12 October 2009

A Cherry Picker's Guide to Global Temperature Trends

Chip Knappenberger has written an informative and fair post that clearly explains how global temperature trends look across different datasets and choice of starting dates, and which is summarized in the figure above. Here is Knappenberger's caption to the figure:
Cherry-Pickers Guide to Global Temperature Trends. Each point on the chart represents the trend beginning in September of the year indicated along the x-axis and ending in August 2009. The trends which are statistically significant (p [greater than]0.05) are indicated by filled circles. The zero line (no trend) is indicated by the thin black horizontal line, and the climate model average projected trend is indicated by the thick red horizontal line.
Based on this analysis Knappenberger calls out Real Climate and Richard Lindzen for blatant cherry picking:

Another use of my Cherry-Pickers Guide besides choosing your own analysis, is to check and see what level of cherry-picking was required to support some statement of the behavior of global temperatures that you saw somewhere.

For instance, in a recent post over at RealClimate.org, Stefan Rahmstorf used about 10-yr to 11-yr trend in the GISS dataset to support the idea that global warming was proceeding pretty much according to plan, concluding “the observed warming over the last decade is 100% consistent with the expected anthropogenic warming trend of 0.2 ºC per decade, superimposed with short-term natural variability.”

A quick check of my Guide would show how carefully Rahmsdorf’s selection was made. Trends a few years longer or a few years shorter that the period selected by Rahmstorf would not have borne out his conclusion with as much conviction.

Another example of careful data selection can be found in recent claims made by Richard Lindzen who is fond of stating that “there has been no statistically significant net global warming for the last fourteen years.” A quick check of my Cherry-Pickers Guide shows Lindzen to be particularly crafty because there is no support for such a statement in any of the five datasets. So how did he arrive at that conclusion? By using annual data values instead of monthly data.
Knappenberger provides some robust conclusions as well as some examples of creative cherrypicking:

Here are a few general statements that can be supported with using my Cherry-Pickers Guide:

• For the past 8 years (96 months), no global warming is indicated by any of the five datasets.

• For the past 5 years (60 months), there is a statistically significant global cooling in all datasets.

• For the past 15 years, global warming has been occurring at a rate that is below the average climate model expected warming

And here are a few more specific examples that the seasoned cherry-picker could tease out:

• There has been no (statistically significant) warming for the past 13 years. [Using the satellite records of the lower atmosphere].

• The globe as been cooling rapidly for the past 8 years. [Using the CRU and satellite records]

Or on the other side of the coin:

• Global warming did not ‘stop’ 10 years ago, in fact, it was pretty close to model projections. [Using the GISS and NCDC records beginning in 1998 and 1999]

• Global warming is proceeding faster than expected. [Using the GISS record staring in 1991 or 1992—the cool years just after the volcanic eruption of Mt. Pinatubo]

I am sure the more creative of you can probably think of many others.


  1. Kudos to Mr. Knappenberger for this extraordinarily useful post. I read a variety of blogs, both skeptical and pro-AGW, and I am often left with my head spinning regarding claims on recent trends. This post appears to sum up the data quite well in an easily understood format. I'll try posting a link at Real Climate and see whether it gets through.

  2. The folks at GISS have long argued that - according to their computer models - Man Made Global Warming is, by far, most evident in northern hemisphere winter warming.

    And yet, even the GISS preferred data sets demonstrate that the continental USA (part of the northern hemisphere) has been in a (minor) winter cooling trend since 1992!

    Of course, that winter cooling trend is even more pronounced over the last 11 years.

    Click here to further examine the current cooling trend. There, you will find instructions for replicating the graphs for yourself (using surface station data -- with a demonstrable warming bias -- from NOAA).

  3. lkdemott - someone called PaulD has got the link through at RC (comment 325 on stefan's "warming pause" thread).
    What's amusing is Gavin's reply:
    "[Response: Vaguely ok, but he has failed to take the clear auto-correlation in the monthly data series into account and so his statements about significance are all biased to be over-definitive. - gavin]"
    In effect, Gavin is therefore defending Richard Lindzen!

