Don’t believe a word of it, guv (part 2)

Ahem. So previously there was a lot of hype and confusion and not much paper. Now that has changed, with Reconstructing sea level from paleo and projected temperatures 200 to 2100AD by Aslak Grinsted, John C. Moore & Svetlana Jevrejeva.

Which says:

We use a physically plausible 4 parameter linear response equation to relate 2000 years of global temperatures and sea level. We estimate likelihood distributions of equation parameters using Monte Carlo inversion, which then allows visualization of past and future sea level scenarios. The model has good predictive power when calibrated on the pre-1990 period and validated against the high rates of sea level rise from the satellite altimetry. Future sea level is projected from IPCC temperature scenarios and past sea level from established multi-proxy reconstructions assuming that the established relationship between temperature and sea level holds from 200-2100 A.D. Over the last 2000 years minimum sea level (-19 to -26 cm) occurred around 1730 AD, maximum sea level (12 to 21 cm) around 1150 AD. Sea level 2090-2099 is projected to be 0.9 to 1.3 m for the A1B scenario, with low probability of the rise being within Intergovernmental Panel on Climate Change (IPCC) confidence limits.

(This abstract is a bit confusing. We use a physically plausible 4 parameter linear response equation to relate 2000 years of global temperatures and sea level. doesn’t mean what it appears to say. They only have 150 (or 300) years of sea level and T together; the 2000 years comes from using the resulting relation to reconstruct sea level over 2000 years. And the results (of the sea level reconstruction) must be dubious, because they earlier say We will, therefore, restrict the use of equation 2 to a relatively short period dominated by sea level rise, and (according to their results) this isn’t true of the last 2 kyr).

OK, so this a now comprehensible: we have some trust in the IPCC model projections of temperature, but we reckon their SLR estimates are too low because they don’t take into account ice sheet melt (see-also RC). So rather than try to model it, which we can’t, we’ll just look at the historical T-SLR relationship and project it into the future. This is a reasonable idea, and Rahmstorf did it in Science in 2007 and defended it. He only did 1880-2001; I’m not sure why he picked that period, it can’t be because he doesn’t trust he HS, perhaps he doesn’t trust the early sea level records.

Grinsted et al. believe that doubt has been cast on the assumptions of Rahmstorf, thought I’m not quite sure why. They do 1700-2007, by use Moberg or the Jones and Mann T reconstruction, and the Amsterdam SL record. Assuming I’m reading their table 2 right, when using only the historical data (1850 onwards) they get basically the same answer (0.32m – 1.34m) as Rahmstorf (0.5m – 1.4m) for SLR to 2100. Which isn’t too surprising, as its a very similar method. Using the longer record from 1700, they get 0.91m – 1.32m from the Moberg reconstruction, or 1.21 – 1.79 from J&M; but the J&M fit isn’t good (fig 7) so they prefer the Moberg version.

Using the 1850-2007 data only, the response time comes out at ~1 kyr. Which doesn’t sound right: I doubt you can determine such a long timescale from a short dataset. As indeed they notice: The simple conclusion is that the calibration time series is too short relative to the response time. Inclusion of the additional pre-1850 data clearly favors faster response and a higher sensitivity (aτ-1) than instrumental observations alone. For Moberg, the response time is much shorter: ~200 years. I wonder if that interacts with Hansens stuff at all? OTOH if you took the response time from the whole Holocene dataset (~2.5 kyr) then SLR at 2100 comes down to ~0.6m.

So where do we end up? Hard to say. The answers are compatible with Rahmstorf, but they effectively reject the Rahmstorf stuff because they don’t believe the response times when restricted to the shorter period. “Don’t believe a word of it” is no longer a fair response; its a reasonable piece of work, though I’ve no idea if its right or not. Boiled down, it amounts to “we’ll probably get more SLR from the ice sheets, but we don’t know how much yet”.

ps: Daniel says, play The Codex of Alchemical Engineering. So do I.

pps: To warm the cockles of Broons heart: when told that VAT was down to 15% from 17.5% D’s first response (having first checked what VAT was) was “oh cool, so I can buy more stuff!”

[Update: Of course, We’re all going to die -W]

[See-also: Aslak Grinsted. Recent global sea level acceleration started over 200 years ago? may also be of interest -W]

13 thoughts on “Don’t believe a word of it, guv (part 2)”

  1. So, to summarize some of these recent papers.

    Kinematic Constraints on Glacier Contributions to 21st-Century Sea-Level Rise

    0.8 m “‘most likely’ starting point.” 2.0 m max by 2100.

    A Semi-Empirical Approach to Projecting Future Sea-Level Rise

    Click to access rahmstorf_science_2007.pdf

    0.5 – 1.4 m by 2100

    Reconstructing sea level from paleo and projected temperatures 200 to 2100AD

    0.9 to 1.3 m by 2100

    High rates of sea-level rise during the last interglacial period

    Click to access Rohlingetal2007.pdf

    Average 1.6 m per century under sustained temperatures 2+ degrees above present.

    Rapid early Holocene deglaciation of the Laurentide ice sheet

    Click to access 2008_Carlson_etal.pdf

    Current projections should be considered “minimum” even without ice sheet dynamics.


