RealClimate logo

Climate Sensitivity Week

Some of you will be aware that there is a workshop on Climate Sensitivity this week at Schloss Ringberg in southern Germany. The topics to be covered include how sensitivity is defined (and whether it is even meaningful (Spoiler, yes it is)), what it means, how it can be constrained, what the different flavours signify etc. There is an impressive list of attendees with a very diverse range of views on just about everything, and so I am looking forward to very stimulating discussions.

There are a number of tweeters there, and it’s likely that we’ll have a joint hashtag for the conference and related materials Update: #ringberg15. Some of what will get presented will be already published, but some will undoubtedly be new or updates of previous work.

To get in the mood, there have many climate sensitivity pieces on RC over the years, the most relevant are On Sensitivity: Part I and Part II, and more recently Climate response estimates from Lewis and Curry and a useful counterpoint from Drew Shindell.

The IPCC AR5 chapter 10 (particularly section 10.8) is probably the most comprehensive background (but dense).

Some of the presenters have also posted some relevant post-AR5 literature for the talks:

This week should be a great opportunity to fill in the background on the this key topic for many people, and hopefully clarify what the sensible discussions are really focussed on. Chime in below if you want to add to that, or ask a question of your own that hasn’t been answered satisfactorily elsewhere.

26 Responses to “Climate Sensitivity Week”

  1. 1
    Eli Rabett says:

    More a philosophical point, but while debating the spread in mean estimates for cliamte sensitivity, attendees might consider Naseem Taleb’s take that it is incoherent to doubt the mean while reducing the variance.

  2. 2
    Jack Barett says:

    My own calculation is in my e-book: Global Warming: The Human Contribution. By assuming that global mean temperature is dependent upon the logarithm of the CO2 concentration (/1 ppmv) a good fit (r = 0.93) to the HADCRUT4 data is obtained. If 2.66 x ln(CO2/ppmv) is removed from the data the remaining values show zero trend. The function gives a sensitivity to CO2 doubling as 1.84 K.

  3. 3
    Matthew R Marler says:

    how it can be constrained

    I am especially interested in how it can be constrained by increases in the rates of advective/convective and evapotranspirative transfer of energy from the surface to the troposphere.

    I am also interested in how long is required for the surface temp to “achieve” 95% of the ECS change: e.g. if climate sensitivity is 2K, how much time is required for the surface temp to increase by 1.9K; and then how much longer for the deep oceans to increase by 1.9K (or whatever 95% of the projected increase in deep ocean temperature works out to.)

  4. 4
    Ray Ladbury says:

    One question I’d find very interesting is why the distribution of climate sensitivity estimates seems to have become bimodal in the last couple of years. It has seemed that estimates fall into one of two modes–below 2.6 degrees per doubling or above 3. This wasn’t the case when I looked at the distribution about 3 years ago.

    Any ideas on why we are seeing this? Are all the sensitivity estimates really looking at the same quantity?

  5. 5
    paulina says:

    Looks intense.

    One comment, meant to be constructive. Maybe you are already planning on this, but just in case.

    Might be worth clarifying some of the very basic concepts (“reasonable,” “extreme,” “more than likely”) a bit if for no other reason than to avoid having some semantics issues masquerade as substantive disagreements during the course of the workshop.

    Workshop presentations are encouraged to help participants:

    collectively explore reasonable bounds on estimates of climate sensitivities (TCR, ECS), i.e., what we might call extreme sensitivities in the sense that they are “more than likely” not to be exceeded.

    But even if we, for instance, were to agree that “reasonable” *estimates* are those that are distinct from extreme estimates, it doesn’t follow that reasonable *bounds* don’t encompass some “unreasonable” estimates, i.e., some subset of extreme estimates. (Assuming that it does follow paves the way to the quagmire of semantics.) So, the “i.e.” in the statement above is unclear (going from “reasonable *bounds*” to a category of “extreme estimates”).

    Further, the clarification (the text following “in the sense that”) is ambiguous and either vague, simply confusing, or possibly very controversial.

    So, it might be worth staking out this conceptual framework a bit more, early on.

