Arguments
Has the rate of surface warming changed? 16 years revisited
Posted on 21 May 2013 by Kevin C
Climate scientists have traditionally looked at climate over long periods - 30 years or more. However the media obsession with short term trends has focussed attention on the past 15-16 years. Short term trends are much more complex because they can be affected by many factors which cancel out over longer periods. In a recent interview James Hansen noted "If you look over a 30-40 year period the expected warming is two-tenths of a degree per decade, but that doesn't mean each decade is going to warm two-tenths of a degree: there is too much natural variability".
Over the winter vacation we produced a video which tried to explain the contributions to the recent temperature trend based on the best evidence available at the time, however the rapid pace of development in this area has thrown significant doubt on the conclusions. While the video has significant educational content, the conclusions do not reflect a scientific consensus, so we will be withdrawing it and will work on an updated version.
The problem
The video was based on an approach pioneered by Lean and Rind (2008) and Foster and Rahmstorf (2011), by determining the contribution of known influences on global temperature to best explain those temperatures. However this approach can give misleading results if significant influences on temperature are missing from the analysis, or if wrong influences are included. Therefore we need a comprehensive list of possible factors which might affect the short term trend. Based on the latest literature, the following should be considered:
Natural influences (included in the 16 years video):
- Volcanic eruptions. The recovery from the Pinatubo eruption is responsible for a short term warming, although in the video the effect is small.
- The solar cycle. The current low solar activity produces a small cooling effect.
- The El Nino oscillation. The recent run of La Niñas produces a moderate cooling effect.
Observational biases (not included in the video):
- Coverage bias. The HadCRUT4 and NOAA temperature records don’t cover the whole planet. Omitting the Arctic in particular produces a cool bias in recent temperatures. (e.g. Hansen et al 2006, Folland et al 2013). The video avoided this problem by using GISTEMP. However the issue affects the Foster and Rahmstorf analysis of the other records.
- Sea surface temperature bias. The GISTEMP and NOAA temperature records don’t include corrections for the transition from warm-biased engine room measurements to buoy measurements over the past 15 years. This produces a cool bias in recent temperature trends, although this result is based on only one study (Kennedy et al 2012).
Other short term influences (i.e. things we are trying to detect):
- Changes in ocean heat uptake. A number of recent papers have found evidence that heat has been going into the oceans rather than the atmosphere recently, see in particular Balmaseda et al (2013), Guemas et al (2013), Nuccitelli et al (2012) and Levitus et al (2012).
- An increase in cooling particles in the atmosphere. While we can assume a near-linear short-term response to long lived greenhouse gasses such as CO2, short-lived atmospheric constituents can violate this assumption. Chinese aerosol emissions have varied significantly over the past 16 years (Klimont et al 2013), however Murphy (2013) suggests the impact has been limited. Neely et al (2013) find a significant cooling contribution from volcanoes.
There may also be contributions from long term ocean cycles such as the AMO and PDO. Any of these factors may contribute to a slowdown in surface warming over the past 16 years.
Here is the problem: We have too many explanations for a recent slowdown, If all of these influences are present, we should probably have seen significant cooling over the past 16 years. And yet if the simple fitting calculation is correct, there has been little or no change in the underlying rate of warming.
How is this discrepancy to be resolved? On the basis of analysis by Troy Masters (part 1, part 2), I believe that the most significant problem lies in fitting the volcanic response. While the magnitude of the volcanic response is fairly well estimated, its duration is not. Changes in the duration of the response translate to a big change in the adjusted trend since 1997. The following animation shows the problem:
If the Pinatubo cooling is long lived then some of the recent warming is due to a recovery from that cooling, leaving less to be explained by greenhouse warming. Or conversely, if there has been a slowdown in warming, the simple fit would account for it by demanding a short Pinatubo response.
What does the literature say about the Pinatubo response? There are papers supporting both a long response (and thus a reduction in the adjusted temperature trend) and a short response (and thus no reduction in trend). Here are a sampling:
| Short Pinatubo Response | Long Pinatubo Response |
| Parker et al 1996 | Soden et al 2002 |
| Lean & Rind 2008 | Bender et al 2010 |
| Foster & Rahmstorf 2011 | Hansen et al 2011 |
| Folland et al 2013 | Rypdal 2012 |
| Merlis et al | |
| Survey by Troy Masters |
The volcanic temperature response from the video is compared to three different estimates from the table above in Figure 2:
Hansen et al (2011) produces the strongest volcanic contribution to recent trends. The impact of this difference can be seen in the adjusted trends since 1997, and the contributions of the various influences.
| 16 years | Hansen | ||
| Observed trend | 0.080 | 0.080 | |
| Volcanic | - | 0.010 | 0.082 |
| Solar | - | -0.014 | -0.012 |
| El Niño | - | -0.041 | -0.045 |
| Adjusted trend | = | 0.124 | 0.056 |
The Hansen paper is an extreme case, combining a strong volcanic forcing with a model with high sensitivity, and so probably provides an upper bound for the volcanic influence on temperature. That effect is however substantial, more than cancelling the effect of the El Niño and solar influences, and producing an adjusted trend which is less than the observed trend.
