RealClimate logo

Unlocking the secrets to ending an Ice Age

Filed under: — group @ 28 April 2012

Guest Commentary by Chris Colose, SUNY Albany

It has long been known that characteristics of the Earth’s orbit (its eccentricity, the degree to which it is tilted, and its “wobble”) are slightly altered on timescales of tens to hundreds of thousands of years. Such variations, collectively known as Milankovitch cycles, conspire to pace the timing of glacial-to-interglacial variations.

Despite the immense explanatory power that this hypothesis has provided, some big questions still remain. For one, the relative roles of eccentricity, obliquity, and precession in controlling glacial onsets/terminations are still debated. While the local, seasonal climate forcing by the Milankovitch cycles is large (of the order 30 W/m2), the net forcing provided by Milankovitch is close to zero in the global mean, requiring other radiative terms (like albedo or greenhouse gas anomalies) to force global-mean temperature change.

The last deglaciation occurred as a long process between peak glacial conditions (from ~26-20,000 years ago) to the Holocene (~10,000 years ago). Explaining this evolution is not trivial. Variations in the orbit cause opposite changes in the intensity of solar radiation during the summer between the Northern and Southern hemisphere, yet ice age terminations seem synchronous between hemispheres. This could be explained by the role of the greenhouse gas CO2, which varies in abundance in the atmosphere in sync with the glacial cycles and thus acts as a “globaliser” of glacial cycles, as it is well-mixed throughout the atmosphere. However, if CO2 plays this role it is surprising that climatic proxies indicate that Antarctica seems to have warmed prior to the Northern Hemisphere, yet glacial cycles follow in phase with Northern insolation (“INcoming SOLar radiATION”) patterns, raising questions as to what communication mechanism links the hemispheres.

There have been multiple hypotheses to explain this apparent paradox. One is that the length of the austral summer co-varies with boreal summer intensity, such that local insolation forcings could result in synchronous deglaciations in each hemisphere (Huybers and Denton, 2008). A related idea is that austral spring insolation co-varies with summer duration, and could have forced sea ice retreat in the Southern Ocean and greenhouse gas feedbacks (e.g., Stott et al., 2007).

Based on transient climate model simulations of glacial-interglacial transitions (rather than “snapshots” of different modeled climate states), Ganopolski and Roche (2009) proposed that in addition to CO2, changes in ocean heat transport provide a critical link between northern and southern hemispheres, able to explain the apparent lag of CO2 behind Antarctic temperature. Recently, an elaborate data analysis published in Nature by Shakun et al., 2012 (pdf) has provided strong support for these model predictions. Shakun et al. attempt to interrogate the spatial and temporal patterns associated with the last deglaciation; in doing so, they analyze global-scale patterns (not just records from Antarctica). This is a formidable task, given the need to synchronize many marine, terrestrial, and ice core records.

The evolution of deglaciation

By analyzing 80 proxy records from around the globe (generally with resolutions better than 500 years) the authors are able to evaluate the changes occurring during different time periods in order to characterize the spatial and temporal structure of the deglacial evolution.

Shakun et al. confirm Ganopolski’s and Roche’s proposition that warming of the Southern Hemisphere during the last deglaciation is, in part, attributable to a bipolar-seesaw response to variations in the Atlantic Meridional Overturning Circulation (AMOC). This is hypothesized to result from fresh water input into the Northern Hemisphere (although it is worth noting that the transient simulations of this sort fix the magnitude of the freshwater perturbation, so this doesn’t necessarily mean that the model has the correct sensitivity to freshwater input).

The bi-polar seesaw is usually associated with the higher-frequency abrupt climate changes (e.g., Dansgaard-Oeschger and Heinrich events) that are embedded within the longer, orbital timescale variations. However, numerous studies have indicated that it also sets the stage for initiating the full deglaciation process. In this scenario, the increase in boreal summer insolation melts enough NH ice to trigger a strong AMOC reduction, which cools the North at the expense of warming the South. The changes in Antarctica are lagged somewhat due to the thermal inertia of the Southern Ocean, but eventually the result is degassing of CO2 from the Southern Ocean and global warming. In particular, CO2 levels started to rise from full glacial levels of about 180 parts per million (ppm), reaching 265 ppm 10,000 years ago (or ~2.1 W/m2 radiative forcing), and with another slow ~15 ppm rise during the Holocene.

Figure 1: Simplified schematic of the deglacial evolution according to Shakun et al (2012). kya = kiloyears ago; NH = Northern Hemisphere

The evolution of temperature as a function of latitude and the timing of CO2 rise are shown below (at two different time periods in part a, see the caption). There is considerable spatial and temporal structure in how the changes occur during deglaciation. There is also long-term warming trend superimposed on higher-frequency “abrupt climate changes” associated with AMOC-induced heat redistributions.

