TCS Daily

What is the Earth's 20th Century Temperature Trend?

By Willie Soon - June 24, 2004 12:00 AM

The following important comments were made by Kary B. Mullis in his autobiography, "Dancing Naked in the Mind Field."

"Science appeared in the seventeenth century. Robert Boyle, who was a Christian and a friend of the English monarch Charles II, made a vacuum pump in the seventeenth century and showed that he could extinguish a candle by pumping air out of the jar wherein the candle was burning. According to Boyle, whatever was left in the jar after the candle went out constituted a vacuum. In the common vernacular, it meant that absolutely nothing was there. Whether God was in there or not was not something Boyle addressed. The Catholics seriously disagreed. But the candle went out. Boyle didn't care whether God was there or not because he couldn't measure God. The religious issue was not as interesting as the issue of what he could measure. That's when science started to take off. Computer modelers of ... the next thousand years of climate could take a lesson from Sir Robert Boyle and his Royal Society. If you can't actually measure something, or make an accurate prediction from a theory, and present it to a group of your fellows, be good enough not to disturb us about it."

Who is Kary Mullis?

Mullis is the scientist and chemist that brought the world the gift of "all the DNA you wanted" by allowing copies to be made through "his invention of the polymerase chain reaction method." For that he was awarded 1/2 of the Nobel Prize in Chemistry in 1993.

Mullis's comments have important ramifications for those of us involved in the scientific enterprise of investigating global warming by CO2. Not only that we should be good enough to resist spreading unfounded fears (which is what I infer by merely browsing over what has been published over the last 10 years or so) of a globally warm Earth; but we better be sure that what we say and publish -- our scientific results -- are accurate and at least can be subjected to the minimal scientific standard of being reproducible.

I have not met Kary Mullis but I sense that Dr. Mullis would be appalled and perhaps even saddened and angered by the misuse of science I will describe below.

The discussion and focus here is on a paper entitled "Estimation and representation of long-term (>40year) trends of Northern-Hemisphere-gridded surface temperature: A note of caution." It is a paper that I wrote with my colleagues, Dr. David Legates and Dr. Sallie Baliunas for the Geophysical Research Letters (GRL) that was printed February 14, 2004.[1] You can download a copy of the actual paper here.

Figure 1: The comparison of the 40-year smoothed Northern Hemisphere temperature series reported by Mann (2002), Mann et al. (2003) and Mann and Jones (2003) [red solid curves in panels a, b and c] with our best attempts to replicate some of these results (panel d). We failed specifically in replicating Mann and Jones (2003) curve in panel c and concluded that the rather alarming change of 1 to 2.5 ºC per decade suggested by the trend curves produced in the short interval of one to two years in these previously published papers are physically implausible but rather they most likely represent artifacts of methodology and procedure of trend smoothing (see further discussion in the text).

Figure 1 illustrates the basic issue for your own inspection. The first three panels in Figure 1 show charts that had been recently published on global warming within the span of a calendar year. The fourth panel summarizes my best attempt to use the same data to recover what I saw.

Figure 1 makes the obvious point that the warming trend represented in the three successive charts had somewhat inexplicably shown warmer and warmer temperature values near the year 2000. That is what prompted my closer look into the issue of defining a temperature trend.

I will now summarize the three key issues we raised in our GRL paper.

First, we wanted to determine what is the 40-year or longer trend in the global ? or for the case of current discussion, the Northern Hemisphere ? averaged temperature?

Figure 2: Various methods of smoothing or filtering the annual temperature changes in order to present the 40-year or longer trend in the record. We note that the trend estimates are not as secure as one would think: for example, the approach used by United Nations IPCC Third Assessment Report (2001) or TAR (2001) involved the subjective padding in of data values from 2001-2020 in order to define the trend values at year 2000 (panel b). We strongly disagree with such an approach and call further caution despite being publicly accused that our work constitutes "an intellectually dishonest approach ... The researcher [referring to Willie Soon] had produced very poor work in the past, and isn't taken very seriously in the climate community. This sounds like another in their installation of bad work. I'm amazed this paper got into print." (see Whipple's January 26, 2004's United Press International article at

Figure 2 shows the results for the 40-year smoothed temperature series from various smoothing or filtering methods. The main point I wish to emphasize in Figure 2 is that to confidently define the 40-year trend value centered at 2000 one will need to know future values and tendencies of the temperature data starting from 2001 till 2020 [i.e., almost full half of the data in the 40-year averaging window is not yet known]. To visualize such a smoothing methodology: consider that fact that when one defines the 40-year smoothed trend at 1980 from the annual temperature record, one is already using up annual values from 1960 to 2000 -- which is the 40-year duration for this averaging or smoothing procedure. So if one wants to define the 40-year trend value centered at 2000, one need to average annual data from 1980 to 2020 (again the 40-year averaging or smoothing interval) but the later chunk of data from 2004-2020 are obviously not available for the operation.

This is why the most conservative estimate (since it does not depend on data outside the realm of actual measurements) of the 40-year smoothed trend for the currently available annual time series from 1856-2002 is given by the top panel (the red curve in panel a) of Figure 2 which starts at 1876 and ends at 1982 or so.