Conclusions
Where does this leave us? In order to reliably interpret surface temperature variations we need a good idea of all the causal factors, including El Niño, solar irradiance, volcanic eruptions, observational biases, changes in ocean circulation and possible long term oscillations. Fitting the surface temperature record is attractive because surface temperatures are easy to understand, and the calculations are easily reproducible by non-specialists. However it may be that the surface temperature record is simply too complex to analyse in this way. The more direct measure of global warming provided by measuring the energy content of the climate system avoids many of these problems, although the observational record is shorter and less complete (e.g. Church et al 2011).
There will undoubtably be new developments in the measurement and attribution of short term trends over the coming months, and we will report new results as they are released. The '16 years' video still contains useful material for showing how natural influences impact global temperatures and so we aim to produce a new version in future. However the conclusions of the current video do not represent a consensus in the peer-reviewed results, and thus we will be withdrawing the current version.
No, no, no. This isn't how the game is played. If some of your evidence is incorrect, you don't withdraw it, you shout it louder.
(I am on a Heartland-friendly site, aren't I?)
[RH] Changed all caps to bold. (Per commenting policies.)
A few months ago I calculated trends and uncertainties for the UAH data. The second page of that PDF has a black line for the trends of the UAH data up to 2012 for different starting years. The red lines are 95% confidence uncertainty bounds which account for autocorrelation with ARMA(1,1) noise. Notice that the larger uncertainty bounds of more recent trends overlap with the smaller uncertainty bounds of the longer trends. This means that there hasn't been a statistically significant change in the surface warming rate. Here's the R code if anyone's interested.
It's even easier to use the SkS trend calculator to confirm that there hasn't been a statistically significant change in the surface warming rate. Here's an example:
GISTEMP, 1990-2000: 0.201 ±0.322 °C/decade
GISTEMP, 2000-2010: 0.096 ±0.256 °C/decade
Note that the error bars overlap, showing that there hasn't been a statistically significant change in the surface warming rate. I've tried many datasets with many potential change points, and so far all their error bars overlap.
As you say, the total energy content of the climate is a more direct measure of global warming. It's also worth pointing out that global land ice and global sea ice continue to decline, absorbing heat without warming as they melt.
Troy's analysis is interesting; I'm still reading it. However, I don't think it's necessary to withdraw the video. In my view, Foster and Rahmstorf 2011 was merely trying to improve the signal-to-noise ratio of the surface temperature record by accounting for some extraneous influences. I even pointed out that other influences like multidecadal oscillations haven't been removed.
But that's not a fatal flaw, because it's impossible to remove all extraneous influences. Similarly, when a better estimate of the long-term effects of Pinatubo becomes available, that will build upon previous work rather than demolishing it.
I agree with Dumb Scientist, that there is no reason to withdraw the video. Don't let the perfect be the enemy of the good!
good question. Has the rate of surface warming changed?
There are many ways of looking at that question. Dumb Scientist above points out one method. Another, more inferior, method is to just compare how longterm trend has actually changed with more data. Using skepticalscience trend calculator and HadCRUT4:
The 1970-2000 trend is 0.169 +- 0.056C/decade
The 1970-present trend is 0.164 +- 0.031C/decade
There's no major change there then.
A better way of looking at it using GISTEMP is this animation from Tamino which shows there is no inconsistency with the prior rate of warming:
http://tamino.files.wordpress.com/2012/12/giss.gif
The important feature of all three of these methods is that they compare two periods. This cuts to the heart of the question "has warming stopped" or "has warming slowed down". To answer that you need to compare some statistic of a recent period with the same statistic of a prior one.
But in contrast the popular method, that pushed by fake skeptics, involves just analyzing a single period. Just sticking a trend line through a recent period (eg 1998-present) and declare it has no warming with no comparison to a prior period (or at least no mention of the confidence ranges).
Fake skeptics have been banging this drum so much that I am afraid even many scientists are being influenced into believing faulty conclusions repeated often based on incorrect methods for assessing warming trends.