Figure 2: Temperature change before increase in CO2 concentration. a, Linear temperature trends in the proxy records from 21.5–19 kyr ago (red) and 19–17.5 kyr ago (blue) averaged in 10° latitude bins with1σ uncertainties. b, Proxy temperature stacks for 30° latitude bands with 1σ uncertainties. The stacks have been normalized by the glacial–interglacial (G–IG) range in each time series to facilitate comparison. From Shakun et al (2012)

What causes the CO2 rise?

The ultimate trigger of the CO2 increase is still a topic of interesting research. Some popular discussions like to invoke simple explanations, such as the fact that warmer water will expel CO2, but this is probably a minor effect (Sigman and Boyle, 2000). More than likely, the isotopic signal (the distribution of 13C-depleted carbon that invaded the atmosphere) indicates that carbon should have been “mined’ from the Southern ocean as a result of the displacement of southern winds, sea ice, and perturbations to the ocean’s biological pump (e.g., Anderson et al., 2009).

This view has been supported by another recent paper (Schmitt et al., 2012) that represents a key scientific advance in dissecting this problem. Until recently, analytical issues in the ice core measurements provided a limitation on assessing the deglacial isotopic evolution of 13C. Because carbon cycle processes such as photosynthesis fractionate the heavy isotope 13C from the lighter 12C, isotopic analysis can usually be used to “trace” sources and sinks of carbon. A rapid depletion in 13C between about 17,500 and 14,000 years ago, simultaneous with a time when the CO2 concentration rose substantially, is consistent with release of CO2 from an isolated deep-ocean source that accumulated carbon due to the sinking of organic material from the surface.

Figure 3: Ice core reconstructions of atmospheric δ13C and CO2 concentration covering the last 24 kyr, see Schmitt et al (2012)

Skeptics, CO2 lags, and all that…

Not surprisingly, several people don’t like this paper because it reaffirms that CO2 is important for climate. The criticisms have ranged from the absurd (water vapor is still not 95% of the greenhouse effect, particularly in a glacial world where one expects a drier atmosphere) to somewhat more technical sounding (like criticizing the way they did the weighting of their proxy records, though the results aren’t too sensitive to their averaging method). There’s also been confusion in how the results of Shakun et al. fit in with previous results that identified a lag between CO2 and Antarctic temperatures (e.g., Caillon et al., 2003).

Unlike the claims of some that these authors are trying to get rid of the “lag,” Shakun et al. fully support the notion that Antarctic temperature change did in fact precede the CO2 increase. This is not surprising since we fully expect the carbon cycle to respond to radical alterations to the climate. Moreover, there is no mechanism that would force CO2 to change on its own (in preferred cycles) without any previous alterations to the climate. Instead, Shakun et al. show that while CO2 lagged Antarctic temperatures, they led the major changes in the global average temperature (including many regions in the Northern Hemisphere and tropics).

It is important to realize that the nature of CO2’s lead/lag relationship with Antarctica is insightful for our understanding of carbon cycle dynamics and the sequence of events that occur during a deglaciation, but it yields very little information about climate sensitivity. If the CO2 rise is a carbon cycle feedback, this is still perfectly compatible with its role as a radiative agent and can thus “trigger” the traditional feedbacks that determine sensitivity (like water vapor, lapse rate, etc). Ganopolski and Roche (2009), for example, made it clear that one should be careful in using simple lead and lags to infer the nature of causality. If one takes the simple view that deglaciation is forced by only global ice volume change and greenhouse feedbacks, then one would be forced to conclude that Antarctic temperature change led all of its forcings! The communication between the NH and Antarctica via ocean circulation is one way to resolve this, and is also supported by the modeling efforts of Ganopolski and Roche. This also helps clear up some confusion about whether the south provides the leading role for the onset or demise of glacial cycles (it apparently doesn’t).

A number of legitimate issues still remain in exploring the physics of deglaciation. For instance, the commentary piece by Eric Wolff references earlier deglaciations and points out that solar insolation may have increased in the boreal summer during the most recent event, but was still not as high as during previous deglacial intervals. It will be interesting to see how these issues play out over the next few years.


  1. P. Huybers, and G. Denton, "Antarctic temperature at orbital timescales controlled by local summer duration", Nature Geoscience, vol. 1, pp. 787-792, 2008.
  2. L. Stott, A. Timmermann, and R. Thunell, "Southern Hemisphere and Deep-Sea Warming Led Deglacial Atmospheric CO2 Rise and Tropical Warming", Science, vol. 318, pp. 435-438, 2007.
  3. A. Ganopolski, and D.M. Roche, "On the nature of lead–lag relationships during glacial–interglacial climate transitions", Quaternary Science Reviews, vol. 28, pp. 3361-3378, 2009.
  4. J.D. Shakun, P.U. Clark, F. He, S.A. Marcott, A.C. Mix, Z. Liu, B. Otto-Bliesner, A. Schmittner, and E. Bard, "Global warming preceded by increasing carbon dioxide concentrations during the last deglaciation", Nature, vol. 484, pp. 49-54, 2012.
  5. D.M. Sigman, and E.A. Boyle, "Glacial/interglacial variations in atmospheric carbon dioxide", Nature, vol. 407, pp. 859-869, 2000.
  6. R.F. Anderson, S. Ali, L.I. Bradtmiller, S.H.H. Nielsen, M.Q. Fleisher, B.E. Anderson, and L.H. Burckle, "Wind-Driven Upwelling in the Southern Ocean and the Deglacial Rise in Atmospheric CO2", Science, vol. 323, pp. 1443-1448, 2009.
  7. J. Schmitt, R. Schneider, J. Elsig, D. Leuenberger, A. Lourantou, J. Chappellaz, P. Kohler, F. Joos, T.F. Stocker, M. Leuenberger, and H. Fischer, "Carbon Isotope Constraints on the Deglacial CO2 Rise from Ice Cores", Science, vol. 336, pp. 711-714, 2012.
  8. N. Caillon, "Timing of Atmospheric CO2 and Antarctic Temperature Changes Across Termination III", Science, vol. 299, pp. 1728-1731, 2003.

