An interesting new paper (behind paywall) has been accepted for publication in the Journal of the Atmospheric Sciences. The paper by Jiansong Zhou and Ka-Kit Tung of the University of Washington, Seattle is titled “Deducing Multi-decadal Anthropogenic Global Warming Trends Using Multiple Regression Analysis”. This paper will add fuel to the recent discussions about the nature of the global warming trend and whether it recently has stabilized or not. The authors by the way conclude it has not. Their main conclusions however is:
When the AMO is included, in addition to the other explanatory variables such as ENSO, volcano and solar influences commonly included in the multiple linear regression analysis, the recent 50-year and 32-year anthropogenic warming trends are reduced by a factor of at least two. There is no statistical evidence of a recent slow-down of global warming, nor is there evidence of accelerated warming since the mid-20th century.
This study is following the same approach as Foster/Rahmstorf 2011 and Lean/Rind 2008 (trying to correct the global temperature for ENSO, solar and volcanoes) but adds the Atlantic Multidecadal Oscillation to their multiple linear regression analysis. This leads to their figure 1b above. What we see is a longterm trend that has hardly changed during the past century.
Now as always this result can be interpreted in many different ways. The century scale trend is still 0.68 degrees Celsius suggesting little of the total trend of 0.8 degrees C can be attributed to solar, volcanic, ENSO and AMO. That’s what the authors seem to suggest as well when they write (bold mine):
The conclusion that we can draw is that for the past 100 years, the net anthropogenic trend has been steady at approximately 0.08 °C/decade.
So for them anything that’s left after filtering out the natural forcings and natural variability is just ‘anthropogenic’. For me this conclusion is rather premature. But before I explain why let’s focus on the other trend lines that the authors show. Just like Foster/Rahmstorf they conclude that there is no slowdown recently:
There is no statistical evidence of a recent slow-down of global warming
However the trend they find for the recent 32 years (0.07ºC/decade) is far lower than that of Foster/Rahmstorf (0.17ºC/decade). If the approach has any validity at all this would suggest that the AMO alone explains the difference between the Zhou/Tung and Foster/Rahmstorf trend.
The paper by Zhou claims that in the last 32 years, the period in which greenhouse gases are supposed to be the dominant forcings, in fact some 60% (0.1ºC of the total 0.17ºC/decade) of the trend can be ‘explained’ by a combination of ENSO, AMO, solar and volcanic forcing). Ergo, only 40% of the trend could be attributed to other factors among which greenhouse gases are of course a logical candidate.
However there are other candidates as well of course. There is ongoing debate about the influence of siting issues on the temperature measurements on land as well as the Urban Heat Island effect and other socio-economic influences. In a controversial and well known paper Michaels/McKitrick estimated that “Using the regression model to filter the extraneous, nonclimatic effects reduces the estimated 1980–2002 global average temperature trend over land by about half.” If true even less of the remaining trend can be attributed to greenhouse gases.
The Zhou study could therefore have serious implications for our estimates of climate sensitivity. The paper though is completely silent about these potential implications, something that reviewers could have raised.
As said above Zhou and Tung call the remaining century long ‘underlying’ trend ‘anthropogenic’. Whether this is ‘right’ could be questioned with their figure 2 (see below). Here one sees that the anthropogenic forcing (green line) seems to underestimate the adjusted trend in the period (1889-1970) while it seems to overestimate the trend thereafter. This suggests that still not all the relevant factors (either natural or anthropogenic forcings or natural variability) are included in the regression analysis. The residuals in figure 2b still show trends which would not be the case, Zhou and Tung write, if the regression analysis would be perfect.
This leaves enough room for all to bend the paper in one’s preferred direction.
The big thing missing in this analysis is possible cooling by aerosols from anthropogenic origin (which happen to peak in 1940 – 1975 and in the past 15 years). Do the authors mention that?
Elmar, aerosols are included in the green line in figure 2a. However, you will see that in discussions of this paper some people will use this result to suggest that the effect of aerosols is not well represented in the anthropogenic forcing term (the green line). Aerosols have been a ‘suitable fudge factor’ (tuning factor) for some time now, also in climate models.
Marcel
So what is the net effect of anthropogenic warming and cooling in this analysis? And what exactly do we know about the amount of cooling by aerosols? There’s quite some uncertainty on the precise aerosol forcing and on ocean mixing rates, as Hansen and colleagues explain in their ‘energy imbalance’ paper:
http://pubs.giss.nasa.gov/abs/ha06510a.html
Is this fudging or tuning, or genuine uncertainty which can hopefully be narrowed down by (better) observations and analysis?