Dumb Scientist: I think you've hit the nail on the head - reducing the uncertainty in the trend is the key part of Foster and Rahmstorf. But I disagree with your conclusion about the video. Here's why:
The Foster and Rahmstorf calculation reduces the trend uncertainty because it takes out a lot of the confounding factors which act as noise in the trend calculation. However, the uncertainty in the result does not arise from the deviations from linearity alone - it must also account for uncertainty in the adjustments which have been made. So if you are uncertain about the duration of the volcanic response, that needs to be figured into the adjusted trend. (Propogating the uncertainties is possible but not easy - I haven't done it.)
So how uncertain are the contributions? The ENSO term seems very robust - it's comes out much the same whatever calculation you do. So the ENSO-removed temperature series will indeed give trends with reduced uncertainties.
Solar is a bit less certain, but it's not very large.
The volcanic signal however can vary a lot with the duration of the effect (depending of the timing of volcanoes with respect to your trend period). As a result, I suspect that applying the volcanic correction actually increases the uncertainty in the trend of the adjusted series, because of the uncertainty in the correction.
This shows up in the huge difference in the adjusted trend between the 16 years and Hansen calculations.
Now, this is a particular problem for the video because the video made two very specific claims concerning uncertainty: Firstly that for the adjusted data the recent trend was not significantly different from the older trend, and secondly that the recent trend was highly significant (i.e. different from zero).
Even without taking the values of the trends into account, these claims are compromised. If the uncertainty in the adjusted trend is large, the first claim is rendered meaningless and the second claim false. That is the basis for withdrawing the video.
I'd really like to wait for new results on coverage and SST bias before redoing the video (it was a lot of work), but it's a shame to lose the educational material about contributions to the temperature trend.
Sadly, I don't know much about this debate over the duration of volcanic aerosol effects. However, I was impressed by how well Pinatubo's aerosol effects were modelled (page 2), which ironically comes from Hansen et al. 2006.
Perhaps the volcanic effect could be viewed as the sum of an immediate effect (which seems to be more certain) and a delayed effect (which seems to be less uncertain). Kind of like how earthquakes have both co-seismic and post-seismic effects.
Your video removed the immediate volcanic effect, which probably reduces the uncertainties on the trend. The delayed effect might have been insufficiently removed, which means that some of the delayed effect joins all the other confounding factors that haven't been removed. That doesn't seem like a reason to withdraw the video.
One thing I didn't like about the video was that you showed the trends with no uncertainties. It might be better to show them as in the SkS trend calculator. But that's just nitpicking on my part.
Regarding the first claim. I think it's important to stress, as I did in my first comment, that the recent trend isn't statistically significantly different from the older trend even when using the unadjusted data.
I doubt that the uncertainty of the adjusted trend is larger than that of the unadjusted trend. First, the uncertainty added by the delayed effect would have to be larger than decreased uncertainty due to removing the more certain immediate effect. I haven't done the calculation, but the graph on page 2 makes me suspect that's not true.
Second, that difference would have to be bigger than the decrease in uncertainty due to removing ENSO and solar variations.
To clarify, this would only be true for timespans that overlap with the immediate volcanic effect.
Dumb Scientist @7
Do you think man contributed to the unsetled period from the 80's to 97 In the introduction to this accepted paper in 98 http://hal.archives-ouvertes.fr/docs/00/31/64/49/PDF/angeo-16-1212-1998.pdf states that the time period disgused coinsides with the same period were large scale weather modification was carried out. Don't you think these experiments would have corrupted the data from this period? This is only one experiment in that time period , does anyone have links to other experiments carried out in the time period being disgused.
jmorpuss: Large scale in terms of the global energy budget? I doubt it, and don't see any such claim in that paper.
Dumb Scientist OK then I'll retract the word large scale and replace it with what was stated large number, Just read it all and come back with something relavent to the paper. You may need a refressure cource in how electric and magnetic fields interact LINK " If the electron enters the field at an angle to the field direction the resulting path of the electron (or indeed any charged particle) will be helical as shown in figure 3. Such motion occurs above the poles of the earth where charges particles from the Sun spiral through the Earth's field to produce the aurorae. " Quote directly from Schoolphisics article. So all that is needed is a manetron, masser or laser to generate and beam electrons into a system and you will increase the magnetic part of the wave and its force.
[RH] Fixed link that was breaking page format.
jmorpuss - I believe this topic was discussed ad nauseum (and dismissed due to actual evaluation of energy levels) in earlier, more appropriate threads.