  2. I am the first author on the study and i would like to comment.

    We do not reject Rahmstorfs predictions – how can we disagree if they are roughly the same? We note that doubts has been cast on his predictions. There are 2 comments published on his study that especially criticize the statistics and claims of significance. As you say, he has answered these comments, but i am sure that his answer has not removed all doubts. If you google for the blog reactions on his study, then you will find some that question the assumption of an infinite response time. So, what we do is leave the response time question open and try to determine it.

    [OK, I didn’t say you rejected R’s results. In fact I has some trouble working out what you did dislike, but I thought it was his method and the response time assumptions, and now you confirm that. My reading of one of your figures (I forget which) is that assuming an infinite response time would get you a rather lower SLR -W]

    You are probably right that Rahmstorf cuts the data at 1880 because he not believe the earlier SL data. He probably found that when he included it then he did not get the same results. That is because he uses a ordinary linear regression which does not take the time-varying uncertainty into account. Obviously the arbitrary cut-off is another weak point that could give rise to doubts, – although it makes sense given the calibration method and that the uncertainty is very large when you go further back. We do take the time-varying (and auto-covarying) uncertainty into account and make predictions similar to Rahmstorf when we only include the data since 1850. That is very comforting, and shows that Rahmstorf has a good feeling for the data. However, we also find that when you only calibrate using the historical data then you get results that imply an implausibly cold little ice age (i.e. colder than both moberg and J&M). That is shown in fig 6c. That is why we prefer Moberg over the historical data alone.

    In the blog-post you write:
    “OTOH if you took the response time from the whole Holocene dataset (~2.5 kyr) then SLR at 2100 comes down to ~0.6m.”
    I assume that then you use 7m/degC. Such a high sensitivity is incompatible with the sea level during the last interglacial (see constraint 4 in the paper). It seems reasonable to assume that the sensitivity (‘a’ in eq2) depends on the total ice volume.

    [All I was doing was reading off fig ?6? which shows SLR against response time -W]

    A minor comment on 2000 years of data:
    From salt marshes and well data we know that sea level varied within +/- 25cm over the last 2000 years (see IPCC). We use this to justify a weak constrait that sea level at t=-2000 was withibn 1m of present day levels. In this sense we do use sea level data for the last 2000 years. This constraint is very important and so this is not a small detail. So i think it is fair to say that we use 2000 years of T and S data.

    [I’m not sure I’m really happy with that, its not what I would expect the text I quoted to mean. But its a minor point.

    What about your assumption that this only works with sea level rise, and your use of it during sea level fall? -W]


  3. Regarding the update. Hilarious how wrong they can get it. I wrote them an email where i pointed out the mistake, but i dont expect anything to come of that.

    [I can’t blame you for the Mirror :-). Newspapers are hopeless 😦 -W]


  4. Wow, solar forcing, SLR, TCs and glacier modeling, all in a few years. You’re out-and-out trendy, Aslak! (Seriously, I’m impressed.)

    cce, Mark Serreze says 1 meter is the new consensus (as a minimum, I think, although given the source I’m not sure about that).


  5. Hi William

    You are reading the figures right. I get a lower SLR if I only use the historical, I get a higher SLR if i use J&M. (This is shown in fig 6d). My interpretation is that the difference is caused by the memory of the LIA: The colder the LIA, the smaller the projection. This is what i have tried to show in figure 6c. So, i think it is important to give the paleoT reconstructions as a kind of model spin-up.

    As the projected SLR declines (fig 6d) for long response times then we are entering a domain where the solutions are constrained by b<5 m. These solutions have much smaller likelihood (fig5) because the inversion is no longer free to choose the parameters as it likes. If you plot the best fitting solutions then you would reject them simply by eye. If this constraint was not there then it seems reasonable to assume that the predictions would be on the tangent. So the projected response is largely independent of the response time (see table 3).

    About the use of the simple model for both rise and fall. The uncertainties in the SL record have such a high serial correlation that i judge it impossible to determine the additional parameters needed for a more complicated model. Further the target series (tide gauge GSL) is dominated by rise, and it would be extremely difficult from this evidence to determine fall-model parameters. Still, i feel it is very comforting that the two, very different, temperature reconstructions result in a very similar SL history. This despite their their big differences in temperature amplitude and response times.

    I think that the power of Rahmstorfs study was the simple and very illustrative method that he used. However, I guess that the published comments will have the consequence that little action is taken on this basis alone. I was trying to tell how Rahmstorfs study might be perceived (not my personal opinion).

    cce: – i copied your review to my personal page. Thanks.

    SteveBloom: Konrad Steffen has also been saying something like 1-1.5m for some time. The press surrounding the recent USGS report had headlines like "4 ft SLR".


  6. Aslak, re the sea level, I was aware of that, but I’ve also seen a few objections to the idea (from scientists, although not anyone who’s any sort of expert on the subject), and Mark’s comment was the first time I’d seen anyone use the “c” word regarding a 1 meter figure.


  7. Here’s a couple of popular articles from 2007 talking about 1 m of sea level rise.


    Chao et. al. estimated that artificial reservoirs have reduced SLR by about half a mm/year over the last half century (or a total of 30 mm). Is this large enough to have an effect on your results?


  8. cce: In some of my test-runs i examined the effect of correcting for reservoir effect based on Chao et al’s estimates. If I correct for that then the predictions are higher, -but only slightly.


  9. This is my first-ever visit and I like very much what I am seeing. Your material is indeed really great to look over, enormously ambrosial as well as ambrosial. I’ll doubtless recommend it to some my friends. They have luck of support as well as custom term paper requirement.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s