    These observations may not at all resonate with workshop participants to whom the conceptual framework may be perfectly clear, but, on the chance that they do, worth posting them here. Thanks for providing this opportunity.

  6. 6
    John L says:

    Looks like a very interesting workshop. As an interested amateur, one issue that I hope is clarified is how to numerically combine different lines of evidence. IPCC famously gives a “likely” 1.5-4.5 interval for the ECS. But it is very hard to understand how such large interval can be derived from the evidence IPCC itself gives: Palaeosens gives a likely 2.2-4.8, CMIP5 models is in the range 2.1-4.7 and the uncertainties in inter-annual feedback observation analysis is mainly from cloud feedbacks which is said to be “likely” positive. Dessler has claimed that it is very unlikely to be below 2 from this analysis.

    Each line of evidence has to be expert-judged for its bias and structural uncertainty (i.e. how much to increase its variance) and when combining, the level of independence has to be judged as well.

    Using conservative values of the judged parameters you may get, say, 1.5-4.5 “likely” intervals for each of the three methods above. Combining you get, perhaps conservatively, a 1.5-4.5 “very likely” interval.

    Remains the modern time period energy balance analysis, which, as I understand it, tends to include lower values. But combining two intervals is not done by taking the average or an interval that “includes” both intervals. For example combining an 1-5 and an 2.5-3.5 (normal distributed) interval results in an interval narrower than 2.5-3.5. This means that the energy balance methods line of evidence cannot really increase the 1.5-4.5 “very likely” interval and certainly not to IPCC:s numbers. (I’ve also learned from this blog that these methods tend to have a low bias and give only weak constraints of ECS).

    This is further confirmed by that the Charney report gave a similar estimate as IPCC AR5 which simply is not reasonable considering the huge amount of research and new data since 1979. OK, you may argue that the Charney report underestimated the uncertainties, but then you have to do that explicitly.

    Finally, another intuitive confirmation is the high certainty in the attribution statement of warming contributions which is difficult to reconcile with such large uncertainties for ECS.

  7. 7
    James C. Wilson says:

    ARGO, GRACE, NOAA MSU and the boreholes permit analysis of the internal energy of the climate system over the last 30 or more years. The hiatus does not appear in internal energy of the climate system. This finding highlights the need to understand the various cycles that impact heat transfer from the oceans to the atmosphere and back. It looks like these efforts are gaining some traction. Looking at GAST since 1880, one sees some famous plateaus that have been ascribed to aerosol and internal variability (ocean cycles).
    How does the discussion of climate sensitivity deal with this information? Average over the ups and flats? Emphasize the steep parts of the curves?

  8. 8
    Thomas says:

    One plea, Is that what feedbacks, and system latencies have been included in a given sensitivity value, be clearly stated. We have things like instantaneous response, Charney (short term sensitivity). Include other feedbacks, and/or thermal reservoir timescales, and yet higher sensitivities are expected. So just labeling a distribution as sensitivity is seriously incomplete, as the details matter.

  9. 9

    Doesn’t “climate sensitivity” in the classical sense depend upon temperature and the feedbacks that rely upon it? Accordingly, is this sensitivity discussion limited to just CO2 doubling, holding all else constant? And what reference temperature is the sensitivity calculated against? Oughtn’t a sensitivity develop along RCPs like other things do? And I presume there’ll be discussions of both ECS and TCS, too, no?

    Finally, I hope the work of Nathan Urban and colleagues is well represented in the discussions, especially their Bayesian methods. Frankly, some of the lobbing of criticism I’ve read back and forth over these questions seem to me to be byproducts of questionable methods. I mean, land ECS is not the same as ocean ECS, and, so, to answer in part a comment made elsewhere here, when they are combined, you do get multiple bumps. Accordingly, representing them as a Gaussian can distort.