139 Responses to “Unlocking the secrets to ending an Ice Age”

  1. 101
    Jim Larsen says:

    96 Radge H says, “So the frustrating thing seems to be here that in the “marketplace of ideas” AGW isn’t selling properly, so it must either be a bad product or badly presented”

    He’s attempting to improve the fortunes of a product which is superior but still not trouncing the opposition. It’s a good thing for a scientist to keep the public debate in his/her mind while choosing work, and much of that is already being done. Answering skeptical scientists’ dubious papers is a prime example. Then there’s Dr Hansen’s paper on extreme temperatures, which takes informing the public to a whole new level. And now, work on clouds will narrow the climate sensitivity range.

  2. 102
    Ray Ladbury says:

    T. Marvell, you are so fricking lost.

    FACT: Whatever the source of CO2, it is depleted in C-13 and C-14–as are fossil fuels–because as CO2 rises, the ratio of C-12 to C-13 and C-14 is rising.

    FACT: Whatever is causing the warming in the troposphere is also cooling the stratosphere–know of anything other than GHGs that does that?

    FACT: The oceans are acidifying–that is absorbing more CO2–so how can they be the source of CO2.

    Dude, please stop now. You are halfway to crank status as it is.

  3. 103
    t marvell says:

    I’m looking at short/intermediate term effects, one to 72 months, largely for the practical reason that the research I know how to do pertains to that range. Climate scientists concentrate more on the long term. We’ve been talking past each other Ladbury obviously is thinking long-term, except that he brought foreward a correlaton that only pertains to the short term.
    I have a feeling that climate scientists should pay even more attention to the intermedate term, say 2-10 years. One reason is to convince the public about the reality of AGW. People can easily judge short term (daily) weather forecasts, but not long-term forecasts. They could also judge intermediate forecasts. That’s a difficult period for climate scientists to handle, but I suspect that they should devote the great bulk of their resources there. (Also, of course, forecasts in this range have immediate practical consequences.)

  4. 104
    Chris Colose says:

    You can detect the signature of ENSO (and Mt. Pinatubo seems to show up well) on CO2 growth rate…it’s small but it’s there. With just some quick looking, I found a good relation between the Nino indices (doesn’t depend very strongly on which one) and Mauna Loa monthly growth rate anomaly (at least a quarter of the variance explained) with the best fit at a ~8-9 month lag between the growth rate anomaly and the Nino index.

  5. 105
    dhogaza says:

    T Marvell:

    The physics couldn’t be clearer – as water warms it can hold less gas, including CO2. If you are contesting outgassing, you are physics deniers.

    Yet the fact that oceans are currently a net sink for CO2 and actively absorbing in the neighborhood of 40% of the CO2 emissions generated by the burning of fossil fuels is well-documented through observations.

    So either the observations are wrong or … perhaps … just something I’ll put on the table …

    You don’t understand the physics.

    I provided a link to a paper addressing CO2 absorbtion by the oceans above, which obviously you didn’t bother to look at.

    Someone else above has done the same, which I doubt you’ll look at.

    Here’s something else which you’ll probably not bother to read, a real pity, because it explains why your claim that “as water warms it can hold less gas” is not correct as it ignores the partial pressure of (in this case) CO2 in the atmosphere. That’s a no-can-do-dude.

  6. 106
    flxible says:

    “I’m looking at short/intermediate term effects, one to 72 months, largely for the practical reason that the research I know how to do …”
    Then you want to be talking to your weatherman about the weather, not to climatologists about long term global climate, your foreshortened perspective does not include the complete complex of cycles that affect the climate, which is in terms of decades, not months.

    “People can easily judge short term (daily) weather forecasts, but not long-term forecasts. They could also judge intermediate forecasts.”

    Again, you need to be talking to meteorologists – I judge their (daily) weather forcasts to be generally close, rarely perfect, or better than my own – I judge their intermediate term forecasts to be coin tossing, even the 7 day forecast changes daily, and at any given time it’s rarely close in retrospect, or again, better than my own. Seasonal forrcasts are pure hand waving – some online US weather site guru said last fall western Canada would have the coldest, snowiest winter in decades …. not even close. As for the long term weather, ask a farmer, they’ll tell you the climate has changed and greater change in a similar direction will not help grow your food.

