RealClimate logo

Climate Feedbacks

Filed under: — group @ 3 August 2006
  • SumoMe

Guest Commentary by Brian Soden (RSMAS, Miami)

Current model estimates of the climate sensitivity, defined as the equilibrated change in global-mean surface temperature resulting from a doubling of CO2, range from 2.6 to 4.1 K, consistent with observational constraints (see previous article). This range in climate sensitivity is attributable to differences in the strength of ‘radiative feedbacks’ between models and is one of the reasons why projections of future climate change are less certain than policy makers would like.

Although radiative forcings and radiative feedbacks both influence the climate by altering the radiative fluxes at the top of the atmosphere, it is important to distinguish between the two. A radiative forcing results from changes that are external to the climate system and may be either natural or anthropogenic in origin. For example, anthropogenic emissions of CO2, changes in solar flux, and the reflection of sunlight from volcanic aerosols are all examples of radiative forcings. A radiative forcing initiates a change in climate that is distinct from the system’s internal variability. A radiative feedback, on the other hand, arises from the response of the climate to either external forcing or internal variability. These responses can either amplify (a positive feedback) or dampen (a negative feedback) the initial perturbation. The exact boundary between a feedback and a forcing depends on what is considered to be part of the ‘system’ and can sometimes be a little fuzzy. This discussion addresses just the feedbacks associated with the atmospheric physical system (see this earlier article for why that is), but other, less well understood, feedbacks (changes in land vegetation, biogeochemical processes, and atmospheric chemical feedbacks – see the NRC 2003 report), while potentially important, are not part of the generally understood definition of ‘climate sensitivity’.

In the absence of radiative forcings, the amount of sunlight absorbed by the earth roughly balances its thermal emission to space; i.e., the earth is in a quasi-steady radiative equilibrium. Doubling the concentration of CO2 decreases the emission of thermal radiation by ~4 W/m2. Because the earth is now emitting less radiation than it absorbs, there is a surplus of energy going into the system and its surface must warm. Because the thermal emission of energy increases as an object warms, the increasing temperature acts to restore radiative equilibrium. In the absence of any feedbacks, a doubling of CO2 would result in an increase in global surface temperature of ~1 K. However, as the climate warms in an attempt to restore radiative equilibrium, other changes occur. These changes can also influence the top-of-atmosphere radiative fluxes and thus act to either decrease (a negative feedback) or to increase (a positive feedback) the radiative surplus. For example, as the climate warms the amount of snow and ice cover decreases which leads to more sunlight being absorbed, thus enhancing the initial radiative surplus and requiring greater warming to restore equilibrium.

There are a number of different radiative feedbacks in the climate system, some more complex than others. Those which are most commonly represented in climate models are feedbacks from water vapor, snow/ice cover, clouds and lapse rate (the change in temperature with height).

Despite the importance of these feedbacks in determining projections of future climate change, there has never been a coordinated intercomparison of their values in GCMs. In a recent issue of the Journal of Climate, Isaac Held and I estimated the range of feedback strengths in current models using an archive of 21st century climate change experiments performed for the upcoming IPCC AR4. The results of this analysis are presented in the figure which expresses the strength of the global mean feedback for each model in terms of their impact on TOA radiative fluxes per degree global warming (units are W/m2/K).

Figure 1 from Soden and Held (2006) showing ranges for each model for each of the key atmospheric feedbacks for the IPCC AR4 models and a comparison with an earlier survey (Colman, 2003).

All models predict the concentrations of water vapor to increase as the climate warms due to the rapid increase in saturation vapor pressure with temperature. Because water vapor is the dominant greenhouse gas, this provides a strong positive feedback in the climate system. In current models, water vapor was found to provide the largest positive feedback in all models and its strength was shown to be consistent with that expected from a roughly constant relative humidity change in water vapor mixing ratio. It should be noted that models are not constrained to conserve relative humidity and significant regional changes in relative humidity are simulated by models in response to atmospheric warming. On the global scale, however, changes in relative humidity are small.

Models do exhibit a range of values for water vapor feedback. This range is not due to departures from constant relative humidity behavior, but rather from intermodel differences in the response of the atmospheric lapse rate to surface warming. All models suggest that the troposphere warms more than the surface (at equilibrium at least – responses are more varied for a short transient period – see the CCSP report). This amplified warming of the troposphere represents a key negative feedback in models because it further increases the thermal emission of energy to space. Models with more surface warming in low latitudes tend to have larger atmospheric warming (and more negative lapse rate feedback) because the surface and free troposphere are more strongly coupled in the tropics than at higher latitudes. Because the water vapor and temperature responses are tightly coupled in the troposphere, models with a larger (negative) lapse-rate feedback also have a larger (positive) water vapor feedback. These act to offset each other. As a result, it is more reasonable to consider the sum of water vapor and lapse-rate feedbacks as a single quantity when analyzing the causes of intermodel variability in climate sensitivity. As shown in the figure, the range for the sum of these two feedbacks is considerably smaller than the range of either the water vapor or lapse rate feedbacks individually.

