It’s a volcano in Iceland which erupted in 2010. Please don’t ask me to pronounce the name.

In comments to the last post, there was speculation (and even evidence) that the CO2 emitted by the eruption was less than the CO2 which was not emitted because of the curtailing of air traffic which it caused. What piqued my interest is that one particular comment suggested that the principal effect of the eruption on atmospheric CO2 might have been the iron content of the volcanic ash, which may have drawn down atmospheric CO2 by “iron fertilization” of the oceans:

The eruption [of Eyjafjallajökull] may have affected atmospheric carbon dioxide levels by fertilizing oceans with iron. According to the Nordic Volcanological Center at the University of Iceland ash samples contained 8 to 12% iron oxide.[147] Observations at the Mauna Loa Observatory show increased carbon dioxide absorption for each of the three months following the eruption compared to the 30 year mean for the same months. Over May, June and July 2010 atmospheric carbon dioxide decreased by a total of 2.40 ppm.[148] The thirty year mean for the same months is 1.66 ppm with a standard deviation of 0.52ppm. The probability of a chance result is less than 8%.”

I agree that it’s not likely to be due only to random fluctuation (but it could be). Still, that doesn’t mean it’s due to iron fertilization, or even to the volcanic eruption, because many things can influence the April-to-July difference in atmospheric CO2.

CO2 concentration generally declines from April to July, because that’s the time during which northern hemisphere land plants grow rapidly, absorbing CO2 to build their tissues during spring and summer. In fall and winter, the decay of land plants returns that CO2 to the air, all part of a notable annual cycle in atmospheric carbon dioxide concentration:

However, the annual cycle itself has not remained constant over time. In fact the size the of the cycle (its amplitude) has increased, mainly during the 1970s. Here’s the semi-amplitude (which is just half the amplitude) estimated from wavelet analysis:

In addition to the overall increase (mostly in the 1970s), there’s also fluctuation in the cycle’s amplitude.

Yet another factor which affects the April-July difference is the timing of the annual cycle. Here’s the time of the annual peak, also estimated by wavelet analysis:

It turns out the the annual cycle in CO2 has been peaking earlier in the year, having migrated about 10 days since we’ve been monitoring it at the Mauna Loa atmospheric observatory.

The overall increase in CO2 (the trend, not the annual cycle) has not been linear — its increase is faster now than when the Mauna Loa data begin. But since about 1995, the increase has been at least approximately linear. Therefore I fit, to the data since 1995, a straight line to model the trend, and a 4th-order Fourier fit to model the annual cycle:

That leaves these residuals (click the graph for a larger, clearer view):

There is indeed a decline (even in the residuals) from April to July 2010. But that seems to be because the April 2010 value was higher than usual, not because the July 2010 value was lower than usual.

I also computed the difference from April to July for the data since 1995:

The latest value (from 2010) is one of the lowest, but is not the lowest, and frankly, it doesn’t strike me as that unusual.

Therefore it’s my opinion that the not-so-extreme decline from April to July 2010, coupled with the higher-than-usual April value, coupled with the changes (both trend and fluctuation) in both the size of and the timing of the annual cycle, are such that there’s insufficient evidence to conclude that the Eyjafjallajökull eruption caused a noticeable change in atmospheric CO2, whether by emissions from the eruption, the lack of emissions from air traffic, or iron fertilization of the oceans. Just my opinion.

41 responses to “Eyjafjallajökull

  1. Pronunciation …. Eh-ya-fyat-la-yuh-cuttle

  2. Tamino and anyone else competent,

    The main reason my drought paper was turned down was that the 1870-2005 data I used wasn’t reliable; not enough area covered in the early part. Dr. Dai at NCAR told me the data was only reliable globally from 1948 on.

    That’s still N = 58. I reran my analysis. Did a lot of statistics. Data-mining. Autoregressions to find appropriate lags by corrected AIC. ADF tests for integration. Engle-Granger tests for cointegration. Granger causality tests. I ran a Monte Carlo simulation 10,000 times. F is a function of past F, temperature anomaly, Dust Veil Index. R^2 = 88%.

    I sincerely hope that somewhere along the line, I made a big, BIG mistake.

    My latest results are that the fraction of Earth’s land surface in severe drought will hit 70%, the limit at which I guess human agriculture will collapse, and human civilization with it, sometime between the years 2022 and 2027.

    That’s 11-16 years from now.

    If I’m right, we’re out of time.

    I don’t know what the hell to do. If I called the White House and asked to speak to the president, no way on God’s green Earth would I be connected. At best some sympathetic flunky would pass the time with me. At worst the Secret Service would investigate me as a possible dangerous nutcase.