  10. 10
    jyyh says:

    I’ve deliberately kept all talk of ‘climate sensitivities’ out of my shares at social media. Locally, knowing what the global sensitivity to whatever forcing is not too important, as the earth warms at different rates depending on the latitude and partly of the location (combination of the longitude and latidtude). “The topics to be covered include how sensitivity is defined (and whether it is even meaningful (Spoiler, yes it is))”, you say, and it might be asked whether using such a number that’s not even defined exactly in scientific communications is useful. yeah, let’s have all the countries negotiating have a different equation to calculate the sensitivity to see if they can agree on something. That said, no doubt such a workshop on poorly defined concepts is a great opportunity to meet with people having different ideas on how to approach the politicians who might use active hearing protectors after hearing the word ‘climate’ (ref.Florida).

    F.e., 1 very obvious error in the image on the feedbacks is the part on vegetation, killing a plant does not take decades, but might happen in a single season due drought etc.

    That said, you have a good workshop. The planting season for this year was delayed a bit by the cool burst last week so I should have some free time to follow it. Apple trees need pruning, though.

  11. 11
  12. 12
    R Bodman says:

    @1 Eli,

    But distribution for ECS is not gaussian – changing spread may well be an asymmetric change, so the mean will change, and the mean is maybe not best estimator anyway – in a skewed distribution median and mode will be lower.

    @4 Ray,

    I think you may be referring to ECS estimate from different approaches – complex models and simple energy balance models/energy balance calculations. Complex models have been fairly consistent with ECS – as an emergent property – of 3 or so. But note these are typically from only a small number of model realisations. Simple models are very different, and are more like an effective sensitivity and may well lack some of the nonlinear dynamics/regional processes that occur in the real world and, in a more limited way, in the complex models.

  13. 13
    Andy Revkin says:

    Gavin or others, Do you know if anyone’s done a plot of the shifts in sensitivity-range curves over time through the 5 IPCC reports? To me, this seems in many ways to be more of a known unknown than an area in which some analytical breakthrough will clarify things in time for policy relevance.

  14. 14
    Ray Ladbury says:

    R. Bodman and Jan Galkowski,
    Yes, I realize that. My puzzlement stems from the way estimates have evolved over time–e.g. that the 5% to 95% range expanded in the most recent IPCC analysis. I noted that during the same period, the distribution had bifurcated into a lower mode and an upper mode. This bimodality was not evident when I did a similar analysis 4 or 5 years ago.

    It would appear that the definition of sensitivity is not consistent from one analysis to the next. I think this is important, as denialists will seize on the lower mode to argue for inaction, while the consequences could be dire indeed if in fact the upper mode comes closer to the truth.

  15. 15
    Racetrack Playa says:

    Here’s an interesting paper that is referenced in some of the listed publications:
    Meraner et al. 2013, Robust increase in equilibrium climate sensitivity under global warming, GRL

    They conclude, based on study of CMIP5 model output, that equilibrium climate sensitivity (ECS) is not a fixed quantity – as temperatures increase, the response is nonlinear, with a smaller effective ECS in the first decades of the experiments, increasing over time. They attribute this increase in ECS to enhanced tropical water-vapor feedback and a rise in the height of the tropopause.

    This also tends to support the general concept that changes in the tropics (30N-30S) are the primary drivers of global climate change; processes in the North Atlantic and North Pacific seem to respond to changes in the tropics, not the reverse – i.e. the concept of climate controlled by a global oceanic conveyor belt driven by a North Atlantic gear mechanism can probably be discarded.

    However, the ECS doesn’t seem to capture biogeochemical feedback processes that could either add or subtract forcing agents from the atmosphere – although most of the feedbacks appear positive – release of organic carbon trapped in permafrost, of methane clathrates in shallow Arctic waters, and drought-related conversion of forests to grasslands or deserts. The most-cited article on the permafrost subject (think Siberian methane blowhole craters) seems to be this one:
    Schuur et al. 2008 Vulnerability of Permafrost Carbon to Climate Change: Implications for the Global Carbon Cycle

  16. 16
    Eli Rabett says:


    The point remains whatever measure of central tendency you use, the estimates of central tendency are all over the place, mean, mode, median, and that means you have to be foolhardy or Nic Lewis/Judith Curry to narrow the probability of outliers.