Very interesting stuff. Here are some slides from a September 2012 presentation.
Thanks Neven; there is one slide that answers Elmar’s question. They said: “When the AMO index is included as a regressor, in addition to ENSO, volcano and solar the net anthropogenic warming trend is steady for the past 100 years, and is 0.07-0.08 K/decade, less than half of the accepted values. This is controversial.” [MC: they should have added that this is controversial for the most recent 50 years; the centennial trend is not controversial] They then continue: “It suggests that the anthropogenic tropospheric aerosols increased along with industrialization faster than that used in current models.” For many mainstream scientists… Lees verder »
Another interesting slide (17) at http://www.tims.ntu.edu.tw/download/talk/20120918_2297.pdf Why doesn’t multidecadal variability play a more prominent role in AR4 models? •Individual models have different periods, different phases and different amplitudes (Medhaug and Furevik (2011)). Ting et al (2009) showed that the observed North Atlantic variability is well outside the range of AR4 model uncertainty/variability. •For the same model, not always the right phase, depending on initiation. One realization of the GFDL model got it right (Delworth and Mann (2000)) •AR4 emphasized forced response, obtained by ensemble averaging different initializations. •AR4 conclusion: the observed temperature variation can be well simulated by forced response;… Lees verder »
Slide 22:
Mechanisms responsible for the ~70-year period of the AMO
•Consider a slightly stronger THC. After a lag of several years makes North Atlantic SST warmer; increased Arctic ice melt; reduced deep water formation.
•Takes about 20 years to slow down the THC a little.
•A slower THC has decreased SST transport with a delay of several years.
•Half cycle of ~35 years.
•A colder Arctic has less ice melt…..
Slide 23: Atlantic Multidecadal Oscillation A brief history •Phrase coined by the Science writer Kerr (2000), who attributed the discovery to Delworth and Mann (2000). •Schlesinger et al (2000) disputed the attribution, and claimed that the credit should go to Schlesinger and Ramankutty (1994), who found two cycles with period 65-70 years. •Actually it should be Folland et al (1984), who found a worldwide temperature fluctuation of 0.6 K with power at 83 years for the period 1856-1981. •In reply Kerr said two cycles did not constitute the discovery of an oscillation. Preferred “half a dozen or more” cycles •Delworth… Lees verder »
I found slide 22 particularly interesting (where they posit a mechanism for enhanced Arctic sea ice melt) as I’ve been wondering a lot about the influence of the AMO on sea ice melt, and whether a switch in the cycle could halt the precipitous drop in Arctic sea ice cover. If I’ve understood correctly they say it has something to do with the strength of the THC. Interesting.
Slide 34:
Conclusion
•The 2.5 cycles of the multi-decadal oscillation that were found in the global mean temperature is likely part of a recurrent oscillation extending at least to the Little Ice Age, is likely natural, and quasi-periodic with 70-year period.
•The observed episodes of warming and cooling in history can now be explained (next two slides).
•A more controversial conclusion: the anthropogenic warming rate during the early 20th century can be detected and it is no different than during the second half. The increasing anthropogenic aerosols likely masked the true greenhouse warming rate during the second half.
Slide 35: Conclusion (conti.) •Accepting that the AMO is real and recurrent allows a more consistent and coherent explanation of many observed multidecadal episodes of warming and cooling in history. Previously different explanation is given for each episode: Recent “stalling” of warming of last 20 years: Kauffmann et al (2011): coal burning in emerging China. Solomon et al (2010): increase in stratospheric water. Solomon et al (2011): variable background stratospheric aerosol. Swanson and Tsonis (2009): climate regime shift. Early 20th Century Warming: Scafetta and West (2005): solar forced. Wood et al (2010): A singular event of internal variability. Cooling period… Lees verder »
No need to thank me. I’m glad that for once you are basing yourself on serious, young scientists, instead of spreading the disinformation of some old white free market fundamentalists.
you’ve just lost your brownie points, neven.
In a controversial and well known paper Michaels/McKitrick estimated that “Using the regression model to filter the extraneous, nonclimatic effects reduces the estimated 1980–2002 global average temperature trend over land by about half.” You should have a look at McKitrick’s socio-economic data. It can be downloaded here: http://www.rossmckitrick.com/uploads/4/8/0/8/4808045/s.bma-gcm.zip Load the file ‘Global7902.csv’ in your spreadsheet, it contains data about literacy, the use of coal, population and GDP et cetera, based on longitude and latitude. Every point on earth in his model is related to the country it belongs to. For instance Alaska (lat = 62.5, lon = -147.5) has 273… Lees verder »
Jos that is serious, did you write McKitrick about this?