  7. 107

    If t marvell actually wants to understand where he is completely missing the mark on the outgassing thing, Skeptical Science has covered that very much in-depth:

    CO2 is coming from the ocean

    The Physical Chemistry of Carbon Dioxide Absorption

    Seawater Equilibria

    Ocean acidification is dealt with, in exquisite detail, in the 18-part OA is not OK series, mercifully summarized in two parts:

    OA not OK part 19: SUMMARY 1/2

    OA not OK part 20: SUMMARY 2/2

  8. 108
    MARodger says:

    Chris Colose @104
    It’s probably best to save your breath & not bother trying to put the record straight with T. Marvell. He has had this particular conversation before, has been shown global temperature & CO2 wobbling to the ENSO beat here at RealClimate over a month ago. But T. Marvell only accepts the evidence of his own finding & even when you lay it all out for him he’ll find some reason why he can ignore what you say (The ‘chart’ mentioned is no longer on line.)
    And since that episode, he has kindly shown us all that the annual melt of Arctic Sea Ice is NH land temperatures. He knows this because he has a marvellous computer package that supplies the truth for him whenever he shovels in piles of data. However helpful you try to be the result is the same T. Marvell just isn’t listening.

  9. 109
    t marvell says:

    Dhogaza (105) – As I said, outgassing and carbon sinking are obviouswly opposing factors. I find only that upward changes in ocean temperatures are followed about 6-12 months later by upward changes in CO2. Beyond that, out to 6 years, I see no connection – both are absent or they cancel each other out. In the long haul, obviously, sinking outweights the outgassing. I’d be very interested in any research about the timing – how quickly does sinking occur as oceans warm and CO2 increases (how many months or years should one expect a noticable adjustment). The same for outgassing – i.e., are my fidings supported or refuted anywhere?

    This short term stuff is not relevant to Shakun or the post above. But I suspect deniers have found this short term link, which is easy to find. It’s best that climate scientists not deny it, but to have an answer as to why it does not explain the CO2/temp connection in the long haul, over the time period that AGW is a concern.

  10. 110

    marvell relies on fake-skeptic websites like climate4you. So there will be no getting him to admit to error as that is anathema to denial.

    However, the silent readership here is to whom we should be pitching our responses.

  11. 111
    Jim Larsen says:

    103 t marvel said, “I have a feeling that climate scientists should pay even more attention to the intermedate term, say 2-10 years. One reason is to convince the public about the reality of AGW.”

    Your suggestion boils down to “Climate scientists and meteorologists should team up to aggressively work on ENSO forecasting.”

    Predicting weather over a year in advance is a daunting challenge, but you’re probably right that advancements in long-term weather forecasting will help convince the public that climate science is on the right track, regardless of the strength of the connection. Here’s a link which shows recent model output.

  12. 112
    dhogaza says:

    Thanks, David Bailey.

    Now that it’s clear to one and all that this dance with T Marvell has been done before, right here at Real Climate, and that he’s not budged a micron from his science denier stance, is impervious to honest attempts to supply him science resources to set him straight, I suppose we should just ignore him.

    Shame on me for paying attention to you, T Marvell. I wasn’t smart enough to recognize that you’d been down this bullshit path months earlier here at real climate.

    If you can’t learn, that’s not my problem.

    If you really think you’re overturning modern physics and statistics, that’s not my problem, either.

    Nor is it a problem for science.

  13. 113
    Ray Ladbury says:

    t marvell, Congratulations. You’ve rediscovered the seasonal cycle of CO2. It has nothing to do with anthropogenic warming. Are you even reading what people write?

  14. 114
    t marvell says:

    Bailey (110), etc. I like climate4you as well as MARodgers’ and dhogaza’s sites. The graphs are entertaining. They are helpful entry points. Climate4you is skeptic, but his charts are interesting. Earlier I suggested that RC add links to the first two. I don’t search to see what the skeptics say, but one cannot avoid hearing about their arguments.

    Just because I take positions you don’t like doesn’t mean I’m a closet skeptic. I don’t hold these positions very strongly (I look at the ice extent charts every day with amusement more than concern). As an outsider I’m not devistated if I am wrong, and there is obviously a lot I don’t know. I get a lot of help from commenters on RC. The adverse comments are what I would expect as a novice and outsider. I’m going on vacation tomorrow, so I won’t be able to play the fool very much.

    Gavin (In 99) colose (104). I ran that rudemtary correlation between lagged CO2 and temperature without years 91-93 and 82-84, and got the same result. So volcanos are not the reason for the negative correlation. I don’t put much stock in correlations unless backed up by regression results.