Not surprisingly, the surface albedo feedback due to changes in snow and ice cover was also found to be positive in all models, although its magnitude is only about 25% of that from the combined “water vapor plus lapse rate” feedback.

Consistent with previous studies, clouds were found to provide the largest source of uncertainty in current models. For the most sensitive models, cloud feedback is positive and comparable in strength to the combined “water vapor plus lapse rate” feedback. For the least sensitive models, cloud feedback is close to neutral. Many specialists and non-specialists alike are sometimes surprised to see that the model-predicted values for cloud feedback ranges from neutral to strongly positive; often believing that cloud feedback is more uniformly distributed between negative and positive values. This confusion may stem, in part, from misinterpretations of the change in the easily-diagnosed “cloud radiative forcing” in model simulations of climate change. This diagnostic, based on the comparison of clear sky and cloudy sky radiation differences (Cess et al, 1996) is related to the more-difficult-to-calculate cloud feedback, but can be negatively biased by correlated changes in water vapor and temperature (Soden et al., 2004). Thus studies that use the “cloud radiative forcing” calculation have reported a more negatively skewed ‘cloud feedback’ then seen here. However, based on these estimates and on a survey of published values of feedback calculations (Colman 2003), there do not appear to be any models for which clouds provide a substantial negative feedback on the climate.

Observational studies do have the potential to help narrow the uncertainties in these individual feedbacks – for instance from studies of the response to Mt. Pinatubo and from long time series of satellite measurements – but observational constraints of cloud feedback remain elusive.

118 Responses to “Climate Feedbacks”

  1. 101
    Chuck Booth says:

    Re# 98 “real world testing in a lab environment” would seem to be a contradiction in terms. And scientists can never “prove” a theory – they can only disprove it, or generate evidence that supports it.

    RE#100 Which models are wrong? And in what way are they wrong?

  2. 102
    julian flood says:

    re 90: not intended as a dismissal of the science, but a mild objection to the terminology used. The oceans will become less alkaline, not acidic.

    re 91: a major oil spill from a tanker is a minor inconvenience. The vast majority of spill comes down sewers — it’s waste oil from human activity, not accident. A quick and ludicrously dirty calculation (deriving in part from an observation of oil on a pond by Benjamin Franklin, which meant at one point I was dealing with teaspoons per fortnight as a unit, so someone who knows about such things might usefully look at the figures again) suggests that the entire surface could be covered every fortnight.

    re 94: increased heat gives increased evaporation, higher salinity in surface waters, saline water sinks and the deep water rises to compensate. I think. It makes me wonder about the Grand Banks and whether the nutrient value of the environment has dropped. That would explain the failure of the fishery to regenerate but is, perhaps, a handwave too far.

    re 92 and 93: I’m not sure these relate to my question. However, if they do, the theory of the cause of CO2 increase does nothing to deny the models — it merely explains the initial fact of CO2 rise. It has always seemed unlikely to me that a few percent increase in output would disturb the overall carbon equilibrium: oil spill and surfactant pollution are small inputs which could produce large effects, which makes me remember the fluorocarbon/ozone problem.

    re 97: are there satellite records of low level cloud cover? Reductions in low-level cloud cover should reduce albedo and increase warming. I believe high level cloud does the opposite.

    Incidentally, I would really like to see some measurements that refute this theory. It suggests, I think, that there should be a mismatch in partial pressure between dissolved and atmospheric CO2 in deep water, not just in the well-mixed shore water. Testable, of course. Does the data exist?

  3. 103
    Chuck Booth says:

    RE#102 “increased heat gives increased evaporation, higher salinity in surface waters, saline water sinks and the deep water rises to compensate.”

    Then how do you explain the permanent thermocline in tropical oceans? You can’t ignore the decrease in seawater density as the ocean surface warms.

  4. 104
    julian flood says:

    re: 103

    You need to look at the bigger picture — it’s not a simple water out the top water in at the bottom system. Evaporation and ppn are thermohaline drivers.

    Do you have any problem with the cloud reduction part of the theory?