    Please, can somebody competent look over my work? Please tell me I screwed up badly somewhere along the line. Find the mistake. Please.

    [Response: Send your latest to me, and to anyone else who shows interest.

    And don’t despair, when we need you most. When it seems the times call for a miracle — believe in our power to forge one.]

  3. BPL,

    “the fraction of Earth’s land surface in severe drought will hit 70%, the limit at which I guess human agriculture will collapse, and human civilization with it, sometime between the years 2022 and 2027.”

    I certainly hope you’re wrong too, and all my qualitative hydrological senses are screaming “this can’t be true”. There should be way to much water in the world to produce that much drought that soon.

    I haven’t seen your methodology properly described but it appears that your model is based on some kind of statistical calibration with past data that is used to forecast future conditions. If this in fact the case, this can be a problem if there isn’t a lot of physics in your model because heavily calibrated models can become unreliable when used as a forecast tool. Such models aren’t necessarily constrained by physics and it’s quite possible that they may perform quite well under historical conditions but fail miserably under conditions that are outside their calibration range.

    Again, I don’t know if this is the case with your work, but it’s always something to keep in mind regardless.

  4. BPL,
    I wouldn’t mind seeing your paper, is there a way that you could post it on your website perhaps or if you prefer I can look for a way to email you.

  5. False alarm!

    I am extremely sorry. I DID, indeed, make a glaring, stupid mistake. The 2020s figures are bogus. We’re all right for the moment.

    I apologize to all and sundry.

    I need to rest…

    • So what decade did you compute for 70% desertification?

    • And that, right there, is science in action.

      Initial work gives rise to predictions of X happening.
      Further analysis reveals a mistake in the calculations, and it’s not so.
      Real scientist admits mistake, and life goes on…

      Of course, in this case, the scientist was relieved there was a mistake (and, frankly, so am I!). I also hope there are some hidden errors in some of the other projections made by climate scientists, but significant errors that substantially change the predicted outcomes are conspicuous in their absence…

    • One of the best things I have found to “debug” a program for data analysis is to sit with the program to explain it to one of your PhD students. You finally “almost” always end up finding your error when you explain the program to another person. It happens a lot of times, BPL, no need to apologize.

      Good luck with the paper.

  6. Let us look at the numbers:
    The purported CO2 difference is (2.40-1.66) = 0.74 ppm.
    0.74 ppm CO2 in the atmosphere = 5.9 x10^15g CO2 = 1.6 x10^15g C = 1.6 Gt C. (My calculations).

    To put this in context, total global oceanic primary production is something like 50 Gt C per year. This classic paper went further and divided the oceans into 57 zones. If we look at the ocean zones affected by ash fallout (zones Arct, Sarc, Nadr, Gfst, Necs) we see that total annual primary production there is a tad under 4 Gt C.

    To spell it out:
    1.6 Gt C extra uptake over 3 months = 4.8 Gt / year. If iron fertilisation were responsible then it would mean that for 3 months the phytoplankton in the relatively small zone of the North Atlantic affected by ash fallout were growing at double the normal rate. Did nobody notice this? I don’t see it in the MODIS chlorophyll images.

    (Actually it would have to be lot more than double the normal rate because phytoplankton are eaten as fast as they grow, and that carbon is quickly released through respiration. But I’ll ignore that here for the sake of argument).

  7. We can also examine the iron fertilization idea from first principles:

    Ocean Iron Fertilisation is an example of Leibig’s Law of the minimum. That is growth is controlled by the scarcest nutrient. Leibig’s law simply says that if your cake recipe calls for 2 eggs then no matter how much of everything else you have if you only have 1 egg then you are limited and the best you can do is make a half size cake.

    In the case of ocean plants, phytoplankton, the controlling nutrients are usually nitrate, phosphate and (for diatoms) silicate. It isn’t that other nutrients are not vital; it is just that these are the ones, especially nitrate, that run out first. This means that for much of the ocean levels of nitrate and/or phosphate are very low; they have been used up by the phytoplankton.

    One of ways in which iron is supplied to surface seawater is by the dissolution of windblown dust. Very little iron is needed and for a long time nobody could work out why there were low levels of phytoplankton but significant levels of nitrate, phosphate and silicate in the Southern Ocean. Why didn’t the phytoplankton grow to use them all up? Various hypotheses (e.g. cold, light) were found unsatisfactory before it was realised that the Southern Ocean is a long way from any sources of dust. (Antarctica is ice-covered and a poor source of dust). The Southern Ocean was ‘anaemic’.