    MARodgers (108) you bring up my “finding” that CO2 changes are followed several months later by changes in ENSO2.4. The relationship is negative and highly significant statistically. I have now replicated that result with the MEI index — CO2 increases are followed by MEI declines some 2-10 months later. A wacko result. If I were not an outsider, I would probably keep mum about it because it makes me look wacko. It’s not due to any seasonal factor, which I control for in several ways. I checked aerosols, and they don’t explain it. I like to view unexpected results as fun detective work rather than annoyances.

    This negative lagged relionship between CO2 and El_Nino might have something to do with the negative correlation between temperature and lagged CO2. More CO2 drives El_Nino down, which drives temperature down. In my regressions there is a small but noticable negative relationship between lagged CO2 and ocean temperature, which goes away when I enter El_Nino as a control variable (EMSO2.4 or MEI). Fun.

  15. 115
    t marvell says:

    my post 89 – I said that land temperature is stationary. That would be an interesting finding, but I now think it is wrong. I used the Augmented Dickey-Fuller test, which includes lags of differences (of the variable in question) as controls. One is supposed to include all significant lags. I concluded that lag 12 was the last significant lag since additional lags up to 15 are not. But now I have checked out further lags, and there are many significant lags out to at least 60. When the extra lags are entered, land temperature no longer is stationary. Moreover, its not stationary with a trend. Nor is total temperature.
    The implication is that temperatures are not stationary (don’t tend to come back to a level or to a trend over the long haul). This makes them difficult variables to deal with. They probably should be always differenced. One can try to deal with stationarity by correcting for autocorrelation, but as far as I know one cannot tell definitely whether this actually removes the non-stationarity. The dangers when using a non-stationary variable in a levels regression are potentially so bad (a nonsense, or spurious regression) that the safe course is to do the regression in differences. The drawback there is that one loses information about long term trends. Perhaps if a differenced regression reaches results consistant with a levels regression, the latter is acceptable.

    Another reason for using differences is to limit collinearity, which is especially likely when entering many lags of a variable.

  16. 116
    Marcus says:

    T Marvell #99:

    “The physics couldn’t be clearer – as water warms it can hold less gas, including CO2. If you are contesting outgassing, you are physics deniers.”

    True only where the CO2 concentration in water is close to saturation.
    When not, dissolving will just slow down a bit and nothing will outgass…
    Think about it.


  17. 117

    They are helpful entry points.

    Only if one purposefully seeks to perpetuate misunderstandings of the science. Climate4you is well-established as a intentional purveyor if misinformation.


  18. 118
    Ray Ladbury says:

    t marvell, would you like some dressing to go with that word salad? Given that you’ve chosen to ignore the FACTS of ocean acidification, delta C-13 and C14 and much else, I see considerable evidence to cast doubt on your bona fides. Don’t forget to sweep up the ashes of your credibility.

  19. 119
    dhogaza says:

    T Marvell:

    Just because I take positions you don’t like doesn’t mean I’m a closet skeptic.

    Which of us ever mentioned the word “closet”?

  20. 120
    t marvell says:

    Marcus (116) Right. As I said earlier, it takes special situations for outgassing, like when warm water mixes with cold water. If I’m not mistaken, the cold water might be saturated because it is neutrant rich, and regularily mixes with warm water in the South Pacific.

    [Response: Outgassing is very regionally dependent since it is a function of the upwelling of deeper (colder) water that is high in dissolved inorganic carbon (mainly from the remineralisation of falling biological matter and the advection of polar water to depth). Effectively, CO2 is absorbed into the ocean in polar regions and is outgassed via the tropics and other upwelling zones. It is not very sensitive to the seasonal cycle, and doesn’t vary hugely as a function of ENSO either. The short term signals in the atm. CO2 record are dominated by tropical deforestation variations, and terrestrial carbon cycle changes (respiration in soils, deciduous plants, droughts etc.). They have very little to do with the long term trend (driven by fossil fuel burning) but are important for understanding the sensitivity of the carbon cycle to changes in climate. – gavin]

  21. 121
    t marvell says:

    (Gavin 120)From my limited point of view, this is very important. As CO2 and temperature are cointegrated, which means a virtually ironclad causal connection between the two. They have to trend together. The only major uncertainty left is the causal direction, which cointegration says nothing about. Outgassing is the only causal connection I can think of in the Temperature-to-CO2 direction (but my knowledge is limited). My regression analysis shows pretty clearly that the temperature-to-CO2 connection is short term, less than a year, with no apparent impact over the next 5 years. It is also small compared to the yearly increase in CO2. Perhaps other research reaches the same result. This is evidence that the part that outgassing plays in creating cointegration is small. This doesn’t address longer causal connections, but if the net impact of temperature on CO2 can be shown to be neutral or in the negative direction over then long term, than cointegration probably means that CO2 is causing global warming. I’m sure there are studies quantifying how much CO2 gets absorbed in the ocean in short and longer terms. That impact can be compared to the short term effect of outgassing.