  17. 17
    Susan Anderson says:

    I hope an effort will be made to clarify the issues for us interested laypeople. OTOH, I’m not sure that is possible. It is all over the map because it is all over the map. The inclusion of the Curry/Lewis paper makes me curious. Is there a possibility that open-mindedness could become a two-way street there?

    I am very interested in the Ladbury query as well.

  18. 18
    Ray Ladbury says:

    I agree that central tendency becomes a fraught concept whenever you are dealing with a pathological distribution. However, a thick-tailed pathology is telling you something fundamentally different than a multimodal distribution. Multiple modes usually reflect a heterogeneous population. This raises the possibility of whether people even know what they are talking about when they discuss sensitivity. The fact that this has gotten worse in the last 5 years rather than better is not comforting.

  19. 19
    Rob Nicholls says:

    Apologies if off-topic; a question I have is why TCR is sometimes said to be more “policy relevant” than ECS. I can understand that approaching equilibrium takes a long, long time, while TCR gives a better measure of what will happen over the next few decades (and that technology and society may be very different in 200 years time); but on the other hand, I thought nations had agreed to try to limit global warming to less than 2 degrees C overall, and not just to limit it to less than 2 degrees C by 2100.

  20. 20
    John L says:

    #18 Ray,
    if there are two different types of methods that consistently gives two different mean values then logically at least one of them must have a bias. But naturally one must first adjust for best estimate/judgement of model bias before combining evidence. For example, some scientists (Shindell, Dessler etc.) argue that the energy balance methods have a low bias. If we assume that they are right and additionally that the bias is about 1K, then a mean at 2.0 means that there is no bimodal problem and no contradiction (in that sense). I think some people seem to do the mistake of interpreting each paper naively from its given nominal mean. For example, if someone publishes a paper with a simplified model that assumes no feedbacks giving a mean ECS at 1.2K, this will not push the combined estimate downwards (regardless of what will be written on the “skeptic” blogs…).

    I’m happy to be corrected by Gavin or others if I’m wrong, otherwise I think that I showed in comment #6 that one big reason for lack of improved interval over time is the analytical mistake of not using a formal procedure from basic statistics for combining evidence, instead relying on vague intuition. That you must subjectively judge some parameters does not mean that the mathematics doesn’t apply. It is not possible to decide exactly which value is correct (but some domain expert average might be the best bet) but you can give a range which it reasonably must be into, therefore it is possible to nicely encapsulate the subjectivity in a formal mathematical model.

  21. 21

    @Ray Ladbury,

    Assuming I understand you correctly, WHY should multimodality in distribution be an indication of pathology? Indeed, all things being possible, it should be expected, and unimodality a happy circumstance for analysis! Ditto the Gaussian, a result from combining the effects of a great many individual phenomena all about equivalent in effect.

    Rather, I should think, if the multimodal/unimodal tension is too much to bear, the resolution should be sought in for what use the posterior density so derived will be put. I doubt policymakers are as interested in the ECS/TCS for the oceans or, by extension, the globe as they are for land. Yet the ECS/TCS for land always trends higher, and that little contingency is curiously ignored.

    I don’t understand why in these semi-technical forums people seem adverse to handling descriptions of ECS up to the details of Professor Ray Pierrehumbert’s Section 3.4.2 (PRINCIPLES OF PLANETARY CLIMATE, page 163ff). Surely there the effects of temperature and feedbacks are fully considered. Or are readers allergic to differentials?

  22. 22
    CM says:

    Jan Galkowski @9, 21: “…land ECS is not the same as ocean ECS, and, so, … when they are combined, you do get multiple bumps.” What? ECS is defined in terms of global mean temperature change, not separately for land and ocean. The different bumps come from different approaches to estimating that one, global, quantity. You must be talking about a different beast.

  23. 23
    Matthew R Marler says:

    Here is my calculation of climate sensitivity, which has been intelligently disputed at Climate Etc by Pat Cassen. It’s what you might call “far out”.

    According to the theory, a doubling of the CO2 concentration will result in an increase in the power carried by the downwelling long wave infrared radiation (DWLWIR), up from approximately 346 W/m^2 (for simplicity I am rounding to the unit place and suppressing the uncertainty) by 4 W/m^2 (2), and the Earth surface will warm until the sum of the upwelling long wave infrared radiation (UWLWIR), the latent heating of the troposphere (LH), and the sensible heating of the troposphere (SH) has increased by 4 W/m^2. How much surface warming might that be? I illustrate by calculating the increase due to a 0.5C increase in surface temperature.