@Hans Erren Yes, and to be honest my source is Steven Mosher: http://judithcurry.com/2012/06/21/three-new-papers-on-interpreting-temperature-trends/#comment-211553 I didn’t think McKitrick’s model would be that simple and so wrong. Being a little skeptic by nature I checked it. Searched McKitrick’s website for the data, downloaded it and checked out the files Global7902.csv and MMJGR07.csv (which are roughly the same). Investigated a lot of datapoints, could not find a more funny point though then the one in the Atlantic Ocean. McKitrick took the Heat Island effect a little too literally. The following remark by Gavin Schmidt is also very interesting: “It’s there because if you… Lees verder »
@Jos Hagelaars I was not aware of this. Steven Mosher is always really sharp and good with data. It’s a pity you react so negative on this. Mosher himself is more neutral and in touch with Ross about this. People make errors, what’s important is that errors will be acknowledged and corrected. I do wonder (just like you I suppose) what happened since June when this was discovered. In one comment Mosher wrote: steven mosher | June 22, 2012 at 3:42 pm | Reply I’m in contact with him and have sent along the data. I think with finer resolution… Lees verder »
@Marcel Crok I’m not science officer Spock, so I do get cynical now and then about things, but it was not directed at you, but at McKitricks work. As you know there’s a long history of accusations over and forth with McKitrick involved. See for instance this quote of him: “I have been probing the arguments for global warming for well over a decade. In collaboration with a lot of excellent coauthors I have consistently found that when the layers get peeled back, what lies at the core is either flawed, misleading or simply non-existent.” In my opinion these three… Lees verder »
I have tried to reproduce the results but without success. I need a bit more time to nail down why my results are similar yet totally different but it is worth looking into. For those interested please see this graph : top left is temperature anomaly, top right AMO, bottom left CO2 radiative forcing. With AMO and CO2 one can explain a whole lot of the temperature graph (bottom right) but to best match the record statistically the slope is 0.5 per unit AMO and 0.5 per unit radiative forcing (see bottom of this post). In other words this little… Lees verder »
For the suspicious folks, here is the outcome of the least squares run with 156 years of temperature ============================================================================== Dependent Variable: T Method: Least Squares Date: Fri, 19 Oct 2012 Time: 22:44:08 # obs: 156 # variables: 3 ============================================================================== variable coefficient std. Error t-statistic prob. ============================================================================== const -0.437494 0.011886 -36.807355 0.000000 AMO 0.509615 0.038816 13.128880 0.000000 CO2 0.511995 0.015455 33.128236 0.000000 ============================================================================== Models stats Residual stats ============================================================================== R-squared 0.897273 Durbin-Watson stat 1.161624 Adjusted R-squared 0.895930 Omnibus stat 2.845848 F-statistic 668.192030 Prob(Omnibus stat) 0.241008 Prob (F-statistic) 0.000000 JB stat 2.596625 Log likelihood 161.824683 Prob(JB) 0.272992 AIC criterion -2.036214 Skew -0.232608 BIC… Lees verder »
I am not an expert in statistics, but something does not quite feel right with this paper. Here is the problem : The AMO itself is defined as the (Kaplan SST) of the Nothern Atlantic, with the LINEAR trend removed. This means that the AMO index not just contains any cyclical signal (if it’s there) but also the NON-LINEAR part of the underlying warming signal. So, if you then remove the AMO index from the global temperature record, by regression, you have just removed not just the syclical part you wanted to remove, but also the NON-LINEAR part of the… Lees verder »
Marcel, The flaw I pointed out above in this paper by Zhou and Tung is pretty basic, and if valid, challenges the conclusions, and potentially nullifies them. Could you please ask Jiang Zhou for a response to my comment ? Also, could you please tell us a bit as to how this publication came about ? For example, which criteria did you use to pick this particular paper for your post ? And which scientists or other experts did you consulted on on Zhou & Tang statistical method before you decided to publish this piece ? By the way, isn’t… Lees verder »
Marcel and others interested, I figured out what is wrong with the study. Please see this graph It shows an idealized situation where temperature is driven by only greenhouse gases (GHG) and AMO. Top left are both these forcings and the trend due solely to GHG, top right is the combined effect of GHG and AMO (resembles loosely the temperature variability over the past century). Now, the authors did not perform a simple multivariate regression but first tried to filter out the natural signal, in this case AMO. That is shown in the bottom left graph, and the residuals are… Lees verder »