    [Response: I pointed you to some of the Kaufmann work on this earlier. But I have to counsel some caution as to your conclusions. Statements about what apparently cointegrated data must do are based on infinitely long time series, while actual time series are finite, and with plenty of auto-correlation and so the ‘true’ level of cointegration is not that certain. Secondly, I can build physically meaningful timeseries of climate that show any number of connections and degrees of I() for each variable for short periods (have CO2 climb quadratically = I(2), but have a deep ML and plenty of noise, SAT will look like I(1), and yet the entire trend will have been caused by the CO2, but with an acceleration term that is not detectable statistically). Your first differencing removes the trends for the large part and so restricts you to short term variability – much of which is related to hidden variables in your analysis (i.e. ENSO, volcanic AOD etc.), thus causality is going to be tricky. People have done a lot of work on the physics of the carbon cycle on short time scales, and outgassing changes are just very small, while terrestrial responses are much larger. – gavin]

  22. 122
    R. Gort says:

    Whenever I look at temperatures curves showing the last several glacial cycles, I’m struck by the asymmetry of the process. The glacial advance is long and with many fluctuations, but when it bottoms out at maximum glacial advance, the reversal is much, much quicker and with fewer and smaller fluctuations. And this is despite what should be a huge increase in the Earth’s albedo with all that ice surface, that I’d expect to positively reinforce the ice extent.

    This discussion is not directly addressing that, but I take away the idea that CO2 increases rapidly enough, by a variety of processes, to greatly amplify what would otherwise occur just due to orbital (Milankovitch) cycles. Oversimplified, but essentially true?

  23. 123
    Dan H. says:

    That is one of the theories; that the Milankovitch cycles start the process, which is then amplified by CO2 releases. The bigger question is what stops the process?

    [Response: Milakovitch *cycles*. Glaciers start melting when northern hemisphere summer insolation increase; eventually northern hemisphere insolation decreases again.–eric]

  24. 124
    Hank Roberts says:

    > insolation decreases again

    And geological processes reducing atmospheric CO2 continue:

    “Erosion and World Climate

    Global climate is a good example—not today’s global warming episode, but long-term climate changes on the scale of many millions of years. This geological climate cycle depends to a large extent on the geological (not biological) carbon cycle…..”

    “PS: Human activities increase erosion; in fact we now play a major role in Earth’s cycle. If strong erosion tends to cool the Earth, might this eventually counteract global warming from the greenhouse effect? Maybe, but not in time. The geological carbon cycle is far slower than the biological one.”

    Andrew Alden’s geology explanations at are well worth reading and pointing to: clear, well cited explanations in a few well chosen words.

  25. 125
    t marvell says:

    I think you understate the importance of cointegration here. With annual data, 50 years is short, but the results are very clear With land temperature, the critical value is -7.1 and the DW 1.94. The probability level has to be less than one in a million, alhtough I’m sure no one has bothered to calculate that. For this kind of analysis results like that are highly unusual, all the more so because the limited number of time periods. I get similar results, slightly less strong, with monthly data and with ocean temperature. Almost all time series analysis assumes an infinate series, but uses finite data. But that assumption is not considered critical (just about any kind of analysis anywhere makes assumptions, and this assumption is less damaging then many), With results this strong there is litte chance that this is an issue.
    The uncertainties you mention pertain to stationarity of temperature time series, not to cointegration. Stationarity is indeed a problem (see my 115 above). The ADF test results are uncertain (I haven’t checked other tests). As I said, it is very risky to assume stationarity, since in levels one cannot know whether the results are spurious.
    You say differencing removes trends – no, the trends go into the intercept. If you are advocating for any other kind of trend being lost by differencing, it is likely to be a trend that causes non-stationarity. It’s not quite right to say that differencing limits the analysis to short term trends. With differencing one can usually enter numerous lagged depenedent variables without risking collinarity (I routinely use up to 72). These lags actually pin-point the location of any lagged impact, unlike levels analysis. Levels analysis, on the other hand, can vaguely suggest longer term relationships, but not if the data are non-stationary.
    Earlier I said that cointegration does not indicate causal direction, so one must rule out a Temperature-to-CO2 relationship, and I had found such a relationship, but only short term. You said that relationship (out gassing) is minor, which I found. The question is whether it is overshadowed by the opposite effects. I suspect it is, but that I don’t know enough about it.
    More important, my analysis earlier strongly suggests that the outgassing is caused by ocean temparature only, and not by land temperature. I don’t know if there is any theoretical argument why land temperates would affect CO2 levels appreciably. I doubt that reverse causation is a factor in the land temperature & CO2 connection.
    I left my copy of Kaufmann at home (I’ve on vacation), but my feeling about it is that it established ccintegration between temperature and CO2, but that it left the causal issue hanging, and probably more important, that relationship got buried with a lot of other analysis.
    It seems to me that the climate scientists have not sufficiently picked up on a big story. Cointegration like this is hugely important. If economists ever get similar results, they’d be dancing in the streets.
    I think that climate scientists are putting too much emphasis in their models (which I would guess involve assumptions that are much more uncertain that cointegration analysis). The models have not been able to persuad people and policy makers to move. Cointegration, although also an tecnical topic, can probably be explained to lots of people, more so than the climate models.