Oops... "less uncertain" should be "less certain." Sorry.
Pathetic. Bob Tisdale makes a meal out of SkS withdrawing the video and the reasons behind it. Does he not realise that a constant incremental adjustment to the consensus position is good science and how our knowledge actually advances? I guess not. A head-in-the-sand, entrenched denial of human-caused warming is much more comfortable.
The key issue here as many of the linear decomposition is the potential hiden cross-talk between factor. Solar signal as at least some linear component in. Depending of the model, volcanic too. The same situation apply to any long period oscillation fitting used by skeptics.
Without physics statistical analysis are pretty limited tools.
@JohnRussell I wouldn't worry too much about what Bob Tisdale says. He's a one-trick pony just going on about ENSO jumping up and down and leaping and cavorting with 'natural' but completely unexplained magical warming (by Bob himself, that is). I've noticed many of the WUWT faithful don't put much credence on him these days. He's missed the main point that Kevin is making altogether in his haste to push his ENSO barrow.
I don't have a view one way or another on the video. I found it useful myself but didn't take it to be the last word on the subject (nothing ever is). Still, I'm happy to wait to see what new science comes out over coming months. Or for Bob T's El Nino - maybe next year? I wouldn't be surprised if it's a doozy when it comes.
I have to say articles like this are refreshing. It forces one to think about things more and realise that there is a lot to consider when it comes to what climate change will bring - and when.
Kevin, I particularly like your animation of the possibly different effects of the volcanoes. I expect that each large-ish volcano can have different effects on climate to some extent, depending on what part of the world it erupts in, time of year etc. So that looking at other eruptions won't necessarily give an answer to what happened in any other case.
DS @2: "As you say, the total energy content of the climate is a more direct measure of global warming."
One thing that bothers me about the way the warming trend is presented to the public in the MSM is that whatever is happening with surface temperatures is always the focus of the headline and the focus first several paragraphs, if not the entire article. If there is any mention of total climate energy content, it is usually buried in a later paragraph and not represented in a prominent figure, such that I would venture to guess that most casual media consumers skim over it, if they even get that far in the article before moving on to something more scintillating. A perfect example is the recent Economist article on climate sensitivity (http://www.economist.com/news/science-and-technology/21574461-climate-may-be-heating-up-less-response-greenhouse-gas-emissions), which, when it finally addresses ocean heat content about midway through the article, focuses on the top 700 m with an accompanying graph showing a flattening (but still positive) temperature trend, and only mentions the Balmaseda et al. study of trends in the total ocean temperature profile as an afterthought.
Do you think that climatologists need to do a better job of conveying the greater long term significance of total climate energy content (as opposed to surface temp fluctuations) to the MSM, so that the MSM might focus more on the significance of total climate energy?
Would it help to use more or better analogies of ocean heat content to something intuitive and mechanical, like Verner Suomi's famous "great and ponderous flywheel," but maybe not so lofty? What about the ocean heat content as a "hand" and surface temperatures as a "yo-yo," which can go up and down but is always tethered to and pulled back towards the hand?
A helpful image could be somebody on an elevator going up while playing with a yo-yo - the height of the yo-yo goes up and down, and might be above or below the hand at any given time, but it's average height is increasing just as quickly as the average height of the hand.
Jdixon1980, is this what you are looking for?
Is it really correct to include both ENSO and ocean heat uptake changes? Aren't these pretty close to measuring the same thing?
jdixon1980: The total energy content of the climate system needs to be stressed much in the media. I've been making this point since 2009, and many of my colleagues have made similar complaints. So I'm inclined to place the blame on the media rather than on scientists.
I proposed this analogy:
Imagine filling a measuring cup at a constant rate while the water sloshes around. Sometimes the water will pile up against the side of the cup that doesn’t have the measuring tick marks. As it piles up, the water level against the tick marks might go down even as the faucet pours water into the cup.
In this analogy, the water level in the cup is the Earth’s total energy and the constant water flow is the extra radiative power added by human emissions. The side of the cup with the tick marks is the Earth’s surface, where most of our temperature sensors are. The other side of the cup is the deep ocean, which we can’t measure as well as the surface.
Water sloshing towards the tickmarks is like a temporarily warm El Nino, while water sloshing away from the tickmarks is like a temporarily cool La Nina.
Humans add extra water to the cup, but it sloshes around the cup naturally.
Humans add extra energy to the Earth, but it sloshes around the Earth naturally.
Keith: I wouldn't include ocean heat uptake in a fitting calculation, but it's the sort of effect you might detect more clearly having removed the other terms. In otherwords, a change in trend.Troy is working on doing this without using fitting methods, which means you don't have to assume the warming signal and thus gives you a much better basis for detecting changes in the warming signal.
