--by Horatio Algeranon
Mouse over image to see what short term trends are about.
Of short-term trends
As means to ends
We've had our fill (and more)
The ups and downs
Of see-saw clowns
Have made our brains quite sore.
This short trend reliance
Is see-saw "science"
Unstable at its core.
The claim that global warming has stopped or abated in recent years seems to be very popular among global warming "skeptics".
Often, this claim is based on the observation that the global average temperature is not as high today as it was at some time in the recent past (eg, 2005). This method is basically useless for gaging even the direction in which temperatures are headed over short time periods, as discussed here.
But sometimes, the above claim is actually based on taking a "trend" of the temperatures over recent years (since 2001, for example) to attempt to "see" if the temperatures have "plateaued" or (possibly) are even decreasing.
Though this approach is better than the previous one, not by much. There is still a significant problem -- not with trend analysis in general, but with estimating the trend in temperatures over a short period of time. Short term trends are very uncertain..
It is a somewhat under-appreciated fact that temperature trends obtained with linear regression on short time periods (eg, 8 years) can be quite highly dependent on the choice of starting and ending points. In some cases, for example, this "apparent" trend changes from positive to negative (from increasing to decreasing) when one simply narrows -- or even "slides" -- the time period under consideration just a little bit.
You can see this effect in the following graphs of trends in the NASA GISS monthly global land-ocean temperature anomalies covering slightly different periods of time.
In the first graph: the blue line shows the (positive) trend from Oct. '99 through Sept. '08 -- 9 full years (the proper way to do trends based on monthly data, to avoid seasonal bias). The red line shows the (negative) trend from Jan. '01 through Sept. '08 -- under 8 years, or slightly more than a year less.
NASA GISS Global Land-Ocean Monthly Temperature Anomalies in deg C
( "Anomaly" is change from 1951-1980 mean )
That may not seem like a very big change, but it is equivalent in magnitude to the trend (change in temperature per decade) for the period from 1975 through 2000 (+0.16 +- 0.05 deg. C/decade, 2-sigma).
Also, the apparent trend from Oct. '99 through Sept. '08 (+ 0.14 deg. C/decade) does not differ appreciably from (is within the uncertainty of) the trend from 1975-2000 . Note: the -0.17 deg. C/decade "swing" in the apparent trend is still well within the uncertainty attached to the trend for such a short (~8 year) period, which is relatively large (+-0.3C/decade, 2-sigma)
There is even a see-saw effect for shortening the period from Oct 2000 through Sept. '08 by just 3 months, as can be seen on the second graph below. The blue line shows the trend from Oct 2000 through Sept. '08 (8 full years) . The red line shows the trend from Jan. '01 through Sept. '08 -- just 3 months less! But what a difference 3 months apparently makes.
NASA GISS Global Land-Ocean Monthly Temperature Anomalies in deg C
( "Anomaly" is change from 1951-1980 mean )
So, what should we make of this see-sawing?
Well, the wise would read this as a warning: "Beware! The short term trend may not actually be what it appears to be!" (why it was referred to above as an "apparent" trend.)
To those who care about not overstating their case, this "see-saw" issue is recognized as a problem that makes (apparent) trends for short time periods unreliable -- and suspect.
This is one of the reasons that it is absolutely critical to "bracket" the range of likely trend values. But there are those who focus on this single (least squares fit) line, as if it somehow represents the "one and only" trend that is consistent with the data.
That is simply not the case. Because of uncertainty associated with the calculated trend (from noise in the data), this line shows only the "apparent" trend. The actual trend might coincide with any one of many lines with differing slopes falling within some range. This range** can only be determined with proper statistics. It can not be done "by eye", for example (by just looking at the temperature graph). Statisticians understand this all too well.
**The 2-sigma uncertainty in the temperature trend for the past 8 years of GISS temperature data is +- 0.3 deg C/decade which means that any claim that "the temperature trend since the beginning of 2001 is negative" is unwarranted. In fact, the trend could actually be as great as + 0.27 deg C/decade (if one uses -0.026 C/decade as the central value) or even + 0.35 C/decade (if one uses 0.047 deg C/decade as the central value -- obtained by trending since October of 2000 rather than January of 2001).
Unfortunately, there seem to be some who see the see-saw effect not as a drawback but as an opportunity. The see-saw can be employed all too effectively on an unsuspecting public by those making claims that go against the scientific consensus.
The possibilities for "convenient adjustments" to the see-saw are virtually limitless:
"Which see-saw position do you want? Up or down?" [Increasing trend or decreasing?]
"Coming right up" [ or down -- depending on what these folk wish to "prove"]
Of course, those who are keen to make such "adjustments" (ie, play with the interval to give them a trend they like) invariably show or specify only a single trend, as if that were somehow the "one true trend" (ie, the only trend that is consistent with the data). They do not show or mention the range of possibilities that statistics suggests.
Whether by design or default, the "global warming has stopped in [insert a date here]" claim seems to have ended up on "Short Trend Road" -- a dead end. The question of whether global warming has stopped or even slowed in recent years can not be reliably answered by short term trends (since 2001, for example).
The much better approach to answering the latter question is to see if the temperatures over the last few years are consistent with the long term trend (which can be gaged more accurately than a short term trend)-- or if they fall outside the normally expected "scatter" about (0.2 deg C above or below) the long-term trend.
In other words, do recent temperatures represent a significant departure from the warming trend experienced by the earth over the past few decades?
Tamino has asked -- and answered -- that very question in the best post ("You bet!") that I have seen anywhere on the web dealing with the issue of whether global warming has stopped or even slowed in recent years.
He shows the following graph of annual global mean temperature anomalies with accompanying text
Here are global average temperature estimates, all set to the same zero point (using the reference period 1950.0 to 1980.0), from NASA GISS, NCDC, and HadCRU: The trend lines are determined from the data covering the time span 1975-2000. The graph is intended to show that the data after 2000 are not inconsistent with the claim that the trend is continuing, in fact they’re following the line with “wiggles” (i.e., noise) that make trends impossible to identify over short time periods but clear over longer time periods. (and indeed that is so)
As Tamino notes, "there’s really no evidence — none whatsoever — that global warming has stopped".
But you can see that for yourself on the graph above: all of the temperatures (for the 3 data sources) since 2001 are consistent with the respective long term (1975-2000) trend .
