Benjamin Laken

Wed 24 February 2016


Monsoon madness

alt text

This story starts around 2012, when I received a request to review a paper on solar activity and the Indian Monsoon from the journal Advances in Space Research. I have reviewed many such papers (perhaps \~20) so this was nothing unusual. A quick read revealed some tortured data, abused statistics, and nonsense prose. All in all a fairly typical manuscript in its field. A version of it can be found at ArXiv. (To get some idea of why I say this is a typical paper, have a look at Barrie Pittocks 1978 piece "Solar Cycles and the Weather: Successful Experiments in Autosuggestion?", although it is, unfortunately, pay-walled.)

The authors were claiming that weak and strong Indian monsoons are related to changes in cosmic rays: energetic particles which ionize the atmosphere. (There is a good review here for those interested in background regarding their possible role in climate.) The study took rainfall data from India and split into two samples, representing the strongest and weakest monsoons recorded. Using composites, they examined the linear change in cosmic rays over a period of 5 months during their samples. The linear change in cosmic rays during the weak and strong monsoon samples were of different sign. Based on their observation that the cosmic ray changes during weak and strong monsoons were 1) anti-correlated, and 2) showing high correlation coefficient values. From this, they argued that cosmic rays affect the monsoon via clouds, and also potentially impact global climate.

Now, something key for you to know about the cosmic rays: they vary with a roughly 11-year period, as they are modulated by the Solar Cycle. Unfortunately for the authors, this means that if you extract a period of several months of cosmic ray data from a time-series, and examine the linear trend, it will very likely show either a decreasing or increasing tendency depending on the phase of the Solar Cycle. If you were to randomly pick a few periods from these data and stack them together (as the authors did with a composite method), you will end up with a trend depending on where the samples were taken from in relation to the phase of the Solar Cycle.

To simplify things, assume the cosmic rays vary like a sine wave over 11-years with no long-term trend. By looking at two small random composite samples and examining the 5-month linear changes you more or less have a 50:50 chance of having either an increasing or decreasing trend. If you don't make an a priori assumption about what direction the trends should be assigned to (e.g. that weak monsoons should correspond to increasing cosmic rays and strong monsoons should correspond to decreasing cosmic rays), and only care that the samples are anti-correlated, your chance of getting a positive result is 50%. However, if you argue that only one specific configuration of samples with changes supports your argument, then the chance of randomly getting this drops to 25%. So that more-or-less takes care of the anti-correlation, as alone it is clearly not evidence of something unusual here.

But were high correlation coefficients unusual in this work? Because the long term-changes in cosmic rays are determined by the Solar Cycle, the correlation coefficient values from samples constructed the way they authors made them tend to be high. In fact high-values are the most common type of values you can obtain from sampling the data in this manner. (This is the core argument I present in my paper). The authors made no real effort to test the probability values of these data in detail, so they did not realise this. They simply applied an out-of-the-box statistical test, and assumed the values were meaningful.

So, after pointing out this core problem with the paper in peer review the editor for Advances in Space Research promptly rejected the paper and thanked me. (I feel like noting that I rarely recommend straight rejections in my reviews, I normally go out of my way to try to give comments that will help people make a stronger analysis. Although the more time goes on, the more empathise with Sisyphus.)

A year or more passed, and I received the same paper again from the Journal of Atmospheric and Solar Terrestrial Physics, unmodified, typos and all**. This is not so uncommon in small fields, and has happened to me several times. So, I submitted my same review and moved on thinking all would be well.

(** This, and other examples, lead to my feeling that in many instances authors are willing to put more effort into pushing articles through journal systems than they are into the underlying science.)

Alas, despite arguments that the study was total rubbish, the editors decided that this paper did indeed prove that the climate system has a heart of cosmic rays and published it. (Again, If the claims of this paper were true they would totally re-shape climate science. The editors were happy to accept this paradigm shattering paper despite a clear explanation of why it was total nonsense!) It didn't take too long for the story to be repeated, e.g. on The Hockey Schtick. Over the course of a year, I protested heavily to the editors, and pursued what I thought was a 'proper' recourse, publishing a direct reply in the same journal as the original article. After a year with the editors however, they were effectively stonewalling my efforts. I don't want to go into the often frustrating and ridiculous details of that, other than to say the end result was a prompt submission and acceptance to the EDP Open Science journal Space Weather and Space Climate.

This story, which out of kindness I only subjected you to an abridged version of, certainly taught me some hard lessons. For one, due to the inordinate amount of time-wasted, I mostly only accept reviews from editors/authors I personally know, to try to ensure they are genuinely motivated and not simply looking to fill quotas, gain rewards, or pad-CVs. I am also trying hard not to get sucked into studies that most sane researchers could spot as junk and simply ignore. Finally, I think more than ever that publishing papers should be seen as the start of a process, not the end of one, and adopting reproducible Open Science methods should be seen as crucial to research credibility.

Go Top

Share this post: · · ·

comments powered by Disqus