    Compared to the Southern Ocean, most of the North Atlantic has low levels of nitrate and high levels of chlorophyll. This strongly suggests that iron is not a limiting nutrient in most of the North Atlantic and suggests that the addition of iron would have little effect. The ash had low levels of N and P so it isn’t likely the ash would have fertilised in any way at all.

  8. Nice analysis the simple explanation is not all that simply proven. It is worth noting that the heavy ashfall on the icecap very close to the volcano led to a decrease in summer ablation during the the warm summer of 2010. This has been observed before where thick ash layers accumulate, though it is not result most people expect. The Icelandic Met Officecompared hydrographs from glacier distant from the volcano and one draining north from the icecaps adjacent to the volcano.

    • But as you noted, and as is clear in the source, that insulation effect is applicable only where the ash is thick – i.e. relatively nearby the eruption site. Presumably the much larger area that received a light dusting would react in the typically expected manner – slightly lower albedo resulting in slightly higher solar absorption.

  9. The graph introduced by “Here’s the time of the annual peak, also estimated by wavelet analysis” seems to back up ecologists’ claims that the “growing season” is getting earlier – eyeballing looks like about 11 days earlier in 55 years.

  10. Pete Dunkelberg

    Barton, for God’s sake hang in there! It is not a big surprise that the trend you are finding would speed up in more recent data, but still it sounds like you may be extrapolating a curve fit leaving physics behind. This does not mean there is no problem! I ask though why don’t we hear more about expanding Hadley cells? Bring physics into it. In addition, there are competing effects: drought and flood. There is a trend (cite???) for more of the total precipitation to be concentration in large storms. Combining these two problems, the cumulative probability for one devastatingly bad year keeps growing…. Keep working on it Barton!

  11. Pete Dunkelberg

    Tamino, thanks for this analysis.

  12. I also find it rather unlikely that iron fertilisation is responsible. Iron fertilisation only works in places where iron is limiting, obviously. These are called HNLC waters (High Nutrient Low Chlorophyll), and are limited mostly to the Southern Ocean, equatorial Pacific, and a smaller area in the sub-Arctic Pacific.

    It would be informative to know if their water samples came from waters that are usually HNLC, and also what kind of phytoplankton were in the samples.

  13. Tamino, the raw daily CO2 data shows that peak annual concentration usually occurs in May. F0r example, in 2009, it was May 26th. That’s day 146.

    Is there a reason the wavelet analysis comes up with such a different number, or have I misunderstood what it’s showing?

    [Response: Good catch! The wavelet analysis gives the time of the peak for a best-fit sinusoid, which happens earlier than the actual peak because the annual cycle is distinctly non-sinusoidal (the rise from minimum to maximum takes quite a bit longer than the fall from maximum to minimum).

    Nonetheless, the wavelet analysis will still reveal (correctly) the change in the phase of the annual cycle, although it gives a biased estimate of the time of the peak — so the result, that the cycle is peaking earlier, is still correct.

    I’ve recently created a new wavelet program which includes multiple harmonics (not just a single sinusoid). That should be able to resolve the absolute peak time (not just the phase of a best-fit sinusoid), and this should make a good “test case.”]

    • Horatio Algeranon

      Seems like the tools for analyzing variable star light curves (this one, for example) would be directly applicable here. ~@:>

  14. Won’t necessarily help with the unpronounceable volcano issue, but the type of analysis you did might have more power if you used the other stations in the SIO network, esp those outside of the tropics where the annual cycle is more extreme


  15. Thanks for your reply above Tamino. I did a completely ham-fisted Excel analysis on the raw data (from 1975 onwards) simply looking up the date of the maximum CO2 each year, and as you say it does confirm that the cycle is peaking earlier.

    The data gives a linear trend of -0.55 days per year, though of course there is significant variation around the trend with R2=0.115.

  16. Again, apologies for getting it wrong and panicking yesterday.

    I’ve fixed the problem. Civilization doesn’t fall until 2057, so we can all relax. :)

  17. OMG!
    My ‘death date’ – using the death date calculator is July 27 2057… Coincidence?

  18. Can the same analysis be performed to test this idea?
    The role of atmospheric nuclear explosions on the stagnation of global warming in the mid 20th century

    This study suggests that the cause of the stagnation in global warming in the mid 20th century was the atmospheric nuclear explosions detonated between 1945 and 1980. The estimated GST drop due to fine dust from the actual atmospheric nuclear explosions based on the published simulation results by other researchers (a single column model and Atmosphere-Ocean General Circulation Model) has served to explain the stagnation in global warming. Atmospheric nuclear explosions can be regarded as full-scale in situ tests for nuclear winter. The non-negligible amount of GST drop from the actual atmospheric explosions suggests that nuclear winter is not just a theory but has actually occurred, albeit on a small scale. The accuracy of the simulations of GST by IPCC would also be improved significantly by introducing the influence of fine dust from the actual atmospheric nuclear explosions into their climate models; thus, global warming behavior could be more accurately predicted

  19. For the definitive guide to pronounciation:

    [edit — please no embedded video]

    Now available in a full length version on Ms Newman’s debut album — youtube with fan video: http://www.youtube.com/watch?v=ULbGsVx0c_8

    • Sorry about that: I pasted in a Youtube link, and WordPress did the embedding automagically. I hadn’t expected that to happen!