    A new theory of global warming at

  5. 105

    Re #101 “Which models are wrong? And in what way are they wrong?”

    The ones that are worong are the radiative convective models and all the general circulation models (GCMs.) In other words, the all the models scientists use to calculate the future climate.

    They are wrong because they cannot replicate the abrupt climate changes which have happened in the past. Moreover they cannot even calculate the cloud base correctly, but this does not matter because the weather men know how to make the correction. More controversially, they do not produce the same lapse rate as that measured by satellites and radiosondes.

    The problem is that they calculate the absorption of outgoing long wave radiation by assuming it is absorbed at all levels of the atmosphere, whereas according to Chapman’s law it all absorbed at the base of the atmosphere. Both methods lead to the same specrtum of long wave radiation as measured by satellites, since both absorb the same amount of radiation overall. Since the current method agrees with the measured spectrum, the scientists are sure that their method is correct.

    But the effect on the atmosphere close to the surface where mankind lives is very different from what their models tell them. See

  6. 106

    Re #98 and “Just out of curiousity, has ANYONE done any real world testing in a lab environment to prove or disprove ANY of these theories? ”

    John Tyndall demonstrated that carbon dioxide absorbed infrared light in 1859. The literature is full of laboratory tests of quantum mechanics, radiative transfer, greenhouse gases, etc., etc. Go to a university library and look at the textbooks. Track down some of the references in the journals. The USAF’s HITRAN project involved testing and recording several MILLION spectral lines.


  7. 107
    Hank Roberts says:

    Julian Flood — you are typing your questions into the wrong box.
    Use the box labeled “Google” at the top of your browser window.

    Here is one of your questions, as typed into Google, with answers, as an example of how Google can help you get real answers you can read and think about.

    Results …about 1,820,000 for: are there satellite records of low level cloud cover?.

  8. 108
    julian flood says:

    re 107: go not to Google for answers, for it will say both no and yes. As will informed opinion even here, looking at no. 93 above. In order for the oceanic surface pollution theory of global warming to be correct then cloud cover should be falling and that should be positive. I’ll go back to the 10^6 hits and start looking: it’s sure to be in there somewhere.

    Thanks for the link. I’ve also searched for partial pressure matching but had no luck, although the RS reports suggests that measurements have been made — but in shallows where wave action is guaranteed.

  9. 109
    Lin Chambers says:

    Re 40: The two sources cited are of very different repute and audience. I am responsible for the second, which is aimed at a K-12 audience. The 30 km value is a historic artifact and I will change it tomorrow to the 20 km value that was determined in the Loeb et al paper. Should have done this in the first place. Mea culpa.

  10. 110

    I overlooked this discussion, as it was started during my trip to Iceland… But there are a few recent studies about cloud behaviour which challenge the positive cloud feedback included in (near all) current climate models.

    Chen and Wielicki (2002) observed satellite based cloud changes in the period 1985-2000, where an increasing SST (+0.085 C/decade) was accompanied with higher insolation (2-3 W/m2), but also higher escape of heat to space (~5 W/m2), with as net result 2-3 W/m2 TOA loss to space for the 30N-30S band. This was caused by faster Walker/Hadley cell circulation, drying up of the upper troposphere and less cirrus clouds.

    In 2005, these findings were expanded by J. Norris with surface based cloud observations in time (from 1952 on for clouds over the oceans, from 1971 on over land) and latitudes. There is a negative trend for upper-level clouds over these periods of 1.3-1.5%. As upper-level clouds have a warming effect, this seems to be an important negative feedback.

    J. Norris has a paper in preparation about cloud cover trends and global climate change.
    On page 58, there is a calculation of cloud feedback, assuming that the observed change in cloud cover is solely a response to increased forcing. The net response is -0.8, which is a very strong negative feedback… Of course this is the response, if nothing else is influencing cloud properties/cover, but important enough for further investigation.

    Even internal oscillations, like an El Nino (1998) leads to several extra W/m2 more net loss of energy to space, due to higher sea surface temperatures. Thus IMHO, if models include a (zero, small or large) positive feedback by clouds, they are not reflecting reality.

  11. 111

    Re: #110 (Ferdinand Engelbeen):

    NASA Langley’s ERBE team, who made the report by Wielicki et al. (2002) in “Science”, has revised their calibration. The increasing decadal trend of outgoing longwave radiation in the tropics still exists, but is much reduced in magnitude. Combined with the decreasing trend of reflected solar radiation (which has changed little by the revision), the net trend is now increase of gain by the earth (~ 1 W/m2 per decade). Analysis of ISCCP data at NASA GISS, though indirect as evaluation of radiation budgets, resulted in similar trend as the revised ERBE trend. See
    and a preprint of a paper by Wong et al. to be published in J. Climate at .

