Arguments about how to smooth a timeseries

When considering the arguments of climate change deniers, my preferred approach is give the benefit of the doubt first, try to understand what they are saying, and evaluate the science. This approach, of course, usually fails immediately because most of the denier arguments are not based on science at all. There are a few of them, though, that claim that science is on their side. One of these guys is Willie Soon, an astrophysicist who has been claiming that the sun causes climate change for the last two decades [LINK].

Background on Willie Soon

The deniers love Soon because he's a real scientist. He's been able to actually publish climate-related papers pretty consistently, too, which gives him a lot of credibility (compared to other prominent deniers). I've been looking into Soon's publications, just for fun, and have noticed a few important aspects of his publication record. I haven't actually found a CV for Soon, but he does have a URL that has a directory called myownPapers-d, which I assume is his archive. This assumption might be wrong, since (1) there are quite a few non-reviewed papers in there (magazine articles and denier-think-tank "reports") and (2) there is at least one paper not credited to Soon by to Richard Mackey. So one thing that is a red flag is that most of the climate papers that Soon has published are in a journal called "Climate Research [LINK]." Why is this of note? Well, because this journal has pretty much been blackballed by the actual climate research community because of the number of dodgy papers that have gotten through "peer review" and published in CR. Now, a lot of the controversy about that journal is related to Soon himself [cf.], so maybe we should give him a pass there. (side note: the typography of CR is pretty nice, even if the content isn't) Well, except that he's also publishing in the notorious Energy & Environment. And the unknown Physical Geography. And New Astronomy. These are not what one would call mainstream science journals. But, in browsing that directory, I also found two papers in GRL, which is a mainstream journal. The second trend in these papers that I noticed was a proclivity to use 'wavelet' analysis; I'm not sure what to make of this, as it is a reasonable approach to time series analysis, but it is more complicated than other methods which are just as valid.

Soon et al 2004 versus Mann 2004

One of the GRL papers that Soon has is from 2004 and has the title: "Estimation and representation of long-term (>40 year) trends of Northern-Hemisphere-gridded surface temperature: A note of caution." [DOI] I am not going to try to simplify their analysis, since it is dead simple to understand. They take a global average temperature record (HadCRUT) and apply three kinds of smoothing using 40-year windows/intervals (running average, Hanning-window, and wavelet). They get different answers for the different methods, and then consider the difference of their estimates compared to other published estimates. They can't match the temperature anomaly at the end of the IPCC TAR at the end of the record (nor the Mann papers), so they try a few ad hoc adjustments to their filtering. They conclude -- and I am not misinterpreting or misrepresenting them -- that since they can't get the same answer then the IPCC must have misreported their methods and that the magnitude of global warming is very sensitive to the method of smoothing.

These results seemed preposterous to me. First, there is nothing novel or interesting about the results, which is a prerequisite to publish in GRL. They show nothing other than that they can't duplicate other people's graphs, which could be interesting if they had done a robust analysis and shown that the previous work had errors. Their point that different smoothing methods gives different answers is very well known, and trivial.

Later that year, Michael Mann published a paper in GRL that is basically a repudiation of the Soon et al work. The paper is titled: "On smoothing potentially non-stationary climate time series." It is more technical than the Soon et al paper, but also easier to understand. The point is to show that there are objective measures for smoothing techniques. He shows one such measure, which was used in his previous work, and shows that it captures the non-smoothed times eries better than the other methods (including the one used in Soon et al 2004). The conclusion is bolstered by comparing to a frequency-domain approach; the two methods agree well. Another example is given, applying the same smoothing methods to a different time series (a measure of the cold season North Atlantic Oscillation). In this case, the method that is best for the northern hemisphere temperature anomaly is the worst match. The point is that this time series does not appear to be as non-stationary (i.e. not such a strong trend at the end of the time series) as the other series, and that an objective measure of the smoothing gives a simple way to evaluate whether the smoothing is appropriate.

The Mann paper makes some interesting points about how to smooth time series that could be non-stationary. More important than that, it explicitly shows that an objective criteria needs to be applied to make any judgements about these kinds of analyses, which essentially blows the Soon paper out of the water because their argument was essentially, 'different methods give different answers, so there's no way to know what is right.' Finally, from reading these two papers (which I encourage you to do), we see the basic difference between doing science and trying blindly to poke holes into science. While the Soon et al paper tries to evoke scientific doubt, it ultimately fails because the methods are sloppy, no hypothesis is actually tested, the conclusions are not robust, and the points they try to make are clearly exaggerated. The Mann paper takes a more objective look at the data and methods, and teaches us something interesting about time series analysis and the nature of two important climatic time series.

If this is the quality of the Soon et al literature when they can get it into mainstream journals, I have to wonder how bad the papers that are hidden away in obscure journals really are.