      [Response: Yeah. I’ve tried disabling that feature, but to no avail.]

  20. And I forgot: one of the great lines in popular song…

    “…because an exploding mountain is always right”.

  21. Using the SIC instead of the AIC makes a big difference. Both give optimal lags of 1 year for F, but while SIC gives 1 year for dT, AIC gives 10 years. When I use 1-year lags, it changes everything–causality runs unequivocally from dT to F and not the other way (heat causes drought. Duh). Proper integration tests show neither series has a unit root, unlike my earlier conclusion. (Oh, Jesus, I need to learn this stuff better.) I’ve also used more up-to-date dT figures from Hadley. Civilization falls in 2056.

    Hmmm. That’s not much of a change. We seem to be converging here.

  22. 1st estimate (cubic, no physics): 2037
    2nd estimate (multiv ariate): 2050-2055, mean 2052
    3rd estimate (bivariate): 2057
    4th estimate (bivariate): 2056

  23. Halldór Björnsson

    In comparison with other Icelandic eruptions, Eyjafjallajökull 2010 was a small-to-medium size, but somewhat unusual in its duration.

    It started with an eruption on the flank of the glacier on March 21st and lasted until April 13th. The flank eruption (also known as the Fimmvörðuháls-eruption, try pronouncing that!) produced mainly lava, and the ash cloud from it rarely rose above 3 km.

    The summit eruption of Eyjafjallajökull (E15 for short) began on April 14th and lasted until May 23rd. This eruption had three distinct phases, with substantial ash production from April 14-18th and again from about May 5th to 20th. In the first phase north westerly winds aloft carried the ash into the crowded European airspace, prompting closures and mayhem for travelers. During the May phase persistent westerly winds were not as frequent and the ash tended circulate in the Atlantic, rarely impinging the continent’s airspace.

    Because of its long duration considerable ash was blown into the Atlantic, especially during May. Even after the eruption ceased, there were episodes of resuspended ash blowing off the glacier and nearby areas onto the ocean. Such episodes are a fairly common occurrence in the aftermath of eruptions, I have been told that following the 1997 Gjálp eruption in Vatnajökull glacier resuspended ash was a reported on and off for several years in the area south of the glacier. As most of the E15 ash fell nearby the volcano, ash blown to sea during those resuspension episodes may be a large part of the total transport to sea.

    Plankton blooms are a regular feature in the springtime ocean near Iceland, so if the ash is foodstuff for the plankton the timing seems to have been fortunate.

    I know there was considerable research interest in how the ash would impact marine life. I did read about planned studies, but I haven’t seen any results yet. However, I know that there was at least one plankton bloom simultaneous with resuspended ash blowing over the coastal waters south of Iceland, see the Modis image in

    This picture does of course not prove much, since blooms do occur in this season. What is needed is a study that examines if the spring 2010 blooms were unusually large. Something like Hamme et. al (2010) paper documenting unusually large blooms in the Pacific following the 2008 Kasatochi eruption (http://www.agu.org/pubs/crossref/2010/2010GL044629.shtml).

    However, even if a link between E15 and anomalous plankton blooms was found, and even if the anomalous uptake of CO2 was calculated (and found to be significant), this is still small potatoes for CO2 budgets in general. At any given moment there are always volcanic eruptions ongoing somewhere
    (see http://www.volcano.si.edu/reports/usgs/ for a list), some of these eruptions may trigger a plankton bloom, others won’t.

    In any case, the climatic influence of E15 seems to have been negligible. Eruptions in Iceland occur every few years ( ~40 in the 20th century, actual number depends on what you call distinct eruptions, and generous counting can raise the number by 5-7), but the last one to have had a climatic impact was probably Askja 1875. That eruption also went on for several months, but the actual ash eruption phase only lasted 8hours (on March 28th 1875). Curiously enough, air traffic was not affected.

  24. Dave Andrews


    Perhaps your biggest mistake was not to realise that you are just a typical religious ‘the end of the world is nigh’ freak and therefore you sought confirmation of your views in the figures.