  12. 112
    Hank Roberts says:

    Is this based on new feedback information?

    “… In his first broadcast interview as president of the American Association for the Advancement of Science, John Holdren told the BBC that the climate was changing much faster than predicted.

    “We are not talking anymore about what climate models say might happen in the future.

    “We are experiencing dangerous human disruption of the global climate and we’re going to experience more,” Professor Holdren said….

    “He added that if the current pace of change continued, a catastrophic sea level rise of 4m (13ft) this century was within the realm of possibility; much higher than previous forecasts. …”

  13. 113
    Alexander Harvey says:

    I do not know if it has bothered or confused others but I have found the definition of terms in the paper less than helpful.

    For instance

    R == change in radiative flux at the top of the atmosphere.

    Now the forcing is internal to the system as defined as everything below the TOA. All external forcings are zero.

    At equilibrium there is zero net flux at the TOA.

    So if equilibrium were obtained in the future at 2xCO2 there would have been no net change in either the inward or outward fluxes (as there is no change in external forcing and no change in net flux) with respect to any other equilibrium condition.

    This can not be what the authors intended.

    The usage of R implies they mean the term to reflect the change in outward flux due their feedbacks that compensates for the loss of outward flux due to increasing CO2. In effect they require R to be a combination of the real change in flux and the change in the forcing. What they have done may be acceptable as a shorthand but is not as clear as it could or perhaps should be.

    Later when they are calculating the effective sensitivity they do need deltaR to equal the real net flux at the TOA as this is just the same thing as the change in total heat of the system (everything below TOA) which combined with the forcing can be used to give a value to the effective sensitivity. This can not be the same use as deltaR is put to earlier.

    Another little piece of shorthand is the equation defining cloud feedback using effective sensitivity. Rearranging this equation and comparing to the earlier equation that defines a total feedback as the sum of individual feedbacks gives the identity

    total feedback = effective sensitivity

    and hence they can give a value to the cloud feedback. This is a little misleading as the sum of the feedbacks is not really the same as the effective sensitivity. That is, it would be incorrect to substitute the total of the four feedbacks (the total feedback) for the left hand side of the Murphy 1995 equation. It is accepted that calculated values for effective sensitivity can be used as estimates of climate sensitivity but they are not to be confused with the real thing.

    On a more general point; feedbacks as used here are not to be confused with the way feedbacks are used in other disciplines where feedback factors or ratios are just that “ratios”.

    Climatic feedbacks have the form of admittances (flux/potential). That is they could more naturally be considered as responses (or sensitivities). It seems common practice to single out one or more of these feedbacks to indicate a standard response and then to treat the other responses as feedbacks modifying the standard response. This is actually quite arbitrary if convenient. Interestingly in Table 1 all the responses are listed as feedbacks including the Planck response.

    In the case of the ~1C value associated with the 4.3W/m^2. This indicated that either the Planck response alone which would give a value ~1.5C or the sum of the Planck and lapse rate feedback (response) has been used. The latter giving ~1C. Depending on which one assumes makes a significant difference to back of an envelop computations as some above may have found.

    When it comes down to it. It is simpler to treat them all as equivalent admittances when you want to do the maths (as they do) even if you find it more convenient to think of them as if they were feedbacks when considering the situation qualitatively.

  14. 114
    Alexander Harvey says:

    Again regarding the Soden & Held Paper

    I simply can not understand how they could start to justify their conclusions on the variance (uncertainty) of the cloud factor.

    From the Abstract:

    “Consistent with previous studies, it is found that the vertical changes in temperature and water vapor are tightly coupled in all models and, importantly, demonstrate that intermodel differences in the sum of lapse rate and water vapor feedbacks are small. In contrast, intermodel differences in cloud feedback are found to provide the largest source of uncertainty in current predictions of climate sensitivity.”

    And from the summary:

    “(ii) clouds provide the largest source of uncertainty in current model predictions of climate sensitivity.”

    Now unlike the other terms, the value for clouds is determined by aggregation, this is important. Comparing the variance of aggregates to the variance individual terms is just the sort of thing we were taught not to do.

    The cloud value is defined as the difference betweeen the effective sensitivity and the sum of the other terms. One would well expect this to give a variance close to the sum of the variances of the aggregated terms (and hence the greatest variance) due to the way it has been calculated. Any difference from the aggregate of the other variances being due to the size and sign of the covariances between the other terms.