It has been suggested that we leave the video up but add annotations to explain where the problems are. This would retain the educational content (and in practice probably enhance it, because learning often progresses through correcting wrong understanding).
But the big advantage is that everywhere the video has been embedded the annotations will now be visible.
However, John is pretty busy with at the moment, so it may take a week or two.
Dumb Scientist @2
I agree with your question about the impact of ice melting on reducing the rate of air temperature increases. I presume that earth energy balance models include the latent heat required to achieve all the ice melting we have seen over the last few years. Does anyone know of any estimates of the effect the melting has had on moderating changes in sensible temperature over, say, the last decade?
Being completely unscientific and speaking anthropogenically, Gaia is fighting back and keeping the temperature constant. A sort of light switch phenomenon. Look out when we have pushed the switch beyone Gaia's capacity to thermo-regulate. The next time we have an El Nino should be interesting. Sept 15, 2016 likewise.
Farmer Dave @24: Last year I noted that the deep ocean's heat capacity is much larger than the cryosphere's. So I think Kevin C was right to focus on the warming deep ocean purely on thermodynamic grounds.
I mentioned melting land and sea ice not because of their total effect on surface temperatures, but because those independent measurements also confirm that the climate is still gaining heat, even over the last 16 years. Here are relevant links and a short movie.
DS @21 I like the water sloshing analogy better than the "flywheel" or the yo-yo because of the built-in illustration of measurement at a single location - I assume you don't mind if I borrow it?
jdixon @27: Sure, just return it when you're done so I can use it again. ;)
I agree with jdixon1980 and Dumb Scientist: Why the heck do climate scientists refer almost exclusively to atmospheric temperature with the words "global temperature" in press releases, interviews, and even sometimes abstracts, without an emphasized disclaimer that atmospheric temperature is not even remotely close to the real global "temperature"--that is, total thermal content? Often the scientists even say publicly that "global warming has slowed" or "paused" or is on "temporary hiatus" when they know perfectly well that it most certainly is not when you properly consider the total energy content.
I asked that of James Hansen last week in his talk at NASA Ames Research Center, but I did not phrase the question clearly enough. He answered that you do not even need to do that, because even surface atmospheric temperature has not paused in its increase; it has merely not been increasing as much as it was in the decades before, and that is entirely expected variability. I think that retort is too technical for most of the public. I think it is much easier for the public to understand that the "temperature" of the entire system must be what is monitored, and clearly it has not stopped increasing. (By the way, Hansen's talk is available online on UStream.)
Apologies for the very late comment. I'd quickly like to add the results published in [Stenchikov et al. 2009] to the debate. In their [Fig.2], OHC and surface temperature changes from GFDL-CM2.1 are juxtaposed. I think it is fairly representative for other GCMs. Given that the temperature signal fades out quickly, I wouldn't be too concerned regarding the impact upon the FR11 methodology. It remains small at all time scales. What remains indeed unaccounted for is the volcanic OHC imprint (which seems responsible for the trend difference between GCMs and FR11 highlighted in [Troy Masters] analysis). However, the restoration of the volcanic OHC imbalance introduces a fairly constant (underlying) trend which is implicitly accounted for by the FR11 method.
As to changes in forcing: We all agree that they won't be detected with the FR11 method (as pointed out in the paper). If one were to assume that non-volcanic OHC anomalies approximately correlate with ENSO (as the results of Balmaseda et al. seem to confirm), one is left with changes in external forcing which the FR11 method would certainly miss, namely anthropogenic aerosols and recent changes in volcanic aerosols. While I agree that the assumption of a constant anthropogenic aerosol forcing over the last 30-40 years is questionable, at least the forcing at the beginning (end of the 1970s) and the end (today) of the analysis is fairly similar to the best of my knowledge. While major volcanic eruptions are accounted for, the more recent smaller eruptions are omitted. I adopted Fig.5 in [Vernier et al. 2011] in order to update the FR11 method until Dec 2012 for GISS temperatures. As a result, the [previous] trend estimate increases [slightly] and the tail end goes up a bit. Apart from 2012, nothing to worry so far. Having seen several unusual cold spells in 2012, the dip is explicable with natural variability. I would be surprised to see another such unexpected dip in 2013.
Bottomline: Currently, I don't see strong evidence for undetected changes in forcing which isn't considered with FR11 (after having accounted for recent volcanic eruptions). The video (which I liked a lot) seems to be as valid as before.