    UWLWIR is proportional to T^4, (2) with emissivity constant, so the increase in UWLWIR, assuming that the global mean surface temperature is equal 288K, works out to delta U = (288.5/288)^4×398 – 398 = 2.8 W/m^2.
    LH results from the hydrologic cycle, cloud formation and precipitation. The review by O’Gorman et al(3) reports that a 1C increase in global mean temperature will result in a 2% – 7% increase in the precipitation rate; the lower values are results of GCM output, and the upper values are results from regressing estimated annual rainfalls on annual mean temperatures. Using the value 4%, a 0.5C increase in global mean temperature will produce an increase of 2% of 88 W/m^2 = 1.8 W/m^2.
    The increase in SH can be estimated from a result reported by Romps et al(4). Their main result was an increase in the cloud-to-lightning ground strike rate by 12% per 1C increase in mean temperature over the US east of the Rocky Mountains. The most important result for this presentation was the estimate of a 12% increase in the power of the process that generated lightning, and that estimate was not confined to the US east of the Rockies. Up to a constant of proportionality, the power of the process generating the lightning was calculated as CAPExPR, where CAPE is “convective available potential energy” (5) and PR was precipitation rate. Precipitation rate was used in the calculation rate not because of the latent energy in the water vapor, but because the precipitation rate was treated as proportional to the rate of transfer of air (with water vapor mixed in) from the surface to the upper cloud level; and the fraction of each kilogram of air that was water vapor was treated as constant. That result depended on the modeled lapse rate and difference between the interior and exterior of the cumulus column. Assuming that their result is widely accurate wherever those can be modeled, and PR rate is proportional to the rate of ascension of air, the increase of SH due to a 0.5C increase of surface mean temperature should be approximately 6% of 24 W/m^2 = 1.4 W/m^2.

    The changes in SH, LH, and UWLWIR sum to approximately 6 W/m^2 (with considerable uncertainty), so the sensitivity of the Earth surface temperature is approximately (4/6)x0.5C = 0.33C (again with considerable uncertainty). This result is lower than most other estimates, and it is approximate and conjectural besides, but the computation is straightforward and based on published research. An obvious omission is a potential increase in the DWLWIR from a warmer upper troposphere (the feedback from the feedback). I hope that many much better estimates can be computed from the actual energy flows in the coming years.

  24. 24
    Doug Allen says:

    I believe climate sensitivity refers to a specific forcing such as the forcing from changes in CO2 or changes in TSI, etc. Without qualification, it’s come to mean the forcing from a double in CO2 including feed backs.
    The difficulty of quantifying the opposing positive and negative feed backs of clouds is well known. How many other opposing feed backs are there- albedo from frozen and liquid precipitation, CO2 fertilization and possible desertification from global warming, and the many others. The conundrum of trying to isolate one climate sensitivity among many, probably, interacting ones should be not only fascinating, but create some of the behaviors that might occur at a head lice convention.

  25. 25
    15120092 says:

    I am especially interested in how climate sensitivity can be constrained by increases in the rates of evapotranspirative transfer of energy from the surface to the troposphere.

    I am also interested in how long is required for the surface temp to “achieve” 95% of the ECS change: e.g. if climate sensitivity is 2K, how much time is required for the surface temp to increase by 1.9K; and then how much longer for the deep oceans to increase by 1.9K (or whatever 95% of the projected increase in deep ocean temperature works out to. (15120092)

  26. 26
    Roscoe Shaw says:

    The CMIP mean ECS is roughly 3.0 and CMIP average temperature forecast has been ~ 2x the observed temp increase. That simple back of envelope calc suggests CS = ~ 1.5k in the absence of other unknowns.

    [Response: “in the absence of other unknowns” – aerosol forcing, ocean heat uptake, internal variability, underestimates of solar and volcanic forcing… – gavin]