  26. 126
    Hank Roberts says:

    It’s a poor sort of memory that only works backwards.

    Cointegration analysis of hemispheric temperature relations
    RK Kaufmann – J. Geophys. Res –
    “… Evidence is mounting that changes in global surface temperature can be attributed to human activities that increase the atmospheric concentration of greenhouse gases and tropospheric sulfates [Sanier et al, 1996a, 1996b]. This evidence comes from two sources: …”

    Cited by 33:

  27. 127
    Ray Ladbury says:

    Allow me to acquaint you with a concept that has evidently escaped your attention: the scientific literature. Do a search. There are thousands of climate scientists. Do you really think nobody has looked at the fricking time series?!

  28. 128
    t marvell says:

    I clearly said that the coinegration between temperature and CO2 is nothing new, and mentioned Kaufmann. My argument is that the cointegration is much more important than seems to be acknowledged. Also I have ten more years of data after Kaufmann, important since the CO2 data series is short.
    I don’t know if there are other papers on the subject. In general I find it difficult to research the literature in climate studies.

  29. 129
    MARodger says:

    t marvell @125 & 128

    You say “Cointegration … can probably be explained to lots of people, more so than the climate models.” and this may be why you rate its importance. Yet people-wise, you have a pretty ‘soft’ target with the folk here but even so your cointegration approach appears to be making very heavy weather of it.

    Do consider the situation if you (or others) ever do show some incredibly strong statistical link. Would it not become but another line in the “Oh yes it is. Oh no it’s not.” pantomime?

    Far more straightforward approaches with nice graphical outputs that smooth out the wobbly rise in global temperatures (eg Foster & Rahmstorf & Neilsen-Gammon ) have failed to impress the ‘great unwashed.’
    So, if your cointegration approach does prove successful (which is something that I consider unlikely), what magic ingredient do you employ to make it irrefutable in the face of reluctant public & political opinion and in the face of aggressive contrarian opposition?

  30. 130
    Hank Roberts says:

    Wait, is this not the Kaufmann article he’s going on about?
    And he’s making claims only from memory?

    Cointegration analysis of hemispheric temperature relations

    [PDF] RK Kaufmann

    Look at what it actually says! Brief quote posted a few responses back.

  31. 131
    t marvell says:

    MARogers – Foster & Rahmstorf don’t address the issue of AGW. They just try to straighten out the temperature curve.
    Cointegration. on the other hand, nails down the CO2-to-temperature connection. It’s much easier for people to accept the notion that temperatures are rising than it is to accept the notion that humans are causing it (as opposed to natural variation).
    To me it is odd that the climate scientists have not emphasized cointegration.
    It’s hard to overstate the importance of cointegration. It is probably the strongest statistical evidence that two variables have a causal relationship. The two economists who created the concept received a Nobel prize for it. I think it is the only Nobel given for a statistical technique. Cointegrated variables move around but in the long run cannot move too far apart. They are like two dogs tied together by an elastic cord. They can run around the field separately, in any direction, but when the distance becomes large enough, they are pulled back such that they move in the same general direction together in the long run.
    The 2002 Kaufmann&Stern article shows this relationship. The article might lose impact because it doesn’t test CO2 specifically, just a concoction of greenhouse gases, and because it includes a lot of additional analysis. Here, to get the AGW message across, I think it is important to focus in on the main message, CO2 and temperature.
    I suspect that nearly every economist and statistician would accept AGW if they accept (replicate) my finding that the cointegration is very strong.

  32. 132
    Ray Ladbury says:

    T. Marvell: “It’s hard to overstate the importance of cointegration.”

    And yet, somehow I have confidence in your ability to overstate it.

  33. 133
    Unsettled Scientist says:

    It kind of cracks me up when economists try to teach math to physicists. I never could get into the study of economics because the rigor so often breaks down into hand-waving, ideology and polemics. You want to talk about politicized science, look to econ.

    T Marvell, cointegration is not the only Nobel in Econ for a statistical technique. Look at the most recent one for vector autoregression, there are probably others but who cares about Nobels anyways. Could it be that cointegration is your hammer, and climate is just another nail to you?

    As for overstating cointegration being the best way to establish causation: Despite its name, Granger causality is not sufficient to imply true causality. If both X and Y are driven by a common third process with different lags, one might still accept the alternative hypothesis of Granger causality. Yet, manipulation of one of the variables would not change the other. Indeed, the Granger test is designed to handle pairs of variables, and may produce misleading results when the true relationship involves three or more variables. A similar test involving more variables can be applied with vector autoregression.

    For other readers, Granger was one of the guys who got the Nobel in 2003 for cointegration who authority T Marvell is appealing to.

  34. 134
    t marvell says:

    Unsettled – you mix up Granger causality with cointegration. The two happened to be created by the same guy. Cointegration is based on stationarity tests of residuals. the Granger causation test is based on lags, and only implies priority in time, as you say, and was not worth a Nobel prize.
    There’s all kinds of math.