  4. Its an interesting post by Knappenberger. However, I'm not sure the logic of the situation allows us to call the situation a "wash" simply because multiple outcomes could be teased from the available data. After all only one group is making the case that the temperatures we are witnessing are within the range of normal naturally occuring variation, and only one group is claiming that we are witnessing something unprecedented. Lindzen isn't attempting to prove the opposite of the AGW theory.

  5. It's worth nothing that Chip shows the trend of "about 2 C/century". In reality, the multi-model mean trend is not a constant, and is supposed to be higher if we fit a least squares trend to temperature starting soon after the eruption of Pinatubo. So, the the seeming "good" agreement for observed and model trends starting in the mid 90s vanishes if we compare to actual model projections.

  6. Would I be accused of cherry picking for examining:

    1) The ongoing 10,000 year cooling trend at Vostok

    2) The ongoing 10,000 year cooling trend on the Greenland ice sheet?

    3) The 140,000 year trend at Vostok (no similar data available yet from Greenland).

    4) The 423,000 year trend at Vostok.

    Click here and here to -- among other things -- verify all the data sources.

    Here is the cold hard fact:
    There is NOTHING even REMOTELY unusual about current temperatures OR current trends!

  7. I am surprised at

    1) the vagueness of the "AGW hypothesis".

    2) the inability to quantitatively describe the level of natural variation in various climatological metrics.

    3) the vagueness of IPCC predictions. ooops -- I guess I mean "projections".

    4) not only are the calculation methods cherry picked, but there is also cherry picking of what metric is an appropriate test of the AGW hypothesis --- global mean temperature, Arctic (but not Antarctic) sea ice, ocean heat content, severe weather incidents.

  8. If this were simply an intellectual exercise then having both viewpoints supported by the same data would make the issue a draw. However AGW has been proposed as justification for an extraordinary change in our society, both in the redistribution of wealth and control of the energy sector.

    That extraordinary claim on behalf of AGW requires extraordinary proof. Having a hypothesis that is consistent with some of the data some of the time falls far short of justifying a trillion dollar tax on an already struggling economy.

    In other news; there is no Geneva Convention protecting innocent data and the English language from teasing and torture.

  9. PaulM-I thought auto-regressive models "lack physical realism" (IPCC)?

    You know, to everyone, 12 years without warming is a rather long stretch from what I can tell.

  10. Roger,

    Off topic, but…
    I thought you might appreciate Thomas Fuller’s latest comments on Joe Romm (and my own).

  11. Gavin is pointing out that the article has contains a statistical error and the significance tests depicted in the diagram at top of your post are incorrect. Perhaps you should update your article so that (more) readers aren't mislead.

  12. -11-andrewt

    Sure, what exactly is the error and how exactly do you propose correcting it?

  13. Roger--
    Likely as not, Gavin think Chip didn't correct for lag-1 autocorrelation. But, I've done that many times. Correcting for that won't change Chip's discussion in any way that matters. Chip setting the projection to 2C/century for all start years is a more substantive issue-- and correcting that will show the observations differ from the models more than Chip indicates.

  14. Per Chip's description, it would seem any reference to temperatures would fall into the "cherry-picking" camp. For example, if using annual data instead of using monthly data results in being called a "blatant" cherry-picker then it has become a meaningless term that is only being used to discredit a differing viewpoint.

  15. C3--
    There are two problems with using annual average data.

    The less important one is that if you use say, 10 years of annual average data instead of 120 months of monthly data, the estimated uncertainty intervals will nearly always be larger for the annual average data. So, by averaging, you lose statistical power. This is never a good choice from a statistical testing point of view. Worse, if the choice is made because you want to say "there is no statistically significant trend", then it you have cherry picked your statistical method to find one that gives you the answer you "like".

    Second, if you use annual average data to give yourself an excuse to ignore recent data, knowing the recent data overturns your result, you are, once gain cherry picking your end date.

    Oddly, while Lindzen may be doing this now there were a number of blog-arguments about the practice last year. The only thing is that last year, it was people who wanted to ignore the cool 2008 monthly data that preferred annual average data to monthly data.

  16. "Misled"? You've got to be kidding me! The significance of the trends is such a tiny part of the issue-I don't even begin to understand what the point of that is.

    In all likelyhood, that would just eliminate significance of all trends after 2000 and back to 95. Why does that matter, when the values would still be the same, in a way that's irrelevant to the main point?

