fall into the same trap. Friis-Christensen and Lassen (Science, 1991) was a notorious paper that purported to link solar-cycle length (i.e. the time between sucessive sunspot maxima or minima) to surface temperatures that is still quoted widely. As discussed at length by Peter Laut and colleagues, the excellent correlation between solar cycle length and hemispheric mean temperature only appeared when the method of smoothing changed as one went along. The only reason for doing that is that it shows the relationship (that they ‘knew’ must be there) more clearly. And, unsurprisingly, with another cycle of data, the relationship failed to hold up.

The potential for self-delusion is significantly enhanced by the fact that climate data generally does have a lot of signal in the decadal band (say between 9 and 15 years). This variability relates to the incidence of volcanic eruptions, ENSO cycles, the Pacific Decadal Oscillation (PDO) etc. as well as potentially the solar cycle. So another neat trick to convince yourself that you found a solar-climate link is to use a very narrow band pass filter centered around 11 years, to match the rough periodicity of the sun spot cycle, and then show that your 11 year cycle in the data matches the sun spot cycle. Often these correlations mysteriously change phase with time, which is usually described as evidence of the non-linearity of the climate system, but in fact is the expected behaviour when there is no actual coherence. Even if the phase relationship is stable, the amount of variance explained in the original record is usually extremely small.

Page 1 of 3 | Next page