    Surely the variance of the cloud factor says more about the variance of, and covariance between, the other terms than anything about the cloud factor itself.

    I think this is pretty fundamental.

    And so I think that the justification for making any statement about the variance of the cloud factor is very thin indeed.

    To be frank I am a bit shocked that the authors have not more thoroughly explored the consequences of their own caveat:

    From Results:

    “Keeping in mind the limitations of computing cloud feedbacks as a residual, and the lack of precise information on radiative forcing in the models, our results are consistent with differences in cloud feedbacks being the largest contributor to intermodel differences in climate sensitivity (Fig. 1).”

    One conclusion that this could have led them to would be:

    It is not safe; so if you can’t specifically justify it then do not do it at all.

    I would be interested to know who cast a critical eye over this, and if it was open at the time.

    Sorry, but I think this is tragic.

    [Response: You are overreaching here. It’s true that the cloud feedback is calculated as a residual – but in the cases where they were able to test this, it works well. It is however the best that can be done with the data at hand. Romeo and Juliet is tragic, Soden and Held is just interesting. – gavin]

  15. 115
    Alexander Harvey says:

    Regarding Forcings versus Feedbacks:

    I would think that the distinction might be more clearly put if the wording was Forcings versus Responses.

    By that I mean responses to a change in temperature.

    If a modifier is principly a function of the temperature then its action is a response to the temperature. Hence it is principly a response (feedback) not a forcing.

    So if its action can plausibly be described as a ratio of “change in flux”/”change in temperature” that is a constant or a function of temperature then it is acting as a response (feedback).

    Here it could be educative to consider clouds.

    If cloud is principly a function of temperature (i.e. all other determining parameters are constant) then its action is a response to temperature (a feedback).

    If some controlled process that is not simply a function of temperature is directly stimulating cloud production (seeding , aerosols etc) then cloud cover is partially a forced condition and its action is partly responsive (a feedback) and partially forced (a forcing).

    [Response: A response that ‘feeds back’ on the original change is a feedback. – gavin]

  16. 116
    Alexander Harvey says:

    Gavin Regarding Feedbacks:

    The Planck response is refered to as the Planck feedback parameter.

    In what shape or form is this response a feedback? What part of the output is fed back to the input?

    If the sun gets hotter causing the earth to get hotter does the planck response act as a feedback. I.E. does it act to restore the temperature of the earth by making the sun cooler.

    No, It is simply a response.

    As I have said before. In form it is an admittance as are the other feedbacks (or responses), whenever they are stated as a parameter (with units of admittance). They are also treated as admittances, albeit some of them as negetive admittances. They are summed and multiplied by a difference in temperature to give a flux or a flux is divided by the sum to give a difference in temperature.

    If they were consistantly treated as feedbacks it would be different. In treating feedbacks one must take account of time delay. One needs to be specific about boundaries to define inputs and outputs. In all the mathematics is much more complicated (we would be talking about a model of the weather system).

    If they were stated as a proportion of an output flux from a system that is delayed and fed back (or added) to the input flux to that system they would be being treated as feed backs and the Planck response could not be included (there is no feedback ratio that can be associated with the Planck response).

    In some cases the delays may be so short as to be irrelevant on climate timespans but I am not sure that this is true of albedo. If it is not or that albedo is not a simple function of temperature, which is likely, then the whole treatment of the albedo as a simple admittance (response or feedback if you will) is questionable. One can only wonder if this is taken into account in the paper. The shift that is made from a differential to a difference equation, begs questions. The ability to treat this effect as an admittance while thinking about it as a feedback might just be clouding the picture. Again a principle issues are delays and the possible dependence of albedo on both temperature and time in some complex way.

    To recap: what I am saying is that effects that originate from feedbacks are commonly treated as admittances and in the case of the Planck response an admittance is erroneously considered a feedback. Yet there is a real difference. Understanding that difference can make any laxity in the use of the mathematics apparent.

    I understand that this is a shorthand that allows complexities to be put to one side but it is important to remind oneself and others that it is just that.

    [Response: Not following you at all. The Planck response (long wave emission) is the principle negative feedback because as T increases, LW increases and moderates the T. All terms in the T equation that include a dependency on T itself (however indirect and including delays) are feedbacks. And yes, we are talking about weather models! – gavin]

  17. 117

    […] Range of estimated magnitudes of major climate feedbacks from the most recent IPCC report and Colman 2003. Figure taken from Soden and Held 2006 (pdf) via Realclimate. […]

  18. 118

Switch to our mobile site