  35. 135
    MARodger says:

    t marvell @131
    Ask yourself – who are the folk who would require the least convincing of the merits of your ‘cointegration approach’? Who would consider ‘nailing down the CO2-to-temperature connection’ a really worthwhile achievement?

    If you cannot convince the audience here, your ‘approach‘ is in trouble!

    And guess what? There appears to be no takers for your ‘approach‘.

    It has been explained to you before that the conventional wisdom is that tbe wobbles of ENSO induces wobbles in both global temperature & in atmospheric CO2. This causality may result in statistical correlations between ENSO-temperature-CO2 being found. Indeed, they can be observed if the three values ENSO, dTemp & dCO2 are plotted out on a graph.
    It may be interesting to know that statistical correlation is stronger between, say, CO2 –> temp or CO2 –> succeeding negative values of ENSO, but does this go anywhere? Is it explained by an evident physical process? Does it pass the Gwilym Jenkins test (ie have you graph it out, my boy!)? The answer is no & no.

    This explains why you are having dificulty convincing folk here. And which ever route you choose to travel, the roads gets stonier away from here.

  36. 136
    t marvell says:

    Climate scientists say “our models show AGW. we are the experts. trust us.” Econometricians are the time-series experts. But you say “I don’t need to understand that. I can just ignore it.” You cannot have it both ways.

    Unsettled (133) – You say that economists’ use of statistics is mixed up with their ideologies. That’s basically true. But the time series math is the province of the econometricians, who as far as I know don’t build ideology into their math.

    What’s this about ENSO causing more CO? Outgassing? Even the circular link you proposed exists, it’s not relevant to cointegration, which takes account of such quirks. For one reason the mechanism you posit is short term, while cointegrationis long term (I’m now trying to estimate the lag between CO2 increases and noticeable increases in temperature, and it seems to be at least 10 years.)

    A major benefit of cointegration is that one can regress temperature and CO2 in levels, whereas ordinarily that would lead to a spurious regression because the variable are not stationary in levels. A levels regression leads to highly significant positive coefficients.

  37. 137
    Hank Roberts says:

    > cointegration

    Seriously, this is the same R.K. Kaufmann — right?
    Cointegration analysis of hemispheric temperature relations[PDF]
    RK Kaufmann – J. Geophys. Res –
    “… Evidence is mounting that changes in global surface temperature can be attributed to human activities that increase the atmospheric concentration of greenhouse gases and tropospheric sulfates [Sanier et al, 1996a, 1996b]. This evidence comes from two sources: …”

    Cited by 33:

    Same R.K. Kaufmann cited for
    Kaufmann and Stern (2002)?

  38. 138
    tamino says:

    t. marvell:

    Time series math is not the province of econometricians. That which is applied to economics is … but there are those of us who are statisticians, pure and applied mathematicians, and yes, practicioners of the physical sciences, and we not only know the math, we know how to apply it to physical sciences better than econometricians. Your claim is not just appeal to authority, it’s repugnant hubris.

    As for your poorly-thought-out devotion to cointegration: it’s true that the drunk and her dog wander all over the place but never stray very far from each other, because they’re cointegrated. It’s also true that CO2 and temperature don’t stray very far from each other. What you fail to get is that they don’t wander all over the place.

    Take a look, for instance, at CO2 concentration from, say, the year 1000 to about 1750 (before the industrial revolution). Try the data from the Law Dome ice core. No, it doesn’t “wander all over the place.” Neither does temperature for that matter.

    The truth is, you’re impressed with how ingenious cointegration is — and well you should be. You’re also impressed with how smart you are, which is ill advised. And, you have elevated admiration of cointegration to worship. A previous commenter was right, that it’s your hammer and you love it so much you think everything is a nail.

    By the way, your earlier statement that Foster & Rahmstorf “just try to straighten out the temperature curve” was both mistaken and insulting. It illustrates how your preconception biases your judgement. What that paper does is estimate the effect of El Nino, volcanic aerosols, and solar variations on global temperature. The only way to believe that they don’t affect temperature is to deny the laws of physics. When those factors are accounted for (not perfectly, but a fair approximation of them), lo and behold! — the temperature curve straightens out dramatically. Imagine that.

    You could learn a lot here, but only if you do two things. First, abandon the notion that your understanding is better than everybody else’s — ’cause it ain’t. Second, accept the fact that you’ve got a lot to learn.

    Until you show some sign of improvement …

  39. 139
    t marvell says:

    Roberts (137). As I have said, there is research in the climate community finding cointegration. All I am saying is that the climate scientists don’t seem to understand its importance.

    tamino – the econometric techniques are all time series, and are used on all kinds of data. What are you saying? That temperature and CO2 are not cointegrated? That cointegration is not important?

    Wandering all over the place – that’s a red herring. Cointegration applies whether there is a lot of wandering or a little, just as long as the two variables tend to wander together.