  17. I think it's important to point out that in Rahmstorf's article, he points out that shorter-term trends do not show the same warming he is talking about. He is not hiding the period of time he uses or the form of data. I wonder if the same can be said for Lindzen?

    Also, I think it's wrong to just refer to "Real Climate" in these kinds of issues. If a post there is by an individual, then that individual is the one to discuss. Unless someone knows that group approval is required for all such posts, or for the posts labeled as group.

  18. Dean,

    What is your reaction to the on-going 10,000 year cooling trends at both poles?

    I documented those trends in this comment.

    Note that the latest warming has NOT broken either of those trends!

    How about the 423,000 year trends?

    Or, would you prefer to do your own cherry picking for trend data?

  19. I really liked Chip Knappenberger's graph, which is very interesting. I posted it at the office for discussion.

    I have a lot of respect for Lucia, and am a regular reader of her blog. It would be very interesting to see a modified graph which addresses the lag-1 autocorrelation issue and also the issue of setting the projection to 2C/century for all start years, to see what the modified graph would look like.

  20. I did a related analysis, but starting trends in January (because that's more convenient for me.) Accounting for lag-1 autocorrelation, I get similar uncertaity intervals as Chip. So, it appears Chip did not make the mistake Andrewt thinks he made based on Gavin's guess. So, it appears it is Andrewt who has been mislead by Gavin's poor guess about what Chip did or did not do.

  21. Hmmm ... GISS 4-year is significantly negative and 10-year is significantly positive. To me that demonstrates two things:

    a)There should probably be full autocorrelation correction (not just AR1), which would broaden the uncertainty intervals considerably for the very short-term portion of the chart.

    b) Short-term linear fits are pretty well meaningless.

  22. deepclimate sez:

    “Short-term linear fits are pretty well meaningless.”

    Is an ON-GOING 10,000 year cooling trend at Vostok, Antarctica “meaningless”?

    Is an ON-GOING 10,000 year cooling trend on the Greenland ice sheet “meaningless”?

    Is the FACT that the Greenland ice sheet was generally warmer and witnessed a MUCH more rapid rise in temperatures circa 1930 “meaningless”?

    Or, is a “meaningless” time frame simply in the eye of the cherry picking alarmist?

    Click here and here to more fully substantiate and explore those two charts.

  23. Deep
    There should probably be full autocorrelation correction (not just AR1), which would broaden the uncertainty intervals considerably for the very short-term portion of the chart.
    You need to learn to think of a larger range of possibilities. The earth has honest to goodness "real" recoveries from temperature drops after eruptions like Pinatubo. So, the positive trend since the temperature drop after Pinatubo is real, large and can be detected even though there is "weather noise" occurring during that period. Once the trend is recovered, the temperature no longer is recovering and that portion of the trend stops contributing. So, the trend changed. (Just as the models suggest it should change.)

    On the short trend end: There is a problem with using the standard mitchel/Quennouille correction for autocorrelation for too short of trends. So, there may actually be a problem for hte 5 year trends. But it's easy to show the problem is small for the 8 year trends.

  24. Lucia,
    Certainly there is a drop in trends from a starting point of 1991 compared to 1994 due to Pinatubo. But I'm discussing the rest of the chart, obviously.

    To me the choice is obvious: either calculate the autocorrelation corrected trend properly in the sub-10 year trends, or don't show them at all.

    I'm pretty sure the three 5-year surface trends would not be significantly negative if properly calculated. The 8-year CRU negative trend may also turn out to be not significant.

    Personally, I don't think there is much point to displaying trends much shorter than 10 years, especially now that we are nearing the 10-year mark for the AR4 projections.

  25. DeepClimate -- upon editing my comment a second time -- sez (at his blog):

    see exchange at Roger Pielke jr's blog

    What “exchange”?

    Deepclimate is STILL dodging the questions. Does he find the FACTS -- as revealed by observational data from directly cited peer reviewed science (as opposed to his demonstrably laughable IPCC computer models) -- too “inconvenient” to even address?

    Yep! Looks that way to me!

    Cross posted in slightly different form -- and probably unpublished -- at the DeepClimate Blog.