A problem of multiplicity

One thing a scientist doesn’t want to mess up is the problem of multiplicity (also known as ‘field significance‘). It’s just like rolling a die 600 times, and then getting excited about getting roughly 100 sixes. However, sometimes it’s much more subtle than just rolling dice.

This problem seems to be an issue in a recent by paper with the title ‘Evidence for solar forcing in variability of temperatures and pressures in Europe‘ by Le Mouel et al. (2009) in the Journal of Atmospheric and Solar-Terrestrial Physics.

In this study, a range of different so-called ‘solar proxies’ (describing the state of the sun – in this case monthly sunspot number, the aa-index, as well as the vertical Z and horizontal H component of the magnetic field measured at Eskdalemuir) is examined and compared with some climate indices – some rather obscure quantities that were assumed to represent the state of European climate, namely ‘mean-squared interannual temperature variations’ (MSITV) and ‘lifetime’.

Moreover, the similarity between the solar proxies and the climate indices were tested for different seasons.

The Le Mouel study found a ‘decent match’ only for one solar proxy and only in winter: the vertical Z component from Eskdalemuir.

Although the paper was not clear on this, it appeared that a number of estimates for MSITV and ‘lifetime’ were explored using different lengths of time intervals (sliding ‘window size’) and for different seasons. In other words, by searching for one particular choice that gives the best match for one solar proxy, they may have ended up (unintentionally) ‘cherry picking’ the data.

There were furthermore substantial differences between the solar proxy and lifetimes of T(2m), SLP, and wind direction before 1940 – and the paper forgot to even point this out.

However, a break-down of correlation outside a limited interval is typical of a fortuitous match – there is plenty of examples through science history of similar alleged links between solar activity and climate that eventually turned out not to hold up.

The mismatch before 1940 further points to my suspicion of a problem of multiplicity. The paper also neglects to discuss the statistical significance levels associated with the analysis – hence the validity of these conclusions is at best an open question.

Another weakness is that the paper offers little discussion on how the solar activity may affect local/regional temperature/pressure. E.g. how does Z affect the wind directions and the ‘lifetime’, and why is there a strong dependency of the ‘signal’ on the season? What is the hypothised physical link? As long as there is no hypothetical mechanism, there is no way of confirming whether the interpretations are correct.

It’s interesting to note that the time evolution of the Z or H components of the magnetic field shows no clear trend over the period 1920-2000, and if anything, it seems as if there are opposite trends in H and Z (these proxies have the highest correlations with the climate indices). Thus, it would be difficult to generalize these results to explain the past global mean temperature trends – as it is to make deductions for the global mean from a regional set of measurements.

Page 1 of 2 | Next page