Figure 3: Four successful replications of previously published results. This is why it is particularly significant that we had failed to reproduce Fig. 2a of Mann & Jones (2003) [see Figure 1 panel c above] and Fig. 2.21 of IPCC TAR (2001) [see Figure 4 below]

The second issue we raised in our GRL paper is that even though we were successful in replicating several published series (see Figure 3) including some of those in IPCC TAR (2001), we disagree with the very subjective data-padding procedures commonly adopted within the global warming research community. The IPCC TAR actually recommended padding artificial data series in order to display the trend values of the Northern Hemisphere temperature record continuously from 1856 to 2000 or so. In other words, arbitrary temperature values from 1836 to 1855 and 2003-2020 were introduced artificially in the temperature trend calculation. We disagree firmly with such a procedure.

The primary author of the curves shown in Figure 1, Professor Michael Mann, admitted to the arbitrary treatments of missing data points in the extensive quote below (Dan Whipple's January 26, 2004's United Press International article at here):

"Mann said he agrees with Soon about the degree of uncertainty, but added his work is based on techniques that have been used by time-series experts over the past decade or more."There are three different ways you can smooth a time series," he said. "Or put another way, the smoothing is non-unique because you don't know what happens to the series outside of the interval where you have data." When a moving-average series bumps into the end of the data, Mann said, "there are additional constraints, called an inverse problem. You don't know exactly what happens, but you can invoke various constraints" to estimate the continuing trend. Such constraints include:

n Setting the missing data points at zero;

n Setting the missing data points at the mean of the available points, and

n Allowing the missing points to preserve some characteristics of the earlier trend. ...

"In past published work, I and other people have described methods where you can use each of these constraints," Mann said. "What you're trying to do is minimize the misfit with the raw data. You can try each of these constraints and see which one minimizes the misfit of the smooth series from the raw series. And the method in the paper -- Mann and Jones 2003 -- where (Soon and colleagues) couldn't figure out which method we used, that's the method we used ... It's an objective method. It finds which of those three choices minimizes the misfit." "

Despite Mann's assertion, no amount of math or "constraints" will allow the highly presumptuous confidence in giving the long-term trend values at the year 2000. Such an open admission and circular explanation by the author strengthens the case for our original concerns discussed in our scientific research paper.

The third and final points we raised in our GRL paper concern our failures in replicating both the smoothed temperature curves shown in Figure 2.21 of the United Nations IPCC Third Assessment Report (2001) or TAR (2001) [see Figure 4 here] and in Figure 2a of Mann and Jones (2003) [see panel c in Figure 1 above]. The failure in reproducing the latter curve for the Northern Hemisphere in Mann and Jones (2003) is especially puzzling since we were able to reproduce the smoothed curve for the Southern Hemisphere (as Figure 2b in Mann and Jones, 2003) from that same paper simultaneously (see the successful replication # 2 shown in Figure 3 above). This fact serves as a strong support that the very high value of > 0.5ºC at year 2000 for the smoothed Northern Hemisphere curve by Mann and Jones (2003) [again see panel c in Figure 1 above] is not a mere typographical error.

We therefore conclude that the quantification of the late 20th century Northern Hemisphere temperature trend is highly sensitive to the choice of methodology, the treatment of data, and the choice of data presentation (i.e., the size of the window of the smoothing filter). We emphasized that padding additional data at the endpoints or the "missing data points" (as done in IPCC TAR, 2001, or in panel b of Figure 2 above) is unphysical and can be misleading and thus should be avoided.

Figure 4: Fig. 2.21 of IPCC TAR (2001) failed the replication test.

What lesson do I take away from this? For that I like to return to Kary Mullis's view of the scientific method. It is clear that a scientific statement or claim does not end with published or peer-reviewed papers. But it is extremely important that a scientific statement or claim can be subjected to replication. I am mindful of what Mullis meant when he suggested that all scientists should be self-restrained especially in offering unsubstantiated claims or results. My small effort was indeed only trying to discuss what I do not understand and was not meant to confront anyone, including the authors of those cited research papers. I simply do not understand how the smoothed temperature curve in Mann and Jones (2003) can indicate a value of 0.55ºC at the 2000 endpoint but the same variable was only showing a value of 0.3 to 0.4ºC less than a year earlier? As in Mullis's world of DNA biochemistry, I would presume that no purpose is served (i.e., no lives saved) when someone along the chain of reasoning makes claims that have no basis in reality. This is why the scientific tenet of replication must be insisted.

[1] Actually the paper was not off to a good start, it was first scheduled to appear on January 27 after being approved by two reviewers and the editor in charge. But upon not seeing our paper appearing as planned and promised and after several rounds of my persistent inquiries, I found out that the publication of our paper was being delayed and the paper may not be printed because of a presumed copyright violation issue. The misunderstanding about and the attempt (indicated below) to prevent the publication of our paper is both alarming and sad but the raw fact is that I had obtained all the necessary copy-right permissions for the purpose of this academic research work prior to the submission of our scientific manuscript to the Geophysical Research Letters (GRL) around November 2003. The matter is now resolved and our paper was published two-and-a half weeks later. Upon my satisfactory resolution with the GRL production office by presenting all the proofs, the GRL production person I spoke to said: "We should not have taken Michael Mann's word for it."


TCS Daily Archives