Guido, The problems with this paper may be far worse than what you suggest. For example, in your analysis and graphs, the AMO is still linearly de-trended, which assumes that (anthropogenic) global warming was linear over the past century, which we KNOW is not correct. Tamino actually did a very good post on why the AMO is not just a “natural signal”, but also includes the non-linear (anthropogenic) global warming signal itself : http://tamino.wordpress.com/2011/01/30/amo/ which significantly accelerated in the second part of the 20th Century. So, could you do your analysis again, but this time, instead of the (linearly de-trended)… Lees verder »
@Rob Dekker Rob, calm down please. I have to do blogging in my spare time, which having a family is scarce. This weekend I was not online. I try to follow the discussions as much as possible and what Guido is doing (trying to replicate the results) is very interesting. We talked on the phone last night and maybe in a few days we can do a joined follow up post in which we raise some issues. Apparently your expectations of a blog post are rather high. In this case I saw this paper, read it and wrote a short… Lees verder »
Rob, I see your point and only now see that you brought this up earlier. I am not sure if I totally understand it, but to me it looks one has to take the detrended data. If not, you would even get a lower trend because basically one would subtract the trend from the trend, leaving only noise. If the AMO contains both a natural and a anthropogenic signal, then to me it makes sense to first subtract that anthropogenic part so that the natural part is isolated as good as possible?! I’d be happy to make a few additional… Lees verder »
I personally find this discussion quite interesting, so I have to thank Marcel for posting it. Note that he is basically just giving an overview of the paper. I think we all know that Marcel is a bit skeptic, and though I sometimes object to his interpretation of events/research I don’t think this post was so bad. I am quite impressed with the points raised against this article though, and it has started to smell a bit fishy in my opinion. I wonder what points were raised in the review process.
It might indeed be a good idea to contact the authors and maybe write a comment to it.
Guido I am not sure if I totally understand it, but to me it looks one has to take the detrended data. If not, you would even get a lower trend because basically one would subtract the trend from the trend, leaving only noise. Let me give an example of the method used by Zhou and Tung, using a hypothetical situation : Suppose that Northern Atlantic sea surface temperatures do not have any cyclical behavior, and just warm with the rest of the planet. And suppose that global warming was only driven by anthropogenic forcing and was observed to be… Lees verder »
Marcel,
I appreciate your honest reply.
I would like to add some comments and context to your response, but it’s too late now (I’m living in California) so that will have to wait till tomorrow.
I owe you guys a summary and a bit of background of what is wrong with Zhou and Tung 2012. This paper is classic case of ‘extreme’ “contamination” in regression analysis. When you want to remove a known variation in a climate forcing from the global temperature record, you always have to be very careful not to mix cause and effect. After all, if your ‘forcing’ contains a global warming signal, then that too will be eliminated from the global warming record ! In extreme cases, if your ‘forcing’ (cause) mainly consists of the global warming signal itself (effect), then… Lees verder »
@Rob Dekker
Interesting, Guido is following the same reasoning.
It would be interesting if you asked Tung to react on your comment.
Marcel
Marcel said
Marcel, YOU put this paper out in the open (with a twist). Your post made it via Anthony Watts, all the way to Marc Morano.
So, I’m sure that if Dr. Tung has something to say, he can do it right here, where the myth started.
@Rob Dekker
Guido is having a detailed look at the paper and hopefully we can come up with a guest post from him in a few days. For now that’s the approach I want to follow.
Marcel
@Rob – thanks for pointing that out. I don’t fully understand though: with detrending the AMO one filters out most of the antropogenic signal, right? In my mind the anthropogenic signal is not linear so one has to do it a bit more careful than just taking the slope out, but if you would bend the AMO downward as in http://tamino.files.wordpress.com/2011/01/sst-giss.jpg then you make the opposite mistake of when you would not detrend it at all, right? Anyway, I have been able to reproduce their results and my earlier conclusion about not including a trend in the MLR was wrong.… Lees verder »
I’m looking forward to Guido’s post on the flaws in the Shou and Tung paper. Marcel, As a journalist, if it turns out that a publication you made is based on flawed science, and/or the conclusions you drew (maybe even the title of the publication) turned out to be an artifact of the flawed method used in the paper you summarized, what would you normally do ? Would you post an update to the publication ? Or a separate new post with corrected results ? Or would you post a retraction ? Or would you just do nothing, and leave… Lees verder »